Thesis

High Performance Hyperspectral Image Classification using Graphics Processing Units. (https://arxiv.org/abs/2106.12942v1)

Authors:
To read the full-text of this research, you can request a copy directly from the author.

Abstract

Real-time remote sensing applications like search and rescue missions, military target detection, environmental monitoring, hazard prevention and other time-critical applications require onboard real time processing capabilities or autonomous decision making. Some unmanned remote systems like satellites are physically remote from their operators, and all control of the spacecraft and data returned by the spacecraft must be transmitted over a wireless radio link. This link may not be available for extended periods when the satellite is out of line of sight of its ground station. Therefore, lightweight, small size and low power consumption hardware is essential for onboard real time processing systems. With increasing dimensionality, size and resolution of recent hyperspectral imaging sensors, additional challenges are posed upon remote sensing processing systems and more capable computing architectures are needed. Graphical Processing Units (GPUs) emerged as promising architecture for light weight high performance computing that can address these computational requirements for onboard systems. The goal of this study is to build high performance methods for onboard hyperspectral analysis. We propose accelerated methods for the well-known recursive hierarchical segmentation (RHSEG) clustering method, using GPUs, hybrid multicore CPU with a GPU and hybrid multi-core CPU/GPU clusters. RHSEG is a method developed by the National Aeronautics and Space Administration (NASA), which is designed to provide rich classification information with several output levels. The achieved speedups by parallel solutions compared to CPU sequential implementations are 21x for parallel single GPU and 240x for hybrid multi-node computer clusters with 16 computing nodes. The energy consumption is reduced to 74% using a single GPU compared to the equivalent parallel CPU cluster. Please cite as: "Hossam, Mahmoud. High Performance Hyperspectral Image Classification Using Graphics Processing Units. May 2021. arxiv.org, https://arxiv.org/abs/2106.12942v1."

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the author.

... With the permanent enhancement of existing and the development of new sensors (Hossam 2015), a wide range of datasets with different potentials became available. This situation manifests itself in a constant development of fusion methods on the one hand and in a broad field of applications on the other hand. ...
Full-text available
Thesis
Increasing awareness is dedicated to environmental impacts of mining activities around the world. In this context, Environmental Hazard Potential (EHP) is considered to be an appropriate way to assess all environmental impacts related to mining activities. A holistic EHP requires detailed knowledge about the spatial extent of the mining area. Even though several studies have already been conducted in the domain of mine detection and mapping, no comparative study has yet been carried out that investigates several remote sensing analyses, which are transferable to other geographic regions. The aim of this study is thus to compare remote sensing analyses that can be applied in order to determine the area that is subject to open-pit mining in different geographic regions. Therefore, this study examines strengths and weaknesses of remote sensing analyses, among them index-based, pixel-based and object-based multi-spectral classifications on single- and multi-source level, as well as crowdsourcing. Data sets comprise freely available Sentinel-2 optical imagery, Aster GDEM V2 elevation model and Sentinel-1 synthetic aperture radar imagery. Four copper or iron ore open-pit mines in Indonesia (Grasberg-Ertsberg Gold Copper Mine), Australia (Hamersley Iron Ore Mines), Canada (Highland Valley Copper Mine) and Brazil (Mariana Iron Ore Complex) constitute the study sites. Index-based, pixel-based and object-based classifications are applied on the datasets for each study site, whereby index-based classifications are conducted on single-source level, pixel-based and object-based classifications on multi-source level. Simultaneously, a crowdsourcing project is launched, where volunteers are asked to digitize the delineation of the four open-pit mines. First, classifications and crowdsourcing are investigated individually by visual interpretation, accuracy assessment and area computation. Secondly, both methods are compared by Intersection over Union (IoU), by area values, accuracy values and visual interpretation. Acquired new findings regarding the implementation of the methods and the achieved results support the final derivation of strengths and weaknesses of classifications and crowdsourcing. Classifications and crowdsourcing can both be applied in order to detect, classify and digitize open-pit mines in different geographic regions with an overall accuracy ≥ 77.41 % and to compute their spatial extent. Overall accuracy ranges for all methods from 77.41 % up to 97.73 %. The comparison of these methods reveals that classification and crowdsourcing results are not congruent, indicated by a mean IoU of 0.49 for all conducted comparisons. Classifications and crowdsourcing results differ among their respective area values, accuracy values and visual impression. Regarding area and accuracy values, crowdsourcing results have an intermediate position between the three considered classifications. Final derivation of strengths and weaknesses, as well as opportunities and threats shows that classifications and crowdsourcing differ further regarding effort, transferability, completeness, implementation, quality and credibility as well as their potential for automatization and further development. This study strongly supports decision making regarding method selection by providing strengths and weaknesses of remote sensing analyses for mine area computation. It contributes thus to the development of a holistic EHP of open-pit mines in different geographic regions. Future research recommendations are primarily related to the detection of unknown mines with classification approaches, to the development of a crowdsourcing project for global mine mapping and to the investigation of the potential of a combined application of classifications and crowdsourcing.
Full-text available
Conference Paper
Watershed transformation based segmentation which is a segmentation based on marker is a special tool used in image processing. Color based image segmentation has been considered an important area since its inception, due to its wide variety of applications in the field of weather forecasting to medical image analysis etc. Due to this color image segmentation is widely researched. This paper analyses the performance of two main algorithms used for image segmentation namely Watershed algorithm and graph based image segmentation. The performance analysis proves that graph based segmentation is better than watershed algorithm in cases where noise is maximum and also the over segmentation problem is removed. Color segmentation with graph based image segmentation gives satisfactory results unlike watershed algorithm.
Full-text available
Conference Paper
The amount of scientific literature on (Geographic) Object-based Image Analysis - GEOBIA or OBIA in short - has been and still is sharply increasing. The term OBIA is herein used for generic image processing tasks. OBIA researchers with backgrounds in geology, geography or mineralogy are joining forces with colleagues from cell biology, molecular biology, medicine, pharmacy or nanotechnology. More narrowly, GEOBIA's objective is the generation of geographic information (in GIS-ready format), from which new spatial knowledge can be obtained. GEOBIA is a relatively new field of research based upon two major concepts: A) dissecting images or any (pseudo- )continuous multidimensional fields of data, and b) allowing for multiple scales when organizing and utilizing the resulting objects. Analyses of the amount of scientific literature reveal that not only the number of articles is increasing, but that the rate of growth is also noticeably accelerating. Finally, it is discussed whether or not GEOBIA can be considered as an important paradigm in remote sensing. Copyright © (2013) by the American Society for Photogrammetry & Remote Sensing.
Full-text available
Article
Hyperspectral remote sensing technology allows one to acquire a sequence of possibly hundreds of contiguous spectral images from ultraviolet to infrared. Conventional spectral classifiers treat hyperspectral images as a list of spectral measurements and do not consider spatial dependences, which leads to a dramatic decrease in classification accuracies. In this paper, a new automatic framework for the classification of hyperspectral images is proposed. The new method is based on combining hidden Markov random field segmentation with support vector machine (SVM) classifier. In order to preserve edges in the final classification map, a gradient step is taken into account. Experiments confirm that the new spectral and spatial classification approach is able to improve results significantly in terms of classification accuracies compared to the standard SVM method and also outperforms other studied methods.
Full-text available
Conference Paper
Hierarchical image segmentation is a well-known image analysis and clustering method that is used for hyperspectral image analysis. This paper introduces a parallel implementation of hybrid CPU/GPU for the Recursive Hierarchical Segmentation method (RHSEG) algorithm, in which CPU and GPU work cooperatively and seamlessly, combining benefits of both platforms. RHSEG is a method developed by National Aeronautics and Space Administration (NASA) which is more efficient than other traditional methods for high spatial resolution images. The RHSEG algorithm is also implemented on both GPU cluster and hybrid CPU/GPU cluster and the results are compared with the hybrid CPU/GPU implementation. For single hybrid computational node of 8 cores, a speedup of 6x is achieved using both CPU and GPU. On a computer cluster of 16 hybrid CPU/GPU nodes, an average speed up of 112x times is achieved over the sequential CPU implementation.
Full-text available
Article
Hyperspectral images show similar statistical properties to natural grayscale or color photographic images. However, the classification of hyperspectral images is more challenging because of the very high dimensionality of the pixels and the small number of labeled examples typically available for learning. These peculiarities lead to particular signal processing problems, mainly characterized by indetermination and complex manifolds. The framework of statistical learning has gained popularity in the last decade. New methods have been presented to account for the spatial homogeneity of images, to include user's interaction via active learning, to take advantage of the manifold structure with semisupervised learning, to extract and encode invariances, or to adapt classifiers and image representations to unseen yet similar scenes. This tutuorial reviews the main advances for hyperspectral remote sensing image classification through illustrative examples.
Full-text available
Technical Report
The recent switch to parallel microprocessors is a milestone in the history of computing. Industry has laid out a roadmap for multicore designs that preserves the programming paradigm of the past via binary compatibility and cache coherence. Conventional wisdom is now to double the number of cores on a chip with each silicon generation. A multidisciplinary group of Berkeley researchers met nearly two years to discuss this change. Our view is that this evolutionary approach to parallel hardware and software may work from 2 or 8 processor systems, but is likely to face diminishing returns as 16 and 32 processor systems are realized, just as returns fell with greater instruction-level parallelism. We believe that much can be learned by examining the success of parallelism at the extremes of the computing spectrum, namely embedded computing and high performance computing. This led us to frame the parallel landscape with seven questions, and to recommend the following: The overarching goal should be to make it easy to write programs that execute efficiently on highly parallel computing systems The target should be 1000s of cores per chip, as these chips are built from processing elements that are the most efficient in MIPS (Million Instructions per Second) per watt, MIPS per area of silicon, and MIPS per development dollar. Instead of traditional benchmarks, use 13 "Dwarfs" to design and evaluate parallel programming models and architectures. (A dwarf is an algorithmic method that captures a pattern of computation and communication.) "Autotuners" should play a larger role than conventional compilers in translating parallel programs. To maximize programmer productivity, future programming models must be more human-centric than the conventional focus on hardware or applications. To be successful, programming models should be independent of the number of processors. To maximize application efficiency, programming models should support a wide range of data types and successful models of parallelism: task-level parallelism, word-level parallelism, and bit-level parallelism. Architects should not include features that significantly affect performance or energy if programmers cannot accurately measure their impact via performance counters and energy counters. Traditional operating systems will be deconstructed and operating system functionality will be orchestrated using libraries and virtual machines. To explore the design space rapidly, use system emulators based on Field Programmable Gate Arrays (FPGAs) that are highly scalable and low cost. Since real world applications are naturally parallel and hardware is naturally parallel, what we need is a programming model, system software, and a supporting architecture that are naturally parallel. Researchers have the rare opportunity to re-invent these cornerstones of computing, provided they simplify the efficient programming of highly parallel systems.
Full-text available
Conference Paper
A major drawback when using the watershed transformation as a segmentation tool comes from the over-segmentation of the image. Over-segmentation is produced by the great number of minima embedded in the image or in its gradient. A powerful technique has been designed to suppress over-segmentation by a primary selection of markers pointing out the regions or objects to be segmented in the image. However, this approach can be used only if we are able to compute the marker set before applying the watershed transformation. But, in many cases and especially for complex scenes, this is not possible and an alternative technique must be used to reduce the over-segmentation. This technique is based on mosaic images and on the computation of a watershed transform on a valued graph derived from the mosaic images. This approach leads to a hierarchical segmentation of the image and considerably reduces over-segmentation. Then, this hierarchical segmentation is redefined by means of a new algorithm called the waterfall algorithm. This algorithm allows the selection of minima and of catchment basins of higher significance compared to their neighborhood. A very powerful implementation of this algorithm using geodesic reconstruction of functions is also presented. Finally, this approach is compared to another powerful tool introduced by M. Grimaud for selecting significant extrema in an image: the dynamics.
Full-text available
Article
An adaptive Markov random field (MRF) approach is proposed for classification of hyperspectral imagery in this letter. The main feature of the proposed method is the introduction of a relative homogeneity index for each pixel and the use of this index to determine an appropriate weighting coefficient for the spatial contribution in the MRF classification. In this way, overcorrection of spatially high variation areas can be avoided. Support vector machines are implemented for improved class modeling and better estimate of spectral contribution to this approach. Experimental results of a synthetic hyperspectral data set and a real hyperspectral image demonstrate that the proposed method works better on both homogeneous regions and class boundaries with improved classification accuracy.
Full-text available
Article
Advances in sensor and computer technology are revolutionizing the way remotely sensed data is collected, managed and analyzed. In particular, many current and future applications of remote sensing in Earth science, space science, and soon in exploration science will require real- or near real-time processing capabilities. In recent years, several efforts have been directed towards the incorporation of high-performance computing (HPC) models to remote sensing missions. A relevant example of a remote sensing application in which the use of HPC technologies (such as parallel and distributed computing) is becoming essential is hyperspectral remote sensing, in which an imaging spectrometer collects hundreds or even thousands of measurements (at multiple wavelength channels) for the same area on the surface of the Earth. In this paper, we review recent developments in the application of HPC techniques to hyperspectral imaging problems, with particular emphasis on commodity architectures such as clusters, heterogeneous networks of computers, and specialized hardware devices such as field programmable gate arrays (FPGAs) and commodity graphic processing units (GPUs). A quantitative comparison across these architectures is given by analyzing performance results of different parallel implementations of the same hyperspectral unmixing chain, delivering a snapshot of the state-of-the-art in this area and a thoughtful perspective on the potential and emerging challenges of applying HPC paradigms to hyperspectral remote sensing problems.
Full-text available
Article
Imaging spectrometers measure electromagnetic energy scattered in their instantaneous field view in hundreds or thousands of spectral channels with higher spectral resolution than multispectral cameras. Imaging spectrometers are therefore often referred to as hyperspectral cameras (HSCs). Higher spectral resolution enables material identification via spectroscopic analysis, which facilitates countless applications that require identifying materials in scenarios unsuitable for classical spectroscopic analysis. Due to low spatial resolution of HSCs, microscopic material mixing, and multiple scattering, spectra measured by HSCs are mixtures of spectra of materials in a scene. Thus, accurate estimation requires unmixing. Pixels are assumed to be mixtures of a few materials, called endmembers. Unmixing involves estimating all or some of: the number of endmembers, their spectral signatures, and their abundances at each pixel. Unmixing is a challenging, ill-posed inverse problem because of model inaccuracies, observation noise, environmental conditions, endmember variability, and data set size. Researchers have devised and investigated many models searching for robust, stable, tractable, and accurate unmixing algorithms. This paper presents an overview of unmixing methods from the time of Keshava and Mustard's unmixing tutorial [1] to the present. Mixing models are first discussed. Signal-subspace, geometrical, statistical, sparsity-based, and spatial-contextual unmixing algorithms are described. Mathematical problems and potential solutions are described. Algorithm characteristics are illustrated experimentally.
Full-text available
Conference Paper
In the past, graphics processors were special purpose hardwired application accelerators, suitable only for conventional rasterization-style graphics applications. Modern GPUs are now fully programmable, massively parallel floating point processors. This talk will describe NVIDIA's massively multithreaded computing architecture and CUDA software for GPU computing. The architecture is a scalable, highly parallel architecture that delivers high throughput for data-intensive processing. Although not truly general-purpose processors, GPUs can now be used for a wide variety of compute-intensive applications beyond graphic.
Full-text available
Article
Remote sensing imagery needs to be converted into tangible information which can be utilised in conjunction with other data sets, often within widely used Geographic Information Systems (GIS). As long as pixel sizes remained typically coarser than, or at the best, similar in size to the objects of interest, emphasis was placed on per-pixel analysis, or even sub-pixel analysis for this conversion, but with increasing spatial resolutions alternative paths have been followed, aimed at deriving objects that are made up of several pixels. This paper gives an overview of the development of object based methods, which aim to delineate readily usable objects from imagery while at the same time combining image processing and GIS functionalities in order to utilize spectral and contextual information in an integrative way. The most common approach used for building objects is image segmentation, which dates back to the 1970s. Around the year 2000 GIS and image processing started to grow together rapidly through object based image analysis (OBIA - or GEOBIA for geospatial object based image analysis). In contrast to typical Landsat resolutions, high resolution images support several scales within their images. Through a comprehensive literature review several thousand abstracts have been screened, and more than 820 OBIA-related articles comprising 145 journal papers, 84 book chapters and nearly 600 conference papers, are analysed in detail. It becomes evident that the first years of the OBIA/GEOBIA developments were characterised by the dominance of 'grey' literature, but that the number of peer-reviewed journal articles has increased sharply over the last four to five years. The pixel paradigm is beginning to show cracks and the OBIA methods are making considerable progress towards a spatially explicit information extraction workflow, such as is required for spatial planning as well as for many monitoring programmes.
Full-text available
Article
Image classification is a complex process that may be affected by many factors. This paper examines current practices, problems, and prospects of image classification. The emphasis is placed on the summarization of major advanced classification approaches and the techniques used for improving classification accuracy. In addition, some important issues affecting classification performance are discussed. This literature review suggests that designing a suitable image-processing procedure is a prerequisite for a successful classification of remotely sensed data into a thematic map. Effective use of multiple features of remotely sensed data and the selection of a suitable classification method are especially significant for improving classification accuracy. Non-parametric classifiers such as neural network, decision tree classifier, and knowledge-based classification have increasingly become important approaches for multisource data classification. Integration of remote sensing, geographical information systems (GIS), and expert system emerges as a new research frontier. More research, however, is needed to identify and reduce uncertainties in the image-processing chain to improve classification accuracy.
Full-text available
Article
Very high resolution hyperspectral data should be very useful to provide detailed maps of urban land cover. In order to provide such maps, both accurate and precise classification tools need, however, to be developed. In this letter, new methods for classification of hyperspectral remote sensing data are investigated, with the primary focus on multiple classifications and spatial analysis to improve mapping accuracy in urban areas. In particular, we compare spatial reclassification and mathematical morphology approaches. We show results for classification of DAIS data over the town of Pavia, in northern Italy. Classification maps of two test areas are given, and the overall and individual class accuracies are analyzed with respect to the parameters of the proposed classification procedures.
Full-text available
Article
Spectral unmixing using hyperspectral data represents a significant step in the evolution of remote decompositional analysis that began with multispectral sensing. It is a consequence of collecting data in greater and greater quantities and the desire to extract more detailed information about the material composition of surfaces. Linear mixing is the key assumption that has permitted well-known algorithms to be adapted to the unmixing problem. In fact, the resemblance of the linear mixing model to system models in other areas has permitted a significant legacy of algorithms from a wide range of applications to be adapted to unmixing. However, it is still unclear whether the assumption of linearity is sufficient to model the mixing process in every application of interest. It is clear, however, that the applicability of models and techniques is highly dependent on the variety of circumstances and factors that give rise to mixed pixels. The outputs of spectral unmixing, endmember, and abundance estimates are important for identifying the material composition of mixtures
Article
This work presents a segmentation method for multidimensional images, therefore it is valid for standard Red, Green, Blue (RGB) images, multi-spectral images or hyperspectral images. On the one hand it is based in a tuned version of watershed transform, and on the other hand it is based on a chromatic gradient that is made through Hyperspherical Coordinates. A remarkable feature of this algorithm is its robustness; it outperforms the natural oversegmentation induced by the standard watershed. Another important property of this algorithm is its robustness respect changes on the intensity: shines and shadows. Inspired on the Human Vision System (HVS) this algorithm provides segmentations according with the user expectations, where homogeneous chromatic regions of an image corespond with homogeneous convex regions of the output.
Conference Paper
ISODATA is a well-known clustering algorithm used in various areas. It employs a heuristic strategy allowing the clusters based on the nearest neighbor rule to split and merge as appropriate. However, since the volume of the data to be clustered in real world is growing continuously, the efficiency of serial ISODATA has become a serious practical issue. The GPU (Graphics Processing Unit) is an emerging high performance computing platform due to its highly parallel multithreaded architecture. In this paper, we propose an efficient parallel ISODATA algorithm based on the latest Kepler GPUs and the dynamic parallelism feature in CUDA (Compute Unified Device Architecture). Performance study shows that our parallel ISODATA can achieve promising speedup ratios and features favorable scalability compared to the original algorithm.
Chapter
The material in many of the previous chapters has presented techniques that can be used to classify an image and create a thematic map. What we have to do now is to see how those techniques can be put into methodologies that make the classification task efficient and accurate. Using an appropriate methodology is probably one of the most important considerations in applying classification to data recorded by imaging sensors, and is often overlooked in practical image labelling. It is the role of this chapter to consider these operational aspects of thematic mapping, including how the accuracy of the final thematic map product can be assessed. We will consider accuracy determination in detail, including guidelines for choosing the number of samples needed to measure map accuracy effectively.
Chapter
Image analysis by photointerpretation is often made easier when the radiometric nature of an image is enhanced to improve its visual characteristics. Specific differences in vegetation and soil type, for example, may be brought out by increasing the contrast of an image. Highlighting subtle differences in brightness value by applying contrast modification, or by assigning different colours to different brightnesses in the method known as colour density slicing, will often reveal features not otherwise easily seen.
Article
A Markov random field is a graphical model that is commonly used to combine spectral information and spatial context into image classification problems. The contributions of the spatial versus spectral energies are typically defined by using a smoothing parameter, which is often set empirically. We propose a new framework to estimate the smoothing parameter. For this purpose, we introduce the new concepts of dynamic blocks and class label cooccurrence matrices. The estimation is then based on the analysis of the balance of spatial and spectral energies computed using the spatial class co-occurrence distribution and dynamic blocks. Moreover, we construct a new spatially weighted parameter to preserve the edges, based on the Canny edge detector. We evaluate the performance of the proposed method on three data sets: a multispectral DigitalGlobe WorldView-2 and two hyperspectral images, recorded by the AVIRIS and the ROSIS sensors, respectively. The experimental results show that the proposed method succeeds in estimating the optimal smoothing parameter and yields higher classification accuracy values when compared with state-of-the-art methods.
Article
This letter presents a Bayesian method for hyperspectral image classification based on the sparse representation (SR) of spectral information and the Markov random field modeling of spatial information. We introduce a probabilistic SR approach to estimate the class conditional distribution, which proved to be a powerful feature extraction technique to be combined with the label prior distribution in a Bayesian framework. The resulting maximum a priori problem is estimated by a graph-cut-based α-expansion technique. The capabilities of the proposed method are proven in several benchmark hyperspectral images of both agricultural and urban areas.
Article
Recent advances in spectral-spatial classification of hyperspectral images are presented in this paper. Several techniques are investigated for combining both spatial and spectral information. Spatial information is extracted at the object (set of pixels) level rather than at the conventional pixel level. Mathematical morphology is first used to derive the morphological profile of the image, which includes characteristics about the size, orientation, and contrast of the spatial structures present in the image. Then, the morphological neighborhood is defined and used to derive additional features for classification. Classification is performed with support vector machines (SVMs) using the available spectral information and the extracted spatial information. Spatial postprocessing is next investigated to build more homogeneous and spatially consistent thematic maps. To that end, three presegmentation techniques are applied to define regions that are used to regularize the preliminary pixel-wise thematic map. Finally, a multiple-classifier (MC) system is defined to produce relevant markers that are exploited to segment the hyperspectral image with the minimum spanning forest algorithm. Experimental results conducted on three real hyperspectral images with different spatial and spectral resolutions and corresponding to various contexts are presented. They highlight the importance of spectral-spatial strategies for the accurate classification of hyperspectral images and validate the proposed methods.
Article
Accelerators have become critical in the process to develop supercomputers with exascale computing capability. In this work, we examine the potential of two latest acceleration technologies, Nvidia K20 Kepler GPU and Intel Many Integrated Core (MIC) Architecture, for accelerating geospatial applications. We first apply a set of benchmarks under 3 different configurations, i.e, MPI+CPU, MPI+GPU, and MPI+MIC. This set of benchmarks include embarrassingly parallel application, loosely communicating application, and intensely communicating application. It is found that the straightforward MPI implementation on MIC cores can achieve the same amount of performance speedup as hybrid MPI+GPU implementation when the same number of processors are used. Further, we demonstrate the potentials of hardware accelerators for advancing the scientific research using an urban sprawl simulation application. The parallel implementation of the urban sprawl simulation using 16 Tesla M2090 GPUs can realize a 155× speedup compared with the single-node implementation, while achieving a good strong scalability.
Article
In this paper, we propose a new framework for spectral-spatial classification of hyperspectral image data. The proposed approach serves as an engine in the context of which active learning algorithms can exploit both spatial and spectral information simultaneously. An important contribution of our paper is the fact that we exploit the marginal probability distribution which uses the whole information in the hyperspectral data. We learn such distributions from both the spectral and spatial information contained in the original hyperspectral data using loopy belief propagation. The adopted probabilistic model is a discriminative random field in which the association potential is a multinomial logistic regression classifier and the interaction potential is a Markov random field multilevel logistic prior. Our experimental results with hyperspectral data sets collected using the National Aeronautics and Space Administration's Airborne Visible Infrared Imaging Spectrometer and the Reflective Optics System Imaging Spectrometer system indicate that the proposed framework provides state-of-the-art performance when compared to other similar developments.
Book
Remote Sensing Digital Image Analysis provides the non-specialist with an introduction to quantitative evaluation of satellite and aircraft derived remotely retrieved data. Each chapter covers the pros and cons of digital remotely sensed data, without detailed mathematical treatment of computer based algorithms, but in a manner conductive to an understanding of their capabilities and limitations. Problems conclude each chapter. This fourth edition has been developed to reflect the changes that have occurred in this area over the past several years.
Article
The rapid increase in the performance of graphics hardware, coupled with recent improvements in its programmability, have made graphics hardware a compelling platform for computationally demanding tasks in a wide variety of application domains. In this report, we describe, summarize, and analyze the latest research in mapping general-purpose computation to graphics hardware.We begin with the technical motivations that underlie general-purpose computation on graphics processors (GPGPU) and describe the hardware and software developments that have led to the recent interest in this field. We then aim the main body of this report at two separate audiences. First, we describe the techniques used in mapping general-purpose computation to graphics hardware. We believe these techniques will be generally useful for researchers who plan to develop the next generation of GPGPU algorithms and techniques. Second, we survey and categorize the latest developments in general-purpose application development on graphics hardware.
Chapter
Hyperspectral image processing has been a very active area in remote sensing and other application domains in recent years. Despite the availability of a wide range of advanced processing techniques for hyperspectral data analysis, a great majority of available techniques for this purpose are based on the consideration of spectral information separately from spatial information information, and thus the two types of information are not treated simultaneously. In this chapter, we describe several spatial-spectral techniques for dimensionality reduction, feature extraction, unsupervised and supervised classification, spectral unmixing and compression of hyperspectral image data. Most of the techniques addressed in this chapter are based on concepts inspired by mathematical morphology, a theory that provides a remarkable framework to achieve the desired integration. Parallel implementations of some of the proposed techniques are also developed to satisfy time-critical constraints in remote sensing applications, using NASA’s Thunderhead Beowulf cluster for demonstration purposes throughout the chapter. Combined, the different topics covered by this chapter offer a thoughtful perspective on future potentials and emerging challenges in the design of robust spatial-spectral techniques for hyperspectral image analysis.
Article
Conventional approaches to training a supervised image classification aim to fully describe all of the classes spectrally. To achieve a complete description of each class in feature space, a large training set is typically required. It is not, however, always necessary to have training statistics that provide a complete and representative description of the classes, especially if using nonparametric classifiers. For classification by a support vector machine, only the training samples that are support vectors, which lie on part of the edge of the class distribution in feature space, are required; all other training samples provide no contribution to the classification analysis. If regions likely to furnish support vectors can be identified in advance of the classification, it may be possible to intelligently select useful training samples. The ability to target useful training samples may allow accurate classification from small training sets. This potential for intelligent training sample collection was explored for the classification of agricultural crops from multispectral satellite sensor data. With a conventional approach to training, only a quarter of the training samples acquired actually made a positive contribution to the analysis and allowed the crops to be classified to a high accuracy (92.5%). The majority of the training set, therefore, was unnecessary as it made no contribution to the analysis. Using ancillary information on soil type, however, it would be possible to constrain the training sample acquisition process. By limiting training sample acquisition only to regions with a specific soil type, it was possible to use a small training set to classify the data without loss of accuracy. Thus, a small number of intelligently selected training samples may be used to classify a data set as accurately as a larger training set derived in a conventional manner. The results illustrate the potential to direct training data acquisition strategies to target the most useful training samples to allow efficient and accurate image classification.
Article
Imaging spectroscopy, also known as hyperspectral imaging, has been transformed in less than 30 years from being a sparse research tool into a commodity product available to a broad user community. Currently, there is a need for standardized data processing techniques able to take into account the special properties of hyperspectral data. In this paper, we provide a seminal view on recent advances in techniques for hyperspectral image processing. Our main focus is on the design of techniques able to deal with the high-dimensional nature of the data, and to integrate the spatial and spectral information. Performance of the discussed techniques is evaluated in different analysis scenarios. To satisfy time-critical constraints in specific applications, we also develop efficient parallel implementations of some of the discussed algorithms. Combined, these parts provide an excellent snapshot of the state-of-the-art in those areas, and offer a thoughtful perspective on future potentials and emerging challenges in the design of robust hyperspectral imaging algorithms.
Article
Hyperspectral imaging, which records a detailed spectrum of light for each pixel, provides an invaluable source of information regarding the physical nature of the different materials, leading to the potential of a more accurate classification. However, high dimensionality of hyperspectral data, usually coupled with limited reference data available, limits the performances of supervised classification techniques. The commonly used pixel-wise classification lacks information about spatial structures of the image. In order to increase classification performances, integration of spatial information into the classification process is needed. In this paper, we propose to extend the watershed segmentation algorithm for hyperspectral images, in order to define information about spatial structures. In particular, several approaches to compute a one-band gradient function from hyperspectral images are proposed and investigated. The accuracy of the watershed algorithms is demonstrated by the further incorporation of the segmentation maps into a classifier. A new spectral–spatial classification scheme for hyperspectral images is proposed, based on the pixel-wise Support Vector Machines classification, followed by majority voting within the watershed regions. Experimental segmentation and classification results are presented on two hyperspectral images. It is shown in experiments that when the number of spectral bands increases, the feature extraction and the use of multidimensional gradients appear to be preferable to the use of vectorial gradients. The integration of the spatial information from the watershed segmentation in the hyperspectral image classifier improves the classification accuracies and provides classification maps with more homogeneous regions, compared to pixel-wise classification and previously proposed spectral–spatial classification techniques. The developed method is especially suitable for classifying images with large spatial structures.
Article
The rapid development of space and computer technologies has made possible to store a large amount of remotely sensed image data, collected from heterogeneous sources. In particular, NASA is continuously gathering imagery data with hyperspectral Earth observing sensors such as the Airborne Visible-Infrared Imaging Spectrometer (AVIRIS) or the Hyperion imager aboard Earth Observing-1 (EO-1) spacecraft. The development of fast techniques for transforming the massive amount of collected data into scientific understanding is critical for space-based Earth science and planetary exploration. This paper describes commodity cluster-based parallel data analysis strategies for hyperspectral imagery, a new class of image data that comprises hundreds of spectral bands at different wavelength channels for the same area on the surface of the Earth. An unsupervised technique that integrates the spatial and spectral information in the image data using multi-channel morphological transformations is parallelized and compared to other available parallel algorithms. The code's portability, reusability and scalability are illustrated by using two high-performance parallel computing architectures: a distributed memory, multiple instruction multiple data (MIMD)-style multicomputer at European Center for Parallelism of Barcelona, and a Beowulf cluster at NASA's Goddard Space Flight Center. Experimental results suggest that Beowulf clusters are a source of computational power that is both accessible and applicable to obtaining results in valid response times in information extraction applications from hyperspectral imagery.
Article
This paper presents an overview of pattern clustering methods from a statistical pattern recognition perspective, with a goal of providing useful advice and references to fundamental concepts accessible to the broad community of clustering practitioners. We present a taxonomy of clustering techniques, and identify cross-cutting themes and recent advances. We also describe some important applications of clustering algorithms such as image segmentation, object recognition, and information retrieval
Article
Hyperspectral analysis algorithms exhibit inherent parallelism at multiple levels, and map nicely on high performance systems such as massively parallel clusters and networks of computers. Unfortunately, these systems are generally expensive and difficult to adapt to onboard data processing scenarios, in which low-weight and low-power integrated components are desirable to reduce mission pay-load. An exciting new development in this field is the emergence of programmable graphics hardware. Driven by the ever-growing demands of game industry, graphics processing units (GPUs) have evolved from expensive, application-specific units into highly parallel and programmable systems which can satisfy extremely high computational requirements at low cost. In this paper, we investigate GPU-based implementations of a morphological endmember extraction algorithm, which is used as a representative case study of joint spatial/spectral techniques for hyperspectral analysis. The proposed implementations are quantitatively compared and assessed in terms of both endmember extraction accuracy and parallel efficiency. Combined, these parts offer a thoughtful perspective on the potential and emerging challenges of implementing hyperspectral imaging algorithms on commodity graphics hardware.
Article
This paper addresses the problem of segmenting an image into regions. We define a predicate for measuring the evidence for a boundary between two regions using a graph-based representation of the image. We then develop an efficient segmentation algorithm based on this predicate, and show that although this algorithm makes greedy decisions it produces segmentations that satisfy global properties. We apply the algorithm to image segmentation using two different kinds of local neighborhoods in constructing the graph, and illustrate the results with both real and synthetic images. The algorithm runs in time nearly linear in the number of graph edges and is also fast in practice. An important characteristic of the method is its ability to preserve detail in low-variability image regions while ignoring detail in high-variability regions.
Article
A new method for segmentation and classification of hyperspectral images is proposed. The method is based on the construction of a minimum spanning forest (MSF) from region markers. Markers are defined automatically from classification results. For this purpose, pixelwise classification is performed, and the most reliable classified pixels are chosen as markers. Each classification-derived marker is associated with a class label. Each tree in the MSF grown from a marker forms a region in the segmentation map. By assigning a class of each marker to all the pixels within the region grown from this marker, a spectral-spatial classification map is obtained. Furthermore, the classification map is refined using the results of a pixelwise classification and a majority voting within the spatially connected regions. Experimental results are presented for three hyperspectral airborne images. The use of different dissimilarity measures for the construction of the MSF is investigated. The proposed scheme improves classification accuracies, when compared to previously proposed classification techniques, and provides accurate segmentation and classification maps.
Article
An important problem in pattern recognition is the effect of limited training samples on classification performance. When the ratio of the number of training samples to the dimensionality is small, parameter estimates become highly variable, causing the deterioration of classification performance. This problem has become more prevalent in remlote sensing with the emergence of a new generation of sensors. While the new sensor technology provides higher spectral and spatial resolution, enabling a greater number of spectrally separable classes to be identified, the needed labeled samples for designing the classifier remain difficult and expensive to acquire. In this thesis, several issules concerning the classification of high dimensional data with limited training samples are addressed. First of all, better parameter estimates can be obtained using a large number of unlabeled samples in addition to training samples under the mixture model. However, the estimation method is sensitive to the presence of statistical out1:iers. In remote sensing data, classes with few samples are difficult to identify and may constitute statistical outliers. Therefore, a robust parameter estima.tion method for the mixture model is introduced. Motivated by the fact that covariance estimates become highly variable with limited training samples, a covariance estimator is developed using a Bayesian formulation. The proposed covariance estimator is advAntageous when the training set size varies and reflects the prior of each class. Finally, a binary tree design is proposed to deal with the problem of varying training sample size. The proposed binary tree can function as both a classifiler and a feature extraction method. The benefits and limitations of the proposed methods are discussed and demonstrated with experiments.
Conference Paper
The image segmentation approach described is a new hybrid of region growing and spectral clustering. This approach produces a specified number of hierarchical segmentations at different levels of detail, based upon jumps in a dissimilarity criterion. A recursive implementation of this segmentation approach on a cluster of 66 Pentium Pro PCs is described, and the effectiveness of this segmentation approach on Landsat Multispectral Scanner data is discussed
Article
The classification of very high resolution remote sensing images from urban areas is addressed by considering the fusion of multiple classifiers which provide redundant or complementary results. The proposed fusion approach is in two steps. In a first step, data are processed by each classifier separately, and the algorithms provide for each pixel membership degrees for the considered classes. Then, in a second step, a fuzzy decision rule is used to aggregate the results provided by the algorithms according to the classifiers' capabilities. In this paper, a general framework for combining information from several individual classifiers in multiclass classification is proposed. It is based on the definition of two measures of accuracy. The first one is a pointwise measure which estimates for each pixel the reliability of the information provided by each classifier. By modeling the output of a classifier as a fuzzy set, this pointwise reliability is defined as the degree of uncertainty of the fuzzy set. The second measure estimates the global accuracy of each classifier. It is defined a priori by the user. Finally, the results are aggregated with an adaptive fuzzy operator ruled by these two accuracy measures. The method is tested and validated with two classifiers on IKONOS images from urban areas. The proposed method improves the classification results when compared with the separate use of the different classifiers. The approach is also compared with several other fuzzy fusion schemes
Article
Developments in the field of image understanding in remote sensing over the past four decades are reviewed, with an emphasis, initially, on the contributions of David Landgrebe and his colleagues at the Laboratory for Applications of Remote Sensing, Purdue University. The differences in approach required for multispectral, hyperspectral and radar image data are emphasised, culminating with a commentary on methods commonly adopted for multisource image analysis. The treatment concludes by examining the requirements of an operational multisource thematic mapping process, in which it is suggested that the most practical approach is to analyze each data type separately, by techniques optimized to that data's characteristics, and then to fuse at the label level.