Hyperspectral Image Analysis - Science topic

USGS High Resolution Spectral Library https://www.usgs.gov/centers/gggsc/science/usgs-high-resolution-spectral-library

JPL Spectral Library https://speclib.jpl.nasa.gov/documents/jpl_desc

Also see http://www.planetary.brown.edu/pds/other.html

Hyperspectral Document Image processing for Writer Identification

Ink Mismatch Detection using Hyperspectral Document Imaging

AN update to my response:

As the PRISMA datasets are available, one can use ENVI 5.6 (if available) for processing PRISMA data. Apart from that ENMAP toolbox in QGIS, EarthBit software from Planetek ITALIA and ERDAS tool for PRISMA ( inbuilt in the latest version), prismaread (https://ranghetti.github.io/prismaread/index.html) can also be explored.

…

Did you get the spectral signature of the ground truth of the Indiana pines data set.

please share it with me. I will be highly grateful to you.@

As Mr Nasir Hameed suggested, if you are looking forward to inspecting individual spectral signatures in a hyperspectral image, the Digital Numbers (DN) are more than adequate to design an autonomous classification, while atmospheric and geographic corrections would be unnecessary for the framework.

Yes, the folder looks like this for a single snap!

I will inbox you as well. Thank you again :)

How to create patches from hyperspectral image without required groundTruth data?

Matthew Mabey Thank so much, Matthew. It's very help full. I appreciate it a lot.

As I know GDAL library cab handle this data. follow below link:

Contact Prof. Ines Gockel at Leipzig University Department of Surgery

Mercury pollution may be picked up by analyzing predatory fish at top of food chain. Mosquito fish (Ambrosia app.) have also been found to accumulate nitrogen. You might find added information from US National Mercury Lab. Wetlands in headwaters of black water streams have also been found to reduce mercury to toxic methyl mercury form.

There are some practical ways to determine the best size, but it all comes to evaluations. The general rule is the optimal size of the hidden layer is usually between the size of the input and size of the output layers.

In sum, for most problems, one could probably get decent performance (even without a second optimization step) by setting the hidden layer configuration using just two rules: (i) number of hidden layers equals one; and (ii) the number of neurons in that layer is the mean of the neurons in the input and output layers.

Jeff Heaton, author of Introduction to Neural Networks in Java offers a few more.

In addition to the recommended articles by Bachir Achour , you can also check out this article (see attached)

You can try this fcn of MATLAB image processing toolbox:

https://www.mathworks.com/help/matlab/ref/multibandread.html

Also, you can try this 3rd party fcns:

https://www.mathworks.com/matlabcentral/fileexchange/27172-envi-file-reader-writer

Good luck.

Ahmet,

Not sure you still need it, but I guess it maybe useful for others.

Please find below a small script for R. There is maybe smarter ways to write it, but at least it works (data and output attached):

#set your directory and file you want to use

setwd("D:/filepath of your working directory")

d1<-read.table("SRI_Chlab_subsampled_RG.csv", head=T, sep = ",", na.string = NA)

dim(d1)

#[1] 72 14

head(d1)

#Create an empty matrix with column and row number corresponding to the #number of wavebands

mat1<-matrix(NA, nrow = ncol(d1[,7:ncol(d1)]), ncol = ncol(d1[,7:ncol(d1)]))

colnames(mat1) <- colnames(d1[7:ncol(d1)])

rownames(mat1) <- colnames(d1[7:ncol(d1)])

#Create a subsample of the initial dataset with just the wavebands:

#number of column of this dataset = number of column the empty matrix

d2 = d1[,7:ncol(d1)]

#Loop to look over the wavebands

for(i in 1:ncol(d2)) {

for(j in 1:ncol(d2)){

if(i == j){

mat1[i,j] <- NA

}

else {

NDx = (d2[,i] - d2[,j]) / (d2[,i] + d2[,j])

#SRx = d2[,i] / d2[,j]

fit1 = lm(d1$Chlab~NDx)

mat1[i,j]<-summary(fit1)$r.squared

}

}

}

#export your data in csv format

write.csv(mat1,"R2-matrix-SRvsChlab.csv")

USGS earth explorer - https://earthexplorer.usgs.gov/ for Landsat and Sentinel2data

Sentinel-1 & 2 data from https://scihub.copernicus.eu/dhus/#/home

Experimental satellite data by registering particular access site by registering with institutional mail ID

You access this link:

to follow previous discussion about MESMA on RG.

For details, please acces

You must: http://di.ulb.ac.be/map/yleborgn/pub/NPL_PCA_07.pdf

I would ask ask a question in line with this conversation topic. I am looking for Land Surface Temperatures at night time with satellite data that covers Europe. Modis offers some interesting data at night with a resolution of 1km, for example water vapour. On the other hand, there are some strategies to obtain LST with Landsat 8 by previously calculating the NDVI and use it in combination with the TIR band. Do you know another strategy with which I can get the result with the greatest rigor and spatial resolution as possible? Thanks in advance.

Multispectral images have a very detailed and strict set of bands, while the Hyperspectral images have a lot of continues bands. The Pros and Cons really depends on what you are interested in observing.

Multispectral images are easier to use, because of the fewer amounts of bands. This means that if you can distinguish the features you are looking for using multispectral images, it makes more sense, since it will narrow down the amount of choices you have to make while using the bands.

Hyperspectral images however can detect other features that the multispectral can not, increasing the potential of making new discoveries, or connections not possible with the multispectral approach. If you can't detect specific features using multispectral images, you should try the hyperspectral images. Keep in mind that while the potential is larger, the options and choices are also more numerous, making it potentially more difficult and/or time-consuming to use.

You can use the below Python library :

https://pypi.org/project/psopy/

Or try to program the swarm optimization yourself... It is quite a work, but with OOP can be done in a reasonable time.

I recently did it for one of my Projects. Let me know if you would need the working python code that can be plugged with any data-set of interest.

www.cecilinstruments.com/introduction-to-uv-visible-spectrophotom...

DigitalGlobe "open Data Program" has very good high resolution images of India check it out https://www.digitalglobe.com/opendata/flooding-in-india/pre-event

And also you can download free data sets from Copernicus scihub (SAR (Sentinel-1A) and optical Sentinel-2) and usgs earthexplorer (Landsat TM, ETM and OLI).

Obviously, as per the MATLAB scaling is needed. I.e. each band image need to be scaled to 256 bins (intensity range 0-255 uint8 format), then the simple "entropy()" can be used to find the entropy. If further the problem exists, let me know, I can share the MATLAB code.

Hi Muneeb,

As very often, several architectures are possible to address your problem.

I suggest you the following publications for a quick but detailed overview:

Regards

Thanks and good luck

Dear Zhu,

Image segmentation whether it is multispectral or hyperspectral is an NP-problem. In other words there is not definite and decisive solution. Your request is to optimize the intra and inter clustering solution ( maximize homogeneity within each cluster and maximize heterogeneity among clusters). There are many ways to classify hyperspectral images such as Spectral Angle Mapper (SAM) [1], Modified Spectral Correlation Angle (SCA) [2], Optimized Spectral Angle Mapper (OSAM) [3], and Extended Spectral Angle Mapper (ESAM) [4]. These algorithms are based on angle measurement between actual spectral signatures and predicted ones. Moreover, some algorithms perform hyperspectral image classification using information measurement such as Spectral Divergence Information (SDI) [5]. Others combine two different approaches to classify hyperspectral images such as combining SAM and SDI [6] and Jeffries–Matusita with SAM (JM-SAM) [7].

Some are available in commercial image processing packages such as SAM, SDI in ENVI.

[1] F. Kruse, A. Lefkoff, J. Boardman, K. Heidebrecht, A. Shapiro, P. Barloon, and A. Goetz, “The Spectral Image Processing System (SIPS) - Interactive Visualization and Analysis of Imaging spectrometer Data.” Remote Sensing of Environment, vol. 44, p. 145 – 163, 1993.

[2]S. Robila, and A. Gershman, “Spectral Matching Accuracy in Processing Hyperspectral Data”, in Proceedings of IEEE International Symposium on Signals, Circuits and Systems, 2005, vol. 1, pp. 163–166.

[3] L. Bertels, D. Bart, K. Pieter, D. Walter, and P. Sam, “Optimized Spectral Angle Mapper Classification of Spatially Heterogeneous Dynamic Dune Vegetation, a Case Study Along the Belgian Coastline”, in Proceedings of the 9th International Symposium on Physical Measurements and Signatures in Remote Sensing (ISPMSRS), Beijing, October 17–19, 2005.

[4] H. Li, W. S. Lee, K. Wang, R. Ehsani, and C. Yang, “Extended Spectral Angle Mapping (ESAM) for Citrus Greening Disease Detection Using Airborne Hyperspectral Imaging”, Precision Agriculture, vol.15, pp. 162–183, 2014.

[5] C. I. Chang, “An Information-Theoretic Approach to Spectral Variability, Similarity, and Discrimination for Hyperspectral Image Analysis”, IEEE Transactions on Information Theory, vol. 46, pp.1927–1932, 2000.

[6] Y. Du, C. I. Chang, H. Ren, C. C. Chang, J. O. Jensen, and F. M. D’Amico, “New Hyperspectral Discrimination Measure for Spectral Characterization”, Optical Engineering, vol. 43, pp. 1777–1786, 2004.

[7] S. Padma, and S. Sanjeevi, “Jeffries Matusita Based Mixed-Measure for Improved Spectral Matching in Hyperspectral Image Analysis”, International Journal of Applied Earth Observation and Geoinformation, vol. 32, pp. 138–151, 2014.

%File path

filepath='D:\image full path';

%Read GTiff

[image,geo]=geotiffread(filepath);

%Read info

info=geotiffinfo(filepath);

imshow(filepath)

Thats ok. I extracted drone spectra from several neighboring pixels.

You need to model multi-variate regression (PLSR/PCR) between each laboratory obtained soil nutrient data of the sample points, keeping as dependent variable (Y) and spectral reflectance of the sampling points obtained from HYPERION data as independent variable (X). One usually predicts the nutrient value from the relationship between X and Y and cross-validates the predicted value through examining the precision of fit in linear regression between the actual and predicted values of nutrient.

You can learn about the MATLAB implementation of PLSR/PCR from the following link:

Hi,

I found this paper that maybe related with your research

”"Variety Identification of Single Rice Seed Using Hyperspectral Imaging Combined with Convolutional Neural Network"

There are also several papers that discuss about plant disease recognition using HSI and CNN.

"Explaining hyperspectral imaging based plant disease identification: 3D CNN and saliency maps "

Links: https://arxiv.org/pdf/1804.08831.pdf

"Classifying Wheat Hyperspectral Pixels of Healthy Heads and Fusarium Head Blight Disease Using a Deep Neural Network in the Wild Field "

Links: https://res.mdpi.com/def50200dc1e333db4fc797d476b6adff985faf79860636a7cc3541a61034885c872bbcc2ce188cdcd5a4a69b06c6186212c0e3e60365998e7ba6cd4bb76b0ca24da6b0201dcd947667f1fe5ae018c36ab81b4484681dd834ad9650a6970e256a06e83d5ca81801ac364d24ad6ef9c908ce5f71327cb89b949ad7cf73756ed6baa30053b3ff8bb1d25471d45ce194f2345f866cf7d1c08aa5fd5009f4c3fbac604826e1f65?filename=&attachment=1

Hope it helps your research.

Hello!

Yes! the careful examination/interpretation of remote sensing images by using various false color composites will help us to collect the signature of land use/land cover classes effectively.

You may refer following article for information about, how to achieve better classification accuracy.

All the best

You can download both multispectral as well as hyperspectral datasets from this link..

Here is a very good source of PCA including presentation and source code Presentation Principal Component Analysis (PCA) : An Overview Article Principal component analysis - a tutorial Alaa

dear checkout these links

regards

In addition to Farouk's post and my own post, more practical details about recovering the endmember basis may be found by jointly checking out:

Hoping it will be helpful to you.

Hi All, I tried to put all the references to the publicly available hyperspectral datasets. I hope it is of help, incase if I'm missing some, please let me know, I can update it. Thanks

best, Vivke

The permanent link to the Urban HYDICE data set is:

By the way, this data looks like it's been atmospherically corrected, but certainly has spectral mismatches. Interesting that people use it unmoving when comparing to better quality measured spectral libraries.

ENVI software has capabilities to find the end members. For this you have to use the PPI than n-Dimensional visualizer. That end members are matched the USGS resampled library and make your own spectral library for classification of image.

You can take a look at: A survey on position-based routing protocols for Flying Ad hoc Networks (FANETs) in which different restrictions about the use of UAVs have been referenced.

You can search the EO Hyperion archive and download products on: https://earthexplorer.usgs.gov/

The satellite was decommissioned in March 2017, which means that no scenes will be addeded after that. https://lta.cr.usgs.gov/ALI

I also find the scene extents of roughly 40x7 km quite unfortunate...

I am not at all performing classification. My concentration is on identifying pure pixels and finding the endmember of those pure pixels identified.

Yes, Image compression are methods to reduce the size of images, there aren't specific data for image compression. you can test your own methods on such data.

I would like to add a couple of broad comments:

1. Keep in mind that 'DN' stands for 'Digital Number' and has nothing to do with an image being calibrated or not calibrated or raw spectral band or ratio or PCA results or anything else.  EVERY image assigns a DN value to the pixels.  What is important is what the DN values represent !  If it is raw Landsat TM or drone image data it is the radiance received at the sensor and adjusted / converted to a digital number range based on the specifications of the sensor typically using gains and offset (or bais) values; if an image has radiometric calibration procedures applied then it can represent TOA radiance, TOA reflectance, surface reflectance, or what ever (even for DEM images the pixel values are DNs that in this case represent elevation).  Again, every image (no matter what they represent) has DN values and I would suggest that 'DN' not be used to imply raw image data only as it is being used in this conversion.

2. Since the early days of Landsat MSS 'broad' spectral bands have been used for digital remote sensing analyses and mapping, so I do not agree that your drone aerial images can only be used for photo interpretation.  I worked with some of the very first Landsat MSS images in early 1972 and have since then used many 'broad' band spectral image data sets for various application.

Pat

The ENVI family of image analysis software offers a full suite of tools to extract all types of vegetation-related information from any type of imagery. And, ENVI can be easily customized to solve your unique challenges, making it a highly flexible solution for agricultural and forestry applications. ENVI easily integrates into GIS workflows, allowing you to quickly and accurately view, manipulate, process, and analyze imagery. Because ENVI products are tightly integrated with ArcGIS® , you can easily exchange data and layer files between the software packages, saving you time and effort.PDF enclosed 

Assesment of ENVI Forest Health Tool in Detection of Dust and Seepage Contaminated Forest Areas

Abstract : ENVI software package is widely used in hyperspectral data analysis. ENVI features a forest health tool, which creates a spatial map showing the overall health and vigor of a forested region. General and quantitative forest health index, capable of detecting different types of forest damages, would be very useful in forest monitoring. The main objective of this study was to assess forest health tool's capability to detect dust and seepage contaminated forest areas. These areas were classified using various combinations of vegetation indices (VIs). The combination of Normalized Difference Vegetation Index, Carotenoid Reflectance Index 1 and Water Band Index produced the best classification results. Overall results of this study shows that ENVI forest health tool can provide valuable information when detecting dust and seepage contaminated forest areas.book ISBN :978-1-4244-2807-6

book e-ISBN :  978-1-4244-2808-3,DOI10.1109/IGARSS.2008.4779612

A different sampling method.

Pixels selected for training, classification model applied on all the images and 50 pixels per image were randomly selected as ground truth.

Dear Cao,

The best merging technique is Price. Please download and read my publication about merging techniques i did apply on Indian satellite data IRS-IC LISS-III.

read also my master thesis:

DOI: 10.13140/RG.2.2.20110.36166

regards

Raed

https://earthexplorer.usgs.gov/...... search here

or contact ESA for CHRIS Proba-multi-angle data

www.mathworks.com:

Syntax

[COEFF,SCORE] = princomp(X)

[COEFF,SCORE,latent] = princomp(X)

[COEFF,SCORE,latent,tsquare] = princomp(X)

[...] = princomp(X,'econ')

 

 

Description

COEFF = princomp(X) performs principal components analysis (PCA) on the n-by-p data matrix X, and returns the principal component coefficients, also known as loadings. Rows of X correspond to observations, columns to variables. COEFF is a p-by-p matrix, each column containing coefficients for one principal component. The columns are in order of decreasing component variance.

princomp centers X by subtracting off column means, but does not rescale the columns of X. To perform principal components analysis with standardized variables, that is, based on correlations, use princomp(zscore(X)). To perform principal components analysis directly on a covariance or correlation matrix, use pcacov.

[COEFF,SCORE] = princomp(X) returns SCORE, the principal component scores; that is, the representation of X in the principal component space. Rows of SCORE correspond to observations, columns to components.

[COEFF,SCORE,latent] = princomp(X) returns latent, a vector containing the eigenvalues of the covariance matrix of X.

[COEFF,SCORE,latent,tsquare] = princomp(X) returns tsquare, which contains Hotelling's T2 statistic for each data point.

The scores are the data formed by transforming the original data into the space of the principal components. The values of the vector latent are the variance of the columns of SCORE. Hotelling's T2 is a measure of the multivariate distance of each observation from the center of the data set.

When n <= p, SCORE(:,n:p) and latent(n:p) are necessarily zero, and the columns of COEFF(:,n:p) define directions that are orthogonal to X.

[...] = princomp(X,'econ') returns only the elements of latent that are not necessarily zero, and the corresponding columns of COEFF and SCORE, that is, when n <= p, only the first n-1. This can be significantly faster when p is much larger than n.

 

Examples

Compute principal components for the ingredients data in the Hald data set, and the variance accounted for by each component.

load hald;

[pc,score,latent,tsquare] = princomp(ingredients);

pc,latent

pc =

-0.0678 -0.6460 0.5673 0.5062

-0.6785 -0.0200 -0.5440 0.4933

0.0290 0.7553 0.4036 0.5156

0.7309 -0.1085 -0.4684 0.4844

latent =

517.7969

67.4964

12.4054

0.2372

The following command and plot show that two components account for 98% of the variance:

cumsum(latent)./sum(latent)

ans =

0.86597

0.97886

0.9996

1

biplot(pc(:,1:2),'Scores',score(:,1:2),'VarLabels',...

{'X1' 'X2' 'X3' 'X4'})

What is the source and the current level (L1 or L2 or L3) of the AVIRIS data that you are using? Do you have any other information in the metadata regarding the units? Is it in raw DN values or Radiance or Reflectance? Depending on the level of the data the units would differ and therefore the min-max data value range. For example, the Surface Reflectance product typically has data values ranging from 0 to 1 (unitless). Attached links might be helpful.

Thanks Nasem and Abdegdir for the reply. MRT process works fine with MOD09 and MOD13. However it couldn't process MOD02 which is a Level 2 product that lacks sinusoidal coordinates (V##h##). I just did  research and found out that the recommended software to process MOD02 is MRTSwath. I want to know the steps to derived Band 1 for daily measurement using MRTSwath Tool. I want to get final values in text file or other format as long as it appears in numbers for calculation purpose.

The observed color at any point on an object is influenced by several factors like  the shape and material of the object, the positions and colors of the light sources, and the position of the viewer.

Ground truth is a common terminology that is widely used in various fields to basically refer to any kind of information provided by direct observation.

In Remote Sensing, Ground truth is the information or data collected on site so that the input data (image) can be related to the actual features and is considered to be much correct and appropriate than the input features. This process compares the pixel on a satellite image to what is there in reality (at the present time) in order to verify the contents of the pixel on the image. Further, Error of Commission and Omission can be calculated for accuracy assessment.

Kindly refer to the paper attached where the researchers has tried creating the ground truth dataset for image algorithms (Ground Truth1.pdf) and the second pdf is a chapter on the same.

Hope it will help u in some way.

Regards,

Rebika Rai

Hi Ayman Hassan 

You can also try The Semi-Automatic Classification Plugin (SCP) for QGIS made by Luca Congedo. This is the link where you can have an overview of it: 

https://fromgistors.blogspot.com/search/label/Tutorial.

It's very nice to use and can help you  to download individual band of Landsat 8/7/4-5/1-5 scenes separately.

There are many factors affecting the accuracy of VNIR hyperspectral imaging. 

First is your samples and the way you do the reference analysis. Let's suppose you are working on kernels to measure for example N content. How many kernels do you use for measuring N content using your ref method? If 5 kernels, then you should use the average spectra of all the 5 kernels to correlate to the reference N (There are more approaches but I think this is the best for linear models). 

Second, your camera properties. I mean resolution and spectral range of your camera? Smaller resolution and wider range are more favourable.

Third, outlier detection and data pre-processing are so important. Dimension reduction and/ band selection also improve the accuracy of the models. 

Fourth, the model you use for calibration. Is that a linear model or a non-linear? Most of the times linear models fit well to your data but sometimes you need to use nonlinear models.  

The number of the samples you are using for calibration is also important more specifically when you are using machine learning models such as ANN.

Once you develop the model, then you can use any pixel for prediction in new samples. Even you can use this system Real time. 

Hope I could add anything to what you already know. 

Hi Tabia,

Common approaches ( https://aviris.jpl.nasa.gov/proceedings/workshops/90_docs/17.PDF , https://aviris.jpl.nasa.gov/proceedings/workshops/95_docs/12.PDF ) to generate hyperspectral ground truthes from spectral libraries are not really accurate (check for http://proceedings.spiedigitallibrary.org/proceeding.aspx?articleid=2571488 ).

As an alternative in a segmentation context, I suggest you look at this paper: http://www5.informatik.uni-erlangen.de/Forschung/Publikationen/2013/Jordan13-MCF.pdf

Here is a excerpt from this paper: "We test our algorithms on the CAVE multispectral image database [8]...  Ground-truth data for evaluation of a global clustering is not available and would be very difficult to obtain. Therefore we mostly rely on qualitative evaluation of the segmentation results."

Hoping it will be helpful to you.

Regards

If you want to generate method  or algorithm for cloud removal, you may learn about atmospheric window and its interaction with electromagnetic wave. Then, check and learn about EM range of the hyperspectral satelit sensor that used.

Hi Haris,

Another reference to corroborate the diversity of noises to take into account when simulating a multispectral / hyperspectral sensor :

http://color.psych.upenn.edu/simchapter/simchapter.pdf (see section §3)

The accuracy of the model will depend on your needs. If your interest focuses on optical lenses, here is a good reference:

Although tree species level identification has always been challenging if your are working at 10 m spatial resolution.

1. So Before going for individual tree species level analysis using Sentinel-2. You must go for some preliminary analysis using Sentinel-2 data. Sentinel -2 provides 10 m spatial resolution for RGB and NIR bands while it provides 20 m spatial resolution for band 5, 6, 7 and 8A in Vegetation RedEdge category. 

2. Since RedEdge bands has capability to discriminate different tree species based upon their spectral variability. Therefore, Sentinel-2 can be a powerful opensource platform for these kind of analysis. 

3. If you do re-sampling of RedEdge bands to 10 m and create the Layer Stack of Red, Green, Blue, NIR and RedEdge Bands and then you can check the spectral signature of different trees species (it could be some homogeneous regions) for different band combinations. 

4. For this you must have some ground data. If you have a then you can cross check these spectral signatures and then you can go ahead for either object based or pixel based supervised classification procedures. 

Dear Mohamed Atef Mosa,

I suggest to you a links and attached files in topics.

- STK - Analytical Graphics, Inc.

-The Nog » Satellite Tool Kit - AGI Blog

-From lynn.wheeler at firstdata.com Tue Jan 1 12:59:28 2002 From ...

Best regards

In near IR the reflection of clear water  is low compared to other land surface features. However in turbid water it may show a higher value.

why do'tn u use matlab or python?

Hi, you can also use: https://www.researchgate.net/publication/293190248_EnGeoMAP_20-Automated_Hyperspectral_Mineral_Identification_for_the_German_EnMAP_Space_Mission

The software is open source and freely available from www.enmap.org . It is integrated in the EnMAP Box software.

Cheers, Chris

Hi,

if you consider not geo-corrected radiance data we can give this approach http://dx.doi.org/10.3390/rs61111082 (most pushbroom sensors are affected by temporal variations of the dark current and related miscalibration) or this approach http://dx.doi.org/10.3390/s110606370 a try else there are some smoothing techniques as Atilio said.

Cheers, Chris

Thank you so much Mam. I will read the paper which you have shared.

Hi Nasser

I don't know of any studies that demonstrate the use of multispectral imagery from UAVs to map iron oxides, in contrast to satellite studies... The extreme difference in spatial extent and resolution between satellite and UAV data might mean that satellite methodologies for detecting iron oxides might not work for UAV data, even with the same band combinations, due to small scale noise and other confounding factors in the UAV imagery...

That said, the iron indices that I know of make use of a narrow blue band, which the Parrot Sequoia lacks, but is available on e.g. the 5-band MicaSense RedEdge multispectral camera for UAVs (and other 4-band cameras that allow for custom band placement as opposed to the Sequoia).

Based on Landsat TM data, the Iron Oxide Ratio = Red / Blue was developed, both of which correspond reasonably well with the MicaSense RedEdge red and blue bands. Based on WorldView-2, the WV New Iron Index = (Green * Yellow) / (Blue * 1000) was developed, for which you need again the narrow blue band, while the yellow could be conceived as a (properly weighted) average of the green and red bands when using the MicaSense RedEdge camera. Last, there's the Iron Feature Depth (IFD), measured as the depth of the 900 nm absorption peak of secondary iron bearing minerals, from the line between the 750 and 1250 nm shoulders. However, I don't think that either the Sequoia or the RedEdge cameras would be suited for this as it requires a SWIR band, and both cameras have red edge and NIR bands placed somewhat too low for the other features in the IFD calculation (the Sequoia even much worse than the MicaSense RedEdge in this respect).

Hope this helps

Klaas 

I'll second what Atiqa Khan says. Vegetation indices are band ratios which by their very nature enhance the information content of imagery.  Just looking at an NDVI image you already know which areas are likely vegetated. That's what is meant by spectral enhancement. Because of this  enhancement aspect, you can classify a stand alone NDVI image by determining thresholds or you can add it as an extra input band when classifying multispectral. 

Hyperspectral data have been used to detect sub-soil features in various disciplines. The MIVIS sensor was used in Archaeology:

Traviglia, A. (2006). MIVIS Hyperspectral sensor for detection of subsoil archaeological sites and interpretation support GIS: an Italian case study. Digital Discovery. Exploring New Frontiers in Human Heritage, 18-26.

ARPA has a project going on mine detection using hyperspectral data:

DePersia, A. T., Bowman, A. P., Lucey, P. G., & Winter, E. M. (1995, June). Phenomenology considerations for hyperspectral mine detection. In SPIE's 1995 Symposium on OE/Aerospace Sensing and Dual Use Photonics (pp. 159-167). International Society for Optics and Photonics.

Hyperspectral data was part of the input to this study on topsoil mapping:

Selige, T., Böhner, J., & Schmidhalter, U. (2006). High resolution topsoil mapping using hyperspectral image and field data in multivariate regression modeling procedures. Geoderma, 136(1), 235-244.

In this article are discussed methods to study and monitor soils with hyperspectral data:

Viscarra Rossel, R. A., Taylor, H. J., & McBratney, A. B. (2007). Multivariate calibration of hyperspectral γ‐ray energy spectra for proximal soil sensing. European Journal of Soil Science, 58(1), 343-353.

Did you solve your problem meanwhile? Extended Morphological Profiles in Hyperspectral imaging means that the datasets available are more precise because of the integration with other datasets. For coding application check Github or this book might be helpful, although is a little outdated http://isp.uv.es/papers/rem_sens_im_proc_12_ch02.pdf

Hi,

The following web link might be helpful for you.

hai first remove bad bands and water vapour 1-7 58-76 165-182 221-242 you have 163 bands around use radiometric calibration tool convert DN to radiance. after that use FLAASH or QUAC my friend.scale factor 0.1 for radiance conversion

In any case contiguous signatures, only can be derived from spectroradiometer handheld/satellite based. If you have handheld spectroradiometer data it is nice, but if you don't have, you can generate signature for a pure patch of single species of size about/more than 30*30 meters using Hyperion based hyperspectral data freely available on earth explorer. The patch size may vary depending upon spatial resolution of the hyperspectral data (space borne/ air borne).