Processing of Hyperspectral Remote Sensing Data

Abstract

Multispectral remote sensing data have been potentially explored in India for various applications. A major limitation of multispectral broadband remote sensing products is that they use average spectral information over broadband widths resulting in loss of critical information available in specific narrow bands. The narrow spectral channels that constitute hyperspectral sensors enable the detection of small spectral features that might otherwise be masked within the broader bands of multi-spectral scanner systems. However, use of hyperspectral remote sensing is still in nascent stage. Keeping in view, recent rapid advances in imaging spectroscopy and opportunities for unique applications hitherto thought to be infeasible using broad-band remote sensing, second short course under the aegis of DST-NRDMS initiative on hyperspectral remote sensing is being organized during Feb 18-27, 2013 to develop trained human resource on hyperspectral remote sensing and its application in agriculture. The publication on “Processing of Hyperspectral Remote Sensing Data” will be a guide book to process and analyze the hyperspectral data collected through spectroradiometer, Fourier Transform Infrared Spectroscopy (FTIR) and EO-1 Hyperion sensor. This includes spectral signature collection through both ground held instruments and their analysis, hyperspectral image processing – pre-processing, classification, spectral library generation and spectral matching etc.

Full text

Processing of Hyperspectral
Remote Sensing Data
Division of Agricultural Physics
Indian Agricultural Research Institute
New Delhi - 110 012
Rabi N. Sahoo, Sourabh Pargal
Sanatan Pradhan, Gopal Krishna, Vinod K. Gupta

73
Rabi N. Sahoo
Sourabh Pargal
Sanatan Pradhan
Gopal Krishna
Vinod K. Gupta
Processing of Hyperspectral
Remote Sensing Data
Division of Agricultural Physics
Indian Agricultural Research Institute
New Delhi - 110 012

Printed : February, 2013
EAuthors
Rabi N Sahoo
Sourabh Pargal
Sanatan Pradhan
Gopal Krishna
Vinod K. Gupta
TB-ICN : 111/2013
Correct Citation
Sahoo R. N., Pargal, S., Pradhan, S., Krishna, G. and Gupta, V.K. 2013. Processing of Hyperspectral
Remote Sensing Data, Division of Agricultural Physics, Indian Agricultural Research Institute,
New Delhi-110 012, India, Pp. 72.
© Copy right by Indian Agricultural Research Institute, New Delhi – 110 012, India
Published by the Division of Agricultural Physics, Indian Agricultural Research Institute, New Delhi – 12,
India and Printed at M/s. Venus Printers and Publishers, B-62/8, Naraina Industrial Area, Phase-II,
New Delhi – 110 028, Ph. : 45576780, 9810089097; E-mail: pawannanda@gmail.com
(ii)

(iii)
Foreword
Ever since satellite remote sensing emerged as potential technology for better managing
natural resources, Indian Agricultural Research Institute took the lead role in establishing
a successful programme to harness this technology for agriculture. The Division of
Agricultural Physics, IARI, New Delhi has been involved for more than five decades in
conducting research and developing human resources in the country on Remote Sensing
and their various applications in Natural Resource Management. Over the years Division
has developed expertise on hyperspectral Remote Sensing and demonstrated various
potential applications of this technique in agriculture. As well Division has contributed
significantly for human resource development on remote sensing and other geospatial
technologies through its Post-Graduate teaching Programme and by conducting series of
training programmes.
With the continued strength and expertise, Division is conducting in series 2nd short
course on “Hyperspectral Remote Sensing for Agriculture” (HYPERAGRI-2013) during
Feb 18-27, 2013 funded by Natural Resource Data Management System (NRDMS),
Department of Science and Technology, Government of India. The training programmed
is well composed by Dr. Rabi N Sahoo, Senior Scientist & Course Director, and Dr. Sanatan
Pradhan, Scientist & Course Coordinator, supported by experienced faculties from the
Institute and other associated organizations. This publication under the title ‘Processing
of Hyperspectral Remote Sensing Data’ describes in details processing of data from both
spectroradiometer and satellite is a welcome contribution of the Division of Agricultural
Physics. I am pleased to record my appreciation to whole team who have contributed to
this publication which will be a useful reference source for many researchers working or
intending to work in the field of hyperspectral remote sensing applications.
(Ravender Singh)

Preface
Multispectral remote sensing data have been potentially explored in India for various
applications. A major limitation of multispectral broadband remote sensing products is
that they use average spectral information over broadband widths resulting in loss of
critical information available in specific narrow bands. The narrow spectral channels that
constitute hyperspectral sensors enable the detection of small spectral features that might
otherwise be masked within the broader bands of multi-spectral scanner systems. However,
use of hyperspectral remote sensing is still in nascent stage. Keeping in view, recent rapid
advances in imaging spectroscopy and opportunities for unique applications hitherto
thought to be infeasible using broad-band remote sensing, second short course under the
aegis of DST-NRDMS initiative on hyperspectral remote sensing is being organized during
Feb 18-27, 2013 to develop trained human resource on hyperspectral remote sensing and
its application in agriculture. The publication on “Processing of Hyperspectral Remote
Sensing Data” will be a guide book to process and analyze the hyperspectral data collected
through spectroradiometer, Fourier Transform Infrared Spectroscopy (FTIR) and EO-1
Hyperion sensor. This includes spectral signature collection through both ground held
instruments and their analysis, hyperspectral image processing – pre-processing,
classification, spectral library generation and spectral matching etc.
We are grateful to the Director and Joint Directors (Research, Education and Extension)
for bestowing responsibility on us for conducting this training course and their
encouragement and support to bring out this publication. We are grateful to the NRDMS,
Department of Science & Technology, Government of India for the financial support. We
are thankful to Venus Printers and Publishers, Naraina Industrial Area, New Delhi for
bringing out the publication to our expected form.
New Delhi Rabi N Sahoo
Sourabh Pargal
Sanatan Pradhan
Gopal Krishna
Vinod K. Gupta
(v)

Contents
Introduction 1
Chapter 1: Spectral Signature Collection Using FieldSpec® 3 Hi-Res Portable
Spectroradiometer and Its Analysis 2
1. Introduction to Spectroradiometer 2
2. Theory of operation and Spectral Data Collection 5
2.1. Dark Current Measurement 6
2.2. White Reference 6
2.3. Collection of Spectral Signature 7
3. Analyzing/ Post-processing the observed Spectra 11
3.1. Viewing graphs of data 12
3.2. Log 1/R (1/T) 13
3.3. 1st Derivative 13
3.4. 2nd Derivative 13
3.5. Lambda Integration 14
3.6. Statistics 14
3.7. ASCII Export 15
4. Post processing of observed spectra using Envi 16
4.1. Building Spectral Library 16
4.2. Procedure for spectral snmixing 20
4.3. Spectral analyst 21
Chapter 2: Fourier transform Infrared (FTIR) Spectroradiometer 24
1. Features & Specifications 24
1.1. Standard Equipment 25
2. Working principal and operational considerations 25
2.1. Michelson Interferrometer 25
2.2. Operational considerations 26
2.2.1. Instrument Considerations 27
2.2.2. Target/Sample considerations 27
3. Software Operation and Spectral Data Collection 28
3.1. File Menu 28
(vii)

3.2. Instrument Menu 29
3.3. Display Menu 29
3.4. Process Menu 29
3.5. Function Keys 31
3.6. Measuring Emmissivity 31
Chapter 3: Hyperspectral Satellite Image Processing 34
1. Hyperspectral Datasets 34
1.1 EO-1 Hyperion Sensor 34
1.2. EO-1 Hyperion Data Products 35
2. Introduction to ENVI 38
2.1. Getting Started 38
2.1.1. Starting ENVI 38
2.1.2. Opening an Image File 38
2.1.3. The Available Band List 39
2.1.4. ENVI File Formats 40
2.1.5. ENVI Header 41
2.2. Basic ENVI Functions 41
2.2.1. Opening External Files 41
2.2.2. Display Curser Location and Value 42
2.2.3. Linking Two Displays 42
2.2.4. Displaying Spectral Profiles 42
2.2.5. Selecting Region of Interest 43
3. Hyperspectral Data Preprocessing 44
3.1. Bad Band Removal 44
3.1.1. Band Selection Using Spectral Subsetting 45
3.2. Along Track Destriping 46
3.3. Atmospheric Corrections 48
3.3.1. Empirical Line Correction 48
3.3.2. Internal Average Relative Reflectance 49
3.3.3. Flat Field Correction 50
3.3.4. FLAASH 50
3.3.4.1. Data Requirements 50
3.3.4.2. Conversion to BIL/BIP Format 51
3.3.4.3. Input Parameters and Settings 51
3.3.4.4. Advanced Settings 55
3.4. Spatial Subset 58
(viii)

4. Advanced Hyperspectral Analysis 58
4.1. Minimum Noise Fraction (MNF) Transform 58
4.2. Pixel Purity Index 60
4.2.1. PPI Images for Endmember Selection 62
4.2.2. N-D Visualizer 62
4.2.3. Defining Classes using n-D Visualizer 63
4.3. Spectral Angle Mapper 63
4.4. Linear Spectral Unmixing 65
4.5. Matched Filtering 67
4.6. Binary Encodimg 67
4.7. Spectra, Feature Fitting and Analysis 68
4.7.1. Recovering Continuum Curve 69
4.7.2. Spectral Feature Fitting 70
(ix)

1
Introduction
Remote Sensing is the science and art of obtaining information about an object, area, or
phenomenon through the analysis of data acquired by a device that is not in contact with the
object, area, or phenomenon under investigation [1]. This is usually in the form of an image
acquired at a distance from the surface. Multispectral remote sensing data have been potentially
explored worldwide for various applications. One of the major limitations of the multispectral
data is that the sensors operate in broad wavelength bands thus limiting the amount of spectral
information available [2]. Hyperspectral sensors record reflected electromagnetic energy from
the Earth surface across the electromagnetic spectrum extending from the visible wavelength
region through the near-infrared and mid-infrared region (0.3m to 2.5m) in tens to hundreds
of narrow (in the order of 10nm) contiguous bands [1]. Such narrow bandwidths results in an
almost continuous and detailed spectral response for each pixel providing accurate and precise
information about its constituents and is clearly an advantage over multispectral imaging. The
high spectral resolution of a hyperspectral sensor allows us to capture small deviations in the
spectral response of the materials thus aiding in their identification. Figure below depicts a
typical Hyperspectral data cube and the spectrum of a single pixel.
Wavelength
Spectral Profile
Reflectance Value
Hyperion Image cube and reflectance spectrum
Hyperspectral imaging or reflectance spectrometry techniques are now in use for over a
decade and come as a rapid and inexpensive mode for taking spectral reflectance measurements.
This course will introduce the participants to the state-of-the-art techniques in hyperspectral
image processing and interpretation with a focus on satellite based and ground based sensors.
The participants will be exposed to the complete hyperspectral processing chain starting from
data acquisition with emphasis on agricultural applications.

2
1
Chapter
Spectral Signature Collection Using FieldSpec®
3 Hi-Res Portable Spectroradiometer
and Its Analysis
1. Introduction to Spectroradiometer
Spectrometer is an optical instrument that uses detectors other than photographic film to
measure the distribution of radiation in a particular wavelength region. A spectroradiometer
is a special kind of spectrometer that can measure radiant energy (radiance and irradiance).
The FieldSpec® 3 Spectroradiometer is a general-purpose spectrometer useful in many
application areas requiring the measurement of reflectance, transmittance, radiance, or
irradiance. It is specifically designed for field environment remote sensing to acquire visible
near-infrared (VNIR) and short-wave infrared (SWIR) spectra. While the most highly
regarded features of the FieldSpec spectroradiometer are performance and field-portability,
this instrument also performs well in the laboratory. The FieldSpec spectroradiometer is a
compact, field portable and precision instrument with a spectral range of 350-2500 nm
and a rapid data collection time of 0.1 second per spectrum. The FieldSpec
spectroradiometer offers spectral data collection in various subsections of the spectral range
as listed in Table 1.1.
Table 1.1 FieldSpec Wavelength Configuration
Wavelength Name Wavelength Range
VNIR-SWIR1-SWIR2 350 - 2500 nm
VNIR only 350 - 1050 nm
VNIR-SWIR1 350 - 1800 nm
SWIR1 only 1000 - 1800 nm
SWIR1-SWIR2 1000 - 2500 nm
SWIR2 only 1800 - 2500 nm
VNIR & SWIR2 350 - 1050 nm & 1800 - 2500 nm
The spectral resolution is:
●3 nm (Full-Width-Half-Maximum) at 700 nm.
●10 nm (Full-Width-Half-Maximum) at 1400 nm.
●10 nm (Full-Width-Half-Maximum) at 2100 nm.

3
The sampling interval is:
●1.4 nm for the spectral region 350-1000 nm.
●2 nm for the spectral region 1000-2500 nm.
The series of figures below show the FieldSpec® 3 Hi-Res Portable Spectroradiometer
and its various accessories and enhancements.
Figure 1. FieldSpec spectroradiometer front-view
showing fiber connection and power output.
Figure 2. FieldSpec spectroradiometer power
supply and cables and power output.
Figure 5. FieldSpec back panel with Ethernet
connection, power switch, and power input jack.
Figure 6. Shielded cross-over
Ethernet cable.
Figure 3. Portable battery pack (one on each side of
hip belt) output.
Figure 4. FieldSpec power connector which plugs
into the instrument output.

4
Figure 7. Enhanced view of the front panel with the
fiber optic cable connector, port for remote trigger,
and accessory power port for probe.
Figure 8. Cable connecting FieldSpec accessory
power port to Contact Probe or other authorized
ASD accessory.
Figure 9. Pistol grip with red-dot scope, Ten degree
field-of-view and trigger (standard) attached.
Figure 10. Laptop Carrier attached to shoulder
straps of the Ergonomic Pro-Pack.
Figure 11. The FieldSpec 3 strapped into the
Ergonomic Pro-Pack.
Figure 12. Routing the battery cable to the
FieldSpec spectroradiometer.

5
2. Theory of Operation and Spectral Data Collection
The FieldSpec spectroradiometer measures the optical energy that is reflected by, absorbed
into, or transmitted through a sample. Optical energy refers to a wavelength range that is
greater than just the visible wavelengths, and is sometimes called electromagnetic radiation
or optical radiation. In its most basic configuration, the spectroradiometer views and detects
the form of radiant energy defined as radiance. With accessories, various set-ups, and
built-in processing of the radiance signal, the FieldSpec spectroradiometer can measure:
●Spectral Reflectance,
●Spectral Transmittance,
●Spectral Absorbance,
●Spectral Radiance, and
●Spectral Irradiance.
Field spectrometry is the quantitative measurement of radiance, irradiance, reflectance
or transmission in the field. It involves the collection of accurate spectra and requires an
awareness of the influences of:
●Sources of illumination.
●Atmospheric characteristics and stability.
●Winds.
●Instrument field-of-view.
●Sample viewing and illumination geometry.
●Instrument scanning time.
●Spatial and temporal variability of the sample characteristics
Figure 13. FieldSpec 3 in Ergonomic Pro-Pack and
instrument controller on the laptop carrier.

6
In order to determine the reflectance or transmittance of a material, two measurements
are required:
●The spectral response of a reference sample.
●The spectral response of the target material.
The reflectance or transmittance spectrum is then computed by dividing the spectral
response of the target material by that of a reference sample. Using this method, all
parameters which are multiplicative in nature and present in both the spectral response of
a reference sample and the target material, are ratio-ed out, such as:
●The spectral irradiance of the illumination source.
●The optical throughput of the field spectrometer.
Note: An inherent assumption when determining the reflectance or transmittance of a material
in the field is that the characteristics of the illumination are the same for the reference and
target materials. Variability of the illumination characteristics between the time the reference
and target materials are measured will result in errors in the resultant spectra.
2.1. Dark Current Measurement
Dark Current (DC) refers to current generated within a detector in the absence of any
external photons. DC is the amount of electrical current that is inherent in the spectrometer’s
detectors and other electrical components and is additive to the signal generated by the
measured external optical radiation.
As Noise is the uncertainty in a given measurement, one channel at a time and noise
by definition is random it can be reduced by using more samples and averaging the
signal. Dark Current is different from noise, because it is relatively stable and can be
characterized. DC is a property of the detector and the associated electronics (not the
light source). DC varies with temperature. In the VNIR region, DC also varies with
integration time.
Whenever DC is taken, a mechanical shutter is used to block off the entrance slit of the
VNIR spectrometer so the signal can be measured. This signal is subtracted from each
subsequent spectrum until another DC is taken. The SWIR spectrometers take and subtract
DC on every scan. The DC measurement can be updated at any time, but should be updated
more frequently in the beginning of a given session while the instrument warms up.
2.2. White Reference
A material with approximately 100% reflectance across the entire spectrum is called a
white reference panel or white reference standard or spectralon. The raw measurement
made by the spectroradiometer is influenced by both the sample and the light source. An
independent measure of the light source illumination on a reference of known reflectance
is required to calculate the reflectance of the sample. The use of a white reference standard
with near 100% reflectance simplifies this calculation.

7
Reflectance and transmittance are inherent properties of all materials and are
independent of the light source. Reflectance is the ratio of energy reflected from a sample
to the energy incident on the sample. Spectral Reflectance is the reflectance as a function
of wavelength. Transmittance is the ratio of the radiant energy transmitted through a sample
to the radiant energy incident on the surface of the sample. Spectral Transmittance is the
transmittance as a function of wavelength.
Relative reflectance is computed by dividing the energy reflected from the sample by
the energy reflected off a white reference panel or standard. Absolute reflectance is
computed by multiplying the relative reflectance by the known reflectance of the white
reference panel. With the reflectance of the reference standard available and known, the
ASD RS3 or Indico applications can compute the absolute reflectance or transmittance for
the material being sampled.
2.3. Collection of Spectral Signature
Before starting to take reading the instrument should be allowed to warm up enough. The
warm-up time of the instrument depends on the environment in which it is used. Minimum
one hour of warm-up time is recommended for radiometric work. Radiometer battery
and laptop battery should be fully charged before use.
Follow the instructions for plugging in the instrument and starting the RS3 software
Step 1: Attach required fore-optic accessories (pistol grip) to fore-optic.
Step 2: Click on Start > All Program > ASD Programs > RS3
RS3 Software opens
Figure 14. RS3 Software user Interface

8
Step 3: Open the Control Configuration in the RS3 application.
Click on Control > Spectrum Save (Give the output Spectrum Name and Path where
files are to be saved).
Figure 15. Set the sample, white reference, and dark current averages to 10 scans
Step 4: Keep the attached accessory (Contact Probe etc) over Spectralon (white reference)
Figure 16. White Reference measurement
Step 5: Click Opt option on the RS3 menu bar

9
Clicking on Opt button perfprms following actions;
●Optimizes the detector sensitivities for the probe and light source currently being used.
●The dark offset and white reference will also be measured and saved.
●Status bars will indicate each process.
●Graph will be displayed as below once optimization is done.
Figure 17. Spectralon reflectance after optimization
Step 6: Now save this optimized file by pressing SPACE BAR.
The FieldSpec 3 spectroradiometer must be re-optimized for:
●Light changes.
●Any atmospheric changes.
●Outdoor solar changes at least every 10 to 15 minutes.
●Indoor use every 30 minutes.
●Changes in temperature.
Note: Conditions can change rapidly or slowly. It all depends on clouds, wind (affecting
temperature), instrument warm up time, etc.
It is important that the position of the reference sample when taking a white
reference is as similar as possible to the position for capturing data from the samples.
When saving reflectance data, point the probe at the Spectralon once every few
measurements for a minute or two with the same viewing geometry. If the relative
reflectance of the Spectralon is less than or greater than one, a new white reference
may be needed. If the relative reflectance of the Spectralon is greater than one, re-
optimization is recommended.

10
Figure 18. Different Status bars on the User Interface
Step 7: Go on taking the signature of desired samples and save it by pressing SPACE BAR
Note: The actual spectrum average will be determined by striking a compromise between
noise reduction through averaging the spectra and the time desired for each spectrum
collection. For instance, if you are using the instrument in the field, are walking a large
area, and are making frequent spectral readings, you will want a shorter average setting
than if you are collecting spectra in-situ and desire the cleanest spectra possible.

11
3. Analysing/ Post-Processing the Observed Spectra
ViewSpec is one of many applications that can post-process the observed data. The spectral
data can be imported into many different applications. When using the RS3 application,
ViewSpec can be used to convert spectral data to ASCII text files. Conversion can be done
one file at a time. Or, several files can be merged into a single text file, which is a useful
feature when inputting data into other analysis programs. ViewSpec also permits viewing
large groups of files.
To reduce the effects of low-frequency noise conditions like those found outdoors it is
recommended to take multiple spectra with spectrum averaging set to 10-25, and then
further averaging of those spectra can be done in post processing using the ViewSpec pro
software.
Figure 19. View Spec Pro Interface
The input directory from where the software picks up the spectra, as well as the directory
where the outputs can be saved is selected from the Setup tab, as shown below:
Figure 20. ViewSpec Pro Menu Bar

12
From the Process pull-down menu, an applicable post-processing option for the selected
spectra file(s) can be selected.
Figure 21. Process pull-down menu.
3.1. Viewing Graphs of the data:
Using this application, one can view the data graphically.
Figure 22. Example of Graph generated for spectral data

13
3.2. Log 1/R (1/T):
Converts reflectance or transmittance to absorbance.
Absorbance = log (1/Transmittance)
A commonly used math pretreatment,
useful for linearizing reflectance data. This
expression is often abbreviated as log(1/R).
In most cases it is possible to find a linear
correlation of log(1/R) data to concentration
of an analyte in the target matrix. However,
a general derivation relating reflectance to
concentration cannot be rigorously derived,
such as, the Bouguer-Lambert-Beer law for
transmittance.
3.3. 1st Derivative:
Takes the first derivative of the data. The
algorithm uses a specified gap distance to
skip that number of points to take the
differences instead of adjacent data points.
The derivative gap dialog box is
displayed wherein; the user is required to
give a derivative gap for calculating the 1st
derivative.
3.4. 2nd Derivative:
Takes the second derivative of the data.
Figure 23.
Figure 24.
Figure 25.

14
3.5. Lambda Integration:
Integrates or averages wavelengths over a certain area that is set by the end user. The
lambda Integration Inputs window is displayed wherein; the user determines the intervals
for integration. The user is required to enter a start and end wavelength of a range and
integrate or average the spectra as per application.
Figure 26.
3.6. Statistics:
This process applies standard statistical functions: Mean, Median and Standard Deviation
to the selected files. Mean, Standard Deviation, and Median distinguishes the noise of
each spectrometer. These statistical operations cannot be performed on single files.
Figure 27. Statistics window

15
3.7. ASCII Export:
This process converts data files into ASCII text files. Files can be exported individually or
similar files can be combined into an array and conveniently output as a single file. Header
data can also be included with the data files or exported independently. Files exported
with this utility can be imported into many analysis, spreadsheet or database programs.
Figure 28. ASCII Export
When ASCII Export is selected, the following Dialog Box is displayed:
Figure 29. ASCII Export window

16
4. Post-processing the observed spectra using ENVI
The composition of the spectral signature taken from the spectroradiometer can be
identified using pure spectra of the probable constituents of the composition. Methodology
adopted for identification of unknown combination of spectral signatures through spectral
analysis using Spectral Angle Mapper (SAM), Spectral Feature Fitting (SFF) and Binary
Encoding (BE) is as shown below in figure 30.
Figure 30.
The mixing of pure spectra of different features is done at an interval of 5% i.e. spectral
mixture of 0%A with 100%B, 5%A with 95%B ….100%A with 0%B. Where A is Target and
B is pure spectra of different materials.
Now the unknown spectra is compared and analyzed with the above mentioned spectral
mixtures using Spectral angle Mapper (SAM) method, Spectral Feature Fitting (SFF) method
and the Binary Encoding method.
4.1. Building Spectral Library
Use the Spectral Library Builder to create ENVI spectral libraries from a variety of spectra
sources, including ASCII files, spectral files produced by ASD spectrometers, other spectral
libraries, ROI and spectral profiles and plots. The procedure for building spectral library
from ASCII files is described below;

17
Figure 31.
The Spectral library Builder window opens. Choose the ASCII file option to give ascii
data file as input.
Figure 32.
A new window opens asking for the file containing the output wavelength region across
which you want to build the library. Choose the ascii file exported from ViewSpe Pro. The first
column of the ASCII file is the wavelength column. Choose the ASCII file and click on Open.
Figure 33.

18
The Input ASCII File window opens. Enter the Wavelength column as ‘1’. Wavelength
units as ‘nanometers’ and y scale factor as ‘1.0000’. Click OK
Figure 34.
The Spectral Library Builder window opens. Click on Importà from ASCII file
Figure 35.
Select the ASCII file again and click open. Import the file and you will see in the Spectral
Library builder window a number of files are listed. These are the individual spectra of
different plant varieties. You can select individual spectra and click on Plot. You can also

19
plot all the spectral profiles in the same window as well by selecting all the spectra and
click on Plot. The spectra will be displayed as shown in the figure 36.
Figure 36.
Under the options menu, there are options for importing spectrum names for all the
spectra from a different ascii file or you can rename them individually. Under the File
menu, there is option for saving the spectra as Spectral Library file.
Figure 37.

20
The Output spectral library window opens. Enter the required information and then
enter the output file name and click OK. The spectral library is created and can be viewed
from Spectralà Spectral Librariesà Spectral LibraryViewer.
4.2 Procedure for Spectral mixing
Figure 38.

21
For viewing the saved Spectral mixture Go to Spectral > Spectral Libraries > Spectral
Library Viewer
Figure 39.
4.3 Spectral Analyst
For identification of the composition of the observed spectra with pure spectra of the library
go to Spectral > Spectral Analyst. Before going for Spectral Analyst open your observed
data in spectral library viewer and follow the procedure –
Figure 40.

22
Figure 41.

23
The analysis is done on the basis of score derived from these three methods. The score
ranges from 0 to 1. Higher is the score nearer is the composition of unknown spectra to the
known mixture. On spectral analysis we can identify that the unknown spectra which has
the maximum matching with the spectral mixture corresponding to the highest score of
SAM, SFF and BE method. A spectral library can be prepared using spectral database
collected from field as well as lab which can further be used to identify the composition of
any unknown spectra. But there is a limitation with this database is that it will be able to
identify only limited no. of samples, otherwise it will result to wrong interpretation.
Figure 42. Spectral Analyst Window

24
2
Chapter
Fourier transform Infrared (FTIR) Spectroradiometer
Fourier Transform Infrared spectroscopy is a technique which is used to obtain an
infrared spectrum of absorption, emission from a solid, liquid or gas. The term Fourier
transform infrared spectroscopy originates from the fact that a Fourier (a mathematical
algorithm) is required to convert the raw data into the actual spectrum. The Hand
Portable FT-IR spectrometer manufactured by D&P instruments is a hand portable,
remote sensing field and industrial instrument designed for field measurement of
spectral radiance from the Earth’s surface and atmosphere in the 3–5-m and 8–14-m
atmospheric windows, with a 6-cm21 spectral resolution [3]. The instrument is
packaged in a small case as shown in figure 36. Portable spectrometers were originally
developed for the battlefield detection of chemical agents. These instruments have
also been used to monitor atmospheric composition, particularly pollution. Small FTIR
spectrometers have also been developed as spaceborne instruments. The use of
spectrometers in space has, in turn, spurred a need for measurements of radiance from
the ground, i.e., the so-called ground-truth measurements, to verify calibration, provide
atmospheric correction data, and to measure the emissivity of terrestrial surface
materials that cannot be measured in the laboratory.
1. Features & Specifications
●Thermally Stabilized Interferometer
●Embedded Pentium® PC Computer
●USB, Ethernet, and VGA ports
●Calibrated Output with Optional Thermally
Stabilized Blackbody
●“Through-the-Lens” Viewing of Targets
●High Sensitivity and Throughput
●Full sun readable LCD screen
●Real-Time On-Screen Spectra and Math
Processing
●Runs on compact battery, 12 volt, or worldwide
universal AC supply.
The Specifications of the field portable FT-IR Spectrometer are
given in Table 2. Figure 43.

25
Table 2. Specifications of Hand Portable FT-IR Spectrometer
Item Paramter Value Units Comments
1 Spectral Range 2 - 16 micrometers Standard IR
2 Spectral Resolution (FWHH) 4 wavenumbers @ 2 μm, Standard, 1 sec. scan
3 Size (WxDxH) 36x20x23 centimeters (14"x8"x9")
4 Weight < 7 kilograms (<15 pounds)
1.1. Standard Equipment
●Optical/Electronic module, including interferometer, drive & sampling electronics,
embedded Pentium® PC computer, WinFT™ processing software running on Windows
XP™, USB, Ethernet, VGA, parallel and serial connections
●Liquid nitrogen (LN2) cooled dual InSb/MCT detector and preamp (2-16 μm)
●1" Ø, 4.8° field-of-view (FOV) and 2", 2.4° FOV fore-optics with through-the-lens viewing
●AC supply and dual battery charger
●12 volt, 7 A-hr battery pack
●Thermoelectrically stabilized blackbody (1" or 2" diameter)
●Diffuse gold plate for down-welling radiance measurements
●Pouring dewar, tripod
2. Working Principle and Operational Considerations
2.1. Michelson interferometer
The core of the spectrometer is the Michelson interferometer. This contains infrared optics,
beam splitter, and a scanning mirror assembly. The high-throughput advantage of the
Michelson interferometer spectrometer resulted in the compact and lightweight
development of the D&P FTIR spectroradiometer. Figure 44 shows the gives the outer and
crossectional view of the Michelson interferometer used inside this spectrometer.
Figure 44.

26
Input light passes through the fore optics, an aperture, and a lens (which also seals the
unit) into the interferometer. The internal mirrors are servo driven at a constant speed,
producing the interference patterns. The output light passes through a focusing lens (which
also seals the unit) onto an infrared detector in a liquid nitrogen (LN2) dewar. The standard
detector is a dual sandwich type, consisting of Indium Antimonide (InSb) over Mercury
Cadmium Telluride (HgCdTe, or MCT). This detector has a spectral range of approximately
2 to 16 micrometers. This must be filled with liquid nitrogen before use. A temperature
controlled laser diode (LD) provides the reference for the servo and sampling electronics,
and wavelength calibration for the spectrum. Figure 45 shows a schematic of the FTIR and
its accessories.
Figure 45.
2.2. Operational Considerations
Removal of environmental factors such as reflected downwelling atmospheric and
background radiance from the measured signal are of paramount importance. Proper
separation of temperature and spectral emissivity is also a key factor in obtaining spectra
of accurate shape and magnitude. The environmental conditions, under which one is
collecting spectra, the time of day, and the target’s thermodynamic properties, will all
have a profound influence on the quality of the collected data.
For the reasons stated above it is imperative that an organization that routinely collects
spectral ground truth data in the field has a protocol to follow that will allow investigators
to produce a spectral library of consistent and repeatable quality. A few of the considerations
are listed below;

27
2.2.1. Instrument Considerations
●Warm-up - If possible, let the instrument run for as long as possible, prior to making
measurements. This “warm-up” period allows ample time for the instrument
components within the enclosure to come to thermal equilibrium.
●An instrument should be controlled to within 0.1°C between calibration and actual
target measurements.
●Black Body - A blackbody calibration should be conducted for every target measured,
or at least every 10 minutes, to reduce the effect of instrument temperature drift.
✓In general the cold blackbody should be set just below ambient (being careful that
condensation does not form on its surface).
✓The warm blackbody should be set just above the sample temperature anticipated.
●Field of View - Be sure to overfill the field-of-view of the device. Samples should be
measured at a distance of no more than 1 meter if possible to minimize the effects of
atmosphere.
●Downwelling radiance should always be measured immediately following the sample
measurement by collecting the reflected radiance off of a diffuse reflective plate, usually
InfraGold or crinkled aluminum foil.
✓Orientation should be the same as the target/sample
✓Direct solar reflection should be avoided (no surface is truly diffuse)
●Time of Measurement - Measurements are best made in the early morning or late
afternoon to avoid the rising thermal currents at the hottest point of the day.
●Instrument and operator must not cast a shadow on the sample
●Any contributing background source must not move or change during the sample and
downwelling scans
✓Operator, other people in the scene
✓Vehicles, other movable objects
✓Clouds, atmospheric conditions
2.2.2. Target / Sample Considerations
●Is the target uniform in composition and makeup?
✓Color
✓Consider why you are collecting the spectra
✓Compositional elements compared to ground and aircraft sensor FOV
✓Individual components may have distinct spectral signatures
●Is the surface multi-faceted?
✓Trees, vegetation
✓Soil/gravel surface
●Is the surface rough or smooth relative to the frequency that measurements are being
made at?

28
✓Painted steel/rusted steel
●Can the background be seen through the target?
✓Transparency
✓Porosity
●Will the sample change temperature during scans?
✓Thermal inertia
3. Software Operation and Spectral Data Collection
The FTIR software WINFT runs on Windows XP platform. All the functions can be easily
accessed by the keyboard, however a USB mouse is also provided to make the operation
quicker. The main menu items are across the top of the screen, and are accessed using the
mouse or using the <Alt> key and the first (underlined) letter of the title,. They are:
File Instrument Display Process Help
Figure 46. WinFT Menu Bar
There are also 10 function keys assigned for the most commonly used repetitive
operations. Their functions are listed in a row of boxes located just below the main menu
titles at the top of the screen. Either the function key or mouse click on the box activates
them. The Function key assignments are:
F1-Acquire F2-Open F3-Save As F4-Refresh F5-Export
F6- TBD F7-Cursor F8-X scale F9-Y scale F10-Status
<alt>F9-Disable Temp Alarm <Alt>F10-Factory Setup
There are quite a few file types used in the software. Raw binary data is saved in one of
six types of files, with distinct extensions. They are:
●Setup File (.INI) used to store instrument setup
●Sample (.SAM) used in all types of processing
●Reference (.REF) used in ratio, difference, absorbance processing
●Cold Blackbody (.CBB) used in radiance and emissivity processing
●Warm Blackbody (.WBB) used in radiance and emissivity processing
●Downwelling Radiance (.DWR) used in emissivity processing
3.1. File Menu
The File menu has 9 items in its submenu, selected by mouse, arrow key, or underlined
letter. They are:

29
Open Save As Save Settings Data File Name and Directory
AutoName Reverse Video Auto Calibrate A-to-D Print Exit
3.2. Instrument Menu
The Instrument menu has 5 items in its submenu, selected by mouse, arrow key, or
underlined letter. They are:
Coadds Resolution Zero Fill FFT Apodization Temperatures
Coadds: sets the number of spectra to be averaged
Resolution: is used to specify the size of the FFT performed, and thus the resolution of the
resulting spectrum. The number of points collected is not changed, and all interferogram
data is stored in the raw data file, so any spectrum can be reprocessed later at a different
resolution.
Zero Fill: is used to generate higher plot resolution at any spectral resolution.
●None no zero filling
●2X interferogram filled out with two times the number of points
●4X interferogram filled out with four times the number of points
●8X interferogram filled out with eight times the number of points
FFT Apodization: sets the weighting function to be used to window the interferogram
data before performing the FFT.
Temperatures: is where all instrument and blackbody temperatures are set and their
respective controllers can be turned on and off manually.
3.3. Display Menu
The Display menu has 3 items in its submenu, selected by mouse, arrow key or underlined
letter. They are:
Plot Type Display Units Plot Scales
Plot Type specifies what type of display to use for data. The choices are:
●Interferogram displays the interferogram full screen in one window
●Spectrum displays the spectrum full screen in one window
●Both displays the interferogram in one window at the top of the screen,
and the spectrum in a separate window at the bottom of the screen.
Display Units is used to set the X and Y axis units for all displays. There are three choices
for the X units:
●Micrometers The X scale is displayed and exported in micrometers (um)
●Nanometers The X scale is displayed and exported in nanometers (nm)

30
●Wavenumbers The X scale is displayed and exported in wavenumbers (cm-1)
Display Scales is used to set the display limits for interferogram and spectral plots. When
this item is activated, a dialog box appears to select Spectrum or Interferogram scales to
set. When displaying a plot, the F9 function key can be used to toggle between Y-axis auto
scaling and the manual limits set here.
3.4. Process Menu
The Process menu normally has 1 item in its submenu, selected by mouse, arrow key, or
underlined letter. If Radiance or Emissivity math is selected, a second item to calibrate a
sample appears. Once a sample has been calibrated, a third item to fit a Planck function to
a calibrated radiance appears. If a Planck function has been fitted to a calibrated radiance,
a fourth menu item appears to remove the Planck function. The four functions are:
Math Calibrate Instrument Fit Planck to Radiance Remove Planck Plot
Math is used to set which math, if any, is to be done on acquired or restored data files. The
choice of Math function also determines what types of data will be acquired or restored
when any of those functions are used. When the Math item is selected, a dialog box appears
with a choice of none, or six different math operations. They are:
●None no math performed, raw data displayed
●Ratio Sam/Ref the Sample file is divided by the Reference file
●Difference Sam-Ref the Reference file is subtracted from the Sample file
●Difference Ref-Sam the Sample file is subtracted from the Reference file
●Inverse 1-Sam/Ref the ratio function is subtracted from 1
●Absorbance -log (Sam/Ref) the logarithm of the ratio is calculated and negated
●Radiance f (cbb,wbb,sam) calibrated Radiance in (Watts/(m2*microns*sr))
●Emissivity f (cbb,wbb,dwr,sam) a calibrated radiance is corrected for sky reflected
energy and divided by a Planck at the estimated
sample temperature
Calibrate Instrument appears under the Process menu if either Radiance or
Emissivity is chosen as the Math function. This operation requires two blackbody data
files (a Cold and a Warm) to be acquired or restored. The calibration function generates
a slope and offset correction at each wavelength, which is then used to take out the
instrument function when displaying sample Radiance or Emissivity. The units of
Radiance will be in [Watts/m2*um*sr]. Selecting this menu item brings up a submenu
with three choices:
Acquire Data and Calibrate to acquire new BB data
Open BB Files and Calibrate to restore previously acquired BB data
Calibrate Now with Open Data Files to use current data in memory

31
Selecting the first option, Acquire Data and Calibrate, will open a dialog box for the
Autocalibrate function. From this screen, the instrument’s blackbody can be controlled,
data acquired, and instrument calibration performed.
3.5. Function Keys
The function keys are linked to frequently used operations. There are 10 function key
assignments listed across the top of the screen, just below the menu bar. Their assignments are:
●F1-Acquire starts the acquisition of a data set
●F2-Open brings up the File Dialog box to open a file (same as menu File-Open)
●F3-Save As brings up the File Dialog box to save a file (same as menu File-Save As)
●F4-Refresh used to refresh the screen after making changes
●F5-Export used to start the ASCII File export process
●F6- TBD
●F7-Cursor puts a cursor on interferogram or spectral plots
●F8-X Scale toggles the X scale between ALL X and USER X scales
●F9-Y Scale toggles the Y scale between AUTO Y and USER Y scales
●F10-Status brings up a status screen showing currently loaded files
●<Alt>F9 turn off audible temperature alarm
●<Alt>F10 factory setup, used to set up instrument operating parameters
3.6. Measuring Emissivity
The instrument has specialized software for the measurement of surface emissivity of
targets in the field. For this, four raw data files are required; three are for calibration and
one is the sample itself. Of the three calibration measurements, two are blackbodies and
one is a measurement of downwelling radiance. The blackbodies are used to calibrate the
target and downwelling radiances. The downwelling radiance sample is collected from a
diffuse reflector placed in the field of view of the instrument. Samples are then placed in
the same location as the diffuse reflector when they are measured. In this way, the reflected
downwelling radiation off the target samples can be subtracted out to give the absolute
emissivity. The algorithm used to compute the emissivity is given by:
es(l) = [Ls(l)-Ldwr(l)]/[B(l,Ts)-Ldwr(l)]; where
es(l) is the surface emissivity of the sample as a function of wavelength;
Ls(l) is the calibrated radiance of the sample;
Ldwr(l) is the calibrated radiance of the downwelling radiance;
B(l, Ts) is a Planck function at the sample temperature.
The downwelling radiance term must be corrected for the emissivity of the diffuse
reflector and its temperature. These two parameters are prompted for as part of the data
acquisition process for a downwelling radiance file. The sample temperature is derived

32
by fitting a Planck function to the calibrated sample radiance. Two methods are provided
in the software for doing this; a manual fit or an automatic fit. When the proper Planck
function is found, this is used in the computation and display of the final emissivity curve.
A step by step procedure for measuring emissivity is discussed below;
●FOUR measurements are required.
●Connect blackbody cable
●Set Math to Emissivity
●Fit blackbody onto fore-optic (twist lock)
●Set Auto-calibrate temperatures under Instrument-Temperature function (10 and 40 to start)
●Go to Process-Calibrate-Acquire Data
●Set Cold, wait for BB Temp to reach set point, then Acquire and Store Cold BB data
●Set Warm, wait for BB Temp to reach set point, then Acquire and Store Warm BB data
●Sight on diffuse gold plate
●Take a “downwelling” spectra
●Sight object to be measured
●Take a “sample” spectra
Collected dpectra of a raw quartz sample is shown in the figures below:
Figure 47.

33
Figure 48. Calibrated Quartz Sand with Planck fit and temperature
Figure 49. Quartz Sand Emissivity 2-16 microns

34
Hyperspectral Satellite Image Processing
1. Hyperspectral Datasets
The availability and use of airborne hyperspectral data has been well studied and
documented with a number of airborne sensors in operation since early eighties. With
the launch of NASA’s Earth Observing 1(EO-1) Hyperion instrument in the year 2000,
a platform was created for exploiting the spaceborne hyperspectral imaging capabilities.
Hyperion was the first hyperspectral sensor to provide a continuous spectral profile
across the broad electromagnetic spectrum ranging from 400nm to 2500nm. The
comparison of an airborne sensor, such as Airborne Visible/Infrared Imaging
Spectrometer (AVIRIS) and Hyperion datasets in terms of spectral information provide
comparable results under optimum acquisition conditions viz. illumination, dark
targets etc. [3]. The spatial resolution of airborne sensors (2-20 m depending upon
flight altitude and sensor resolution) is however comparatively higher than that of
space borne sensors (30 m in the case of Hyperion). The low spatial resolution of the
hyperion sensor causes a problem of mixed pixels, a pixel which is formed when spectra
of different underlying substances are combined into a mixture spectrum. In spite of
the limitations on the spatial resolution there are quite a few arguments which go in
favour of space borne sensors. Firstly, they allow regular and repeated coverage over
wider and restricted areas. Secondly, variations and distortions arising due to aircraft
motion are reduced [4].
Due to the continuous spectrum for each pixel, the high-dimensional data space
generated by hyperspectral sensors poses challenges in image processing and data
analysis and is quite different from multispectral processing where there are only a few
discrete bands. Also the space borne hyperspectral remote sensing images are more
affected by noise due to the narrow bandwidths, which can hamper the image
interpretation and information extraction processes. Hyperspectral datasets are spectrally
overestimated and there is a lot of redundant information present. So there is need for
exploration of dimensionality reduction (DR) and end member extraction (EE) methods
which can effectively reduce noise in hyperspectral datasets and aid in the spectral
unmixing.
1.1 EO-1 Hyperion Sensor
Hyperion instrument onboard NASA’s Earth Observation-1 (EO-1), launched on 21st
November 2000 as part of NASA’s New Millennium Program, is the first space borne
Hyperspectral sensor for Earth Observation studies. It orbits the Earth in a sun-
3
Chapter

35
synchronous orbit at an altitude of 705km. The Hyperion is a Push-broom scanner
with a high spectral resolution. It has 242 spectral bands spanning a spectral range
from 0.4 to 2.5 m, with a sampling interval of 10nm. The Spatial resolution is 30m
with a swath width of 7.7 km and covers an area of 7.7 × 100 square km per image with
high radiometric resolution (16 bit).
The Hyperion sensor has two spectrometers
operating over different spectral ranges. One operates
in Visible and near Infrared region (VNIR) i.e. 0.355
m to 1m having 70 bands and the other operates in
Shortwave Infrared region (SWIR) i.e. 0.9 to 2.5m
having 172 bands. The overlap region between the two
spectrometers between 0.9 to 1m allows for cross
calibration between two spectrometers. Figure 50
shows an image of the Hyperion Sensor onboard
EO-1.
1.2. EO-1 Hyperion Data Products
The data in the form of cubes is put into Hierarchical Data Format (HDF) format written
as band-interleaved-line (BIL) files stored as 16-bit signed integer radiance values. The
SWIR bands have a scaling factor of 80 and the VNIR bands have a scaling factor of 40
applied. The actual radiance values vary from zero to approximately 32,767. The various
formats in which the Hyperion sensor data is made available by USGS to the users are
listed below [5]:
•Level 1R (L1R)
The Level 1 Radiometric product is only radiometrically corrected and not geometrically
resampled. The L1R product ia available only in HDF format. The data product consists of
the HDF data file (.L1R), a metadata file (.MET), a header file (.hdr and a auxiliary file (.AUX).
•Level 1Gs (L1Gst)
The Level 1 Gs product is radiometrically corrected and geometrically resampled and
is registered to a geographic map projection. The image is also terrain corrected i.e. ortho-
rectified using digital elevation models (DEM) for correcting parallax error due to
topographic relief. Each image band in the L1G product is provided in a separate file. The
L1Gst product is available in two formats: HDF v 4.1 and GeoTIFF.
L1Gst (HDF)
The Hyperion product includes a metadata file (_MTL.L1T), an HDF header file
(_HDF.L1T), a Federal Geographic Data Committee (FGDC) metadata file (.fgdc)
and multiple image bands (_B###.L1T).
L1Gst (GeoTIFF)
GeoTIFF defines a set of public domain TIFF tags that describe all cartographic and
geodetic information associated with geographic TIFF imagery. This Hyperion
Figure 50. Hyperion Sensor

36
product includes a metadata file (_MTL_L1T.TIF), an FGDC metadata file (.fgdc)
and multiple image bands (_B###_L1T.TIF).
The USGS products are packaged in Hierarchical Data Format (HDF) with the
datasets being image data, spectral center wavelengths, spectral bandwidths, gain
coefficients and a flag mask. The file naming convention utilizes an entity ID with
acquisition target. In this course work we will be using the L1R product for hands on
practical classes. Table 3.1 describes the naming convention for Hyperion datasets and
Table 3.2 lists the scene characteristics of the Hyperion L1R image of Jalandhar area, to
be used in this course work.
Table 3.1: Data File Name Description
EO1SPPPRRRYYYYDDDXXXML_GGG_VV
EO1 Earth Observing 1 mission
SSensor, A = ALI, H = Hyperion
PPP Target WRS path of the product
RRR Target WRS row of the product
YYYY Acquisition year of the image
DDD Acquisition Julian day of year
XHyperion, ALI, Atmospheric Corrector (AC), (1 = sensor on, 0 = sensor off)
MPointing Mode, P = Pointed within path/row, K = Pointed outside path/row, N = Nadir
LScene identifier which may be 0-9 or an upper or lower case alpha character.
GGG Ground/ Receiving Station
VV Version Number
Table 3.2: Scene Characteristics of Hyperion Image of Jalandhar Area (Source:
http://edcsns17.cr.usgs.gov/NewEarthExplorer)
Data Attribute Attribute Value Data Attribute Attribute Value
Entity ID EO1H1480382008133110PW_PF1_01 Scene Start Time 2008:133:05:19:04.228
Acquisition Date 5/12/2008 Scene Stop Time 2008:133:05:19:20.228
Site coordinates Date Entered 5/13/2008
NW Corner 31°41’01.22"N, 75°36’50.63"E Target Path 148
NE Corner 31°40’06.36"N, 75°41’26.14"E Target Row 38
SW Corner 30°46’41.15"N, 75°22’15.21"E Sun Azimuth 115.528803
SE Corner 30°45’46.71"N, 75°26’48.18"E Sun Elevation 64.876435
Cloud Cover 0 to 9% Cloud Cover Satellite Inclination 98.13
Receiving Station PF1 Look Angle 3.3236

37
More about E01-Hyperion
• The Hyperion sensor on board the EO-1 satellite is the first Hyperspectral sensor to
operate from space.
• The Hyperion provides high resolution Hyperspectral data having 242 spectral
bands (from 0.35 to 2.5 μm) with a 30-meter resolution.
• Hyperion data is initially processed by the EO-1 product generation system (EPGS)
and distributed in different processing levels (.L1R & .L1T)
• Hyperion is a push broom type sensor; characteristically these sensors have poorly
calibrated detectors. These detectors cause high frequency errors in the VNIR or
SWIR regions, which can be identified as vertical strips in the image bands.
Data Description
Samples 1041
Lines 3531
Bands 242
Header offset 0
File type HDF Scientific Data
Data type 2
Interleave BIL
Sensor type HYPERION
Byte order 1
Read procedures HDF read spatial
Subset procedure HDF read scroll
Hyperion Sensor Characteristics
EO-1 SATELLITE-HYPERION DATA CHARACTERISTICS
Sensor altitude 705 Km
Spatial resolution 30 m
Radiometric resolution 16 Bits
Swath 7.2 Km
IFOV (mrad) 0.624 degrees (256 pixels)
Imaging Technology Pushbroom
(Samples) No. of rows 1041
(Lines) No. of columns 3531
VNIR range 0.35-1.35 m
SWIR 1.40-2.48 m
No. of Bands 242
Scaling factor (VNIR) 40
Scaling factor (SWIR) 80
Band width 10 nm

38
• Due to poorly calibrated detectors, mean and slandered deviation of the data values
for particular band will be effected.
• During this realignment, actual SWIR bands are shifted across track by -1 (FOV) in
X direction and +1 pixel down track in Y direction.
• The digital values of the Level-1 products are in 16-bit radiances and stored as a 16-bit
signed integer.
• The Level 1 radiometric (L1R) product used in the study has 242 bands; only 198 of
them are calibrated (band 8 to 57 for visible-to-near-infrared (VNIR) and 77 to 242
in shortwave-infrared (SWIR) regions).
Disadvantages
• Large volume,
• Band selection,
• Computational complexities
2. Introduction to ENVI
ENVI (Environment for Visualizing Images) is a software tool for image visualization,
analysis and interpretation. The whole package includes advanced state of the art of image
processing tools for spectral analysis, geometric correction, terrain analysis, radar image
analysis, vector and raster GIS capabilities and much more. This section will discuss about
some of the basic concepts of image processing in ENVI software and few of its key features
with emphasis on Hyperspectral Remote Sensing. ENVI includes a full suite for
Hyperspectral Image analysis and Interpretation.
2.1. Getting Started
2.1.1. Starting ENVI
• For running ENVI in Windows machines select Startà Programs  ENVI v.v  ENVI
(where v.v is the version number)
2.1.2. Opening an Image file
1. From theENVI menu bar, select File  Open Image File
2. Locate the Hyperspectral Image file and click Open
3. The HDF Dataset Selection Window will open, select the file (.L1R) and click Open
4. Select the data format (BSQ, BIL and BIP) and click OK
5. To load a Gray scale image, From the available band list select radio button Gray Scale
and select a band and then click Load Band
6. To load a colour image select RGB Color radio button along with the desired band
combination.
7. Click Display#1 and select New Display and then click Load Band to load RGB image.
As you click on Load Band a group of windows will open allowing you to analyse the
image. The group of windows is collectively referred to as “Display Group” and consists
of an Image Window, a Scroll Window and a Zoom Window as shown in Figure 51.

39
You can choose which combinations of windows should appear on the screen by right-
clicking on any image window and selecting a style from the Display Window Style
submenu.
From the ENVI main menu bar select File  Preferences  Display Defaults tab to
change the default settings for which windows you wish to display and where you wish
to position them. A wide variety of other types of ENVI windows may also be displayed,
such as scatter plots, spectral profiles, spectral plots, and vector windows.
Figure 51. Display Group
2.1.3. The Available Band List
The available band list as shown in Figure 52 is an ENVI dialog box containing a list of
image bands available from all open files or any associated map information. The available
band list dialog box can be for loading both colour and gray scale images into multiple
displays simultaneously.
The File and Options menu at the top in the Available Band List dialog box provides
various functionalities such as Opening and closing the Image files or bands, wavelength

40
locator, show currently displayed band. By right clicking on the available band list or on a
file displays a range of functions for editing the header file, computing statistics of the
image, and for loading true colour and colour infrared images into the display.
2.1.4. ENVI File Formats
The ENVI uses a general raster data format consisting of a simple flat-binary file and an
associated ASCII header file. This file format permits ENVI to use nearly any image file,
including those that contain their own embedded header information. The general raster
data is stored in ENVI Flat Binary Format as a binary stream of bytes either in Band
Sequential Format (BSQ), Band Interleaved by Pixel Format (BIP), or Band Interleaved by
Line Format (BIL) formats.
•BSQ is the simplest format, with each line of data followed immediately by the next
line of the same spectral band. BSQ format is optimal for spatial (x,y) access to any part
of a single spectral band.
•BIP format provides optimal spectral processing performance. Images stored in BIP
format have the first pixel for all bands in sequential order, followed by the second
pixel for all bands, followed by the third pixel for all bands, etc., interleaved up to the
number of pixels. This format provides optimum performance for spectral (z) access
of the image data.
Figure 52. Available Band List

41
•BIL format provides a compromise in performance between spatial and spectral
processing and is the recommended file format for most ENVI processing tasks. Images
stored in BIL format have the first line of the first band followed by the first line of the
second band, followed by the first line of the third band, interleaved up to the number
of bands. Subsequent lines for each band are interleaved in similar fashion.
ENVI also supports a number of generic image formats such as ASCII, BMP, HDF,
JPEG, JPEG 2000, PICT, PNG, TIFF (GeoTIFF), TIFF world (.tfw), SRF, XWD. Also it supports
a variety of data types such as byte, integer, unsigned integer, long integer, unsigned long
integer, floating-point, double-precision floating-point, complex, double-precision complex,
64-bit integer, and unsigned 64-bit integer.
2.1.5. ENVI Header
The ENVI Header file contains information about
the dimensions of the image, data format or any
other information which ENVI uses to read the
image data file. The Header file can be created
the first time a particular data file is read by ENVI.
It can also be created or edited manually by the
user itself.
1. From the ENVI menu bar, select File  Edit
Envi header or alternatively right click on file
name in Available band List  select Edit
Header
2. Header Info: dialog box will appear as shown
in Figure 53.
3. Click the Cancel button to close the header
file.
2.2. Basic ENVI Functions
This section will discuss about some of the basic functions which we can perform in ENVI.
2.2.1. Opening External Files
This option is used for opening a number of standard file types including some sensor
specific formats, military formats and other generic image formats.
1. From the ENVI menu bar, select File  Open External File  file_type  file_format.
(where: file_type is the type of external file (for example, Landsat; file_format is the
format of the external file (for example, HDF)
2. Select a file to open.
3. Click Open. ENVI automatically extracts the necessary header information, including
the associated georeferencing information, and places the filename and bands in the
Available Bands List.
Figure 53. Header Info Dialog Box

42
2.2.2. Display Cursor Location and Value
1. To display the cursor location and value, select Window à Cursor Location/Value from
either the ENVI main menu bar or the Display group menu bar. You can also right-
click in the Image window and select Cursor Location/Value.
2. Double-click in the Image window to display or to hide the Cursor Location/Value
dialog
2.2.3. Linking Two Displays
Link the two displays together for comparison. When you link two displays, any action
you perform on one display (scrolling, zooming, etc.) is echoed in the linked display. To
link the two displays on your screen, do the
following.
1. From the Display group menu bar, select
Tools  Link  Link Displays. You can
also right-click in the Image window and
select Link Displays.
2. The Link Displays dialog box opens
3. Select the displays to be linked and click
OK in the Link Displays dialog to
establish the link.
4. Scroll and zoom in on display group and
observe as the changes are mirrored in
the second display.
2.2.4. Displaying Spectral Profiles
1. From the Display group menu bar, select Tools  Profiles  Z Profile to display a
spectral plot. You can also open a Z profile from the right-click menu in any Image
window.
2. The spectral profile window will open as shown in Figure 55.
Figure 54. Link Displays
Figure 55. Spectral Profile of a Pixel

43
2.2.5. Selecting Regions of Interest
ENVI lets you define regions of interest (ROIs) in your images. ROIs are typically used to
extract statistics for classification, masking, and for deriving average spectra from a group
of pixels. You can define as many ROI’s as you want in a displayed image.
1. From the Display group menu bar, select Overlay  Region of Interest or right-click
in the Image window and select ROI Tool.
2. Draw a polygon that represents the region of interest by clicking the left mouse button
in the Image window to establish the first point of the ROI polygon, then selecting
further border points in sequence by clicking the left button again. Close the polygon
by clicking the right mouse button, then accept the polygon by clicking the right mouse
button again. The middle mouse button deletes the most recent point, or (if you have
closed the polygon) the entire polygon.
ROIs can also be defined in the Zoom and Scroll windows by selecting the appropriate
window radio button in the ROI Tool dialog.
When you have finished defining an ROI, it is shown in the dialog table, with the
name, region color, number of pixels enclosed, and other ROI properties. ROIs can
also be defined as polylines or as a collection of individual pixels by selecting the desired
ROI type from the ROI_Type pull-down menu in the ROI Tool as shown in Figure 2.6.
3. Click the New Region button.
4. Select an ROI by clicking in a cell of the far left column of the ROI Tool table. An ROI is
selected when its entire row is highlighted. An asterisk next to the row also signifies
the currently active ROI. Multiple ROIs can be selected by using Shift-click or Ctrl-
click. All the ROIs can be selected by clicking the Select All button. Click and type to
edit the values in the cells of the ROI Tool table. Change the name for the region and
select a new color.
Figure 56. ROI Tool

44
5. Hide ROIs by selecting them in the table and clicking the Hide ROIs button. Use the
Show ROIs button to re-display these hidden ROIs.
6. Go to an ROI in the ENVI display by selecting it and clicking the Goto button.
7. View the statistics for one or more ROIs by selecting them in the table and clicking the
Stats button.
8. Grow an ROI to its neighboring pixels within a specified threshold by selecting it and
clicking the Grow button.
9. Pixelate polygon and polyline ROIs by selecting them in the table and clicking the
Pixel button. Pixelated objects become a collection of editable points.
10. Delete ROIs by selecting them in the table and clicking the Delete button.
The ROI Tool table also allows you to view and edit various ROI properties, such as
name, color, and fill pattern. Menu options available at the top of the ROI Tool dialog
let you perform other various tasks, such as calculate ROI means, save your ROI
definitions, and load saved definitions. ROI definitions are retained in memory after
the ROI Tool dialog is closed, unless you explicitly delete them. ROIs are available to
other ENVI functions even if they are not displayed.
11. Close the ROI Tool using the menu at the top of the table, select File  Cancel.
3. Hyperspectral Data Preprocessing
This section discusses about the various pre-processing steps which are applied on the
Hyperion dataset. The dataset used in this research work is the Hyperion level L1R dataset
of the Jalandhar area and its surroundings. The Hyperion is a push-broom sensor with
242 contiguous, narrow bandwidth bands. Because of the huge volume of spectral data
available, and the noise present the spaceborne hyperspectral dataset, it requires careful
pre-processing for managing the noise. The pre-processing of dataset can be considered
as the first step towards further interaction with the dataset.
The pre-processing approach adopted in this thesis involves:
• Bad band removal i.e. removing the bands with no information,
• Along track destriping and
• Atmospheric corrections to convert the radiance to reflectance.
3.1. Bad Band Removal
Hyperion level L1R data has 242 bands out of which only 198 are nonzero i.e. a few were
intentionally left unused (Bands 1 to 7 and 225 to 242) and others fall in the overlap region
of the two spectrometers (Bands 58 to 76). Among the non zero bands, four band are still
in the overlap region of the two spectrometers i.e. bands 56, 57 and 77, 78 out of which
bands 77 and 78 were eliminated because of the higher noise levels present in those bands
[6], which left us with 196 unique bands.
Then there are water vapour absorption bands which needs to be eliminated and are
identified as bands120 to 132 (1346nm to 1467 nm), bands 165-182 (1800 to 1971 nm) and

45
bands 221 (above 2356) and higher. Water vapour absorption bands absorb all the incident
solar energy and can be easily identified visually. The number of bands to be used for
further analysis is decided by the user based on the application. However, the list of bands
which are eliminated including the water absorption bands is given below in Table 3.3.
Table 3.3: List of Unused Bands of the Hyperion Sensor, L1R product
Bands Description
1 to 7 Not Illuminated
58 to 78 Overlap Region
120 to 132 Water Vapour Absorption Band
165 to 182 Water Vapour Absorption Band
185 to 187 Identified by Hyperion Bad Band List
221 to 224 Water Vapour Absorption Band
225 to 242 Not Illuminated
3.1.1. Band Selection Using Spectral Subsetting
The steps for band selection are given below and are also illustrated in Figure 57.
1. From the ENVI menu bar select Basic Tools  Resize Data(Spatial Spectral)
2. In the Resize Data Input File window select the File and click Spectral Subset
3. In the File Spectral Subset dialog box Select the desired bands manually or click Apply
BBL and then click OK
4. Click OK in Resize Data Input File window and give the output filename.
Figure 57. Spectral Subset

46
3.2. Along Track Destriping
There are a number of corrupted pixels and dark vertical stripes in the Hyperion datasets
that are caused by calibration differences in Hyperion detector array and temporal
variations in the detector response [7]. The vertical stripes are in the along-track direction
and appear as a series of stripes either along the whole length of the image or intermittently
and are also referred to as striping noise. These vertical stripes and the corrupted pixels
are referred to as abnormal pixels [8]. These abnormal pixels must be accounted for and
corrected before further processing.
According to Han et al. [8] majority abnormal pixels in the Hyperion images appear as
vertical stripes and can be classified into 4 categories:
• Class1 - continuous with atypical DN values - extremely small DN values, usually
zero
• Class2 - continuous with low DN values - low DN values compared to adjacent columns
• Class3 - intermittent with atypical DN values - extremely small DN values
• Class4 - intermittent with lower DN values - low DN values compared to neighbouring
pixels
The figures below show examples of different types of abnormal pixels in the Hyperion
data. Figure 58. a) shows the Class 1 type of abnormal pixels by taking a spatial subset
from the Hyperion image and Figure 58. b) shows the corrected image after correcting the
image using Hyperiontools.sav.
(a) Original Band (b) Band after correction
Figure 58. a) Class 1 Abnormal pixels: Continuous with atypical DN values, Band 99 and
b) Band after correction using Hyperion tools.sav

47
Figure 60. a) Class 2 Abnormal pixels: Continuous with low DN values, Band 10,
b) Band after correction using Hyperion tools.sav
Figure 59. a) Class 4 Intermittent pixels: Intermittent with atypical DN values, Band 14 and
b) Band after correction using Hyperion tools.sav

48
The level L1R Hyperion dataset contains a number of bands containing a series of
vertical stripes which are left for the user to correct according to its convenience. While
generating the bad band list the hyperiontools.sav utility of ENVI uses the flag mask
correction for detecting and correcting the continuous vertical stripes and the abnormal
pixels with atypical values. Figures 58 (b), 59 (b) and 60 (b) show the output of the
hyperiontools.sav utility, for band number 99, 14 and 10 respectively.
3.3. Atmospheric Corrections – Converting Radiance to Reflectance
The electromagnetic signals recorded by the space borne or airborne hyperspectral sensors
are a combination of the signals from earth’s surface, atmospheric constituents and sensor
errors. Thus for quantitative analysis of earth reflectance, these atmospheric effects need
to be removed from the acquired signal and the procedure is called atmospheric correction
or compensation. Atmospheric corrections transform the hyperspectral data to apparent
surface reflectance. Atmospheric corrections are required for matching the image
endmember spectra with the reference spectral libraries or ground data.
The choice of atmospheric correction method depends upon various factors such as
nature of the problem, type of the sensor and data available, historical atmospheric
information of the area, etc. Different atmospheric correction techniques which are available
to process hyperspectral datasets are listed in Table 3.4.
Table 3.4: Atmospheric correction: Methods and Models
Empirical Approaches (Statistical Based) Atmospheric Models (Physics Based)
Empirical line FLAASH
IARR ATREM
Flat Field ATCOR
Output = relative reflectance Output = absolute reflectance
3.3.1. Empirical Line Correction (ELC)
The Empirical Line correction methods involves computing a empirical relation between
the radiance and reflectance using a dark and bright target from the study area, both of
which are well defined by field as well as image spectra. For an optimal representation
these targets should be acquired in the field during the over flight of satellite. A linear
regression is applied for all wavelengths which equates DN to reflectance.
First, the radiance spectrum is derived from a dark and a bright target from the scene
by either using the ROI tool or by selecting a single pixel.
1. From the ENVI main menu bar, select Basic Tools  Preprocessing  Calibration
Utilities  Empirical Line  Compute Factors and Calibrate.
2. The Empirical Line Input File dialog appears.

49
3. Select the input hyperion file and click OK.
4. Empirical Line Spectra dialog box opens.
5. In the Empirical Line Spectra dialog, click Data
Spectra  Import Spectra.
6. The Data Spectra Collection dialog box opens.
7. Collect spectra using the Import menu
8. After the data spectra are selected, click Apply.
The spectra names are entered into the Empirical
Line Spectra dialog.
9. In the Empirical Line Spectra dialog, click Field
Spectra  Import Spectra.
10. Collect spectra using the Import menu
11. Click Apply to enter the spectra names
12. In the Empirical Line Spectra dialog, select the
data spectrum name at the top list.
13. In the bottom list, select the corresponding field
spectrum name.
14. Click Enter Pair to associate the two spectra. The
paired spectra are listed in the Selected Pairs
field.
15. Repeat the selection process for as many data and field spectra pairs as desired.
16. Click OK. The Empirical Line Calibration Parameters dialog appears.
17. Enter filename in Output Calibration Filename field and clock OK.
18. The calibration factors are plotted in a plot window and ENVI adds the resulting output
to the Available Bands List.
3.3.2. Internal Average Relative Reflectance (IARR)
The Internal Average Relative Reflectance (IARR) calibration technique is used to normalize
images to a scene average spectrum. This is particularly effective for reducing imaging
spectrometer data to “relative reflectance” in an area where no ground measurements
exist and little is known about the scene. It works best for arid areas with no vegetation.
The IARR calibration is performed by calculating an average spectrum for the entire AVIRIS
scene and using this as the reference spectrum. Apparent reflectance is calculated for each
pixel of the image by dividing the reference spectrum into the spectrum for each pixel.
IARR requires no user input.
1. From the ENVI menu bar select Spectral  Preprocessing  Calibration Utilities
 IAR Reflectance or Basic tools  Preprocessing  Calibration Utilities  IAR
Reflectance
2. The Calibration Input File dialog box will open
3. Select the Hyperion file and take a spectral subset of 158 bands and click OK
Figure 61.

50
4. IARR Calibration Parameters dialog box opens
5. Enter Output Filename and click OK
3.3.3. Flat Field Correction
Flat field calibration produces relative reflectance by dividing the mean spectrum of a
user-defined ROI into the spectrum of each pixel in the image. ROIs you define should be
a spectrally flat material within the wavelength range of the sensor. Beach sand and concrete
are popular choices. Materials with spectral features, such as vegetation, are a poor choice.
Since the mean spectrum of the ROI is divided into each pixel, the relative reflectance for
pixels within the ROI will be flat and have a value around 1.0.
1. To perform flat field calibration, first define a region of interest.
2. In the ENVI display window select Tools  Region of Interest à ROI Tool
3. The ROI Tool dialog box opens.
4. Set in the ROI tool the target window to the Zoom Window and ROI_TYPE as Polygon.
5. Draw a polygon in the Zoom Window over settlement area of the Image.
6. When satisfied with your polygon, close the polygon by clicking the right mouse button.
Click again with the right mouse button to capture the pixels inside the polygon.
7. Save the ROI (File  Save in ROI tool) to disk and exit the ROI tool.
8. From the ENVI menu bar select Spectral  Preprocessing  Calibration Utilities
 Flat Field or Basic tools  Preprocessing  Calibration Utilities  Flat Field
9. The Calibration Input File dialog box will open
10. Select the Hyperion file and take a spectral subset of 158 bands and click OK
11. Flat Field Calibration Parameters dialog box opens
12. Enter Output Filename and click OK.
3.3.4. FLAASH (Fast Line-of-Sight Atmospheric Analysis of the Spectral Hyper cubes)
ENVI’s FLAASH module is a model for retrieving spectral reflectance from hyperspectral
radiance images and was developed by Spectral Sciences, Inc., under the sponsorship of
the U.S. Air Force Research Laboratory. It compensates for atmospheric effects and corrects
wavelengths in the visible region of electromagnetic spectrum through NIR and SWIR
region. FLAASH has inbuilt support for hyperspectral sensors such as Hyperion, AVIRIS,
HYDICE, HYMAP,Probe-1, CASI and multispectral sensors such as Landsat, SPOT, IRS,
AVHRR, ASTER etc.
3.3.4.1. Data Requirements
• The input to FLAASH atmospheric correction module must be a radiometrically
calibrated radiance image in BIL or BIP format.
• For water retrieval the image bands must cover at least one of the following ranges at
15nm or better spectral resolution:
770 – 870 nm (for the 820 nm water feature)

51
870 – 1020 nm (for the 940 nm water feature)
1050 – 1210 nm (for the 1135 nm water feature)
• Wavelengths, FWHM values must be available in ENVI header files or as separate
ASCII files.
• Scale factors in ASCII format to convert radiance image into floating point values (units:
W/cm2 nm sr) which is the required FLAASH input data format
3.3.4.2. Conversion to BIP/BIL Format
1. From ENVI main menu bar select Basic Tools  Convert Data (BSQ, BIL, BIP)
2. Convert File Input File window opens.
3. Select the Hyperspectral Image file (ENVI Format) and click OK
4. Convert File Parameters Window Opens
5. Select BIL or BIP option in Output Interleave, enter the output filename and click OK
3.3.4.3. FLAASH Input parameters and Settings
1. From the ENVI menu bar select either
Spectral  FLAASH
Basic tools  Preprocessing  Calibration Utilities FLAASH
The FLAASH Atmospheric Correction Model Input Parameters dialog box appears
as shown in Figure 62.
Figure 62. FLAASH Atmospheric Correction Model Input Parameters dialog box

52
2. In the FLAASH Atmospheric Correction Model Input Parameters dialog box, to select
the input radiance image, click Input Radiance Image and select the Hyperion (BIL/
BIP) image file.
3. Select the option, Read array of scale factors (1 per band) from ASCII file, from the
Radiance Scale Factors dialog box. Then Locate the scale factor file.
4. The various Input and Output parameters required to be entered are listed below
•Input radiance image (in BIL or BIP format)
•Output filename
•Output directory – Directory to which FLAASH results are stored
•Root name – prefix which is appended to all the output FLAASH filenames
•Latitude and Longitude of the centre of the scene
•Sensor type – select the name of the sensor which acquired the radiance data
•Sensor altitude (km) – altitude of the sensor when the image was collected,
•Ground elevation (km) of the area – average scene elevation
•Pixel size (m) – image pixel sixe used for adjacency effect correction
•Flight date and Time GMT (HH:MM:SS)
•Atmospheric Model – The Atmospheric Model is selected is selected based on
seasonal latitude surface temperature model as shown in Table 3.5.
Table 3.5: Atmospheres Based on Latitudinal/Seasonal Dependence of Surface Temperature
Latitude (oN) January March May July September November
80 SAW SAW SAW MLW MLW SAW
70 SAW SAW MLW MLW MLW SAW
60 MLW MLW MLW SAS SAS MLW
50 MLW MLW SAS SAS SAS SAS
40 SAS SAS SAS MLS MLS SAS
30 MLS MLS MLS T T MLS
20 TTTTTT
10 TTTTTT
0TTTTTT
-10 TTTTTT
-20 T T T MLS MLS T
-30 MLS MLS MLS MLS MLS MLS
-40 SAS SAS SAS SAS SAS SAS
-50 SAS SAS SAS MLW MLW SAS
-60 MLW MLW MLW MLW MLW MLW
-70 MLW MLW MLW MLW MLW MLW
-80 MLW MLW MLW SAW MLW M LW

53
Where,
SAW - Sub-Arctic Winter
MLW - Mid-Latitude Winter
SAS - Sub-Arctic Summer
MLS - Mid-Latitude Summer
T - Tropical
•Water Retrieval - FLAASH includes a method for retrieving the water amount for
each pixel. This technique produces a more accurate correction than using a constant
water amount for the entire scene. To use this water retrieval method, the image must
have bands that span at least one of the following ranges at a spectral resolution of 15
nm or better:
◆1050-1210 nm (for the 1135 nm water feature)
◆870-1020 nm (for the 940 nm water feature)
◆770-870 nm (for the 820 nm water feature)
For most of the multispectral sensor types, the Water Retrieval setting is No because
these sensors do not have the appropriate bands to perform the retrieval. The Water
Retrieval options are as follows:–
◆Yes: Perform water retrieval.
The 1135 nm feature is recommended if the appropriate bands are available. If you
select 1135 nm or 940 nm, and the feature is saturated due to an extremely wet
atmosphere, then the 820 nm feature is automatically used in its place if bands
spanning this region are available.
◆No: Use a constant column water vapor amount for all pixels in the image.
In this case, the column water vapor amount is determined according to the standard
column water vapor amount for the selected Atmospheric Model, multiplied by an
optional Water Column Multiplier. Set the Water Column Multiplier value
accordingly.
•Aerosol Model - The model choices are as follows:
◆Rural: Represents aerosols in areas not strongly affected by urban or industrial
sources. The particle sizes are a blend of two distributions, one large and one small.
◆Urban: A mixture of 80% rural aerosol with 20% soot-like aerosols, appropriate for
high-density urban/industrial areas.
◆Maritime: Represents the boundary layer over oceans or continents under a
prevailing wind from the ocean. It is composed of two components, one from sea
spray and another from rural continental aerosol (that omits the largest particles).
◆Tropospheric: Applies to calm, clear (visibility greater than 40 km) conditions over
land and consists of the small-particle component of the rural model.

54
•Aerosol Retrieval
◆None: When you select this option, the value in the Initial Visibility (tm) field is
used for the aerosol model (described in the following section).
◆2-Band (K-T): Use the aerosol retrieval method. If no suitable dark pixels are found,
then the value in the Initial Visibility field is used.
•Initial Visibility (Km)
◆In the Initial Visibility field, enter an estimate of the scene visibility in kilometers.
An estimate of visibility during different conditions is given in Table 3.6.
Table 3.6: Scene Visibility options
Weather Conditions Scene Visibility
Clear 40 – 100 Km
Moderate Haze 20-30 Km
Thick Haze 15 km or less
•Spectral Polishing - Polishing is a term for a linear renormalization method that reduces
spectral artifacts in Hyperspectral data using only the data itself.
The basic assumptions are as follows: –
◆The artifacts may be removed by applying a uniform linear transformation (that is,
channel-dependent gain factors and offsets) to the spectra.
◆Spectrally smooth reference pixels (for example, soil or pavement) can be found
within the scene from which the transformation can be derived.
◆The true spectra of the reference pixels can be approximated by applying a spectral
smoothing operation.
◆Click the Spectral Polishing toggle button to select one of the following options:
Yes : Spectrally polish the reflectance image.
No : Output the unaltered modeled reflectance.
◆In the Width (number of bands) field, enter the width of the smoothing window to
be used in the FLAASH spectral polishing algorithm.
The reference pixels are selected as follows:
◆For each pixel, a smoothed spectrum is calculated.
◆The difference between the smoothed spectrum and the unsmoothed spectrum is
calculated for each measured wavelength.
◆The RMS of the smoothed-unsmoothed difference at each wavelength is calculated,
and normalized by dividing by the mean reflectance over all wavelengths for the
pixel.
◆The pixels with the lowest ten percent of normalized RMS differences are selected
from among the cloud-free, non-blank, and non-vegetated pixels.

55
◆The normalized RMS differences for the selected pixels are histogrammed and the
pixels that fall into the lower half of the histogram are selected to be reference pixels
for the calculation of the gain factor. This eliminates artifacts often found using
EFFORT with a vegetated scene.
◆The gain factor for the linear transformation is computed as the ratio of the RMS
smoothed to RMS un-smoothed spectrum for these pixels. There is no offset in the
transformation. For this reason, very dark pixels such as water are essentially
unaffected by the FLAASH polishing.
A larger number generates more smoothing. A value of 9 is recommended for typical
10 nm-resolution hyperspectral sensors (such as AVIRIS). A value of 2 provides minimal
smoothing but removes odd-even spectral band imbalances. Odd polishing widths are
slightly more computationally efficient. Spectral polishing requires hyperspectral input
data, and is therefore disabled when a multispectral sensor type is selected.
•Wavelength Recalibration - An accurate wavelength calibration is critical for
atmospherically correcting Hyperspectral data. Even slight errors in the locations of
the band center wavelengths can introduce significant errors into the water retrieval
process, and reduce the overall accuracy of the modeled surface reflectance results. To
minimize such errors, FLAASH includes a method for identifying and correcting
wavelength miscalibrations. AVIRIS, HYDICE, HYMAP, HYPERION, CASI, and AISA
sensors are automatically supported for wavelength recalibration.
◆Yes: Automatically adjust the wavelength calibration prior to computing the water
retrieval.
◆No: Use the input file’s wavelengths.
3.3.4.4. FLAASH Advanced Settings
The FLAASH advanced settings are arranged into three categories: modeling parameters,
viewing geometry and FLAASH processing controls.
1. Click on Advanced Settings in FLAASH Atmospheric Correction Model Input
Parameters window.
2. FLAASH Advanced Settings window opens as shown in Figure 63.
3. FLAASH Advanced Parameters
◆In the Aerosol Scale Height (km) field, enter the effective 1/e height of the aerosol
vertical profile in km. Typical values are 1 to 2 km. The default value is 1.5 km.
◆In the CO2 Mixing Ratio (ppm) field, enter the carbon dioxide (CO2) mixing ratio
in parts per million by volume. In 2001, the value was approximately 370 ppm. For
best results, add 20 ppm to the actual value.
◆Click the Use Square Slit Function toggle button to select Yes for images that were
derived by averaging together adjacent bands (for example, LASH). This better
models the spectral response of the derived bands.
◆Click the Use Adjacency Correction toggle button to specify whether or not to
use adjacency correction. Unlike most atmospheric correction models, the

56
FLAASH model accounts for both the radiance that is reflected from the surface
that travels directly into the sensor and the radiance from the surface that is
scattered by the atmosphere into the sensor. The distinction between the two
accounts for the adjacency effect (spatial mixing of radiance among nearby pixels)
caused by atmospheric scattering. More accurate reflectance retrievals result when
adjacency is enabled; however, there may be occasions when it is desirable to
ignore this effect.
• Click the Reuse MODTRAN Calculations toggle button to specify whether or not to
reuse previous MODTRAN calculations. Reusing previous MODTRAN calculations
is useful for rapidly processing multiple images taken under essentially identical
conditions; the identical illumination, viewing geometries, and visibility are assumed,
but the water vapor retrieval is performed again. It is also useful for rapidly generating
images with and without polishing, or with different polishing widths. Following is
an explanation of the options:
◆No: FLAASH computes a new set of MODRTRAN radiative transfer calculations
for the selected image and model parameters
◆Yes: FLAASH performs the atmospheric correction using the MODTRAN
calculations from the previous FLAASH run. This change causes FLAASH to perform
a water retrieval; therefore, the Water Retrieval toggle button on the main FLAASH
Input Parameters dialog is automatically set to Yes . You cannot reset Water Retrieval
to No until you set Reuse MODTRAN Calculations to No.
• In the Modtran Resolution drop-down list, select a resolution. The Modtran Resolution
setting controls the MODTRAN spectral resolution and the trade-off of speed versus
accuracy for the MODTRAN4 portion of the calculation. Lower resolution yields
proportionally better speed but less accuracy. The main differences in accuracy are
Figure 63. FLAASH Advanced Settings

57
seen near 2000 nm and beyond. The 5 cm-1 resolution is the default value when you
select a hyperspectral sensor as input, but it changes to 15 cm-1 when you select a
multispectral sensor. If aerosol is being retrieved, there are two MODTRAN4 runs
performed at 15 cm-1 resolution followed by one MODTRAN4 run at the resolution
you select. If aerosols are not being retrieved, the first two runs are omitted.
• In the Modtran Multiscatter Model drop-down list, select a multiple-scattering
algorithm to be used by MODTRAN4. FLAASH offers three options for multiscatter
models:–
◆Isaacs : The Isaacs 2-stream method is fast but oversimplified.
◆Scaled DISORT : The Scaled DISORT method provides near-DISORT accuracy with
almost the same speed as Isaacs.
◆DISORT : The DISORT model provides the most accurate shortwave (less than ~
1000 nm) corrections, however it is very computationally intensive. DISORT
multiscatter model dramatically increases FLAASH processing time, and is rarely
necessary for accurate atmospheric corrections. DISORT with 8 streams usually
takes about 30 times longer than using Isaacs.
◆The default is Scaled DISORT with eight streams.
◆In the Scaled DISORT model, DISORT and Isaacs calculations are performed at a
small number of atmospheric window wavelengths.
4. After entering all the required parameters click on Apply. Now, the Radiance Hyperion
Data is converted into Reflectance Image. The final output will be listed on the Available
Band List Dialog Box.
5. Display and compare the Reflectance and the Radiance spectral profiles of the same
features in separate windows before and after atmospheric corrections as shown in
Figure 64.
The data specific parameters used in the FLAASH atmospheric correction module for
the Jalandhar Hyperion data can be found in the table given in Appendix A.
Figure 64. Spectral profile (Z-profile) of a randomly selected pixel, a) before Atmospheric
corrections and b) after Atmospheric corrections

58
3.4. Spatial Subsetting
The spatial subset of the image is usually taken to extract the area of interest of the user
from the Hyperion Image.
1. From ENVI mail menu bar select Basic Tools  Resize Data (Spatial/Spectral)
2. Resize Data Input File dialog box will open
3. Select the Hyperion file and click Spatial Subset
4. Select Spatial Subset dialog box opens.
5. The spatial subset can be selected by using one the following methods:
◆Entering samples and line values
◆Selecting interactively from the image
◆Entering map coordinates
◆Using the same spatial subset that was previously used on another file
◆Using the image shown in the meta scroll window
◆Using the bounding box around a region of interest
6. Choose one of the methods for Subsetting the data and click OK
7. Choose the Resampling technique in the Resize Data Parameters dialog box.
8. Enter the Output Filename and click OK.
4. Advanced Hyperspectral Analysis
In this section the user will be introduced to the advanced concepts and procedures for
the analysis of hyperspectral data starting from dimensionality reduction techniques to
spectral unmixing. We will use the Hyperion data of Jalandhar, India for our analysis.
4.1. Minimum Noise Fraction (MNF) Transform
Switzer & Green [9] , and Green et. al. [10] proposed the MNF transform which chooses the
new components to maximize the SNR and orders them according to increasing image
quality or decreasing noise. Minimum noise fraction (MNF) [10] computes the noise
statistics information for effectively removing the noise from the dataset and for
determining the inherent dimensionality of the dataset. MNF can be treated as two cascaded
Principal Component Transformations; the first is the transformation of the noise covariance
matrix to an identity matrix also called as the noise whitening step. The second is the
standard principal component transformation of the noise whitened dataset maximizing
the signal to noise ratio (SNR) and thus segregating the signal from the noise. The noise
statistics are calculated using the shift difference method also known as nearest neighbour
difference [10].
MNF splits and projects the input image into two subspaces based on visual analysis
of the images and associated eigenvalues: The first one is the Signal Subspace (signal plus
noise) corresponding the largest eigenvalues and the second is the noise subspace
corresponding to the lower eigenvalues. MNF images (eigen images) are used to evaluate

59
the dimensionality of the data. Eigenvalues for bands that contain information will be an
order of magnitude larger than those that contain only noise. The corresponding images
will be spatially coherent, while the noise images will not contain any spatial information.
In ENVI, MNF transform is used to remove noise from data by performing a forward
MNF transform. ENVI assumes that each pixel contains both signal and noise, and that
adjacent pixels contain the same signal but different noise. The best noise estimate is
gathered using the shift-difference statistics from a homogeneous area rather than from
the whole image.
1. From the ENVI menu bar, select either Transform  MNF Rotation  Forward MNF
 Estimate Noise Statistics From Data or Spectral  MNF Rotation  Forward
MNF  Estimate Noise Statistics From Data
2. The input File dialog box appears
3. Select an input file and perform optional Spatial Subsetting, Spectral Subsetting, and/
or Masking, then click OK. The Forward MNF Transform Parameters dialog appears.
4. Click Shift Diff Subset to select a spatial subset or an area under an ROI/EVF/and so
forth on which to calculate the statistics. You can then apply the calculated results to
the entire file (or to the file subset if you selected one when you selected the input file).
5. In the Enter Output Noise Stats Filename [.sta] field, enter a filename for the noise
statistics.
6. In the Output MNF Stats Filename [.sta] field, enter an output file for the MNF statistics.
Be sure that the MNF and noise statistics files have different names.
7. Select output to File or Memory.
8. To select the number of output bands without examining the eigenvalues, select No
from the Select Subset from Eigenvalues toggle button, then set the Number of Output
MNF Bands.
9. To select the number of output MNF bands by examining the eigenvalues, use the
following steps:
◆Select Yes from the Select Subset from Eigenvalues toggle button.
◆Click OK. ENVI calculates the statistics and the Select Output MNF Bands dialog
appears, with each band listed with its corresponding eigenvalue. Also listed is the
cumulative percentage of data variance contained in each MNF band for all bands.
◆Set the Number of Output MNF Bands. For the best results, and to save disk space,
output only those bands with high eigenvalues. Images with eigenvalues close to 1
are mostly noise.
10. Click OK. When ENVI finishes processing, the MNF Eigenvalues plot window appears
as shown in Figure 65 and the MNF bands are added to the Available Bands List.
Display the MNF bands from the Available Bands List and compare with the MNF
Eigenvalue plot to determine which bands contain data and which bands contain
predominantly noise. In subsequent processing of this data, spectrally subset the MNF
bands to only include those bands where the images appear spatially coherent and the

60
eigenvalues are above the break in slope of the MNF Eigenvalue plot. We should only
include the first ten to twelve MNF bands because these bands contain 95% of the total
information.
4.2. Pixel Purity Index
Pixel purity index (PPI) [11] algorithm, is one of the most widely endmember extraction
algorithm used for hyperspectral image analysis. PPI is a means of finding the most
“spectrally pure,” or extreme, pixels in the hyperspectral images. First the dataset is
transformed onto lower dimensions by using either PCA or MNF as the assumption here
is that the endmembers lie in the first few principal components. The endmember pixels
are obtained by repeatedly projecting the transformed data onto randomly projected vectors
(k) in n-dimensional space. As the vectors are randomly generated the results depend
upon the number of random projections. Pixels lying at the extremes of a random vector
are assigned a purity value. The values are updated after each projection and the pixels
having values more than a set threshold (t) are considered as “pure” pixels. The extreme
pixels in each projection are recorded and the total number of times each pixel is marked
as extreme is noted. A Pixel Purity Index (PPI) image is created in which the DN of each
pixel corresponds to the number of times that pixel was recorded as extreme. ENVI employs
a FAST PPI method which the image data into memory and performs the computations in
memory, which is much faster than the disk-based PPI.
1. From the ENVI main menu bar, select Spectral  Pixel Purity Index  New Output
Band or [FAST] New Output Band.
2. The Input File dialog appears.
3. Select the input file (PCA or MNF transform file) and perform optional Spatial
Subsetting, Spectral Subsetting, then click OK.
Figure 65. MNF Eigenvalue Plot

61
4. Click OK. The FAST Pixel Purity Index Parameters
dialog appears, as shown in Figure 66.
5. Enter a Number of Iterations value.
The iterations in the PPI Parameters dialog
designate the number of times the data will be projected
onto the random vector. After a certain number of
iterations, the PPI result will stabilize.
6. Enter a Threshold Factor value in data units for
extreme pixel selection.
The threshold is a measure for the extremeness of
the pixels. For example, a threshold of 2 marks all
pixels greater than two digital numbers (DN) from
the extreme pixels (both high and low) as being
extreme. This threshold selects the pixels on the ends
of the projected vector. The threshold should be approximately 2-3 times the noise
level in the data.
7. Enter the Output Filename and click OK
8. A dialog appears that indicates the amount of memory needed and prompts you to
continue if that amount of memory is acceptable.
A processing status dialog appears with the PPI plot as shown in Figure 67. This plot
shows the total number of extreme pixels satisfying the threshold criterion found by the
PPI processing as a function of the number of iterations. It should asymptotically approach
a flat line (zero slope) when all of the extreme pixels are found.
Figure 66. PPI Parameters
Figure 67. Pixel Purity Index Plot

62
4.2.1. PPI Images for Endmember Selection
1. Display the PPI image. Brighter pixels represent more spectrally pure and extreme
hits. Darker pixels are less spectrally pure.
2. Select Window  Cursor Location/Value from the ENVI main menu bar, or select
Tools  Cursor Location/Value from the Display group menu bar to determine the
range of values present in the image.
3. From the Display group menu bar, select Overlay  Region of Interest to open the
ROI Tool dialog.
4. From the ROI Tool menu bar, select Options  Band Threshold to ROI to create an
ROI containing only the pixels with high PPI values (i.e. the pure pixels)
5. This ROI contains the pixel locations of the purest pixels in the image regardless of the
endmember to which they correspond. The n-Dimensional Visualizer will be used in
the next section to extract the specific pure endmembers.
4.2.2. n-D Visualizer
Spectra can be thought of as points in an n-dimensional scatter plot, where n is the number
of bands [11]. The coordinates of the points in n-space consist of “n” values that are simply
the spectral radiance or reflectance values in each band for a given pixel. The distribution
of these points in n-space can be used to estimate the number of spectral endmembers and
their pure spectral signatures. ENVI’s n-Dimensional Visualizer provides an interactive
tool for selecting the endmembers in n-space.
1. From the ENVI main menu bar, select Spectral  n-Dimensional Visualizer 
Visualize with New Data.
2. Select the file to extract the n-D scatter plots from (typically an MNF file).
3. Spectrally subset the MNF data to exclude noise bands determined by reviewing the
eigenimages and eigenvalue plot.
Figure 68. n-D Visualizer (left) and n-D Controls dialog (right)

63
4. For speed and clarity, an ROI is used to limit the number of pixels that are input into
the n-D Visualizer. If only one ROI is present for the input image, it is automatically
used as input to the n-D Visualizer. If more than one ROI is present, the n-D Visualizer
Input ROI dialog appears. Select the ROI to use.
5. A status box appears while the ROI is loaded. The n-D Visualizer and n-D Controls
dialogs appear.
6. In the n-D Controls dialog, click the band numbers to be projected in the n-D Visualizer.
Select three or more bands for rotation to be possible.
7. Visually identify and distinguish the purest endmembers in the image by selecting the
pixels lying at the tip of a corner in the data cloud and assign a different colour to each
corner which corresponds to a different and spectrally unique endmember.
4.2.3. Defining classes using n-D Visualizer
Classes are defined when groups of pixels stay together during rotation and are separated
from the rest of the pixels. Multiple classes can be defined at the same time.
1. Click Stop in the n-D Controls dialog to stop the rotation when a group of pixels is
isolated from the main body of pixels plotted in the n-D Visualizer. Or, use the arrow
buttons to go to a particular projection view.
2. Highlight the desired pixels on the n-D Visualizer by left-clicking to set vertices, and
right-clicking to close the polygon.
3. From the n-D Controls menu bar, select Class and choose a color for the class.
To automatically use the next available class color for the next ROI; select Class 
New from the n-D Controls menu bar (or right-click in the n-D Visualizer and select
New Class).
4. Click Start to rotate the scatter plot until additional groups of pixels are isolated, and
repeat the class definition process.
5. Export your best set of classes to ROIs by right clicking in the N-D visualize window
and select Export ALL.
6. From the N-D controls dialog box select Options  Class Control
7. Extract the average spectra for the different ROIs using either the “Stats” or “Mean”
tabs in the n-D class controls dialog box and compare it with the image spectra.
4.3. Spectral Angle Mapper
The spectral angle mapper (SAM) as explained in [2] computes the spectral similarity
between a test (or pixel) spectrum, t, and the reference spectrum (target spectrum or
laboratory spectrum or another pixel spectrum), r, and is expressed in terms of vector
angle, , as:
Cos =jån
i=1tr
ii
ÖÖ
ån
i=1ti
2ån
i=1ri
2(4.1)
where, - spectral angle, t- test or pixel spectrum
r- reference spectrum, n- number of bands

64
SAM assumes that the data has been converted to apparent reflectance. While
computing the SAM each spectrum is considered a vector in the n-dimensional space, The
output of spectral angle mapping for each pixel is an angular difference between the test
and the reference spectrum measured in radians, ranging from zero radians to Ð/2. The
smaller the spectral angle more is the similarity between the test and the reference spectrum.
Figure 69 gives an example of the spectral angle between a pixel and the reference or
target spectrum.
The spectral angle distance is preferred over other distance metrics as it is insensitive
to illumination differences in a pixel. Any illumination change will change the magnitude
of the vector but not the direction.
For each reference spectrum chosen in the analysis of a hyperspectral image, the spectral
angle, á, is determined for every image spectrum (pixel). This value, in radians, is assigned
to the corresponding pixel in the output SAM image, one output image for each reference
spectrum. The derived spectral angle maps form a new data cube with the number of
bands equal to the number of reference spectra used in the mapping.
The SAM algorithm implemented in ENVI takes as input a number of “training classes”
or reference spectra from ASCII files, ROIs, or spectral libraries. It calculates the angular
distance between each spectrum in the image and the reference spectra or “endmembers”
in n-dimensions. The result is a classification image showing the best SAM match at each
pixel and a “rule” image for each endmember showing the actual angular distance in
radians between each spectrum in the image and the reference spectrum. Darker pixels in
the rule images represent smaller spectral angles, and thus spectra that are more similar
to the reference spectrum. The rule images can be used for subsequent classifications using
different thresholds to decide which pixels are included in the SAM classification image.
1. From ENVI menu bar, select Classification  Supervised  Spectral Angle
Mapper or Spectral  Mapping Methods  Spectral Angle Mapper.
2. The Classification Input File dialog box appears. Select a file and click OK.
Figure 69. Spectral angle between target and the reference spectra

65
3. In Endmember Collection: SAM dialog box, select Import  source of spectra
from the drop down menu and click Apply.
4. The Spectral Angle Mapper Parameters dialog appears.
5. Select the Thresholding options (Maximum angle), enter the Output Filename
and Output Rule Filename and click OK.
6. The results are added to the Available Bands List dialog box.
4.4. Linear Spectral Unmixing
Natural surfaces are rarely composed of a single uniform material. Spectral mixing occurs
when two of more materials with spectrally distinct qualities are represented by a single
image pixel. If the scale of mixing is large (macroscopic), mixing occurs in a linear fashion.
For microscopic or intimate mixtures, the mixing is generally nonlinear. The linear model
assumes no interaction between materials. If each photon only “sees” one material, these
signals add (a linear process). Multiple scattering involving several materials can be thought
of as cascaded multiplications (a non-linear process).
The simplest and the most commonly assumed model for a mixed spectrum is a linear
model. A single pixel can be portrayed as a checkerboard mixture, as illustrated in Figure
70 and assuming that there is no multiple scattering between components, then the spectral
response of the pixel is a linear combination of the fractional abundances (area covered by
each endmember in the pixel) of the individual substances [13], hence the term Linear
Mixture Model (LMM).
Figure 70. Mixing model illustration, a) Linear mixing (no multiple scattering) and
b) Non Linear mixing scenario (multiple bounces due to intimate mixture)
If there are  endmembers, then the linear mixture model can be expressed as
x =  misii + wi = Ms + w, j = 1, 2, … …, N (4.2)

66
where, x - L × 1 received pixel spectra
M - L × p matrix, whose columns are L×1 endmembers.
s - abundance fraction of each endmember in a pixel
w - L × 1 additive noise
N - number of pixels in the image
To be physically meaningful the linear mixture model is subjected to following two
constraints; the first is the non negativity constraint,
sii  0
and the second is the full additivity constraint,
 sii = 1
Spectral unmixing can defined as the process of determination of the number of image
endmembers and their pure signatures and the amount in which they appear in the given
mixed pixel. The whole process of end to end spectral unmixing can be presented as a
sequence of three consecutive procedures [13]:
•Dimensionality Reduction: Reduce the dimension of the data in the scene. This
step is optional and is only invoked by some algorithms to reduce the computational
load of subsequent steps.
•Endmember Determination: Estimate the set of distinct spectra (endmembers) that
constitute the mixed pixels in the scene.
•Spectral Unmixing: Estimate the fractional abundances of each mixed pixel from
its spectrum and the endmember spectra.
The procedure of spectral unmixing in ENVI after the first two steps is given below:
1. From the ENVI main menu bar, select Spectral  Mapping Methods  Linear
Spectral Unmixing. The Input File dialog appears.
2. Select the input file in Unmixing Input File dialog box and click OK.
3. In Endmember Collection: Unmixing dialog box, select Import  source of spectra
to match from the drop down menu and click Apply.
4. To apply a Unit-Sum Constraint in the unmixing, use the Toggle Button to select
Yes and enter a Weight value. This weight is added to the system of simultaneous
equations in the unmixing inversion process. Larger weights cause the unmixing
to honor the unit-sum constraint more closely.
5. Enter the Output Filename and click OK.
The result of the unmixing step will be the fraction abundance images for each
representative class. The pixel values of these images indicate the fraction of the pixel that
contains the endmember material corresponding to that image.

67
4.5. Matched Filtering
Matched Filtering is a signal processing technique in which the response of a known
endmember is maximized and the response of the composite unknown background id
suppressed, thus matching the known spectra [14]. It is used to find the abundances of
user-defined endmembers using a partial unmixing. It provides a rapid means of detecting
specific materials based on matches to library or image endmember spectra and does not
require knowledge of all the endmembers within an image scene. This technique produces
images similar to the unmixing, but with significantly less computation and without the
requirement to know all the endmembers.
1. From the ENVI main menu bar, select Spectral à Mapping Methods  Matched
Filtering. The Input File dialog box appears.
2. Select the input file in Matched Filter Input File dialog box and click OK.
3. In Endmember Collection: Matched Filter dialog box, select Import  source of spectra
to match from the drop down menu and click Apply.
4. Use the toggle button to select Compute New Covariance Stats and enter an Output
Statistics Filename.
5. Enter the Output Filename and click OK.
The results of MF appear as a series of gray scale images, one for each selected
endmember. Floating-point results provide a means of estimating the relative degree of
match to the reference spectrum and approximate sub-pixel abundance, where 1.0 is a
perfect match.
4.6. Binary Encoding
The binary encoding classification technique encodes the data and endmember spectra
into zeros and ones, based on whether a band falls below or above the spectrum mean,
respectively. An exclusive OR function compares each encoded reference spectrum with
the encoded data spectra and produces a classification image. All pixels are classified to
the endmember with the greatest number of bands that match, unless you specify a
minimum match threshold, in which case some pixels may be unclassified if they do not
meet the criteria.
1. From the ENVI main menu bar, select Classification  Supervised à Binary Encoding
or Spectral  Mapping Methods  Binary Encoding
2. The Classification Input File dialog box appears
3. Select the input file in Classification Input File dialog box and click OK
4. In the Select Classes from Regions list, select ROIs and/or vectors as training classes.
The ROIs listed are derived from the available ROIs in the ROI Tool dialog.
5. Select one of the following thresholding options from the Set Minimum Encoding
Threshold area:
•None: No threshold.

68
•Single Value: Use a single threshold for all classes. Enter a decimal percentage
value (from 0.0 to 1.0) in the Minimum Encoding Threshold field. The percentage
value represents the number of bands that must match.
•Multiple Values: Enter a different threshold for each class. Use this option as follows:
◆In the list of classes, select the class or classes to which you want to assign different
threshold values and click Multiple Values. The Assign Minimum Encoding
Threshold dialog appears.
◆Select a class, and enter a threshold value in the field at the bottom of the dialog.
If you do not enter a minimum value, ENVI classifies all pixels. Repeat for each
class. Click OK when you are finished.
6. Enter the Output Class Filename
7. Toggle button to select whether or not to create rule images. Rule images are created
for intermediate classification of image results before final assignment of classes.
8. Enter the Output Rule Filename
9. Click OK. ENVI adds the resulting output to the Available Bands List. If you selected
to output rule images, ENVI creates rule images for each class with the pixel values
equal to the percentage (0-100%) of bands that matched that class.
4.7. Spectral Feature Fitting and Analysis
Spectral Feature Fitting (SFF) is an absorption-feature-based method for matching image
spectra to reference endmembers. Most methods used for analysis of hyperspectral data
still do not directly identify specific materials. They only indicate how similar the material
is to another known material or how unique it is with respect to other materials. Techniques
for direct identification of materials, however, via extraction of specific spectral features
from field and laboratory reflectance spectra have been in use for many years.
These methods require that data
be reduced to reflectance and that a
continuum be removed from the
reflectance data prior to analysis. A
continuum is a mathematical function
used to isolate a particular absorption
feature for analysis. Spectra are
normalized to a common reference
using a continuum formed by
defining high points of the spectrum
(local maxima) and fitting straight
line segments between these points.
The continuum is removed by
dividing it into the original spectrum
(Figure 71). Figure 71. Example of a fitted continuum and a
continuum removed spectrum

69
Continuum removal can be performed on data files or on individual spectra in a plot
window:
1. From the ENVI main menu bar, select Spectral  Mapping Methods  Continuum
Removal. The Input File dialog appears.
2. Select the Input file and click OK.
3. The Continuum Removal Parameters dialog appears.
4. Enter the Output File name and click OK. ENVI adds the resulting output to the
Available Bands List.
4.7.1. Recovering the Continuum Curve
The Continuum Removal tool only outputs the continuum-removed spectra; it does not
show you the curve of the calculated continuum.
1. Display a hyperspectral image.
2. From the Display group menu bar, select Tools  Profiles  Z Profile (Spectrum). A
Spectral Profile window appears.
3. From the Spectral Profile menu bar, select Options  New Window with Plots. An
ENVI Plot Window appears.
4. From the ENVI Plot Window menu bar, select Plot_Function  Continuum Removed.
5. From the ENVI main menu bar, select Spectral  Spectral Math. The Spectral Math
dialog appears.
6. In the Enter an expression field, enter the following:
• float(S1) / (S2)
7. Click OK. The Variables to Spectra Pairings dialog appears.
8. Map S1 to the original spectrum (in the Spectral Profile window), and map S2 to the
continuum-removed plot (in the ENVI Plot Window).
9. Ensure that the Output Result to toggle button is set to Same Window.
10. Click OK. The continuum curve plots over the original spectrum as in Figure 72.
Figure 72. Continuum Curve

70
4.7.2. Spectral Feature Fitting
Spectral feature fitting requires that reference endmembers be selected from either the
image or a spectral library, that both the reference and unknown spectra have the
continuum removed, and that each reference endmember spectrum be scaled to match
the unknown spectrum. A “Scale” image is produced for each endmember selected for
analysis by first subtracting the continuum-removed spectra from one, thus inverting
them and making the continuum zero. A single multiplicative scaling factor is then
determined that makes the reference spectrum match the unknown spectrum.
Assuming that a reasonable spectral range has been selected, a large scaling factor is
equivalent to a deep spectral feature, while a small scaling factor indicates a weak
spectral feature.
1. From the ENVI main menu bar, select Spectral  Mapping Methods  Spectral
Feature Fitting. The Input File dialog appears.
2. Select the input file and click OK.
3. In the Input File dialog box, select Spectral Subset. The File Spectral Subset dialog
appears.
4. Select bands to subset around the region containing the absorption features of interest
and click OK.
5. Click OK in the Input File dialog. The Endmember Collection:Feature Fitting dialog
box appears.
6. Import the reference spectra and click Apply.
7. The Spectral Feature Fitting Parameters dialog box appears.
8. Use the toggle button to switch between Output separate Scale and RMS Images or
Output Combined (Scale/RMS) Image.
9. Click OK. ENVI adds the resulting output to the Available Bands List.
A scale image and RMS image or a combined “fit” (scale/RMS) image is output for
each reference spectrum. The image is a measure of absorption feature depth, which is
related to material abundance. The brighter pixels in the scale image indicate a better
match to the reference material in those pixels (for areas with a low rms error). However,
a large scale value (> 1) can result if incorrect reference endmembers are input or if the
incorrect wavelength range is used. The image and reference spectra are compared at
each wavelength in a least-squares sense, and the RMS error is calculated for each
reference spectrum. Dark pixels in the rms error image indicate a low error. You can use
the RMS errors and scale image results to locate areas that best match the reference
spectrum.

71
References
[1] T. Lillesand, R. W. Kiefer, and J. W. Chipman, Remote sensing and image interpretation, 5th ed.
New York: Wiley, 2004.
[2] F. D. van der Meer and S. M. de Jong, Imaging spectrometry/ : basic principles and prospective
applications. Dordrecht: Kluwer Academic Publishers, 2001.
[3] A. R. Korb, P. Dybwad, W. Wadsworth, and J. W. Salisbury, “Portable Fourier transform
infrared spectroradiometer for field measurements of radiance and emissivity,” Applied Optics,
vol. 35, no. 10, p. 1679, Apr. 1996.
[4] F. A. Kruse, J. W. Boardman, and J. F. Huntington, “Comparison of airborne hyperspectral
data and EO-1 Hyperion for mineral mapping,” Geoscience and Remote Sensing, IEEE
Transactions on DOI - 10.1109/TGRS.2003.812908, vol. 41, no. 6, pp. 1388–1400, 2003.
[5] P. Varshney, Advanced image processing techniques for remotely sensed hyperspectral data. Berlin/ ;;
New York: Springer, 2004.
[6] K. Simon, “Hyperion Level 1G (L1GST) Product Output Files Data Format Control Book
(DFCB).” Department of the Interior U.S. Geological Survey, Apr-2006.
[7] B. Datt, T. R. McVicar, T. G. Van Niel, D. L. B. Jupp, and J. S. Pearlman, “Preprocessing eo-1
hyperion hyperspectral data to support the application of agricultural indexes,” IEEE Trans.
Geosci. Remote Sensing, vol. 41, no. 6, pp. 1246–1259, Jun. 2003.
[8] N. Acito, M. Diani, and G. Corsini, “Subspace-Based Striping Noise Reduction in
Hyperspectral Images,” IEEE Trans. Geosci. Remote Sensing, 2010.
[9] T. Han, D. G. Goodenough, A. Dyk, and J. Love, “Detection and correction of abnormal pixels
in Hyperion images,” in IEEE International Geoscience and Remote Sensing Symposium, Toronto,
Ont., Canada, pp. 1327–1330.
[10] P. Switzer and A. A. Green, “Min / Max autocorrelation factors for multivariate spatial
imagery,” Department of Statistics, Stanford University, Technical Report No. 6, Apr. 1984.
[11] A. A. Green, M. Berman, P. Switzer, and M. D. Craig, “A transformation for ordering
multispectral data in terms of image quality with implications for noise removal,” Geoscience
and Remote Sensing, IEEE Transactions on DOI - 10.1109/36.3001, vol. 26, no. 1, pp. 65–74, 1988.
[12] J. W. Boardman, F. A. Kruse, and R. O. Green, “Mapping Target Signatures Via Partial
Unmixing Of Aviris Data,” Center for the Study of Earth from Space Cooperative Institute
for Research in the Environmental Sciences, Campus Box 216 University of Colorado, Emdder,
CO S03W-0216, 1995.
[13] N. Keshava and J. F. Mustard, “Spectral unmixing,” Signal Processing Magazine, IEEE DOI -
10.1109/79.974727, vol. 19, no. 1, pp. 44–57, 2002.
[14] N. Keshava, “Algorithm taxonomy for hyperspectral unmixing,” in Proceedings of SPIE,
Orlando, FL, USA, 2000, pp. 42–63.
[15] J. Chen and I. Reed, “A Detection Algorithm for Optical Targets in Clutter,” IEEE Trans. Aerosp.
Electron. Syst., vol. AES-23, no. 1, pp. 46–59, Jan. 1987.

72
Appendix – A
Scene center location Lat:- 75.38439941 Long:- 31.76269913
Sensor altitude 705 Km
Ground elevation 0.100 Km
Pixel size 30 m
Flight date May 12th, 2008
Flight Time 0.22
Atmospheric Model Mid Latitude Summer
Water Retrieval Yes
Water Absorption Feature 820 nm
Aerosol Model Urban
Aerosol Retrieval 2-Band (K-T)
Initial Visibility 40 Km
Spectral Polishing Yes
Width (No. of bands 9.00
Wavelength Recalibration No
Aerosol Scale Height 2 Km
CO2 mixing ratio (ppm) 390 ppm
Use Square Slit Function No
Use Adjacency Correction Yes
Reuse MODTRAN Calculation No
Modtran Resolution 15 cm-1
MODTRAN Multiscatter Model Scaled Distort
No of Distort Streams 8.00
Zenith Angle 7.50
Azimuth Angle 115:31:43:68
Use Tiled Processing Full Scene
Automatically Save Template File Yes
Output Reflectance Scale Factor 1000.00
Output Diagnostic Files No

Processing of Hyperspectral
Remote Sensing Data
Division of Agricultural Physics
Indian Agricultural Research Institute
New Delhi - 110 012
Rabi N. Sahoo, Sourabh Pargal
Sanatan Pradhan, Gopal Krishna, Vinod K. Gupta