This PDF file contains the front matter associated with SPIE Proceedings Volume 6966, including the Title Page, Copyright information, Table of Contents, Introduction (if any), and the Conference Committee listing.

Subspace methods for hyperspectral imagery enable detection and identification of targets under unknown environmental conditions (i.e., atmospheric, illumination, surface temperature, etc.) by specifying a subspace of possible target spectral signatures (and, optionally, a background subspace) and identifying closely fitting spectra in the image. The subspaces, defined from a set of exemplar spectra, are compactly expanded in singular value decomposition basis vectors or, less commonly, endmember basis spectra, linear combinations of which are used to fit the image data. In the present study we compared detection performance in the thermal infrared using several different constrained and unconstrained basis set expansions of low-dimensional subspaces, including a method based on the Sequential Maximum Angle Convex Cone (SMACC) endmember algorithm. Constrained expansions were found to provide a modest improvement in algorithm robustness in our test cases.

Anomaly detection for hyperspectral imaging is typically based on the Mahalanobis distance. The sample statistics for Mahalanobis distance are not resistant to the anomalies that are present in the sample pixels. Consequently, the sample statistics do not estimate the corresponding population parameters accurately. In this paper, we will present an algorithm for hyperspectral anomaly detection based on the Mahalanobis distance computed using robust statistics which are estimated based on the minimum generalized variance of the sample pixels. Numerical results based on actual hyperspectral images will be presented.

This paper describes a new adaptive spectral matched filter and a modified RX-based anomaly detector that incorporates the idea of regularization (shrinkage). The regularization has the effect of restricting the possible matched filters (models) to a subset which are more stable and have better performance than the non-regularized adaptive spectral matched filters. The effect of regularization depends on the form of the regularization term and the amount of regularization is controlled by so called regularization coefficient. In this paper the sum-of-squares of the filter coefficients is used as the regularization term and several different values for the regularization coefficient are tested. A Bayesian-based derivation of the regularized matched filter is also provided. Experimental results for detecting and recognizing targets in hyperspectral imagery are presented for regularized and non-regularized spectral matched filters and RX algorithm.

An adaptive algorithm is described for deriving constant false alarm rate (CFAR) detection thresholds based on statistically motivated models of actual spectral detector output distributions. The algorithm dynamically tracks the distribution of detector observables and fits the observed distribution to a suitable mixture density model function. The fitted distribution model is used to compute numerical detection thresholds that achieve a constant probability of false alarm (Pfa) per pixel. Typically gamma mixture densities are used to model outputs of anomaly detectors based on quadratic decision statistics, while normal mixture densities are used for linear matched filter type detectors. In order to achieve the computational efficiency required for real-time implementations of the algorithm on mainstream microprocessors, a robust yet considerably less complex exponential mixture model was recently developed as a general approximation to common long-tailed detector distributions. Within the region of operational interest, namely between the primary mode and the far tail, this approximation serves as an accurate model while providing significant reduction in computational cost. We compare the performance of the exponential approximation against the full-blown gamma and normal models. We also demonstrate the false alarm regulation performance of the adaptive CFAR algorithm using anomaly and matched detector outputs derived from actual VNIR-band hyperspectral imagery collected by the Civil Air Patrol (CAP) Airborne Real time Cueing Hyperspectral Enhanced Reconnaissance (ARCHER) system.

The paper outlines a new method for band selection derived from a multivariate normal mixture anomaly detection method. The method consists in evaluating detection performance in terms of false alarm rates for all band configurations obtainable from an input image by selecting and combining bands according to selection criteria reflecting sensor physics. We apply the method to a set of hyperspectral images in the visible and near-infrared spectral domain spanning a range of targets, backgrounds and measurement conditions. We find optimum bands, and investigate the feasibility of defining a common band set for a range of scenarios. The results suggest that near optimal performance can be obtained using general configurations with less than 10 bands. This may have implications for the choice of sensor technology in target detection applications. The study is based on images with high spectral and spatial resolution from the HySpex hyperspectral sensor.

The design and performance characteristics of a novel Acousto Optic Tunable Filter (AOTF) are presented. Particular attention has been paid to the reduction of optical side lobes, maximising the light throughput and achieving efficient wideband RF matching of a device for use in hyperspectral imaging systems.

Tomographic spectral imagers owe much of their development to sophisticated numerical processing. In order to reduce system size and complexity, mechanical detail has often been replaced with ever-increasing algorithm sophistication. In developing the Field Multiplexed Dispersive Imaging Spectrometer (FMDIS), the processing has been broken down into two steps; one which deconvolves the solution spatially and a second which deconvolves the solution spectrally. The first step is characterized by large inversion matrices of a few iterations (typically less than 10), while the second requires small matrices and a large number of iterations (hundreds to millions). Iterative processing has been employed due to the physical nature of the data. Inversions must be robust to moderate amounts of noise and calibration uncertainty. In this paper we present a deterministic pseudo inversion technique to replace the second iterative processing step in FMDIS datacube generation. It is shown to be within required limits of accuracy and can speed up processing by an order of magnitude or more. While not intended to replace the iterative solution technique, it provides a fast means of processing data when speed is more important than accuracy. Implementation of the solution algorithm is discussed relative to the over-all solvability of the under determined system of equations. Several results are shown from a visible instrument with 33 colors which contrast the two techniques.

We have proposed a new method for illumination suppression in hyperspectral image data. This involves transforming the data into a hyperspherical coordinate system, segmenting the data cloud into a large number of classes according to the radius dimension, and then demeaning each class, thereby eliminating the distortion introduced by differential absorption in shaded regions. This method was evaluated against two other illumination-suppression methods using two metrics: visual assessment and spectral similarity of similar materials in shaded and fully illuminated regions. The proposed method shows markedly superior performance by each of these metrics.

A 3-D spectral/spatial DFT represents an image region using a dense sampling in the frequency domain. An alternative approach is to represent a 3-D DFT by its projection onto a set of functions that capture specific orientation, scale, and spectral attributes of the image data. For this purpose, we have developed a new model for spectral/spatial information in images based on three-dimensional Gabor filters. This model achieves optimal joint localization in space and frequency and provides an efficient means of sampling a three-dimensional frequency domain representation of HSI data. Since 3-D Gabor filters allow for a large number of spectral/spatial quantities to be used to represent an image region, the performance and efficiency of algorithms that use this representation can be improved if methods are available to reduce the dimensionality of the model. Thus, we have derived methods for selecting filters that emphasize the most significant spectral/spatial differences between the various classes in a scene. We demonstrate the utility of the new model for region classification in AVIRIS data.

In this work, we study unsupervised classification algorithms for hyperspectral images based on band-by-band scalar histograms and vector-valued generalized histograms, obtained by vector quantization. The corresponding histograms are compared by dissimilarity metrics such as the chi-square, Kolmogorov-Smirnorv, and earth mover's distances. The histograms are constructed from homogeneous regions in the images identified by a pre-segmentation algorithm and distance metrics between pixels. We compare the traditional spectral-only segmentation algorithms C-means and ISODATA, versus spectral-spatial segmentation algorithms such as unsupervised ECHO and a novel segmentation algorithm based on scale-space concepts. We also evaluate the use of complex features consisting of the real spectrum and its derivative as the imaginary part. The comparison between the different segmentation algorithms and distance metrics is based on their unsupervised classification accuracy using three real hyperspectral images with known ground truth.

In this paper, an algorithm that extracts regional texture information by computing spectral difference histograms over window extents in hyperspectral images is presented. The spectral angle distance is used as the spectral metric and different window sizes are explored for computing the histogram. The histograms are used in a semi-supervised learning framework that uses both labeled and unlabeled samples for training the support vector machine classifier, which is then tested with unlabeled samples. Results are presented with real and synthetic hyperspectral images. The method performs well with high spatial resolution images. The algorithm performs well under different noise levels.

A processing algorithm to classify hyperspectral images from an imaging spectroscopic sensor is investigated in this paper. In this research two approaches are followed. First, the feasibility of an analysis scheme consisting of spectral feature extraction and classification is demonstrated. Principal component analysis (PCA) is used to perform data dimensionality reduction while the spectral interpretation algorithm for classification is the K nearest neighbour (KNN). The performance of the KNN method, in terms of accuracy and classification time, is determined as a function of the compression rate achieved in the PCA pre-processing stage. Potential applications of these hyperspectral sensors for foreign object detection in industrial scenarios are enormous, for example in raw material quality control. KNN classifier provides an enormous improvement in this particular case, since as no training is required, new products can be added in any time. To reduce the high computational load of the KNN classifier, a generalization of the binary tree employed in sorting and searching, kd-tree, has been implemented in a second approach. Finally, the performance of both strategies, with or without the inclusion of the kd-tree, has been successfully tested and their properties compared in the raw material quality control of the tobacco industry.

Thermal Infrared Multispectral Scanner (TIMS) data are processed to yield surface temperatures over the lava tube system of Kilauea Volcano, Hawaii. TIMS is a 6-band airborne longwave infrared (8 μm to 12 μm) multispectral imaging system built and operated by the National Aeronautics and Space Administration (NASA). The data analyzed were collected in 1988 and are part of the Compiled Volcanology Data Set collection of Glaze et al., (1992). The primary goal of the analyses is to utilize the TIMS-derived surface temperatures to estimate lava tube roof thickness (LTRT). There is a paucity of studies that have utilized remotely-sensed imaging spectrometry data to estimate LTRT - a component important to understanding (and modeling) the thermal field of lava tube systems. Lava tube systems, in turn, are important to the emplacement of areally extensive lava flows on earth and on other planets. An in-scene atmospheric compensation method was applied to the data followed by a normalized emissivity method temperature/emissivity separation algorithm to obtain surface temperature. Surface temperature measurements are then compared to modeled temperatures in order to estimate lava tube roof thickness. Modeled temperatures are calculated via finite element analysis. Boundary conditions of the finite element models are derived from analyses of the TIMS data, independent knowledge of lava liquidus and solidus temperatures, and crustal heat-flow geophysical data. A TIMS plus modeling-derived LTRT agrees with estimates based on field observations. The TIMS data are described as are all processing and analysis methods. The thermal modeling is also described as is an effort to build a lookup table for LTRTs to be used in conjunction with surface temperature measurements. Archived data such as those exploited here provide a historical context particularly for terranes which may undergo relatively rapid change - such as the lava flow fields of Kilauea Volcano.

Since Magnetic Resonance (MR) images can be considered as multispectral images where each spectral band image is acquired by a particular pulse sequence, this paper investigates an application of a technique that is widely used in multispectral image processing, referred to as Linear Spectral Unmixing (LSU), in MR image analysis where two types of LSU, unconstrained LSU and constrained LSU are considered. Due to a limited number of MR images acquired by MR sequences, the ability of the LSU cannot be fully explored and utilized. In order to mitigate this dilemma, a band expansion process is introduced to expand an original set of MR images to an augmented set of multsipectral images by including additional spectral band images that can be generated from the original MR images using a set of nonlinear functions. In order to demonstrate the utility of the LSU in MR image analysis, two sets of MR images, synthetic MR images available on website and real MR images, are used for experiments. Experimental results show that the LSU can be a very effective technique in quantifying MR substances to calculate their partial volumes for further MR image analysis.

Electromagnetic signatures of terrain exhibit significant spatial heterogeneity on a range of scales as well as considerable temporal variability. A statistical characterization of the spatial heterogeneity and spatial scaling algorithms of terrain electromagnetic signatures are required to extrapolate measurements to larger scales. Basic terrain elements including bare soil, grass, deciduous, and coniferous trees were studied in a quasi-laboratory setting using instrumented test sites in Hanover, NH and Yuma, AZ. Observations were made using a visible and near infrared spectroradiometer (350 - 2500 nm) and hyperspectral camera (400 - 1100 nm). Results are reported illustrating: i) several difference scenes; ii) a terrain scene time series sampled over an annual cycle; and iii) the detection of artifacts in scenes. A principal component analysis indicated that the first three principal components typically explained between 90 and 99% of the variance of the 30 to 40-channel hyperspectral images. Higher order principal components of hyperspectral images are useful for detecting artifacts in scenes.

Due to the complex nature of hyper-spectra imaging, there are diversified noises in different bands of hyper-spectra image. Without proper pre-processing, these noises will lead to false target detection results in application. Furthermore, because of low signal to noise ratio, some bands, such as bands affected by water vapor in the infrared wavelengths, cannot be utilized in the target detection task. To improve the performance of hyper-spectra applications, many noise removal technologies have been developed. Most traditional denoising approaches either take only single band image into account at a time or only consider spectra shape at one location a time. But these approaches could not deal effectively with the common noises in hyper-spectra image that change from band to band and from one spatial spot to another. Also most generalized smooth filters without local adaptation will lead to losses in spatial details at band images. We propose a denoising approach that is based on bilateral filtering, which takes both spectra and spatial information into account. By locally adapt to adjacent spectra distribution, this approach will have the advantage of effective noise removal while keeping the spatial details in the band images. We also proposed parameter estimation method for hyperspectral image bilateral filtering. The experiment results show that this approach deliver better performance under various noises than other approach, the low signal to noise ratio in some band images have been significantly improved.

An expert system, the "Spectral Expert" has been implemented for identification of materials based on extraction of key spectral features from visible/near infrared (VNIR) and shortwave infrared (SWIR) reflectance spectra and hyperspectral imagery (HSI). Spectral absorption features are automatically extracted from a spectral library and each is analyzed to determine diagnostic features and characteristics - the "rules". An expert optionally analyzes spectral variability and separability to create refined rules for identification of specific materials. The rules can be used by a non-expert to identify materials by matching individual feature parameters or with a rule-controlled RMS approach. The result for a single spectrum is a score between 0.0 (no-match) and 1.0 (perfect-match) for each specific material in the spectral library, or for hyperspectral data, a classified image showing the predominate material on a per-pixel basis and a score image for each material. A feature-based-mixture-index (FBMI) score or image is also created, which alerts the analyst to possible problem spectra and mixing. This can be used to determine iterative expert system processing requirements for determination of secondary materials and assemblages and to point the analyst towards supplementary analyses using other non-feature-based methods. A geologic example demonstrates simplest case Spectral Expert analysis - application to minerals with a laboratory spectral library and well-defined spectral features. An example for an urban site demonstrates application and results where no previous spectral library exists. The approach, methods, and algorithms have been implemented in a software plug-in to the popular "ENVI" image processing and analysis software.

This paper extends the field of hyperspectral anomaly and target detection by introducing a new approach for preprocessing hyperspectral image data. In this study, we investigate the Median-Spectral-Spatial Transformation as an approach to draw out the sub-pixel difference characterizations of anomalous spectra. By implementing this preprocessing step, we have realized a significant improvement in false alarm reduction with increased probability of detection for sub-pixel targets. Sub-pixel anomalies contain target information consisting of only a small fraction of an image pixel's surface reflected material content. To demonstrate the efficacy of our approach, we compare results from RX anomaly detection across multiple HSI images.

For multi-sensor registration, previous techniques typically use mutual information (MI) rather than the sum-of-the-squared difference (SSD) as the similarity measure. However, the optimization of MI is much less straightforward than is the case for SSD-based algorithms. A new technique for image registration has recently been proposed that uses an information theoretic measure called the Cross-Cumulative Residual Entropy (CCRE). In this paper we show that using CCRE for multi-sensor registration of remote sensing imagery provides an optimization strategy that converges to a global maximum with significantly less iterations than existing techniques and is much less sensitive to the initial geometric disparity between the two images to be registered.

The variability of panchromatic and multispectral images, vector data (maps) and DEM models is growing. Accordingly, the requests and challenges are growing to correlate, match, co-register, and fuse them. Data to be integrated may have inaccurate and contradictory geo-references or not have them at all. Alignment of vector (feature) and raster (image) geospatial data is a difficult and time-consuming process when transformational relationships between the two are nonlinear. The robust solutions and commercial software products that address current challenges do not yet exist. In the proposed approach for Vector-to-Raster Registration (VRR) the candidate features are auto-extracted from imagery, vectorized, and compared against existing vector layer(s) to be registered. Given that available automated feature extraction (AFE) methods quite often produce false features and miss some features, we use additional information to improve AFE. This information is the existing vector data, but the vector data are not perfect as well. To deal with this problem the VRR process uses an algebraic structural algorithm (ASA), similarity transformation of local features algorithm (STLF), and a multi-loop process that repeats (AFE-VRR) process several times. The experiments show that it was successful in registering road vectors to commercial panchromatic and multi-spectral imagery.

Arguably the single greatest confound for change detection algorithms is the misregistration of the two images in which changes are being sought. On the other hand, since the effects of misregistration are exhibited over the entire image, there is reason to hope that algorithms which are designed to deal with pervasive effects (such as illumination differences in the scene, or calibration drifts in the sensor) will be less sensitive to the inevitable misregistration errors that occur when comparing two images. This work will describe some controlled experiments in which change detection performance is evaluated as a function of how misregistered the images are. The performance is observed to degrade quite rapidly with the amount of misregistration (so that any practical system for automated change detection will require accurate image registration), but algorithms that are more adaptive to pervasive differences are less sensitive to the effect of misregistration.

This paper covers the impact of registration errors between two images on chronochrome and covariance equalization predictors used for hyperspectral change detection. Hyperspectral change detection involves the comparison of data collected of the same spatial scene on two different occasions to try to identify anomalous man-made changes. Typical change detection techniques employ a linear prediction method followed by a subtraction step to identify changes. These linear predictors rely upon statistics from both scenes to determine a respective gain and offset. Chronochrome and covariance equalization remain two common predictors used in the change detection process. Chronochrome relies upon a cross-covariance matrix for prediction whereas covariance equalization relies solely upon the individual covariance matrices. In theory, chronochrome seems more susceptible to image misregistration issues as joint statistic estimates may suffer with registration error present. This paper examines the validity of this assumption. Using a push-broom style imaging spectrometer mounted on a pan and tilt, visible to near infrared data of scenes suitable for change detection analysis are gathered. The pan and tilt system ensures initial misregistration of the data is minimal. Using simple translations of the scenes, misregistration impacts upon prediction error and change detection are examined for varying degrees of shift.

High spatial resolution sounding observations will improve initialization and assimilation into the next generation forecast models and validation of the next generation of climate models. One such advanced sounder concept for low earth orbit is the Advanced Remote-sensing Imaging Emission Spectrometer (ARIES) which proposes to provide high spatial hyperspectral resolution observations in the mid to longwave infrared. This paper explores the effects of spatial resolution on the errors expected from the combined use of models and observations for representing scene information. We calculate the frequency response of the instrument and model and determine the error at any given spatial frequency. The results show that it is vital to have observations match the spatial resolution of models to minimize the uncertainty in the representation of the scene contents.

The AIRS Science Team Version 5.0 retrieval algorithm became operational at the Goddard DAAC in July 2007 generating near real-time products from analysis of AIRS/AMSU sounding data. This algorithm contains many significant theoretical advances over the AIRS Science Team Version 4.0 retrieval algorithm used previously. Two very significant developments of Version 5 are: 1) the development and implementation of an improved Radiative Transfer Algorithm (RTA) which allows for accurate treatment of non-Local Thermodynamic Equilibrium (non-LTE) effects on shortwave sounding channels; and 2) the development of methodology to obtain very accurate case by case product error estimates which are in turn used for quality control. These theoretical improvements taken together enabled a new methodology to be developed which further improves soundings in partially cloudy conditions. In this methodology, longwave CO2 channel observations in the spectral region 700 cm^-1 to 750 cm^-1 are used exclusively for cloud clearing purposes, while shortwave CO₂ channels in the spectral region 2195 cm^-1 to 2395 cm^-1 are used for temperature sounding purposes. This allows for accurate temperature soundings under more difficult cloud conditions. This paper further improves on the methodology used in Version 5 to derive surface skin temperature and surface spectral emissivity from AIRS/AMSU observations. Now, following the approach used to improve tropospheric temperature profiles, surface skin temperature is also derived using only shortwave window channels. This produces improved surface parameters, both day and night, compared to what was obtained in Version 5. These in turn result in improved boundary layer temperatures and retrieved total O₃ burden.

The AIRS instrument is currently the best space-based tool to simultaneously monitor the vertical distribution of key climatically important atmospheric parameters as well as surface properties, and has provided high quality data for more than 5 years. AIRS analysis results produced at the GODDARD/DAAC, based on Versions 4 & 5 of the AIRS retrieval algorithm, are currently available for public use. Here, first we present an assessment of interrelationships of anomalies (proxies of climate variability based on 5 full years, since Sept. 2002) of various climate parameters at different spatial scales. We also present AIRS-retrievals-based global, regional and 1x1 degree grid-scale "trend"-analyses of important atmospheric parameters for this 5-year period. Note that here "trend" simply means the linear fit to the anomaly (relative the mean seasonal cycle) time series of various parameters at the above-mentioned spatial scales, and we present these to illustrate the usefulness of continuing AIRS-based climate observations. Preliminary validation efforts, in terms of intercomparisons of interannual variabilities with other available satellite data analysis results, will also be addressed. For example, we show that the outgoing longwave radiation (OLR) interannual spatial variabilities from the available state-of-the-art CERES measurements and from the AIRS computations are in remarkably good agreement. Version 6 of the AIRS retrieval scheme (currently under development) promises to further improve bias agreements for the absolute values by implementing a more accurate radiative transfer model for the OLR computations and by improving surface emissivity retrievals.

We apply the method of Vanishing Partial Derivatives (VPD) to AIRS spectra to retrieve daily the global distribution of CO₂ at a nadir geospatial resolution of 90 km x 90 km without requiring a first-guess input beyond the global average. Our retrievals utilize the 15 μm band radiances, a complex spectral region. This method may be of value in other applications, in which spectral signatures of multiple species are not well isolated spectrally from one another.

Recent work has demonstrated the feasibility of neural network estimation techniques for atmospheric profiling in partially cloudy atmospheres using combined microwave (MW) and hyperspectral infrared (IR) sounding data. In this paper, the retrieval performance in problem areas (over land, near the poles, elevated terrain, etc.) is examined. Retrieval performance has been improved by stratifying the neural network training data into distinct groups based on geographical (latitude, for example), geophysical (atmospheric pressure, for example), and sensor geometrical (scan angle, for example) considerations. The spectral information content of cloud signatures in Infrared Atmospheric Sounding Interferometer (IASI) data is also explored. A Principal Components Analysis is presented that indicates that most variability due to clouds is contained in the first two eigenvectors. A novel statistical method for the retrieval of atmospheric temperature and moisture (relative humidity) profiles has been developed and evaluated with sounding data from the Atmospheric InfraRed Sounder (AIRS) and the Advanced Microwave Sounding Unit (AMSU). The present work focuses on the cloud impact on the AIRS radiances and explores the use of stochastic cloud clearing mechanisms together with neural network estimation. A stand-alone statistical algorithm will be presented that operates directly on cloud-impacted AIRS/AMSU data, with no need for a physical cloud clearing process. The algorithm is implemented in three stages. First, the infrared radiance perturbations due to clouds are estimated and corrected by combined processing of the infrared and microwave data using a Stochastic Cloud Clearing (SCC) approach. The cloud clearing of the infrared radiances was performed using principal components analysis of infrared brightness temperature contrasts in adjacent fields of view and microwave-derived estimates of the infrared clear-column radiances to estimate and correct the radiance contamination introduced by clouds. Second, a Projected Principal Components (PPC) transform is used to reduce the dimensionality of and optimally extract geophysical profile information from the cloud-cleared infrared radiance data. Third, an artificial feedforward neural network (NN) is used to estimate the desired geophysical parameters from the projected principal components. The performance of this method was evaluated using global (ascending and descending) EOS-Aqua orbits co-located with ECMWF fields for a variety of days throughout 2003, 2004, 2005, and 2006. Over 1,000,000 fields of regard (3x3 arrays of footprints) over ocean and land were used in the study. The method requires significantly less computation than traditional variational retrieval methods, while achieving comparable performance. Retrieval accuracy will be evaluated using ECMWF atmospheric fields as ground truth. The accuracy of the neural network retrieval method will be compared to the accuracy of the AIRS Level 2 (Version 5) retrieval method.

More than five years of CO retrievals from the Atmospheric InfraRed Sounder (AIRS) onboard NASA's Aqua satellite reveal variations in tropospheric CO on timescales from twelve hours to five years and on spatial scales from local to global. The shorter timescales are invaluable to monitor daily variations in CO emissions, to enable three-dimensional tracking of atmospheric motions, and to enhance insights into atmospheric mixing. Previous studies have utilized AIRS CO retrievals over the course of days to weeks to track plumes from large forest fires. On the local scale, we will present AIRS observations of pollution from several northern hemisphere Megacities. On the regional scale, we will present AIRS observations of the Mexico City pollution plume. We will illustrate global scale AIRS CO observations of interannual variations linked to the influence of large-scale atmospheric perturbations from the El Nino Southern Oscillation (ENSO). In particular, we observe a quasi-biennial variation in CO emissions from Indonesia with varying magnitudes in peak emission occurring in 2002, 2004, and 2006. Examining satellite rainfall measurements over Indonesia, we find the enhanced CO emission correlates with occasions of less rainfall during the month of October. Continuing this satellite record of tropospheric CO with measurements from the European IASI instrument will permit construction of a long time-series useful for further investigations of climatological variations in CO emissions and their impact on the health of the atmosphere.

The Spaceborne Infrared Sounder for Geosynchronous Earth Orbit (SIRAS-G) was developed by Ball Aerospace & Technologies Corp (BATC) under NASA's 2002 Instrument Incubator Program. SIRAS-G was a technology development program focused on next-generation IR imaging spectrometers for sounding of the atmosphere. SIRAS-G demonstrated that the dispersive grating spectrometer is a suitable instrument architecture for this application. In addition to providing atmospheric temperature and water vapor profiles, SIRAS-G can provide trace gases concentrations, land and ocean surface temperatures and the IR mineral dust aerosol signature from satellite. The 3-year SIRAS-G IIP development effort included the successful cryogenic testing of the SIRAS-G laboratory demonstration spectrometer operating in the 2083 to 2994 cm^-1 frequency range. The performance of the demonstration instrument has been quantified including measurement of keystone distortion, spectral smile, MTF, and the spectral response function (SRF). Development efforts associated with this advanced infrared spectrometer technology provides the basis for instrumentation to support future Earth science missions.

Aerosols in the atmosphere affect the Earth's radiation budget in complicated ways, depending on their physical and optical characteristics and how they interact with solar and terrestrial radiation or affect cloud nucleation. While the Arctic atmosphere is generally very clean, spring incursions of haze and dust from Eurasia are known to perturb the surface radiation balance. Recent analyses (based on "Radiative impact of boreal smoke in the Arctic: Observed and modeled", Stone, et al., to be referred to throughout this ms as Stone2008) also reveal that smoke plumes from boreal forest fires can have significant effects during summer. Once aloft, upper-level winds can transport this smoke long distances. In late June and July 2004 fires raged across eastern Alaska and the Yukon and the resulting smoke was advected across the Arctic, reaching as far as Europe. The long-range transport was tracked using a dispersion model combined with various in situ measurements along its path, all showing enhancements in aerosol opacity. The measurements made at Barrow, Alaska, documented just a portion of the transport and the radiative impact of smoke. The comprehensive measuring systems in place near Barrow (NOAA/GMD and DoE/ARM) presented a unique opportunity to characterize the smoke aerosol both physically and optically, and therefore permit quantification of the upwelling radiance (outgoing shortwave radiance - OSR, 0.28 to 4.0 μm) as observed by NASA satellites: Clouds and the Earth's Radiant Energy System (CERES) 5, coupled with data from Moderate Resolution Imaging Spectroradiometer (MODIS).

The atmosphere is a critical factor in remote sensing. Radiance from a target must pass through the air column to reach the sensor. The atmosphere alters the radiance reaching the sensor by attenuating the radiance from the target via scattering and absorption and by introducing an upwelling radiance. In the thermal infrared, these effects will introduce errors in the derived apparent temperature of the target if not properly accounted for. The temperature error is defined as the difference between the target leaving apparent temperature and observed apparent temperature. The effects of the atmosphere must be understood in order to develop methods to compensate for this error. Different atmospheric components will affect the radiation passing through it in different ways. Certain components may be more important than others depending on the remote sensing application. The authors are interested in determining the actual temperature of the superstructure that composes a mechanical draft cooling tower (MDCT), hence water vapor is the primary constituent of concern. The tower generates a localized water vapor plume located between the target and sensor. The MODTRAN radiative transfer code is used to model the effects of a localized exhaust plume from a MDCT in the longwave infrared. The air temperature and dew point depression of the plume and the thickness of the plume are varied to observe the effect on the apparent temperature error. In addition, the general atmospheric conditions are varied between two standard MODTRAN atmospheres to study any effect that ambient conditions have on the apparent temperature error. The Digital Imaging and Remote Sensing Image Generation (DIRSIG) modeling tool is used to simulate the radiance reaching a thermal sensor from a target after passing through the water vapor plume. The DIRSIG results are validated against the MODTRAN results. This study shows that temperature errors of as much as one Kelvin can be attributed to the presence of a localized water vapor plume.

The Air Force Institute of Technology's Center for Directed Energy (AFIT/CDE) developed the High Energy Laser End-to-End Operational Simulation (HELEEOS) model in part to quantify the performance variance in laser propagation created by the natural environment during dynamic engagements. As such, HELEEOS includes a fast-calculating, first principles, worldwide surface-to-100 km, atmospheric propagation and characterization package. This package enables the creation of profiles of temperature, pressure, water vapor content, optical turbulence, atmospheric particulates and hydrometeors as they relate to line-by-line layer transmission, path and background radiance at wavelengths from the ultraviolet to radio frequencies. Physics-based cloud and precipitation characterizations are coupled with a probability of cloud free line-of-sight algorithm for all possible look angles. HELEEOS was developed under the sponsorship of the High Energy Laser Joint Technology Office. In the current paper an example of a unique high fidelity simulation of a bi-static, time-varying five band multispectral remote observation of laser energy delivered on a test object is presented. The multispectral example emphasizes atmospheric effects using HELEEOS, the interaction of the laser on target and the observed reflectance and subsequent hot spot generated. A model of a sensor suite located on the surface is included to collect the diffuse reflected in-band laser radiation and the emitted radiance of the hot spot in four separate and spatially offset MWIR and LWIR bands. Particular care is taken in modeling the bidirectional reflectivity distribution function (BRDF) of the laser/target interaction to account for both the coupling of energy into the target body and the changes in reflectance as a function of temperature. The architecture supports any platform-target-observer geometry, geographic location, season, and time of day; and it provides for correct contributions of the sky-earth background. The simulation accurately models the thermal response, kinetics, turbulence, base disturbance, diffraction, and signal-to-noise ratios.

The degrading effect of the atmosphere on hyperspectral imagery has long been recognised as a major issue in applying techniques such as spectrally-matched filters to hyperspectral data. There are a number of algorithms available in the literature for the correction of hyperspectral data. However most of these approaches rely either on identifying objects within a scene (e.g. water whose spectral characteristics are known) or by measuring the relative effects of certain absorption features and using this to construct a model of the atmosphere which can then be used to correct the image. In the work presented here, we propose an alternative approach which makes use of the fact that the effective number of degrees of freedom in the atmosphere (transmission, path radiance and downwelling radiance with respect to wavelength) is often substantially less than the number of degrees of freedom in the spectra of interest. This allows the definition of a fixed set of invariant features (which may be linear or non-linear) from which reflectance spectra can be approximately reconstructed irrespective of the particular atmosphere. The technique is demonstrated on a range of data across the visible to near infra-red, mid-wave and long-wave infra-red regions, where its performance is quantified.

We continue previous work that generalizes the traditional linear mixing model from a combination of endmember vectors to a combination of multi-dimensional affine endmember subspaces. This generalization allows the model to handle the natural variation that is present is real-world hyperspectral imagery. Once the endmember subspaces have been defined, the scene may be demixed as usual, allowing for existing post-processing algorithms (classification, etc.) to proceed as-is. In addition, the endmember subspace model naturally incorporates the use of physics-based modeling approaches ('target spaces') in order to identify sub-pixel targets. In this paper, we present a modification to our previous model that uses affine subspaces (as opposed to true linear subspaces) and a new demixing algorithm. We also include experimental results on both synthetic and real-world data, and include a discussion on how well the model fits the real-world data sets.

This paper presents a full algorithm to compute the solution for the unsupervised unmixing problem based on the positive matrix factorization. The algorithm estimates the number of endmembers as the rank of the matrix. The algorithm has an initialization stage using the SVD subset selection algorithm. Testing and validation with real and simulated data show the effectiveness of the method. Application of the approach to environmental remote sensing is shown.

This paper presents a ground truth data collection effort along with its use in evaluating unmixing algorithms. Unmixing algorithms are typically evaluated using synthetic data generated by selecting endmember spectrums and adding them in different amounts and with added noise. Going from synthetic to real data poses many problems. One of the greatest is the amount of data to be collected. Also, there will be many unmodeled variations in real data. These include greater variation of the endmembers, additional endmembers that are a very small percentage of the image, and nonlinear effects in the data that are not modeled. The data collation effort produced a high resolution class map along with spectral measurements of 153 different sampling sites to validate the map. The methodology for using this high resolution class map for generating the ground truth data for use in the unmixing algorithms is presented. Specifically, a 1m class map is used to generate the endmember abundances for every pixel in a 30m Hyperion image of the Enrique Reef in Southwest Puerto Rico. The results using two unmixing algorithms, one with a sum to one constraint and the other with a non-negative constraint are presented. The unmixing results for each endmember are presented along with a newly developed unmixing parameter called the Correct Unmixing Index (CUI).

Spectral unmixing of hyperspectral images is a process by which the constituent's members of a pixel scene are determined and the fraction of the abundance of the elements is estimated. Several algorithms have been developed in the past in order to obtain abundance estimation from hyperspectral data, however, most of them are characterized by being highly computational and time consuming due to the magnitude of the data involved. In this research we present the use of Graphic Processing Units (GPUs) as a computing platform in order to reduce computation time related to abundance estimation for hyperspectral images. Our implementation was developed in C using NVIDIA^(R) Compute Unified Device Architecture (CUDA^TM). The recently introduced CUDA platform allows developers to directly use a GPU's processing power to perform arbitrary mathematical computations. We describe our implementation of the Image Space Reconstruction Algorithm (ISRA) and Expectation Maximization Maximum Likelihood (EMML) algorithm for abundance estimation and present a performance comparison against implementations using C and Matlab. Results show that the CUDA technology produced results around 10 times better than the fastest implementation done on previous platforms.

Endmember extraction has received considerable interest in recent years. Many algorithms have been developed for this purpose and most of them are designed based on convexity geometry such as vertex or endpoint projection and maximization of simplex volume. This paper develops statistics-based approaches to endmember extraction in the sense that different orders of statistics are used as criteria to extract endmembers. The idea behind the proposed statistics-based endmember extraction algorithms (EEAs) is to assume that a set of endmmembers constitute the most un-correlated sample pool among all the same number of signatures with correlation measured by statistics which include variance specified by 2nd order statistics, least squares error (LSE) also specified by 2nd order statistics, skewness 3rd order statistics, kurtosis 4th order statistics, kth moment and statistical independency specified by infinite order of statistics measured by mutual information. In order to substantiate proposed statistics-based EEAs, experiments using synthetic and real images are conducted for demonstration.

The inherent dimensionality of a spectral image can be estimated in a number of ways, primarily based on statistical measures of the data cloud in the hyperspace. Methods using the eigenvalues from a Principal Components Analysis, a Minimum Noise Fraction transformation, or the Virtual Dimensionality algorithm are widely used as applied to entire images typically with the goal of reducing the dimensionality of an image in its entirety. However, it is desirable to understand the dimensionality of individual components within a hyperspectral scene, as there is no a priori reason to expect all distinct material classes in the scene to have the same inherent dimensionality. Additionally, in complex scenes containing non-natural materials, the lack of multivariate normality of the data set implies that a statistically based estimation is less than optimal. Here, a geometric approach is developed based on the local estimation of dimensionality in the native data hyperspace. It will be shown that the dimensionality of a collection of data points (k) in the full n dimensions (where n is the number of spectral channels measured) can be estimated by calculating the change in point density as a function of distance in the full n dimensional hyperspace. Simple simulated examples to demonstrate the concept will be shown, as well as applications to real hyperspectral imagery collected with the HyMAP sensor.

Dimensionality Reduction (DR) has found many applications in hyperspectral image processing, e.g., data compression, endmember extraction. This paper investigates Projection Pursuit (PP)-based data dimensionality reduction where three approaches are developed. One is to use a Projection Index (PI) to produce projection vectors that can be used to generate Projection Index Components (PICs). It is a common practice that PP generally uses random initial conditions to produce PICs. As a result, when the same PP is performed in different times or different users at the same time, the resulting PICs are generally not the same. In order to resolve this issue, two approaches are proposed. One is referred to as PI-based PRioritized PP (PI-PRPP) which uses a PI as a criterion to prioritize PICs that are produced by any component analysis, for example, Principal Components Analysis (PCA) or Independent Component Analysis. The other approach is called Initialization-Driven PP (ID-PP) which specifies an appropriate set of initial conditions that allows PP to not only produce PICs in the same order but also the same PICs regardless of how many times PP is run or who runs the PP.

In our previous paper, it has been demonstrated that principal component analysis (PCA) can outperform discrete wavelet transform (DWT) in spectral coding for hyperspectral image compression and a superior rate distortion performance can be provided in conjunction with 2-dimensional (2D) spatial coding using JPEG2000. The resulting compression algorithm is denoted as PCA+JPEG2000. In this paper, we further investigate how the data size (i.e., spatial and spectral size) influences the performance of PCA+JPEG2000 and provide a rule of thumb for PCA+JPEG2000 to perform appropriately. We will also show that using a subset of principal components (PCs) (the resulting algorithm is denoted as SubPCA+JPEG2000) can always yield a better rate distortion performance than PCA+JPEG2000 with all the PCs being preserved for compression.

The Space Dynamics Laboratory (SDL) has developed an FPGA-based hyperspectral demonstration compression system. The system consists of two boards: the first board performs a decorrelation process using a 5/3 wavelet; the second board performs the JPEG 2000 image compression. The hardware and firmware design of this system is described here and data is presented that shows the results of compressed hyperspectral data cubes containing various types of image content. This paper presents the importance of bit rate control among the individual spectral bands. Some of the theory for basing bit rate control on JPEG 2000 compression, bit rate control based on the 5/3 wavelet, as well as advantages and disadvantages of each method are discussed. Concepts for developing hyperspectral image compression technology for systems that can be used for remote sensing in real applications are also presented.

Component Analysis (CA) has found many applications in remote sensing image processing. Two major component analyses are of particular interest, Principal Components Analysis (PCA) and Independent Component Analysis (ICA) which have been widely used in signal processing. While the PCA de-correlates data samples via 2nd order statistics in a set of Principal Components (PCs), the ICA represents data samples via statistical independency in a set of statistically Independent Components (ICs). However, in order to for component analyses to be effective, the number of components to be generated, p must be sufficient for data analysis. Unfortunately, in MultiSpectral Imagery (MSI) p seems to be small, while p in HyperSpectral Imagery (HSI) seems too large. Interestingly, very little has been reported on how to deal with this issue when p is too small or too large. This paper investigates this issue. When p is too small, two approaches are developed to mitigate the problem. One is Band Expansion Process (BEP) which augments original data band dimensionality by producing additional bands via a set of nonlinear functions. The other is a kernel-based approach, referred to as Kernel-based PCA (K-PCA) which maps features in the original data space to a higher dimensional feature space via a set of nonlinear kernel. While both approaches make attempts to resolve the issue of a small p using a set of nonlinear functions, their design rationales are completely different, particularly they are not correlated. As for a large p such as HSI, a recently developed Virtual Dimensionality (VD) can be used for this purpose where the VD was originally developed to estimate number of spectrally distinct signatures. If we assume one spectrally distinct signature can be accommodated by one component, the value of p can be actually determined by the VD. Finally, experiments are conducted to explore and evaluate the utility of component analyses, specifically, PCA and ICA using BEP and K-PCA for MSI and VD for HSI.

Hyperspectral imagery (HSI) is a relatively new technology capable of relaying intensity information gathered from both visible and non-visible ranges of the electromagnetic spectrum. HSI images can contain hundreds of bands, which present a problem when an image analyst must select the most relevant bands from such an image for visualization, particularly when the bands that are within the range of human vision are either not present or heavily distorted. It is proposed here that two-dimensional principal component analysis (2DPCA) can aid in the automatic selection of the bands from an HSI image that would best reflect visual information. The method requires neither prior knowledge of the image contents nor the association between spectral bands and their center wavelengths.

A traditional linear-mixing model with a structured background used in the hyperspectral imaging literature often assumes Gaussianity of the error term. This assumption is often questioned, but to the best of our knowledge, we are not aware of a definite answer on how well such a model may reflect the real hyperspectral images. One difficulty is in the correct identification of the background signatures. The lack of Gaussianity in the error term might be due to missing one of the significant background signatures. In this paper, we investigate this issue using an AVIRIS hyperspectral image. We obtain the projections of the pixel spectra on the orthonormal basis system obtained through the singular value decomposition, and then we measure their Gaussianity using three different methods. We identify the subspace for the structured part of the model based on two criteria - the contribution to the image variability and non-Gaussianity of the marginal distribution. The subspace orthogonal to the structured part of the model forms the subspace of residuals, which is then investigated for multivariate Gaussianity. The resulting model forms a reasonable approximation of the hyperspectral image, and can be successfully used in a variety of applications such as unmixing and target detection. At the same time, it is clear that further improvements are possible by better modeling of the error term distribution.

Hyperspectral imaging sensors operating in the visual, near IR, and thermal IR bands are sufficiently advanced to become a standard component of surveillance sensor suites. The output of these sensors contains a wealth of spectral and spatial information that can improve target detection and recognition performance. However, the large volume and complex features of hyperspectral data are challenges to automatic target recognition (ATR) algorithm development, and a simulation of hyperspectral sensing is therefore essential in evaluating algorithm performance. This paper describes the Infrared Hyperspectral Scene Simulation (IRHSS), an accurate, non-real-time large-scene simulation tool for hyperspectral imagers operating in the thermal IR bands. The simulation contains models for target and background spectral radiance, atmospheric propagation, and sensor processing. It uses a new hyperspectral version of the Multi-service Electro-optical Signature (MuSES) model to compute scene temperatures and hyperspectral radiances. IRHSS is able to handle very large terrain and feature databases by selective use of radiation view factors. It provides end-to-end simulation starting with scene models built from COTS simulation databases with faceted terrain and targets, and optional overlays of visual high-resolution texture imagery. IRHSS can be run as a standalone application via its Windows-based graphical user interface (GUI) or as a plug-in to existing software using the IRHSS application programming interface (API). Some screen images of the IRHSS GUI and example hyperspectral image cubes generated by IRHSS are included herein.

The calibration of data from hyperspectral sensors to spectral radiance enables the use of physical models to predict measured spectra. Since environmental conditions are often unknown, material detection algorithms have emerged that utilize predicted spectra over ranges of environmental conditions. The predicted spectra are typically generated by a radiative transfer (RT) code such as MODTRAN^TM. Such techniques require the specification of a set of environmental conditions. This is particularly challenging in the LWIR for which temperature and atmospheric constituent profiles are required as inputs for the RT codes. We have developed an automated method for generating environmental conditions to obtain a desired sampling of spectra in the sensor radiance domain. Our method provides a way of eliminating the usual problems encountered, because sensor radiance spectra depend nonlinearly on the environmental parameters, when model conditions are specified by a uniform sampling of environmental parameters. It uses an initial set of radiance vectors concatenated over a set of conditions to define the mapping from environmental conditions to sensor spectral radiance. This approach enables a given number of model conditions to span the space of desired radiance spectra and improves both the accuracy and efficiency of detection algorithms that rely upon use of predicted spectra.

Many hyperspectral imaging algorithms are available for applications such as spectral unmixing, subpixel detection, quantification, endmember extraction, classification, compression, etc and many more are yet to come. It is very difficult to evaluate and validate different algorithms developed and designed for the same application. This paper makes an attempt to design a set of standardized synthetic images which simulate various scenarios so that different algorithms can be validated and evaluated on the same ground with completely controllable environments. Two types of scenarios are developed to simulate how a target can be inserted into the image background. One is called Target Implantation (TI) which implants a target pixel by removing the background pixel it intends to replace. This type of scenarios is of particular interest in endmember extraction where pure signatures can be simulated and inserted into the background with guaranteed 100% purity. The other is called Target Embeddedness (TE) which embeds a target pixel by adding this target pixel to the background pixel it intends to insert. This type of scenarios can be used to simulate signal detection models where the noise is additive. For each of both types three scenarios are designed to simulate different levels of target knowledge by adding a Gaussian noise. In order to make these six scenarios a standardized data set for experiments, the data used to generate synthetic images can be chosen from a data base or spectral library available in the public domain or websites and no particular data are required to simulate these synthetic images. By virtue of the designed six scenarios an algorithm can be assessed objectively and compared fairly to other algorithms on the same setting. This paper demonstrates how these six scenarios can be used to evaluate various algorithms in applications of subpixel detection, mixed pixel classification/quantification and endmember extraction.

Hyperspectral ground to ground viewing perspective presents major challenges for autonomous window based detection. One of these challenges has to do with object scales uncertainty that occur when using a window-based detection approach. In a previous paper, we introduced a fully autonomous parallel approach to address the scale uncertainty problem. The proposed approach featured a compact test statistic for anomaly detection, which is based on a principle of indirect comparison; a random sampling stage, which does not require secondary information (range or size) about the targets; a parallel process to mitigate the inclusion by chance of target samples into clutter background classes during random sampling; and a fusion of results at the end. In this paper, we demonstrate the effectiveness and robustness of this approach on different scenarios using hyperspectral imagery, where for most of these scenarios, the parameter settings were fixed. We also investigated the performance of this suite over different times of the day, where the spectral signatures of materials varied with relation to diurnal changes during the course of the day. Both visible to near infrared and longwave imagery are used in this study.

Most hyperspectral (HS) anomaly detectors in the literature have been evaluated using a few HS imagery sets to estimate the well-known ROC curve. Although this evaluation approach can be helpful in assessing detectors' rates of correct detection and false alarm on a limited dataset, it does not shed lights on reasons for these detectors' strengths and weaknesses using a significantly larger sample size. This paper discusses a more rigorous approach to testing and comparing HS anomaly detectors, and it is intended to serve as a guide for such a task. Using randomly generated samples, the approach introduces hypothesis tests for two idealized homogeneous sample experiments, where model parameters can vary the difficulty level of these tests. These simulation experiments are devised to address a more generalized concern, i.e., the expected degradation of correct detection as a function of increasing noise in the alternative hypothesis.

Kernel-based approaches have recently drawn considerable interests in hyperspectral image analysis due to its ability in expanding features to a higher dimensional space via a nonlinear mapping function. Many well-known detection and classification techniques such as Orthogonal Subspace Projection (OSP), RX algorithm, linear discriminant analysis, Principal Components Analysis (PCA), Independent Component Analysis (ICA), have been extended to the corresponding kernel versions. Interestingly, a target detection method, called Constrained Energy Minimization (CEM) which has been also widely used in hyperspectral target detection has not been extended to its kernel version. This paper investigates a kernel-based CEM, called Kernel CEM (K-CEM) which employs various kernels to expand the original data space to a higher dimensional feature space that CEM can be operated on. Experiments are conducted to perform a comparative analysis and study between CEM and K-CEM. The results do not show K-CEM provided significant improvement over CEM in detecting hyperspectral targets but does show significant improvement in detecting targets in multispectral imagery which provides limited spectral information for the CEM to work well.

A method for trace gas detection in hyperspectral data is demonstrated using the wavelet packet transform. This new method, the Wavelet Packet Subspace (WPS), applies the wavelet packet transform and selects a best basis for pattern matching. The wavelet packet transform is an extension of the wavelet transform, which fully decomposes a signal into a library of wavelet packet bases. Application of the wavelet packet transform to hyperspectral data for the detection of trace gases takes advantage of the ability of the wavelet transform to locate spectral features in both scale and location. By analyzing the wavelet packet tree of specific gas, nodes of the tree are selected which represent an orthogonal best basis. The best basis represents the significant spectral features of that gas. This is then used to identify pixels in the scene using existing matching algorithms such as spectral angle or matched filter. Using data from the Airborne Hyperspectral Imager (AHI), this method is compared to traditional matched filter detection methods. Initial results demonstrate a promising wavelet packet subspace technique for hyperspectral trace gas detection applications.

The long-wave infrared (LWIR) hyperpectral sensing modality is one that is often used for the problem of detection and identification of chemical warfare agents (CWA) which apply to both military and civilian situations. The inherent nature and complexity of background clutter dictates a need for sophisticated and robust statistical models which are then used in the design of optimum signal processing algorithms that then provide the best exploitation of hyperspectral data to ultimately make decisions on the absence or presence of potentially harmful CWAs. This paper describes the basic elements of an automated signal processing pipeline developed at MIT Lincoln Laboratory. In addition to describing this signal processing architecture in detail, we briefly describe the key signal models that form the foundation of these algorithms as well as some spatial processing techniques used for false alarm mitigation. Finally, we apply this processing pipeline to real data measured by the Telops FIRST hyperspectral (FIRST) sensor to demonstrate its practical utility for the user community.

A processing methodology based on Support Vector Machines is presented in this paper for the classification of hyperspectral spectroscopic images. The accurate classification of the images is used to perform on-line material identification in industrial environments. Each hyperspectral image consists of the diffuse reflectance of the material under study along all the points of a line of vision. These images are measured through the employment of two imaging spectrographs operating at Vis-NIR, from 400 to 1000 nm, and NIR, from 1000 to 2400 nm, ranges of the spectrum, respectively. The aim of this work is to demonstrate the robustness of Support Vector Machines to recognise certain spectral features of the target. Furthermore, research has been made to find the adequate SVM configuration for this hyperspectral application. In this way, anomaly detection and material identification can be efficiently performed. A classifier with a combination of a Gaussian Kernel and a non linear Principal Component Analysis, namely k-PCA is concluded as the best option in this particular case. Finally, experimental tests have been carried out with materials typical of the tobacco industry (tobacco leaves mixed with unwanted spurious materials, such as leathers, plastics, etc.) to demonstrate the suitability of the proposed technique.

We describe a novel approach to produce color composite images from hyperspectral data using weighted spectra averages. The weighted average is based on a sequence of numbers (weights) selected using pixel value information and interband distance. Separate sequences of weights are generated for each of the three color bands forming the color composite image. Tuning of the weighting parameters and emphasis on different spectral areas allows for emphasis of one or other feature in the image. The produced image is a distinct approach from a regular color composite result, since all the bands provide information to the final result. The algorithm was implemented in high level programming language and provided with a user friendly graphical interface. The current design allows for stand-alone usage or for further modifications into a real time visualization module. Experimental results show that the weighted color composition is an extremely fast visualization tool.