This PDF file contains the front matter associated with SPIE Proceedings Volume 8048, including the Title Page, Copyright information, Table of Contents, and Conference Committee listing.

The purpose of this paper is to introduce a general type of detection fusion that allows combining a set of basic detectors into one, more versatile, detector. The fusion can be performed based on the spectral information contained in a pixel, global characteristics of the background and target spaces, as well as spatial local information. The new approach shown in this paper is especially promising in the context of recent geometric and topological approaches that produce complex structures for the background and target spaces. We show specific examples of generalized fusion and present some results on false alarm rates and probabilities of detection of fused detectors. We show that continuum fusion is a special case of generalized fusion. Our new framework allows better understanding of continuum fusion, as well as other useful types of fusion, such as discrete fusion proposed in this paper. We also explain the relationship between the generalized likelihood-ratio detectors and various fusion detectors.

Continuum fusion methods define a new design approach for multivariate detection algorithms, hyperspectral applications being only one example. However, the high dimensions in which such detectors operate can challenge human intuition. We show how certain low-dimensional representations can be used to understand the performance of many existing discrimination algorithms, with special emphasis on newer CF methods. We also give examples illustrating how the interplay between analytical and geometrical interpretations can be used to inform the process of designing special purpose detectors, such as for eliminating sensor artifacts.

The potential of a new class of detection algorithms is demonstrated on an object of practical interest. The continuum fusion (CF) [1] methodology is applied to a linear subspace model. A new algorithm results from first invoking a fusion interpretation of a conventional GLR test and then modifying it with CF methods. Usual performance is enhanced in two ways. First the Gaussian clutter model is replaced by a Laplacian distribution, which is not only more realistic in its tail behavior but, when used in a hypothesis test, also creates decision surfaces more selective than the hyperplanes associated with linear matched filters. Second, a fusion flavor is devised that generalizes the adaptive coherence estimator (ACE) [2, 3] algorithm but has more design flexibility. An IDL/ENVI user interface has been developed and will be described.

Hyperspectral imaging is particular useful in remote sensing to identify a small number of unknown man-made objects in a large natural background. An algorithm for detecting such anomalies in hyperspectral imagery is developed in this article. The pixel from a data cube is modeled as the sum of a linear combination of unknown random variables from the clutter subspace and a residual. Maximum likelihood estimation is used to estimate the coecients of the linear combination and covariance matrix of the residual. The Mahalanobis distance of the residual is dened as the anomaly detector. Experimental results obtained using a hyperspectral data cube with wavelengths in the visible and near-infrared range are presented.

A new method for hyperspectral change detection derived from a parametric radiative transfer model was recently developed. The model-based approach explicitly accounts for local illumination variations, such as shadows, which act as a constant source of false alarms in traditional change detection techniques. Here we formally derive the model-based approach as a generalized likelihood ratio test (GLRT) developed from the data model. Additionally, we discuss variations on implementation techniques for the algorithm and provide results using tower-based data and HYDICE data.

The challenge of finding small targets in big images lies in the characterization of the background clutter. The more homogeneous the background, the more distinguishable a typical target will be from its background. One way to homogenize the background is to segment the image into distinct regions, each of which is individually homogeneous, and then to treat each region separately. In this paper we will report on experiments in which the target is unspecified (it is an anomaly), and various segmentation strategies are employed, including an adaptive hierarchical tree-based scheme. We find that segmentations that employ overlap achieve better performance in the low false alarm rate regime.

Change detection with application to wide-area search seeks to identify where interesting activity has occurred between two images. Since there are many different classes of change, one metric may miss a particular type of change. Therefore, it is potentially beneficial to select metrics with complementary properties. With this idea in mind, a new change detection scheme was created using mean-shift and outlier-distance metrics. Using these metrics in combination should identify and characterize change more completely than either individually. An algorithm using both metrics was developed and tested using registered sets of multispectral imagery.

Many spectral algorithms that are routinely applied to spectral imagery are based on the following models: statistical, linear mixture, and linear subspace. As a result, assumptions are made about the underlying distribution of the data such as multivariate normality or other geometric restrictions. Here we present a graph based model for spectral data that avoids these restrictive assumptions and apply graph based metrics to quantify certain aspects of the resulting graph. The construction of the spectral graph begins by connecting each pixel to its k-nearest neighbors with an undirected weighted edge. The weight of each edge corresponds to the spectral Euclidean distance between the adjacent pixels. The number of nearest neighbors, k, is chosen such that the graph is connected i.e., there is a path from each pixel x_i to every other. This requirement ensures the existence of inter-cluster connections which will prove vital for our application to change detection. Once the graph is constructed, we calculate a metric called the Normalized Edge Volume (NEV) that describes the internal structural volume based on the vertex connectivity and weighted edges of the graph. Finally, we demonstrate a graph based change detection method that applies this metric.

A Telops Hyper-Cam Fourier-transform spectrometer (IFTS) was used to collect infrared hyper-spectral imagery of the smokestack plume from a coal-burning power facility to assess the influence of turbulence on spectral retrieval of temperature (T) and pollutant concentrations (C_i ). The mid-wave (1.5-5.5 μm) system features a 320x256 InSb focal-plane array with a 326 μrad instantaneous field-of-view (IFOV). The line-of-sight distance to the 76mtall smokestack exit was 350m(11.4 x 11.4 cm² IFOV). Approximately 5000 interferogram cubes were collected in 30 minutes on a 128x128 pixel window corresponding to a spectral resolution of 20 cm^-1. Radiance fluctuations due to plume turbulence were observed on a time scale much shorter than hyper-spectral image acquisition rate, suggesting scene change artifacts (SCA) would be present in the Fourier-transformed spectra. Time-averaging the spectra minimized SCA magnitudes, but accurate T and C_i retrieval requires a priori knowledge of the statistical distribution of temperature and other stochastic flow field parameters. A method of quantile sorting in interferogram space prior to Fourier-transformation is presented and used to identify turbulence throughout the plume. Immediately above the stack exit, T and CO₂ concentration estimates from the median spectrum are 395 K and 6%, respectively, which compare well to in situ measurements. Turbulence is small above the stack exit and introduced systematic errors in T and C_i on the order of 0.5 K and 0.01%, respectively. In some plume locations, turbulent fluctuations introduced errors in T and C_i on the order of 8 K and 1%, respectively. While more complicated radiance fluctuations precluded straightforward retrieval of the temperature probability distribution, the results demonstrate the utility of additional information content associated with multiple interferogram quantiles and suggest IFTS may find use as a tool for non-intrusive flow field analysis.

In the task of automated anomaly detection, it is desirable to find regions within imagery that contain man-made structures or objects. The task of separating these signatures from the scene background and other naturally occurring anomalies can be challenging. This task is even more difficult when the spectral signatures of the man-made objects are designed to closely match the surrounding background. As new sensors emerge that can image both spectrally and polarimetrically, it is possible to utilize the polarimetric signature to discriminate between many types of man-made and natural anomalies. One type of passive imaging system that allows for spetro-polarimetric data to be collected is the pairing of a liquid crystal tunable filter (LCTF) with a CCD camera thus creating a spectro-polarimetic imager (SPI). In this paper, an anomaly detection scheme is implemented which makes use of the spectral Stokes imagery collected by this sensing system. The ability for the anomaly detector to find man-made objects is assessed as a function of the number of spectral bands available and it is shown that low false alarm rates can be achieved with relatively few spectral bands.

There are numerous anomaly detection algorithms proposed for hyperspectral imagery. Robust parameter design (RPD) techniques have been applied to some of these algorithms in an attempt to choose robust settings capable of operating consistently across a large variety of image scenes. Typically, training and test sets of hyperspectral images are chosen randomly. Previous research developed a frameworkfor optimizing anomaly detection in HSI by considering specific image characteristics as noise variables within the context of RPD; these characteristics include the Fisher's score, ratio of target pixels and number of clusters. This paper describes a method for selecting hyperspectral image training and test subsets yielding consistent RPD results based on these noise features. These subsets are not necessarily orthogonal, but still provide improvements over random training and test subset assignments by maximizing the volume and average distance between image noise characteristics. Several different mathematical models representing the value of a training and test set based on such measures as the D-optimal score and various distance norms are tested in a simulation experiment.

In this paper we present a new methodology for automated target detection and identification in hyperspectral imagery. The standard paradigm for target detection in hyperspectral imagery is to run a detection algorithm, typically statistical in nature, and visually inspect each high-scoring pixel to decide whether it is a true detection or a false alarm. Detection filters have constant false alarm rates (CFARs) approaching 10^-5, but these can still result in a large number of false alarms given multiple images and a large number of target materials. Here we introduce a new methodology for target detection and identification in hyperspectral imagery that shows promise for hard targets. The result is a greatly reduced false alarm rate and a practical methodology for aiding an analyst in quantitatively evaluating detected pixels. We demonstrate the utility of the method with results on data from the HyMap sensor over the Cooke City, MT.

Analyzing flow-like patterns in images for image understanding is an active research area but there have been much less attention paid to the process of enhancement of those structures. The completion of interrupted lines or the enhancement of flow-like structures is known as Coherence-Enhancement (CE). In this work, we are studying nonlinear anisotropic diffusion filtering for coherence enhancement. Anisotropic diffusion is commonly used for edge enhancement by inhibiting diffusion in the direction of highest spatial fluctuation. For CE, diffusion is promoted along the direction of lowest spatial fluctuation in a neighborhood thereby taking into account how strongly the local gradient of the structures in the image is biased towards that direction. Results of CE applied multispectral and hyperspectral images are presented.

The small island nation of Haiti was devastated in early 2010 following a massive 7.0 earthquake that brought about widespread destruction of infrastructure, many deaths and large-scale displacement of the population in the nation's major cities. The World Bank and ImageCat, Inc tasked the Rochester Institute of Technology's (RIT) Wildfire Airborne Sensor Platform (WASP) to gather a multi-spectral and multi-modal assessment of the disaster over a seven-day period to be used for relief and reconstruction efforts. Traditionally, private sector aerial remote sensing platforms work on processing and product delivery timelines measured in days, a scenario that has the potential to reduce the value of the data in time-sensitive situations such as those found in responding to a disaster. This paper will describe the methodologies and practices used by RIT to deliver an open set of products typically within a twenty-four hour period from when they were initially collected. Response to the Haiti disaster can be broken down into four major sections: 1) data collection and logistics, 2) transmission of raw data from a remote location to a central processing and dissemination location, 3) rapid image processing of a massive amount of raw data, and 4) dissemination of processed data to global organizations utilizing it to provide the maximum benefit. Each section required it's own major effort to ensure the success of the overall mission. A discussion of each section will be provided along with an analysis of methods that could be implemented in future exercises to increase efficiency and effectiveness.

The Information Products Laboratory for Emergency Response (IPLER) is a new initiative led by the Rochester Institute of Technology (RIT) to develop and put into use new information products and tools derived from remote sensing data. This effort involves technical development and outreach to the user community having the two-fold objective of providing new information tools to enhance public safety and fostering economic development. Specifically, this paper addresses the demonstration of the collection and delivery of geo-referenced overhead imagery to local (county level) emergency managers in near realtime. The demonstration proved valuable to county personnel in showing what is possible and valuable to the researchers in highlighting the very real constraints of operatives in local government. The demonstration consisted of four major elements; 1) a multiband imaging system incorporating 4 cameras operating simultaneously in the visible (color), shortwave infrared, midwave infrared and long wave infrared, 2) an on-board inertial navigation and data processing system that renders the imagery into geo-referenced coordinates, 3) a microwave digital downlink, and 4) a data dissemination service via FTP and WMS-based browser. In this particular exercise, we successfully collected and downloaded over 700 images and delivered them to county servers located in their Emergency Operations Center as well as to a remote GIS van.

On April 28, 2010, the Environmental Protection Agency's (EPA) Airborne Spectral Photometric Environmental Collection Technology (ASPECT) aircraft was deployed to Gulfport, Mississippi to provide airborne remotely sensed air monitoring and situational awareness data and products in response to the Deepwater Horizon oil spill disaster. The ASPECT aircraft was released from service on August 9, 2010 after having flown over 85 missions that included over 325 hours of flight operation. This paper describes several advanced analysis capabilities specifically developed for the Deepwater Horizon mission to correctly locate, identify, characterize, and quantify surface oil using ASPECT's multispectral infrared data. The data products produced using these advanced analysis capabilities provided the Deepwater Horizon Incident Command with a capability that significantly increased the effectiveness of skimmer vessel oil recovery efforts directed by the U.S. Coast Guard, and were considered by the Incident Command as key situational awareness information.

The conservation and efficient use of natural and especially strategic resources like oil and water have become global issues, which increasingly initiate environmental and political activities for comprehensive recycling programs. To effectively reutilize oil-based materials necessary in many industrial fields (e.g. chemical and pharmaceutical industry, automotive, packaging), appropriate methods for a fast and highly reliable automated material identification are required. One non-contacting, color- and shape-independent new technique that eliminates the shortcomings of existing methods is to label materials like plastics with certain combinations of fluorescent markers ("optical codes", "optical fingerprints") incorporated during manufacture. Since time-resolved measurements are complex (and expensive), fluorescent markers must be designed that possess unique spectral signatures. The number of identifiable materials increases with the number of fluorescent markers that can be reliably distinguished within the limited wavelength band available. In this article we shall investigate the reliable detection and classification of fluorescent markers with specific fluorescence emission spectra. These simulated spectra are modeled based on realistic fluorescence spectra acquired from material samples using a modern VNIR spectral imaging system. In order to maximize the number of materials that can be reliably identified, we evaluate the performance of 8 classification algorithms based on different spectral similarity measures. The results help guide the design of appropriate fluorescent markers, optical sensors and the overall measurement system.

This paper presents the Image Mapping Spectrometry a new snapshot hyperspectral imaging platform for variety of applications. These applications span from remote sensing and surveillance use to life cell microscopy implementations and medical diagnostics. The IMS replaces the camera in a digital imaging system, allowing one to add parallel spectrum acquisition capability and to maximize the signal collection (> 80%). As such the IMS allows obtaining full spectral information in the image scene instantaneously at real time imaging rates. Presented implemention provides 350x350x48 datacube (x,y,λ) and spectral sampling of 2 to 6 nm in visible spectral range but is easily expandable to larger cube dimensions and other spectral ranges. The operation of the IMS is based on redirecting image zones through the use of a custom-fabricated optical element known as an image mapper. The image mapper is a complex custom optical component comprised of high quality, thin mirror facets with unique 2D tilts. These mirror facets reorganize the original image onto a single large format CCD sensor to create optically "dark" regions between adjacent image lines. The full spectrum from each image line is subsequently dispersed into the void regions on the CCD camera. This mapping method provides a one-to-one correspondence between each voxel in the datacube and pixel on the CCD camera requiring only a simple and fast remapping algorithm. This paper provides fundamentals of IMS operations and describes an example design. Preliminary imaging results for gas detection acquired at 3 frames / second, for 350x350x48 data cubes are being presented. Real time unmixing of spectral signatures is also being discussed. Finally paper draws perspective of future directions and system potential for infrared imaging.

We demonstrate a Fourier transform spectrometer (FTS) using a Fabry-Perot interferometer with the gap between its partially reflecting layers varying orthogonal to the optical axis to produce a gradient in optical path difference at a detector. The gradient produces a period fringe pattern that can be analyzed with standard FTS techniques. Experiments in the visible and IR demonstrate the feasibility of this method for spectroscopy.

The EMAS-HS or Enhanced MODIS Airborne Simulator is an upgrade to the solar reflected and thermal infrared channels of NASA's MODIS Airborne Simulator (MAS). In the solar reflected bands, the MAS scanner functionality will be augmented with the addition of this separate pushbroom hyperspectral instrument. As well as increasing the spectral resolution of MAS beyond 10 nm, this spectrometer is designed to maintain a stable calibration that can be transferred to the existing MAS sensor. The design emphasizes environmental control and on-board radiometric stability monitoring. The system is designed for high-altitude missions on the ER-2 and the Global Hawk platforms. System trades optimize performance in MODIS spectral bands that support land, cloud, aerosol, and atmospheric water studies. The primary science mission driving the development is high altitude cloud imaging, with secondary missions possible for ocean color. The sensor uses two Offner spectrometers to cover the 380-2400 nm spectral range. It features an all-reflective telescope with a 50° full field-of-view. A dichroic cold mirror will split the image from the telescope, with longer radiation transmitted to the SWIR spectrometer. The VNIR spectrometer uses a TE-cooled Si CCD detector that samples the spectrum at 2.5 nm intervals, while the SWIR spectrometer uses a Stirling-cooled hybrid HgCdTe detector to sample the spectrum at 10 nm per band. Both spectrometers will feature 1.05 mRad instantaneous fields-of-view registered to the MAS scanner IFOV's.

A standing wave spectrometer is turned into a wavelength tunable band-pass filter by the addition of a reflective coating. It results in the standing wave filter (SWF), a miniaturized Fabry-Perot band-pass filter with a semi-transparent detector that can be constructed into a pixel-tunable focal plane array, suitable for hyperspectral imaging applications. The asymmetric Fabry-Perot cavity is formed between the reflective coating and a tunable mirror, originally part of the spectrometer. The predicted performance of the SWF is optimized through modeling based on the matrix formalism used in thin film optics and with FDTD simulations. The SWF concept is taken from an ideal device to a focal plane array design that was fabricated with 40 micron pixels using semi-conductor processing technology. First-light spectra measured from the 100 pixel Standing Wave Filter array agree with predictions and prove the concept.

Current development of optical sensors has lead to their increased utility and potential. Applications for these imagers encompass not just single regions of the electromagnetic spectrum but indeed all parts of the thermal radiation spectrum, ultraviolet through long-wave infrared, indicative for instance of Earth's atmosphere. Accordingly, these multispectral imagers mandate the development of entirely new test methods and test hardware to measure and calibrate the benchmarks of their performance; such as SNR, uniformity, sensitivity, linearity, and dynamic range. The role of the test hardware is thus driven not only to provide high-resolution, uniform, and stable output but also to provide multispectral output to minimize the amount of measurement equipment required and in order to demonstrate their full functionality. Multispectral imagers require that test hardware be capable of producing an output that matches high daylight down through low light/starlight irradiance levels. This paper explores the characterization, testing, and advantages and drawbacks of various types of multispectral sources spanning UV through SWIR over a high dynamic range of output.

With the advent of the commercial 3D video card in the mid 1990s, we have seen an order of magnitude performance increase with each generation of new video cards. While these cards were designed primarily for visualization and video games, it became apparent after a short while that they could be used for scientific purposes. These Graphical Processing Units (GPUs) are rapidly being incorporated into data processing tasks usually reserved for general purpose computers. It has been found that many image processing problems scale well to modern GPU systems. We have implemented four popular hyperspectral processing algorithms (N-FINDR, linear unmixing, Principal Components, and the RX anomaly detection algorithm). These algorithms show an across the board speedup of at least a factor of 10, with some special cases showing extreme speedups of a hundred times or more.

A significant topic in many image processing systems is the derivation of a threshold to actuate the automated analysis of outputs from spectral filters and/or anomaly filters, the detection of targets and/or classes of objects which are different than the local background clutter. There are cases where the signals of interest have contrast locally against their immediate surroundings but the application of a global threshold over the entire image produces poor results with missed detections and numerous false alarms. In such cases an adaptive or local threshold operator offers a more robust solution. One local threshold function is the conditional dilation which produces a reference image via a series of dilations which are conditioned on not exceeding the signal levels in the original image. In the limit this reference image becomes a threshold surface where only areas or objects exhibiting contrast locally remain after application of the threshold. Algorithms have been introduced which enable use of conditional dilation in realtime systems by reducing the unbounded series of dilations to a small, fixed number of operations. In the present work we present an adaptation of this algorithm to both single CPU systems and also to systems which incorporate a GPGPU device which enables a highly parallel version of the algorithm subject to the unique architecture constraints of the GPGPU. Execution timings for comparison are introduced: The GPGPU offers somewhat better performance than the single CPU system despite the GPGPU architecture not being suitable for implementation of a neighborhood process.

Hyperspectral sensors can collect hundreds of images taken at different narrow and contiguously spaced spectral bands. This high-resolution spectral information can be used to identify materials and objects within the field of view of the sensor by their spectral signature, but this process may be computationally intensive due to the large data sizes generated by the hyperspectral sensors, typically hundreds of megabytes. This can be an important limitation for some applications where the detection process must be performed in real time (surveillance, explosive detection, etc.). In this work, we developed a parallel implementation of three state-ofthe- art target detection algorithms (RX algorithm, matched filter and adaptive matched subspace detector) using a graphics processing unit (GPU) based on the NVIDIA® CUDA™ architecture. In addition, a multi-core CPUbased implementation of each algorithm was developed to be used as a baseline for the speedups estimation. We evaluated the performance of the GPU-based implementations using an NVIDIA ® Tesla® C1060 GPU card, and the detection accuracy of the implemented algorithms was evaluated using a set of phantom images simulating traces of different materials on clothing. We achieved a maximum speedup in the GPU implementations of around 20x over a multicore CPU-based implementation, which suggests that applications for real-time detection of targets in HSI can greatly benefit from the performance of GPUs as processing hardware.

Manifold learning, also called nonlinear dimensionality reduction, affords a way to understand and visualize the structure of nonlinear hyperspectral datasets. These methods use graphs to represent the manifold topology, and use metrics like geodesic distance, allowing embedding higher dimension objects into lower dimension. However the complexities of some manifold learning algorithms are O(N³), therefore they are very slow (high computational algorithms). In this paper we present a CUDA-based parallel implementation of the three most popular manifold learning algorithms like Isomap, Locally linear embedding, and Laplacian eigenmaps, using CUDA multi-thread model. The result of this dimensionality reduction was employed in segmentation using active contours as an application of these reduced hyperspectral images. The manifold learning algorithms were implemented on a 64-bit workstation equipped with a quad-core Intel® Xeon with 12 GB RAM and two NVIDIA Tesla C1060 GPU cards. Manifold learning outperforms significantly and achieve up to 26x speedup. It also shows good scalability where varying the size of the dataset and the number of K nearest neighbors.

The paper describes the georeferencing part of an airborne hyperspectral imaging system based on pushbroom scanning. Using ray-tracing methods from computer graphics and a highly efficient representation of the digital elevation model (DEM), georeferencing of high resolution pushbroom images runs in real time by a large margin. By adapting the georeferencing to match the DEM resolution, the camera field of view and the flight altitude, the method has potential to provide real time georeferencing, even for HD video on a high resolution DEM when a graphics processing unit (GPU) is used for processing.

"ProSpecTIR" Imaging spectrometer (hyperspectral imagery or "HSI") data were collected for the city of Las Vegas, Nevada, USA at 10:55 PM July 28, 2009 for the purposes of identification, characterization, and mapping of urban lighting based on spectral emission lines unique to specific lighting types. The ProSpecTIR sensor measures the spectrum in 360 spectral bands between 0.4 and 2.5 micrometers at approximately 5nm spectral resolution, and for this flight, at 1.2m spatial resolution. Spectral features were extracted from the data and compared to a spectral library of known lighting measurements. Specific lighting types identified based on spectral signatures using the ProSpecTIR data included blue and red neon, high pressure sodium, and metal halide lights. A binary encoding method was used to map the spatial distribution of lighting types based on simplified spectral signatures. Results were overlain on a Quickbird panchromatic 0.6m spatial resolution image. The observed locations of specific light types were compared to a 3-D Las Vegas building model, and airborne signatures validated against spectral library measurements. The ProSpecTIR data successfully identified and mapped different lighting types and distributions, allowing determination of the nature and spatial associations of specific lights. Results illustrate the potential for using imaging spectrometer data to characterize urban development.

Understanding the capabilities of satellite sensors with spatial and spectral characteristics similar to those of MODIS for Maritime Domain Awareness (MDA) is of importance because of the upcoming NPOES with 100 minutes revisit time carrying the MODIS-like VIIRS multispectral imaging sensor. This paper presents an experimental study of ship detection using MODIS imagery. We study the use of ship signatures such as contaminant plumes in clouds and the spectral contrast between the ship and the sea background for detection. Results show the potential and challenges for such approach in MDA.

This paper presents a new method for finding the direction of a dust storm in satellite images including the 5-band NOAA-AVHRR imagery that were used in our previous work. The previous methods for obtaining the prominent direction of the dust storms involved the combination of edge detectors and local spectral-domain classification techniques applied to subimages/blocks. These approaches produced promising results but have the limitation of not providing consistent results among the subimages that overlap the dust storm region. In this paper, other algorithms like wavelets and state-of-the-art directional filters, based on the contourlet transform, are used to help us determine the direction with more precision and consistency among the relevant subimages. Before applying the directional filtering to the candidate region of the multispectral image, a preprocessing step involves passing the image through a nonsubsampled pyramid selective amplification, this preprocessing step is required in order to enhance the image and improve its directional streaks, in turn, this will help improve the performance of the directional filter to get better and more consistent results. For AVHRR images, our methodology involves applying directional filtering on bands 4 or 5 since these wavelengths highlight the absorption and subsequent emission of thermal radiation by the silicate particles in the dust storms. Directional filtering is applied on these image bands at different angles where energy measurements are computed to find the prominent direction of the dust storm. The presence of a prominent direction in the texture of the candidate region of the dust storm can be used as a verification of its presence in an automated detection system.

Worldview-2 imagery acquired over Duck, NC and Camp Pendleton, CA were analyzed to extract Bidirectional Reflectance Distribution Functions (BRDF) for 8 visible/near-infrared spectral bands. Images were acquired at 15 azimuth/elevation positions at ten-second intervals during the Duck, NC orbit pass. Ten images were acquired over Camp Pendleton, CA. Orthoready images were coregistered using first-order polynomials for the two image sequences. BRDF profiles have been created for various scene elements. MODTRAN simulations are presented to illustrate atmospheric effects under varying collection geometries. Results from analysis of the Camp Pendleton, CA data are presented here.

This paper describes a novel approach for the detection and classification of man-made objects using discriminating features derived from higher-order spectra (HOS), defined in terms of higher-order moments of hyperspectral-signals. Many existing hyperspectral analysis techniques are based on linearity assumptions. However, recent research suggests that significant nonlinearity arises due to multipath scatter, as well as spatially varying atmospheric water vapor concentrations. Higher-order spectra characterize subtle complex nonlinear dependencies in spectral phenomenology of objects in hyperspectral data and are insensitive to additive Gaussian noise. By exploiting these HOS properties, we have devised a robust method for classifying man-made objects from hyerspectral signatures despite the presence of strong background noise, confusers with spectrally similar signatures and variable signal-to-noise ratios. We tested classification performance hyperspectral imagery collected from several different sensor platforms and compared our algorithm with conventional classifiers based on linear models. Our experimental results demonstrate that our HOS algorithm produces significant reductions in false alarms. Furthermore, when HOS-based features were combined with standard features derived from spectral properties, the overall classification accuracy is substantially improved.

In this paper, we present a new approach to filtering high spatial resolution multispectral (MSI) or hyperspectral imagery (HSI) for the purpose of classification and segmentation. Our approach is inspired by the bilateral filtering method that smooths images while preserving important edges for gray-scale and color images. To achieve a similar goal for MSI/HSI, we build a nonlinear tri-lateral filter that takes into account both spatial and spectral similarities. Our approach works on a pixel by pixel basis; the spectrum of each pixel in the filtered image is the combination of the spectra of its adjacent pixels in the original image weighted by the three factors: geometric closeness, spectral Euclidean distance and spectral angle separation. The approach reduces small clutter across the image while keeping edges with strong contrast. The improvement of our method is that we use both spectral intensity differences together with spectral angle separation as the closeness metric, thus preserving edges caused both by material as well as by similar materials with intensity differences. A k-means classifier is applied to the filtered image and the results show our approach can produce a much less cluttered class map. Results will be shown using imagery from the Digital Globe Worldview-2 multispectral sensor and the HYDICE hyperspectral sensor. This approach could also be expanded to facilitate feature extraction from MSI/HSI.

Automatic clustering of spectral image data is a common problem with a diverse set of desired and potential solutions. While typical clustering techniques use first order statistics and Gaussian models, the method described in this paper utilizes the spectral data structure to generate a graph representation of the image and then clusters the data by applying the method of optimal modularity for finding communities within the graph. After defining and identifying pixel adjacencies to represent an image as an adjacency matrix, a recursive splitting is performed to group spectrally similar pixels using the method of modularity maximization. The careful selection of pixel adjacencies determines the success of this spectral clustering technique. The modularity maximization process uses the eigenvector of the modularity matrix with the largest positive eigenvalue to split groups of pixels with non-linear decision surfaces and uses the modularity measure to help estimate the optimal number of clusters to best characterize the data. Using information from each recursion, the end result is a variable level of detail cluster map that is more visually useful than previous methods. Additionally, this method outperforms many typical automatic clustering methods such k-means, especially in highly cluttered urban scenes. The optimal modularity technique hierarchically clusters spectral image data and produces results that more reliably characterize the number of clusters in the data than common automatic spectral image clustering techniques.

The current extent of publicly available space-based imagery and data products is unprecedented. Data from research missions and operational environmental programs provide a wealth of information to global users, and in many cases, the data are accessible in near real-time. The availability of such data provides a unique opportunity to investigate how information can be cascaded through multiple spatial, spectral, radiometric, and temporal scales. A hierarchical image classification approach is developed using multispectral data sources to rapidly produce large area landuse identification and change detection products. The approach derives training pixels from a coarser resolution classification product to autonomously develop a classification map at improved resolution. The methodology also accommodates parallel processing to facilitate analysis of large amounts of data. Previous work successfully demonstrated this approach using a global MODIS 500 m landuse product to construct a 30 m Landsat-based classification map. This effort extends the previous approach to high resolution U.S. commercial satellite imagery. An initial validation study is performed to document the performance of the algorithm and identify limitations in the process. Results indicate this approach is scalable and has broad applications to target and anomaly detection applications. In addition, discussion is focused on how information is preserved throughout the processing chain, as well as situations where the data integrity could break down. This work is part of a larger effort to deduce practical, innovative, and alternative ways to leverage and exploit the extensive low-resolution global data archives to address relevant civil, environmental, and defense objectives.

The Multi-class Convex-FUMI (Multi-class C-FUMI) method is developed and described. The method is capable of learning prototypes for multiple target classes from hyperspectral imagery. Multi-class C-FUMI is a non-traditional supervised learning method based on the Functions of Multiple Instances (FUMI) concept. The FUMI concept differs significantly from traditional supervised by the assumption that only functions of target patterns are available. Moreover, these functions are likely to involve other non-target patterns. In this paper, data points which are convex combinations of multiple target and several non-target prototypes are considered. Multi-class C-FUMI learns the target and non-target patterns, the number of non-target patterns, and the weights (or proportions) of all the prototypes for each data point. For hyperspectral image analysis, the target and non-target prototypes estimated using Multi-class C-FUMI are the endmembers for the target and non-target (background) materials. For this method, training data need only binary labels indicating whether a data point contains or does not contain some proportion of a target endmember; the specific target proportions for the training data are not needed. After learning the target prototype using the binary-labeled training data, target detection is performed on test data. Results showing sub-pixel target detection on highly mixed simulated hyperspectral data generated from the ASTER spectral library are presented.

The Landsat Data Continuity Mission (LDCM), a partnership between the National Aeronautics and Space Administration (NASA) and the Department of Interior (DOI) / United States Geological Survey (USGS), is scheduled for launch in December, 2012. It will be the eighth mission in the Landsat series. The LDCM instrument payload will consist of the Operational Land Imager (OLI), provided by Ball Aerospace and Technology Corporation (BATC) under contract to NASA and the Thermal Infrared Sensor (TIRS), provided by NASA's Goddard Space Flight Center (GSFC). This paper outlines the present development status of the two instruments.

The Landsat Data Continuity Mission consists of a two-sensor platform with the Operational Land Imager and Thermal Infrared Sensor (TIRS). Much of the success of the Landsat program is the emphasis placed on knowledge of the calibration of the sensors relying on a combination of laboratory, onboard, and vicarious calibration methods. Rigorous attention to NIST-traceability of the radiometric calibration, knowledge of out-of-band spectral response, and characterizing and minimizing stray light should provide sensors that meet the quality of Landsat heritage. Described here are the methods and facilities planned for the calibration of TIRS which is a pushbroom sensor with two spectral bands (10.8 and 12 micrometer) and the spatial resolution 100 m with 185-km swath width. Testing takes place in a vacuum test chamber at NASA GSFC using a recently-developed calibration system based on a 16-aperture black body source to simulate spatial and radiometric sources. A two-axis steering mirror moves the source across the TIRS field while filling the aperture. A flood source fills the full field without requiring movement of beam providing a means to evaluate detector-to-detector response effects. Spectral response of the sensor will be determined using a monochromator source coupled to the calibration system. Knowledge of the source output will be through NIST-traceable thermometers integrated to the blackbody. The description of the calibration system, calibration methodology, and the error budget for the calibration system shows that the required 2% radiometric accuracy for scene temperatures between 260 and 330 K is well within the capabilities of the system.

The Landsat Data Continuity Mission (LDCM) focuses on a next generation global coverage, imaging system to replace the aging Landsat 5 and Landsat 7 systems. The major difference in the new system is the migration from the multi-spectral whiskbroom design employed by the previous generation of sensors to modular focal plane, multi-spectral pushbroom architecture. Further complicating the design shift is that the reflective and thermal acquisition capability is split across two instruments spatially separated on the satellite bus. One of the focuses of the science and engineering teams prior to launch is the ability to provide seamless data continuity with the historic Landsat data archive. Specifically, the challenges of registering and calibrating data from the new system so that long-term science studies are minimally impacted by the change in the system design. In order to provide the science and engineering teams with simulated pre-launch data, an effort was undertaken to create a robust end-to-end model of the LDCM system. The modeling environment is intended to be flexible and incorporate measured data from the actual system components as they were completed and integrated. The output of the modeling environment needs to include not only radiometrically robust imagery, but also the meta-data necessary to exercise the processing pipeline. This paper describes how the Digital Imaging and Remote Sensing Image Generation (DIRSIG) model has been utilized to model space-based, multi-spectral imaging (MSI) systems in support of systems engineering trade studies. A mechanism to incorporate measured focal plane projections through the forward optics is described. A hierarchal description of the satellite system is presented including the details of how a multiple instrument platform is described and modeled, including the hierarchical management of temporally correlated jitter that allows engineers to explore impacts of different jitter sources on instrument-to-instrument and band-to-band registration. The capabilities of a new, non-imaging instrument to simulate the measurement of platform ephemeris is also introduced. Finally, the geometric and radiometric foundations for modeling clouds in the DIRSIG model will be described and demonstrated as one of the more significant challenges in registering multi-spectral pushbroom sensor data products.

The Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) are two new sensors being developed by the Landsat Data Continuity Mission (LDCM) that will extend over 35 years of archived Landsat data. In a departure from the whiskbroom design used by all previous generations of Landsat, the LDCM system will employ a pushbroom technology. Although the newly adopted modular array, pushbroom architecture has several advantages over the previous whiskbroom design, registration of the multi-spectral data products is a concern. In this paper, the Digital Imaging and Remote Sensing Image Generation (DIRSIG) tool was used to simulate an LDCM collection, which gives the team access to data that would not otherwise be available prior to launch. The DIRSIG model was used to simulate the two-instrument LDCM payload in order to study the geometric and radiometric impacts of the sensor design on the proposed processing chain. The Lake Tahoe area located in eastern California was chosen for this work because of its dramatic change in elevation, which was ideal for studying the geometric effects of the new Landsat sensor design. Multi-modal datasets were used to create the Lake Tahoe site model for use in DIRSIG. National Elevation Dataset (NED) data were used to create the digital elevation map (DEM) required by DIRSIG, QuickBird data were used to identify different material classes in the scene, and ASTER and Hyperion spectral data were used to assign radiometric properties to those classes. In order to model a realistic Landsat orbit in these simulations, orbital parameters were obtained from a Landsat 7 two-line element set and propagated with the SGP4 orbital position model. Line-of-sight vectors defining how the individual detector elements of the OLI and TIRS instruments project through the optics were measured and provided by NASA. Additionally, the relative spectral response functions for the 9 bands of OLI and the 2 bands of TIRS were measured and provided by NASA. The instruments were offset on the virtual satellite and data recorders used to generate ephemeris data for downstream processing. Finally, potential platform jitter spectra were measured and provided by NASA and incorporated into the simulations. Simulated imagery generated by the model was incrementally provided to the rest of the LDCM team in a spiral development cycle to constantly refine the simulations.

The Thermal Infrared Sensor (TIRS) on board the Landsat Data Continuity Mission (LDCM) is a two-channel, push-broom imager that will continue Landsat thermal band measurements of the Earth. The core of the instrument consists of three Quantum Well Infrared Photodetector (QWIP) arrays whose data are combined to effectively produce a linear array of 1850 pixels for each band with a spatial resolution of approximately 100 meters and a swath width of 185 kilometers. In this push-broom configuration, each pixel may have a slightly different band shape. An on-board blackbody calibrator is used to correct each pixel. However, depending on the scene being observed, striping and other artifacts may still be present in the final data product. The science-focused mission of LDCM requires that these residual effects be understood. The analysis presented here assisted in the selection of the three flight QWIP arrays. Each pixel was scrutinized in terms of its compliance with TIRS spectral requirements. This investigation utilized laboratory spectral measurements of the arrays and filters along with radiometric modeling of the TIRS instrument and environment. These models included standard radiometry equations along with complex physics-based models such as the MODerate spectral resolution TRANsmittance (MODTRAN) and Digital Imaging and Remote Sensing Image Generation (DIRSIG) tools. The laboratory measurements and physics models were used to determine the extent of striping and other spectral artifacts that might be present in the final TIRS data product. The results demonstrate that artifacts caused by the residual pixel-to-pixel spectral non-uniformity are small enough that the data can be expected to meet the TIRS radiometric and image quality requirements.

This work describes numerical methods for the joint reconstruction and segmentation of spectral images taken by compressive sensing coded aperture snapshot spectral imagers (CASSI). In a snapshot, a CASSI captures a two-dimensional (2D) array of measurements that is an encoded representation of both spectral information and 2D spatial information of a scene, resulting in significant savings in acquisition time and data storage. The double disperser coded aperture snapshot imager (DD-CASSI) is able to capture a hyperspectral image from which a highly underdetermined inverse problem is solved for the original hyperspectral cube with regularization terms such as total variation minimization. The reconstruction process decodes the 2D measurements to render a three-dimensional spatio-spectral estimate of the scene, and is therefore an indispensable component of the spectral imager. In this study, we seek a particular form of the compressed sensing solution that assumes spectrally homogeneous segments in the two spatial dimensions, and greatly reduces the number of unknowns. The proposed method generalizes popular active contour segmentation algorithms such as the Chan-Vese model and also enables one to jointly estimate both the segmentation membership functions and the spectral signatures of each segment. The results are illustrated on a simulated Hubble Space Satellite hyperspectral dataset, a real urban hyperspectral dataset, and a real DD-CASSI image in microscopy.

While able to measure the red, green, and blue channels, color imagers are not true spectral imagers capable of spectral measurements. In a previous paper, it was demonstrated that an estimate of a low resolution visible spectra of a naturally illuminated outdoor scene can be estimated from RGB values measured by a conventional color imager. In this paper we present a refined algorithm and document results in a study to estimate visible source spectra from solar illumination scenes using reflectance spectra generated from the USGS data base.

Currently, the MODIS instrument on the Aqua satellite has a number of broken detectors resulting in unreliable data for 1.6 micron band (band 6) measurements. Damaged detectors, transmission errors, and electrical failure are all vexing but seemingly unavoidable problems leading to line drop and data loss. Standard interpolation can often provide an acceptable solution if the loss is sparse. Interpolation, however, introduces a-priori assumptions about the smoothness of the data. When the loss is significant, as it is on MODIS/Aqua, interpolation creates statistically or physically implausible image values and visible artifacts. We have previously developed an algorithm to recreate the missing band 6 data from reliable data in the other 500m bands using a quantitative restoration. Our algorithm uses values in a spectral/spatial neighborhood of the pixel to be estimated, and proposes a value based on training data from the uncorrupted pixels. In this paper, we will present extensions of that algorithm that both improve the performance and robustness of the algorithm. We compare with prior work that just restores band 6 from band 7, and present statistical evidence that data from bands 3, 4, and 5 are also pertinent. We will demonstrate that the increased accuracy from our multi-band statistical estimate has significant consequences at the product level. As an example we show that the restored band 6 has potential benefit to the NASA snow mask for MODIS/Aqua when compared with using band 7 as a replacement for the damaged band 6.

The Advanced Baseline Imager (ABI) on GOES-R will help NOAA's objective of engaging and educating the public on environmental issues by providing near real-time imagery of the earth-atmosphere system. True color satellite images are beneficial to the public, as well as to scientists, who use these images as an important "decision aid" and visualization tool. Unfortunately, ABI only has two visible bands (cyan and red) and does not directly produce the three bands (blue, green, and red) used to create true color imagery. We have developed an algorithm that will produce quantitative true color imagery from ABI. Our algorithm estimates the three tristimulus values of the international standard CIE 1931 XYZ colorspace for each pixel of the ABI image, and thus is compatible with a wide range of software packages and hardware devices. Our algorithm is based on a non-linear statistical regression framework that incorporate both classification and local multispectral regression using training data. We have used training data from the hyper-spectral imager Hyperion. Our algorithm to produce true color images from the ABI is not specific to ABI and may be applicable to other satellites which, like the ABI, do not have the ability to directly produce RGB imagery.

Hyperspectral imaging becomes an important technique that increases the valuable information enclosed within the image. Spectral cube produced by this type of imaging introduces a new material signature known as "spectral signature". This signature is unique for each material as it depends on the molecular composition of the material surface. To produce the spectral cube, a spectrometer should be used in the imagery device to split the electromagnetic energy at different wavelengths before its projection on the imaging array. This spectrometer may be a dispersive element, such as prism and grating, or an electronically tuneable filter. Some of dispersive spectrometers, such as Fourier transform interferometer (FTIR) and image multi-spectral imaging (IMSS), are based on sliding the lenses, or mirrors, along the optical axis which may result in a slightly out-of-focus blurring. Blind deconvolution techniques have been successfully used to decrease this blurring but at the expense of edge sharpening which may be a problem in some applications such as target detection and recognition. In this paper, we introduce a new method to deblurr the hyperspectral images keeping edges as sharp as possible. This is done by firstly detecting the edges locations and then applying a class of morphological filtering. Motivated by the success of threshold decomposition, gradient-based operators are used to detect the locations of these edges followed by an adaptive morphological filter to sharpen these detected edges. Experimental results demonstrate that the performance of the proposed deblurring filter is superior to that of the blind deconvolution methods.

In this paper, sparse kernel-based ensemble learning for hyperspectral anomaly detection is proposed. The proposed technique is aimed to optimize an ensemble of kernel-based one class classifiers, such as Support Vector Data Description (SVDD) classifiers, by estimating optimal sparse weights. In this method, hyperspectral signatures are first randomly sub-sampled into a large number of spectral feature subspaces. An enclosing hypersphere that defines the support of spectral data, corresponding to the normalcy/background data, in the Reproducing Kernel Hilbert Space (RKHS) of each respective feature subspace is then estimated using regular SVDD. The enclosing hypersphere basically represents the spectral characteristics of the background data in the respective feature subspace. The joint hypersphere is learned by optimally combining the hyperspheres from the individual RKHS, while imposing the l1 constraint on the combining weights. The joint hypersphere representing the most optimal compact support of the local hyperspectral data in the joint feature subspaces is then used to test each pixel in hyperspectral image data to determine if it belongs to the local background data or not. The outliers are considered to be targets. The performance comparison between the proposed technique and the regular SVDD is provided using the HYDICE hyperspectral images.

Hyperspectral pixels are acquired in hundreds of narrow and continuous spectral bands, and the hyperspectral data cubes typically contain hundreds of megabytes. Analysis and processing of the high-dimensional hyperspectral data are computationally expensive and memory inefficient. However, there is a large amount of redundancy between neighboring spectral bands and the hyperspectral pixels lie in a much lower dimensional subspace. Therefore, numerous techniques can be applied to reduce the dimensionality while maintaining the structure of the data. This would lead to a significant reduction of the complexity of the imaging system, as well as an improvement of the computational efficiency of the detection algorithms. In this paper, we explore the use of several dimensionality reduction techniques that can be easily integrated into the imaging sensors. We also investigate their effect on the performance of classical target detection techniques for hyperspectral images, including spectral matched filters (SMF), matched subspace detectors (MSD), support vector machines (SVM), and RX anomaly detection algorithm. Specifically, each N-dimensional spectral pixel is embedded to an M-dimensional measurement space with M « N by a linear transformation (e.g., random measurement matrices, uniform downsampling, PCA). The SMF, MSD, SVM, and RX detectors are then applied to the M-dimensional measurement vectors to detect the targets of interests and their detection performances are compared to those obtained from the entire N-dimensional spectrum by the receiver operating characteristics curves. Through extensive experiments on several HSI datasets, we demonstrate that only 1/5 to 1/3 measurements (i.e., the compression ratio M/N is 1/5 ~ 1/3 ) are necessary to achieve detection performance comparable to that obtained by exploiting the full N-dimensional pixels.

The detection of gaseous chemical plumes in long-wave infrared hyperspectral images is often accomplished with algorithms derived from linear radiance models, such as the matched filter. While such algorithms can be highly effective, deviations of the physical radiative transfer process from the idealized linear model can reduce performance. In particular, the steering vector employed in the matched filter will never exactly match the observed plume signature, the estimated background covariance matrix will often suffer some contamination by the plume signature, and the plume and background will typically be spatially correlated to some extent. In combination, these effects can be worse than they are individually. In this paper, we systematically vary these factors to study their impact on detection using a data set of synthetic plumes embedded into measured background data.

The central parameter in the quantification of chemical vapor plumes via remote sensing is the mean concentrationpath length (CL) product, which can lead to estimates of the absolute gas quantity present. The goal of this paper is to derive Cramer-Rao lower bounds on the variance of an unbiased estimator of CL in concert with other parameters of a general non-linear radiance model. These bounds offer a guide to feasibility of CL estimation that is not dependent on any given algorithm. In addition, the derivation of the bounds yields great insight into the physical and phenomenological mechanisms that control plume quantification.

This paper shows how to use a public domain raytracer POV-Ray (Persistence Of Vision Raytracer) to render multiand hyper-spectral scenes. The scripting environment allows automatic changing of the reflectance and transmittance parameters. The radiosity rendering mode allows accurate simulation of multiple-reflections between surfaces and also allows semi-transparent surfaces such as plant leaves. We show that POV-Ray computes occlusion accurately using a test scene with two blocks under a uniform sky. A complex scene representing a plant canopy is generated using a few lines of script. With appropriate rendering settings, shadows cast by leaves are rendered in many bands. Comparing single and multiple reflection renderings, the effect of multiple reflections is clearly visible and accounts for 25% of the overall apparent canopy reflectance in the near infrared.

The utility of a hyperspectral image for target detection can be measured by synthetically implanting target spectra in the image and applying detection algorithms.1 In this paper we apply this method, called the target implant method, for the purpose of determining the top performing algorithms for a given image and given target and for determining the relative difficulty for detection of targets in a given image with a given detector. Our tests include variations on the matched filter, adaptive coherence/cosine estimator and constrained energy minimization detection algorithms. This enables one to predict the fill fraction at which a given target can be detected and the best detection algorithm in a given image under ideal circumstances. Comparison of predictions from this method to detection performance on real target pixels shows that the target implant method does provide accurate relative predictions in terms of both target difficulty and detector performance, but reliably predicting the actual number of false alarms for a given target at a given fill fraction is difficult or impossible. In our tests we used images from the Cooke City Collection^2,3 and from the Forest Radiance Collection.⁴ The Cooke City Collection was taken with the HyMap sensor on July 4, 2006. This imagery has 126 bands ranging from 453.8 to 2496.3 nm at a ground sample distance of approximately 3 meters. Seven flightlines were collected, six of which contain 4 fabric target panels and 3 vehicles with known spectra. The Forest Radiance imagery had 210 spectral bands (145 good bands) ranging from 397.4nm to 2496.5 with a ground sample distance of approximately 1.9 meters.

Data dimensionality (DR) is generally performed by first fixing size of DR at a certain number, say p and then finding a technique to reduce an original data space to a low dimensional data space with dimensionality specified by p. This paper introduces a new concept of dynamic dimensionality reduction (DDR) which considers the parameter p as a variable by varying the value of p to make p adaptive compared to the commonly used DR, referred to as static dimensionality reduction (SDR) with the parameter p fixed at a constant value. In order to materialize the DDR another new concept, referred to as progressive DR (PDR) is also developed so that the DR can be performed progressively to adapt the variable size of data dimensionality determined by varying the value of p. The advantages of the DDR over SDR are demonstrated through experiments conducted for hyperspectral image classification.

Historically, much of spectral image analysis revolves around assumptions of multivariate normality. If the background spectral distribution can be assumed to be multivariate normal, then algorithms for anomaly detection, target detection, and classification can be developed around that assumption. However, as the current generation sensors typically have higher spatial and/or spectral resolution, the spectral distribution complexity of the data collected is increasing and these assumptions are no longer adequate, particularly image-wide. However, large portions of the imagery may be accurately described by a multivariate normal distribution. A new empirical method for assessing the multivariate normality of a hyperspectral distribution is presented here. This method assesses the multivariate normality of individual spectral image tiles and is applied to the large area search problem. Additionally, the methodology is applied to a selection of full hyperspectral data sets for general content evaluation. This information can be used to indicate the degree of multivariate normality (or complexity) of the data or data regions and to determine the appropriate algorithm to use globally or locally for spatially adaptive processing.

Visualization of the high-dimensional data set that makes up hyperspectral images necessitates a dimensionality reduction approach to make that data useful to a human analyst. The expression of spectral data as color images, individual pixel spectra plots, principal component images, and 2D/3D scatter plots of a subset of the data are a few examples of common techniques. However, these approaches leave the user with little ability to intuit knowledge of the full N-dimensional spectral data space or to directly or easily interact with that data. In this work, we look at developing an interactive, intuitive visualization and analysis tool based on using a Poincaré disk as a window into that high dimensional space. The Poincaré disk represents an infinite, two-dimensional hyperbolic space such that distances and areas increase exponentially as you move farther from the center of the disk. By projecting N-dimensional data into this space using a non-linear, yet relative distance metric preserving projection (such as the Sammon projection), we can simultaneously view the entire data set while maintaining natural clustering and spacing. The disk also provides a means to interact with the data; the user is presented with a "fish-eye" view of the space which can be navigated and manipulated with a mouse to "zoom" into clusters of data and to select spectral data points. By coupling this interaction with a synchronous view of the data as a spatial RGB image and the ability to examine individual pixel spectra, the user has full control over the data set for classification, analysis, and instructive use.

Simulated imagery has been and will continue to be a great resource to the remote sensing community. It not only fills in the gaps when real imagery is not available, but allows the user to know and control every aspect of the scene. Over the last 20 years we have seen its value in algorithm development, systems level design trade studies and phenomenology investigation. The realism of this data is often linked to its radiometric accuracy. The Rochester Institute of Technology's Digital Imaging and Remote Sensing (DIRS) Laboratory has done extensive work on making simulations more realistic for years, while developing our in house image generator, DIRSIG. In the past we have invested hundreds of man-hours to painstakingly build large scale scenes of real locations with manual methods. Recently, new procedural tools and open source geometry repositories have allowed the creation of similar scenes with improved scene clutter in significantly less time. It is now possible to assemble and build large city-scale scene geometries with a more automated workflow over the course of a few hours. Even with these advances, an observer viewing these high resolution, complex, spectrally and spatially textured simulated images is still visually aware that they are nothing but simulations, albeit radiometrically and spectrally accurate. This paper will investigate the above concern regarding simulated imagery by looking at the utility, evolution and future of image simulations.

Spectral pixels in a hyperspectral image are known to lie in a low-dimensional subspace. The Linear Mixture Model states that every spectral vector is closely represented by a linear combination of some signatures. When no prior knowledge of the representing signatures available, they must be extracted from the image data, then the abundances of each vector can be determined. The whole process is often referred to as unsupervised endmember extraction and unmixing. The Linear Mixture Model can be extended to Sparse Mixture Model R=MS + N, where not only single pixels but the whole hyperspectral image has a sparse representation using a dictionary M made of the data itself, and the abundance vectors (columns of S) are sparse at the same locations. The endmember extraction and unmixing tasks then can be done concurrently by solving for a row-sparse abundance matrix S. In this paper, we pose a convex optimization problem, then using simultaneous sparse recovery techniques to find S. This approach promise a global optimum solution for the process, rather than suboptimal solutions of iterative methods which extract endmembers one at a time. We use l1l2 norm of S to promote row-sparsity in simultaneous sparse recovery, then impose additional hyperspectral constraints to abundance vectors (such as non-negativity and sum-to-one).

In this paper, we propose a joint sparsity model for target detection in hyperspectral imagery. The key innovative idea here is that hyperspectral pixels within a small neighborhood in the test image can be simultaneously represented by a linear combination of a few common training samples, but weighted with a different set of coefficients for each pixel. The joint sparsity model automatically incorporates the inter-pixel correlation within the hyperspectral imagery by assuming that neighboring spectral pixels usually consists of similar materials. The sparse representations of the neighboring pixels are obtained by simultaneously decomposing the pixels over a given dictionary consisting of training samples of both the target and background classes. The recovered sparse coefficient vectors are then directly used for determining the label of the test pixels. Simulation results on several real hyperspectral images show that the proposed algorithm based on the joint sparsity model outperforms the classical hyperspectral target detection algorithms, such as the popular spectral matched filters, matched subspace detectors, adaptive subspace detectors, as well as binary classifiers such as support vector machines.

Linear spectral unmixing and endmember selection are two of the many tasks that can be accomplished using hyperspectral imagery. The quality of the unmixing results depends on an accurate estimate of the number of endmembers used in the analysis. Too many estimated endmembers produce over fitting of the spectral unmixing results; too few estimated endmembers produce spectral unmixing results with large residual errors. Several statistical and geometrical approaches have been developed to estimate the number of endmembers, but many of these approaches rely on using the global dataset. The global approach does not take into consideration local endmember variability, which is of particular interest in high-spatial resolution imagery. Here, the number of endmembers within local image tiles is estimated by using a novel, spatially adaptive approach. Each pixel is unmixed using the locally identified endmembers and global abundance maps are generated by clustering these locally derived endmembers. Comparisons are made between this new approach and an established global method that uses PCA to estimate the number of endmembers and SMACC to identify the spectra. Multiple images with varying spatial resolution are used in the comparison of methodologies and conclusions are drawn based on per-pixel residual unmixing errors.

Linear Spectral Mixture Analysis (LSMA) is a theory developed to perform spectral unmixing where three major LSMA techniques, Least Squares Orthogonal Subspace Projection (LSOSP), Non-negativity Constrained Least Squares (NCLS) and Fully Constrained Least Squares (FCLS) for this purpose. Later on these three techniques were further extended to Fisher's LSMA (FLSMA), Weighted Abundance Constrained-LSMA (WAC-LSMA) and kernel-based LSMA (KLSMA). This paper combines both approaches of KLSMA and WACLSMA to derive a most general version of LSMA, Kernel-based WACLSMA (KWAC-LSMA) which includes all the above-mentioned LSMAs as its special cases. The utility of the KWAC-LSMA is further demonstrated by multispectral and hyperspectral experiments for performance analysis.

Hyperspectral sensors are delivering a data cube consisting of hundreds of images gathered in adjacent frequency bands. Processing such data requires solutions to handle the computational complexity and the information redundancy. In principle, there are two different approaches deployable. Data compression merges this imagery to some few images. Hereby only the essential information is preserved. Small variations are treated as disturbances and hence removed. Band selection eliminates superfluous bands, leaving the others unmodified. Thus even minor deviations are preserved. In our paper, we present a novel band selection method especially developed for surveillance purposes. Hereby, the capability to detect even small variations poses an essential requirement, only fulfilled by the second approach. The computational complexity and the performance of such an algorithm depend on the available information. If complete knowledge about the targets and the background is available, contrast maximization establishes a perfect band selection. Without any knowledge the selection has to be performed by exploiting the band attributes often resulting in a poor choice. In order to avoid this, the developed algorithm incorporates the accessible information from the monitoring scene. In particular, features (e.g. anomalies) based on proximity relations are extracted in each band. Subsequently, an assessment of their suitability is accomplished by means of the value margins and the associated distributions. The final selection is then based on the inspection of the variations caused by the illumination and other external effects. We demonstrate and evaluate the appropriateness of this new method with a practical example.

In this paper, we propose a denoising method for hyperspectral images using a joint bilateral filter. The joint bilateral filter with the fused image of hyperspectral image bands is applied on the noisy image bands. This fused image is a single grayscale image that is obtained by the weighted summation of hyperspectral image bands. It retains the features and details of each hyperspectral image band. Therefore the joint bilateral filter with the fused image is powerful in reducing noise while preserving the characteristics of the individual spectral bands. We evaluated the performance of the proposed noise reduction method on hyperspectral imaging systems, which we developed for visible and near-infrared spectral regions. Experimental results show that the proposed method outperforms the conventional approaches, such as the basic bilateral filter.

In recent years, miniature spectrometers have been found useful in many applications to resolve spectrum signature of objects or materials. In this paper, algorithms for filter-array spectrum sensor to realize miniature spectrometers are investigated. Conventionally, the filter-array spectrum sensor can be modeled as an over-determined problem, and the spectrum can be reconstructed by solving a set of linear equations. On the contrary, we model the spectrum reconstruction process as an under-determined problem, and bring up the concept of template-selection by sparse representation. L1-minimization algorithm is tested to achieve a high reconstruction resolution. Simulation results show superior quality of spectrum reconstruction can be made possible from this under-determined approach.

Hyperspectral data due to its higher information content afforded by higher spectral resolution is increasingly being used for various remote sensing applications including information extraction at subpixel level. There is however usually a lack of matching fine spatial resolution data particularly for target detection applications. Thus, there always exists a tradeoff between the spectral and spatial resolutions due to considerations of type of application, its cost and other associated analytical and computational complexities. Typically whenever an object, either manmade, natural or any ground cover class (called target, endmembers, components or class) gets spectrally resolved but not spatially, mixed pixels in the image result. Thus, numerous manmade and/or natural disparate substances may occur inside such mixed pixels giving rise to mixed pixel classification or subpixel target detection problems. Various spectral unmixing models such as Linear Mixture Modeling (LMM) are in vogue to recover components of a mixed pixel. Spectral unmixing outputs both the endmember spectrum and their corresponding abundance fractions inside the pixel. It, however, does not provide spatial distribution of these abundance fractions within a pixel. This limits the applicability of hyperspectral data for subpixel target detection. In this paper, a new inverse Euclidean distance based super-resolution mapping method has been presented that achieves subpixel target detection in hyperspectral images by adjusting spatial distribution of abundance fraction within a pixel. Results obtained at different resolutions indicate that super-resolution mapping may effectively aid subpixel target detection.