Hyperspectral sensors produce large quantities of data when operating on uninhabited aerial vehicles (UAV) that can overwhelm available data links. Technical Research Associates, Inc. designed, developed, and implemented a data compression approach that is capable of reducing this data volume by a factor of 100 or more with no loss in the tactical utility of the data. This algorithm, Full Spectrum Wavelet, combines efficient coding of the spectral dimension with a wavelet transformation of the spatial dimension. The approach has been tested on a wide variety of reflection band and thermal band hyperspectral data sets. In addition to such traditional measures as the error introduced by the compression, the performance of the algorithm was evaluated using application-oriented measures such as Receiver Operating Curves (ROC) and terrain categorization maps. Comparisons between these products showed little or no degradation of performance out to compression factors of 100. The evaluation procedure provided results directly relevant to tactical users of the data
There has been interest in overhead tracking of automobiles on our roadways using optical sensors. Tracking of multiple
vehicles can be accomplished with a single band high-resolution sensor as long as the vehicles are continuously in view.
However, in many cases the vehicles pass through or behind blackouts, such as through tunnels or behind tall buildings.
In these cases, the vehicles of interest must be reacquired and recognized from the collection of vehicles present after the
blackout. The approach considered here is to add an additional sensor to assist a single band high-resolution tracking
sensor, where the adjunct sensor measures the vehicle signatures for recognition and reacquisition. The subject of this
paper is the recognition of targets of interest amongst the observed objects and the reacquisition after a blackout. A
Generalized Likelihood Ratio Test (GLRT) algorithm is compared with the Spectral Angle Mapper (SAM) and Euclidian
distance algorithms. All three algorithms were evaluated on a database of signatures created by measuring samples from
old automobile gas doors. The GLRT was the most successful in recognizing the target after a blackout and could
achieve a 95% correct reacquisition rate. The results show the feasibility of using a hyper spectral sensor to assist a multi
target tracking sensor by providing target recognition for reacquisition.
Proc. SPIE. 8048, Algorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XVII
KEYWORDS: Principal component analysis, Independent component analysis, Video acceleration, Detection and tracking algorithms, Visualization, Image processing, Video, Data processing, Video processing, C++
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.
The spectral emissivity of soils in the region of thermal emission from 8 - 14 micrometers is a combination of
the spectral emission of the mineral and other components in the soil, as well as their physical arrangement and the
thermal state of the soil (presence of thermal gradients). In this paper, we will outline the procedure for producing a
spectral model of a mixed soil, and show examples of model soils compared to measured soils with the two major soil
constituents: quartz and clay. The predictions of this theory are then compared to field measurements made with a
LWIR Spectrometer of disturbed and undisturbed soil.
Hyperspectral imagers tend to have lower spatial resolution than multispectral ones. This often results in a (sometimes
difficult) trade-off between spectral and spatial resolution. We have developed a technique, called CRISP, that combines
low-resolution hyperspectral data and high-resolution multispectral data to produce high quality, high-resolution
hyperspectral data. This technique shows good quantitative performance when applied to realistic applications such as
land cover estimation and anomaly detection. As a test of this technique, we have performed an experiment using
HyMap hyperspectral data and multispectral instruments over the coast waters of Oahu, Hawaii. The accuracy of the
CRISP sharpening approach when used for coastal applications such as depth mapping is assessed.
Over the past five years, advances have been made in the spectral detection of surface mines under minefield
detection programs at the U. S. Army RDECOM CERDEC Night Vision and Electronic Sensors Directorate
(NVESD). The problem of detecting surface land mines ranges from the relatively simple, the detection of large
anti-vehicle mines on bare soil, to the very difficult, the detection of anti-personnel mines in thick vegetation.
While spatial and spectral approaches can be applied to the detection of surface mines, spatial-only detection
requires many pixels-on-target such that the mine is actually imaged and shape-based features can be exploited.
This method is unreliable in vegetated areas because only part of the mine may be exposed, while spectral detection
is possible without the mine being resolved. At NVESD, hyperspectral and multi-spectral sensors throughout the
reflection and thermal spectral regimes have been applied to the mine detection problem. Data has been collected
on mines in forest and desert regions and algorithms have been developed both to detect the mines as anomalies and
to detect the mines based on their spectral signature. In addition to the detection of individual mines, algorithms
have been developed to exploit the similarities of mines in a minefield to improve their detection probability. In this
paper, the types of spectral data collected over the past five years will be summarized along with the advances in
We have developed a new and innovative technique for combining a high-spatial-resolution multispectral image with a
lower-spatial-resolution hyperspectral image. The approach, called CRISP, compares the spectral information present
in the multispectral image to the spectral content in the hyperspectral image and derives a set of equations to
approximately transform the multispectral image into a synthetic hyperspectral image. This synthetic hyperspectral
image is then recombined with the original low-spatial-resolution hyperspectral image to produce a sharpened product.
The result is a product that has the spectral properties of the hyperspectral image at a spatial resolution approaching
that of the multispectral image. To test the accuracy of the CRISP method, we applied the method to synthetic data
generated from hyperspectral images acquired with an airborne sensor. These high-spatial-resolution images were used
to generate both a lower-spatial-resolution hyperspectral data set and a four-band multispectral data set. With this
method, it is possible to compare the output of the CRISP process to the 'truth data' (the original scene). In all of these
controlled tests, the CRISP product showed both good spectral and visual fidelity, with an RMS error less than one
percent when compared to the 'truth' image. We then applied the method to real world imagery collected by the
Hyperion sensor on EO-1 as part of the Hurricane Katrina support effort. In addition to multiple Hyperion data sets,
both Ikonos and QuickBird data were also acquired over the New Orleans area. Following registration of the data sets,
multiple high-spatial-resolution CRISP-generated hyperspectral data sets were created. In this paper, we present the
results of this study that shows the utility of the CRISP-sharpened products to form material classification maps at four-meter
resolution from space-based hyperspectral data. These products are compared to the equivalent products
generated from the source 30m resolution Hyperion data.
Over the years several methods have been used to determining the best bands for a visible near IR multi-spectral
sensor. The most popular method, the committee method, places scientists with differing opinions on the phenomena
and the sensor mission in one room, and a compromise set is developed. To avoid this, there have been several
methods to automate this selection process. We have developed a method to examine hyperspectral data to find the
best multi-spectral band set (whether 3, 4, 5 or 6 bands) based on the background, on the premise that, with the target
unknown, the band set that best separates the background materials is the best. We start with a hyperspectral data set
of a background area without any targets. We then run a program for determining the spectral endmembers. Any
endmembers that look like they are due to sensor artifacts or an anomalous point on the ground (junk) are discarded
from the list. The resulting hyperspectral endmembers are then input to an exhaustive search program. The goal of the
exhaustive search is to find a set of N (say 4) multi-spectral bands that maximizes the spectral angles between all of
the endmembers. Thus, at each trial the multi-spectral bands are made by binning the hyperspectral (to four bands in
this case) and the spectral angles calculated between endmembers 1 and 2, 1 and 3, 1 and 4, 2 and 3, 2 and 4 etc. The
endmembers in each case have been binned to four multi-spectral bands. We save the average of these spectral angle
calculations. After examining often millions of combinations, the multi-spectral band set that maximizes the spectral
separation is judged to be the best. We have applied this method to the selection of multi-spectral bands sets for
Creating a minefield requires disturbing the soil. This disturbance alters the soil properties and processes in a measurable way. The U.S. Army is investigating techniques to exploit the altered properties of disturbed soil to assist in the detection of buried landmines. The differential quartz reststrahlen signatures between disturbed and undisturbed soil at the long wave infrared (LWIR) region have shown promise in past field tests.(1,3)We have initiated ground-based measurements using a non-imaging spectral sensor to investigate the phenomenology of LWIR disturbed soil signature. Our primary goal is to develop rainfall-dependent models to predict the degradation of the differential reststrahlen signature for varying soil types. A bare soil test site with strong quartz reststrahlen signature was selected for our initial investigation. The disturbed and undisturbed soil spectral signatures at the LWIR regions were obtained after multiple rain events using a Design and Prototypes field portable Fourier transform infrared (FTIR) spectrometer. The intensity and total amount of rainfall were recorded using a high-resolution tipping-bucket rain gauge. In addition to these measurements, photomicrographs of the disturbed soil were obtained after rainfall events, and X-ray diffraction analyses were conducted to obtain detailed soil mineralogy of the test site. We present these results and discuss the changes in the spectral characteristics of disturbed soil as a function of rainfall amount and intensity.
Hyperspectral imagers tend to have lower spatial resolution than multispectral ones. This often results in a (sometimes difficult) trade-off between spectral and spatial resolution. One means of addressing this spatial/spectral resolution trade-off is to acquire both multispectral and hyperspectral data simultaneously, and then combine the two to produce a hyperspectral image with the high spatial resolution of the multispectral image. This process, called 'sharpening', results in a product that fuses the rich spectral content of a hyperspectral image with the high spatial content of the multispectral image. The approach we have been investigating compares the spectral information present in the multispectral image to the spectral content in the hyperspectral image and derives a set of equations to approximately transform the multispectral image into a synthetic hyperspectral image. This synthetic hyperspectral image is then recombined with the original low-spatial-resolution hyperspectral image to produce a sharpened product. We have evaluated this technique against several types of data for terrain classification and it has demonstrated good performance across all data sets. The spectra predicted by the sharpening algorithm match truth spectra in synthetic image tests, and performance with detection algorithms show little, if any, degradation of detection performance.
Multispectral sharpening of hyperspectral imagery fuses the spectral content of a hyperspectral image with the spatial and spectral content of the multispectral image. The approach we have been investigating compares the spectral information present in the multispectral image to the spectral content in the hyperspectral image and derives a set of equations to approximately transform the multispectral image into a synthetic hyperspectral image. This synthetic hyperspectral image is then recombined with the original low-resolution hyperspectral image to produce a sharpened product. We evaluate this technique against several types of data, showing good performance across with all data sets. Recent improvements in the algorithm allow target detection to be performed without loss of performance even at extreme sharpening ratios.
Hyperspectral imaging is an important technology for the detection of surface and buried land mines from an airborne platform. For this reason, hyperspectral was included with SAR sensors in the two deployments that were executed by the CECOM RDEC Night Vision and Electronic Systems Directorate (NVESD) in Fall 2002 and in Spring 2003. The purpose of these deployments was to bring together a wide variety of airborne sensors for the detection of mines, with well ground-truthed targets. The hyperspectral sensors included the Airborne Hyperspectral Imager (AHI), a University of Hawaii LWIR HSI sensor and the Compact Airborne Spectral Sensor (COMPASS), an NVESD VNIR/SWIR sensor. Both a high frequency SAR and a ground penetrating radar were also flown. These experiments were carried out at sites where an extensive array of buried and surface mines were deployed. At the first location, on the east coast, the mines were deployed against several different backgrounds ranging from bare dirt to long grass. At the second location in the desert southwest, the mines were placed on backgrounds ranging from loose sand to mixed sand and vegetation. The COMPASS and AHI sensors were both placed on the Twin Otter aircraft, and data was collected with the airplane as low as 700 ft and as high as 4000 ft. In this paper, the data collected on surface mines will be reviewed, and specific examples from each background type presented. Spectral detection algorithms will be applied to the data and the results of the algorithm processing will be presented.
Robust, timely, and remote detection of mines and minefields is central to both tactical and humanitarian demining efforts, yet remains elusive for single-sensor systems. Here we present an approach to jointly exploit multisensor data for detection of mines from remotely sensed imagery. LWIR, MWIR, laser, multispectral, and radar sensor have been applied individually to the mine detection and each has shown promise for supporting automated detection. However, none of these sources individually provides a full solution for automated mine detection under all expected mine, background and environmental conditions. Under support from Night Vision and Electronic Sensors Directorate (NVESD) we have developed an approach that, through joint exploitation of multiple sensors, improves detection performance over that achieved from a single sensor. In this paper we describe the joint exploitation method, which is based on fundamental detection theoretic principles, demonstrate the strength of the approach on imagery from minefields, and discuss extensions of the method to additional sensing modalities. The approach uses pre-threshold anomaly detector outputs to formulate accurate models for marginal and joint statistics across multiple detection or sensor features. This joint decision space is modeled and decision boundaries are computed from measured statistics. Since the approach adapts the decision criteria based on the measured statistics and no prior target training information is used, it provides a robust multi-algorithm or multisensor detection statistic. Results from the joint exploitation processing using two different imaging sensors over surface mines acquired by NVESD will be presented to illustrate the process. The potential of the approach to incorporate additional sensor sources, such as radar, multispectral and hyperspectral imagery is also illustrated.
The Compact Airborne Spectral Sensor (COMPASS) has been flying for over a year and has gathered data in support of a variety of missions. While COMPASS is an array imaging spectrometer, the quality of the spectrometer optics and the alignment of the instrument during assembly have removed many of the sources of error often present in array imaging spectrometers, such as spectral band mis-registration, smile and keystone. Since COMPASS has begun flying, we have been studying new procedures for improving the calibration of the COMPASS sensor and array imaging spectrometers, in general. The use of the on-board calibration sources was compared to using a combination of on-board sources and a scene average, and also compared to using laboratory calibration sources. In addition, different methods for finding and removing bad detectors were investigated. The coupling of the bad detector replacement procedure with the flatfielding was also studied. We have found that bracketing the light levels in the scene is the key to reducing the effect of bad detectors. An effective method of bracketing the scene is to use the scene average for each detector as the white and the on-board dark. Alternative methods using multiple white sources are also attractive. Several examples from collected scene data will be presented and evaluated in terms of image quality in particular bands and Principal Components.
The NVESD COMPASS instrument is an airborne dispersive hyperspectral imager that covers the VNIR through SWIR bands and incorporates a real-time data processing system. The processing system consists of a Data Processing Computer (DPC) and an Operator Display/Control Computer (ODC). The high-performance DPC executes real-time sensor calibration and multiple spectral detection algorithms on 13 G4-processors in a Race++ switched backplane. The DPC sends three-band pseudo-color hyperspectral data, high-resolution target chips, and GPS/INS data to the ODC. The ODC outputs a geo-registered display of HSI color imagery with detection cue overlays. The COMPASS detection algorithms, which are particularly well suited to CC&D targeting applications, include the SSRX spectral anomaly detector, the NFINDR/STD spectral unmixing-based anomaly detector, (3) a supervised spectral matched filter (SSMF), and (4) Healey's invariant subspace detector. The DPC airborne component is VME-based in a compact, ruggedized chassis. The COMPASS real-time processor is a second generation system based on NRL-sponsored WarHORSE demonstrations. This paper reviews the DPC system design, capabilities and performance.
More and more hyperspectral sensors are now employing two-dimensional focal plane arrays to simultaneously record the spectra for a line of points on the ground. Since a large number of spectra are obtained simultaneously, the instantaneous data rate can be much higher than that achieved with a flying spot scanner. Unfortunately, the use of more than one detector per band means that there are many new sources of sensor pattern that must be removed during preprocessing. These sources of pattern usually are the limitation to the performance of imaging array spectrometers. One of the more troublesome problems with focal plane arrays is the existence of dead or bad detectors. For an imaging system, the effect of these detectors is removed by interpolation with neighbors. The problem is much more difficult to solve when the array is used as the focal plane in a hyperspectral instrument. If the bad detectors are ignored, the result is a stripe down the image in a particular band. Simple interpolation in the spectral direction can be atempted, but often the interpolation itself is the source of stripes in the image. The effect of inaccurate interpolation is particularly noticeable in the vicinity of atmospheric absorption features, where the spectral variation with wavelength is far from linear. The bad detector is replaced by the average of its two neighboring point in the spectra, which fails to match the proper value for that point. While this value is a better match than the uncorrected detector, the result is still a stripe in the image. This stripe will show up in many hyperspectral analysis operations. One solution is to move the atmospheric compensation step into the preprocessing to remove the rapidly changing spectral features before the bad detectors are removed by interpolation. This would be a rearrangement of the normal division between level 1 and level 2 processing. In this paper an alternative procedure to minimize the effect of bad detectors is discussed. This procedure avoids the atmospheric correction in preprocessing.
Hyperspectral imaging is an important technology for the passive optical detection of surface and buried land mines from an airborne platform. Hyperspectral remote sensing can exploit many different potential mine observables in the visible and infrared portions of the spectrum. The primary surface mine observable is a spectral difference between the mine body and the background. With a high quality VNIR/SWIR hyperspectral sensor, it is possible to detect these mines as spectral anomalies using techniques that have been previously applied to the detection of military targets. Algorithms developed for the military surveillance application can be directly applied to the surface mine problem. In this paper, two different spectral anomaly approaches are explored. The first is a local spectral anomaly detection algorithm, which examines the color of each pixel for differences with its surroundings. The second is a global spectral anomaly detection algorithm that measures the color of each pixel relative to its occurrence in the whole scene. Both algorithms were developed for the problem of detecting military targets in complex backgrounds and are applied here to the problem of detecting surface mines.
The University of Hawaii's Efficient Materials Mapping program aims to automatically and rapidly produce material maps from hyperspectral scenes. The program combines an end- member determination algorithm and a material identification algorithm to produce context maps in real time without user intervention. The material identification algorithm is a combination of a spectral databse and analytic code; each spectrum in the library augmented with computer readable diagnostic instructions. At present, the material library consists of over three hundred different spectra, generally geological materials from the USGS digital spectral library, however selected spectra from other libraries have been incorporated. Our method has been applied to an AVIRIS sceme taken over Kaneohe Bay, Hawaii. This scene contains large expanses of ocean, developed and undeveloped land, thus providing a good test bed for the program. The results of applying this methodolgy were verified by ground truth where possible by team equipped with hand held spectrometer. Algorithm derived archetypical en-member locations were well matched well by the material identification database, however the end-member determination itself operated sub- optimally on this scene. These results will guid progress with respect to the continued development of this program.
There has been considerable interest in the application of real-time processing techniques to the problem of hyperspectral scene analysis. Recent satellite and aircraft systems can produce data at a rate far faster than the data can be analyzed by interactive computer procedures. Automated and fast procedures for preparing the data for analyst inspection are required for even laboratory use of the large quantities of data. In addition, there are several real-time applications where the data must be processed as it is being acquired. A typical application is a computing system on-board an airplane for operator analysis of the scene as the hyperspectral sensor collects data. In this paper the possible tradeoffs fore rapid analysis are discussed, including choice of algorithm, possible dimensionality reduction, and reduced display level. A real time anomaly detection processing system based on the N- FINDR algorithm has been designed and implemented for the Night Vision Imaging Spectrometer (NVIS). The N-FINDR algorithm is a linear unmixing based algorithm that automatically finds spectral endmembers. The algorithm works by inflating a simplex inside the data, beginning with a random set of pixels. Once these endmember spectra have been found, the image cube can be unmixed using a least-squares approach into a map of fractional abundances of each endmember material in each pixel. In addition to the N-FINDR algorithm, the real-time processing system performs calibration, bad pixel removal, and display of selected fraction planes. The real-time processor is implemented in a commercial Pentium IV computer.
Hyperspectral imaging systems are assuming a greater importance for a wide variety of commercial and military systems. The reason for this increased interest is the fact that a hyperspectral sensor of a give4n spatial resolution or pixel sized will reveal information on the scene that are not obtainable by single band or multi-spectral sensors. There have been several approaches to using a single higher spatial resolution band to improve the spatial resolution fo the hyperspectral data. In this paper, a new technique for improving the spatial resolution of hyperspectral image data will be presented. This technique, called Joint End-member Determination and Unmixing, combines a high-resolution image with a lower spatial resolution hyperspectral image to produce a product that has the spectral properties of the hyperspectral image at a spatial resolution approaching that of the panchromatic image. Instead of using statistical methods to sharpen hyperspectral imagery, a physical model is used where the data present in both the hyperspectral and high-resolution data are assumed to follow linear mixing model. In this paper, the new mixture model based resolution enhancement approach will be compared to the statistical approach using data from NASA/JPL AVIRIS hyperspectral sensor.
Many new imaging spectrometers have been developed over the past several years that use a two-dimensional detector array to simultaneously record the spectra for a line of points on the ground. The second spatial dimension is built up over time by motion of the sensor. The motion of the sensor can be in the direction of the platform motion or at right angles to it. Since a large number of spectra are obtained simultaneously, the instantaneous data rate can be much higher than that achieved with a flying spot scanner. The result can be both greater angular coverage and higher spatial resolution at the same signal level. There are many consequences of using this type of sensor. The two dimensional design of the optical system and its effect on the data must be considered. In addition, since the same detector is not used to build up the image in each spectral band, there are new sources of pattern noise in the data that are not normally seen in data from a flying spot sensor. In this paper, the effects of some of these design considerations are discussed from the point of view of their impact on classification and anomaly detection. Design recommendations for sensor design are made from the processing point of view.
The AHI sensor consists of a long-wave infrared pushbroom hyperspectral imager and a boresighted 3-color visible high resolution CCD linescan camera. The system used a background suppression system to achieve good noise characteristics (less than 1(mu) fl NESR). Work with AHI has shown the utility of the long-wave infrared a variety of applications. The AHI system has been used successfully in the detection of buried land mines using infrared absorption features of disturbed soil. Recently, the AHI has been used to examine the feasibility active and passive hyperspectral imaging under outdoor and laboratory conditions at three ranges. In addition, the AHI was flown over a coral reef ecosystem on the Hawaiian island of Molokai to study fresh water intrusion into coral reef ecosystems. Theoretical calculations have been done propose extensions to the AHI design in order to produce an instrument with a higher signal to noise ratio.
The COMPACT Airborne Spectral Sensor (COMPASS) design is intended to demonstrate a new design concept for solar reflective hyper spectral systems for the Government. Capitalizing from recent focal plane developments, the COMPASS system utilizes a single FPA to cover the 0.4-2.35micrometers spectral region. This system also utilizes an Offner spectrometer design as well as an electron etched lithography curved grating technology pioneered by NASA/JPL. This paper also discusses the technical trades, which drove the design selection of COMPASS. When completed, the core COMPASS spectrometer design could be used in a large variety of configurations on a variety of aircraft.
The Compact Airborne Spectral Sensor (COMPASS) is a hyperspectral sensor covering the 400 to 2350 nm spectral region using a single focal plane and a very compact optical system. In addition, COMPASS will include a high-resolution panchromatic imager. With its compact design and its full spectral coverage throughout the visible, near infrared and SWIR, COMPASS represents a major step forward in the practical utilization of hyperspectral sensors for military operations. COMPASS will be deployed on a variety of airborne platforms for the detection of military objects of interest. There was considerable interest in the development of an on-board processor for COMPASS. The purpose of this processor is to calibrate the data and detect military targets in complex background clutter. Because of their ability to operate on truly hyperspectral data consisting of a hundred or more bands, linear unmixing algorithms were selected for the detection processor. The N-FINDR algorithm that automatically finds endmembers and then unmixes the scene was selected for real-time implementation. In addition, a recently developed detection algorithm, Stochastic Target Detection (STD), which was specifically designed for compatibility with linear unmixing algorithms, was chosen for the detection step. The N-FINDR/STD algorithm pair was first tested on a variety of hyperspectral data sets to determine its performance level relative to existing hyperspectral algorithms (such as RX) using Receiver Operator Curves (ROC) as the basis. Following completion of the testing, a hardware implementation of a real time processor for COMPASS using commercial off-the-shelf computer technology was designed. The COMPASS on-board processor will consist of the following elements: preprocessing, N-FINDR endmember determination and linear unmixing, the STD target detection step, and the selection of a High Resolution Image Chip covering the target area. Computer resource projections have shown that these functions, along with supporting interactive display functions, can operate in real-time on COMPASS data using multi-processor Pentium III class processors.
Hyperspectral data rates and volumes challenge analysis approaches that are not highly automated and efficient. Derived products from hyperspectral data, which are presented in units that are physically meaningful, have added value to analysts who are not spectral or statistical experts. The Efficient Materials Mapping project involves developing an approach that is both efficient in terms of processing time and analyzed data volume and produces outputs in terms of surface chemical or material composition. Our approach will exploit the typical redundancy inherent in hyperspectral data of natural scenes to reduce data volume. This data volume reduction is combined with an automated approach to extract chemical information from spectral data. The results will be a method to produce maps of chemical quantities that can be readily interpreted by analysts specializing in characteristics of terrains and targets rather than photons and spectra.
Proc. SPIE. 4381, Algorithms for Multispectral, Hyperspectral, and Ultraspectral Imagery VII
KEYWORDS: Hyperspectral imaging, Short wave infrared radiation, Detection and tracking algorithms, Sensors, Calibration, Image processing, Spectroscopy, Night vision, Data processing, Software development
The US Army's Night Vision and Electronic Sensors Directorate (NVESD) has developed software tools for processing, viewing, and analyzing hyperspectral data. The tools were specifically developed for use with the U.S. Army's NVESD Night Vision Imaging Spectrometer (NVIS), but they can also be used to process hyperspectral data in a variety of other formats. The first of these tools is the NVESD Hyperspectral Data Processor, which is used to create a calibrated datacube from raw hyperspectral data files. It can calibrate raw NVIS data to spectral radiance units, perform spectral re-alignment, and can co-register imagery from NVIS's VNIR and SWIR subsystems. The second tool is the NVESD Hyperspectral Viewer, which can display focal plane data, generate images, and compute spatial and temporal statistics, produce data histograms, estimate spectral correlation, compute signal-to-clutter ratios, etc. Additionally, this software tool has recently been modified to utilize the INS/GPS data that is currently embedded into NVIS data as well as the high-resolution imagery (HRI) that is collected simultaneously. Furthering its capabilities, Technical Research Associates (TRA) has added the following detection algorithms to the Viewer: N-FINDR, PC and MNF Transformations, Spectral Angle Mapper, and R-X. The purpose of these software developments is to provide the DoD and other Government agencies with a variety of tools, which are not only applicable to NVIS data but also can be applied to other hyperspectral data.
While reflection band hyperspectral instruments have been in use for over a decade, only recently has data from airborne thermal IR hyperspectral instruments become available. One such instrument is the Airborne Hyperspectral Imager (AHI). AHI is a pushbroom sensor developed by the University of Hawaii that spans the 8 to 11.5 micrometer spectral band with 32 spectral bands and 256 simultaneous spatial channels. While many analysis techniques used for reflection band hyperspectral processing can be applied to the thermal band, new procedures had to be developed. In particular, sensor noise and sensor non-linearity induced spectral artifacts are a greater problem than for the VNIR and SWIR. The process begins with calibration, with different calibration files being used to optimize the reduction of sensor artifacts such as shading and striping. Once the data has been calibrated to radiance units, the absorption and path radiance effects of the atmosphere can be removed, if atmospheric truth is available. Following this step, the apparent emissivity is calculated for every pixel in each band. The data is now in a form that is analogous to the apparent reflectance images developed for reflection band data. At this point spectral analysis techniques can be applied to classify the image. The procedure used here was to use an automated endmember determination algorithm, N- FINDR, to determine spectral endmembers and unmix the data cube into fractional abundances. Since some endmembers are likely to result from residual sensor and cultural artifacts, the automated endmember determination and unmixing procedure is performed interactively to optimize results. Both the fractional abundance planes and the endmember spectra themselves are then reviewed for artifacts. Selected abundance planes that correspond to real minerals can then be combined into a classification map. In this paper, AHI data collected for two applications: the detection of buried land mine application and a geological remote sensing application will be presented using similar processing steps.
The AHI sensor consists of a long-wave infrared pushbroom hyperspectral imager and a boresighted 3- color visible high resolution CCD linescan camera. The system used a background suppression system to achieve good noise characteristics (less than 1µfl NESR). Work with AHI has shown the utility of the longwave infrared a variety of applications. The AHI system has been used successfully in the detection of buried land mines using infrared absorption features of disturbed soil. Gas detection was also shown feasible, with gas absorption being clearly visible in the thermal IR. This allowed the mapping of a gas release using a matched filter. Geological mapping using AHI can be performed using the thermal band absorption features of different minerals. A large-scale geological map was obtained over a dry lake area in California using a mosaic of AHI flightlines, including mineral spectra and relative abundance maps.
The Marsokhod Field Experiment performed in Silver Lake, California, was designed to test the use of a “robotic geologist” for future unmanned Mars missions. The University of Hawaii’s Airborne Hyperspectral Imager was included in the experiment to provide geologic context information. The AHI sensor was flown over a 3 by 3 km area, imaging in the long-wave infrared. The hyperspectral data was then processed with the N-FINDR algorithm to produce estimates of constituent material spectra and mineral abundance maps. The derived mineral spectra were identified by comparison to library spectra and found generally consistent with the geology of the area.
In the past 3 years, US Army’s Night Vision and Electronic Sensors Directorate has worked in conjunction with Navy SPAWAR on DARPA's Adaptive Spectral Reconnaissance Program (ASRP). The Night Vision Imaging Spectrometer (NVIS), which is a solar reflective (0.4-2.35um) hyperspectral imaging device, has played a major role in the ASR Program. As with all spectral imaging devices, there exist a certain number of imperfections in the NVIS device. If not handled properly, these imperfections can have an impact upon the performance of certain detection algorithms. This paper will describe the overall measured sensor performance parameters of the NVIS, its imperfections and the effect they may have on algorithm performance. There will also be a discussion concerning the processing tools and methods that have been developed in the past year, and have allowed the imperfections to be removed to some level.
Recently, new hyperspectral sensors have become available that provide both high spatial resolution and high spectral resolution. These characteristics combined with high signal to noise ratio allow the differentiation of vegetation or mineral types based upon the spectra of small patches of the surface. In this paper, automated endmember determination methods are applied to high spatial and spectral resolution data from two new sensors, TRWIS III and NVIS. Both of these sensors are high quality low noise pushbroom imaging spectrometers that acquire data at 5 to 6 nm resolution from 400 to 2450 nm. The data sets collected will be used for two different applications of the automated determination of endmembers: scene material classification and the detection of spectral anomalies. The NVIS hyperspectral data was collected from approximately 6000 ft above ground level over Cuprite, Nevada, resulting in a footprint of approximately two meters. The TRWIS III data was collected from 1500 meters altitude over mixed agriculture backgrounds in Ventura County, California, a largely agricultural area about 100 km from Los Angeles. After calibration and other preprocessing steps, the data in each case was processed using the N-FINDR algorithm, which extracts endmembers based upon the geometry of convex sets. Once these endmember spectra are found, the image cube can be "unmixed" into fractional abundances of each material in each pixel. The results of processing this high spatial and spectral resolution data for these two different applications will be presented.
The Airborne Hyperspectral Imager (AHI) system is a long- wave infrared imaging spectrometer originally designed to detect the presence of buried land mines. Subsequent work with AHI has shown the utility of the long-wave infrared for other applications. The AHI system has been used successfully in the detection of buried land mines using infrared absorption features of disturbed soil. Gas detection was also shown to be feasible, with gas absorption being clearly visible in the thermal IR. This allowed the mapping of a gas release using a matched filter. Geological mapping using AHI can be performed using the thermal band absorption features of different minerals. A large-scale geological map was obtained over a dry lake area in California using a mosaic of AHI flightlines, including mineral spectra and relative abundance maps.
The use of hyperspectral sensors for geological, agricultural and other remote sensing applications is continually increasing. In addition to airborne sensors, there are now at least four hyperspectral satellite sensors under development. These sensors will be producing a near continual stream of high dimensional data, leading to an obvious analysis bottleneck. Much of the planned analysis of hyperspectral image cubes requires the determination of certain basis spectra called 'end-members.' Once these spectra are found, the image cube can be 'unmixed' into fractional abundances of each material in each pixel. There exist several techniques for accomplishing the determination of these end-members, most of which require the intervention of a trained geologist. This process and the associated computations are often time- consuming. There is a need for automated techniques to allow the quick review of data collected by the sensors. Several different approaches to finding end-members in data will be reviewed, including the Pixel Purity Index, Orasis, and the Iterative Error Estimation methods. A new method, called N- FINDR, which extracts end-members based upon the geometry of convex sets, will be discussed in detail. End-member spectra and abundance maps will be compared to USGS results on AVIRIS data. Data examples from AVIRIS will also be used to compare several of the algorithms.
Hyperspectral data provides the opportunity to perform a classification of scene data by either deterministic or stochastic techniques. A typical deterministic technique is linear unmixing. This involves finding certain basis spectra called 'end-members' within the scene. Once these spectra are found, the image cube can be unmixed into a map of fractional abundances of each material in each pixel. The N-FINDR algorithm autonomously finds these end-member spectra within the data and then unmixes the scene by determining the fraction of each end-member in each pixel. A stochastic technique for characterizing spectral classes is the Stochastic Expectation Maximization (SEM) approach. This is a spectral clustering technique for classifying spectral terrain data that involves iterative estimation of a Gaussian mixture fit to spectral data. Both techniques can be misled by commonly occurring sensor defects. This is a particular problem with the new class of pushbroom hyperspectral sensors that use a two-dimensional focal plane. These defects are often caused by errors in the calibration process and bad detectors. They manifest themselves in the data as spectrally dependent shading and/or striping and are usually the limit to the performance of the sensor. It is the purpose of this paper to investigate the effect of these sensor defects on the two different classes of algorithms using the N-FINDR and SEM algorithms. Results from actual data are presented.
In recent years a number of techniques for automated classification of terrain from spectral data have been developed and applied to multispectral and hyperspectral data. Use of these techniques for hyperspectral data has presented a number of technical and practical challenges. Here we present a comparison of two fundamentally different approaches to spectral classification of data: (1) Stochastic Expectation Maximization (SEM), and (2) linear unmixing. The underlying background clutter models for each are discussed and parallels between them are explored. Parallels are drawn between estimated parameters or statistics obtained from each type of method. The mathematical parallels are then explored through application of these clutter models to airborne hyperspectral data from the NASA AVIRIS sensor. The results show surprising similarity between some of the estimates derived from these two clutter models, despite the major differences in the underlying assumptions of each.
Recently a new class of instruments, that uses a detector array to measure spectra for multiple points on the ground, has become available. These instruments build up an image by a pushbroom technique. The realization of the maximum potential of these Array Imaging Spectrometers is dependent on the ability to correct pixel to pixel variations in gain, bias, dark current and linearity of the detector array. Residual calibration striping in each band along the temporal axis, is usually the limiting noise source in these sensors, not photon or system electronic noise. Indeed, performance calculations or measurements which do not take this important noise source into account seriously overestimate the performance of Array Imaging Spectrometers. In this paper, the calibration requirements for Array Imaging Spectrometers, in general, and the procedure used to calibrate the Airborne Hyperspectral Imager (AHI) will be discussed. Examples from the Airborne-Hyperspectral Imager (AHI) will be used to illustrate the residual error and characterize its effect.
In October 1997, the TRWIS III sensor was mounted into a small airplane for the purpose of collecting hyperspectral data over a variety of scenes in Ventura County, California, a largely agricultural area about 100 km from Los Angeles. The resulting hyperspectral 384 band data was analyzed using two different approaches. The first was a physically based procedure using the ratios of spectra selected based upon ground truth. Ratios between images in different bands is a way to emphasize the spectral difference and minimize the effect of illumination. The spectral bands selected were in the vicinity of the near infrared 'red edge' chlorophyll feature. The second procedure is an image processing procedure to transpose the image cube using an orthogonal subspace projection of the hyperspectral data cube. In general, a transformation over the full spectral region of the data (from approximately 400 nanometers to 2.45 micrometers) did not give results separating the tree types as well as the physically based ratio method. However, if the spectral region was restricted to 20 to 30 bands in vicinity of the red edge feature, then similar vegetation separation was achieved. In this paper, the analysis using both procedures will be discussed and compared.
The AHI (Airborne Hyperspectral Imager) system was designed to detect the presence of buried land mines from the air through detection of along wave IR observable associated with mine installation. The system is a helicopter-borne LWIR hyperspectral imager with real time on-board radiometric calibration and mine detection. It collects hyperspectral imagery from 7.5 to 11.5 μm in either 256 or 32 spectral bands. At all wavelengths the AHI noise equivalent delta (NEΔT) temperature is less than 0.1K at 300K and the NESR is less than .02 watts/m2-sr-μm.
The ability to detect weak targets of low contrast or signal-to- noise ratio (SNR) is improved by a fusion of data in space and wavelength from multispectral/hyperspectral sensors. It has been demonstrated previously that the correlation of the clutter between multiband thermal infrared images plays an important role in allowing the data collected in one spectral band to be used to cancel the background clutter in another spectral band, resulting in increased SNR. However, the correlation between bands is reduced when the spectrum observed in each pixel is derived from a mixture of several different materials, each with its own spectral characteristics. In order to handle the identification of objects in this complex (mixed) clutter, a class of algorithms have been developed that model the pixels as a linear combination of pure substances and then unmix the spectra to identify the pixel constituents. In this paper a linear unmixing algorithm is incorporated with a statistical hypothesis test for detecting a known target spectral feature that obeys a linear mixing model in a mixture of background noise. The generalized linear feature detector utilizes a maximum likelihood ratio approach to detect and estimate the presence and concentration of one or more specific objects. A performance evaluation of the linear unmixing and maximum likelihood detector is shown by comparing the results to the spectral anomaly detection algorithm previously developed by Reed and Yu.
Multispectral and hyperspectral infrared (IR) sensors have been utilized in the detection of ground targets by exploiting differences in the statistical distribution of the spectral radiance between natural clutter and targets. Target classification by hyperspectral sensors such as the Spatially Modulated Imaging Fourier Transform SPectrometer (SMIFTS) sensor, a mid-wave infrared imager, depends on exploiting target phenomenology in the infrared. Determination of robust components from hyperspectral IR sensors that are useful for discriminating targets is a key issue in classification of ground targets. Both synthetic aperture radars (SAR) and IR imagers have been utilized in the target detection and recognition processes. Improved target classification by sensor fusion depends on exploitation of target phenomenology from both of these sensors. Here we show the results of an investigation of the use of hyperspectral infrared and low-frequency SAR signatures for the purpose of target recognition. Features extracted from both sensors on similar targets are examined in terms of their usefulness in separating between various classes of targets. Simple distance measures are computed to determine the potential for classifying targets based on a fusion of SAR and hyperspectral infrared data. These separability measures are applied to measurements on similar vehicle targets obtained from separate experiments involving the SMIFTS hyperspectral imager and the Stanford Research Institute SAR.
Under the sponsorship of the DARPA Hyperspectral Mine Detection program, a series of both non-imaging and imaging experiments have been conducted to explore the physical basis of buried object detection in the visible through thermal infrared. Initially, non-imaging experiments were performed at several geographic locations. Potential spectral observables for detection of buried mines in the thermal portion of the infrared were found through these measurements. Following these measurements with point spectrometers, a series of hyperspectral imaging measurements was conducted during the summer of 1995 using the SMIFTS instrument from the University of Hawaii and the LIFTIRS instrument from Lawrence Livermore National Laboratory. The SMIFTS instrument (spatially modulated imaging Fourier transform spectrometer) acquires hyperspectral image cubes in the short-wave and mid-wave infrared and LIFTIRS (Livermore imaging Fourier transform infrared spectrometer) acquires hyperspectral image cubes in the long-wave infrared. Both instruments were optimized through calibration to maximize their signal to noise ratio and remove residual sensor pattern. The experiments were designed to both explore further the physics of disturbed soil detection in the infrared and acquire image data to support the development of detection algorithms. These experiments were supported by extensive ground truth, physical sampling and laboratory analysis. Promising detection observables have been found in the long-wave infrared portion of the spectrum. These spectral signatures have been seen in all geographical locations and are supported by geological theory. Data taken by the hyperspectral imaging sensors have been directly input to detection algorithms to demonstrate mine detection techniques. In this paper, both the non-imaging and imaging measurements made to date will be summarized.
The detection and recognition of targets in infrared wide area surveillance systems is made difficult by clutter background and low resolution. Recent advances in technology have made available small and lightweight hyperspectral imaging sensors. Hyperspectral sensors can facilitate the detection of targets in clutter because natural vegetation clutter has a different statistical distribution of radiant energy in the spectral bands than targets. Natural clutter from vegetation can be characterized as a grey body, but man made objects (i.e. targets) are selective radiators. Compared to blackbody radiators, targets emit radiation more strongly at some wavelengths than at others. The approach taken in this paper is to partition the bands into two groups. The targets exhibit substantial color signatures in one group but look like grey bodies in the other group. A generalized formation for combining the hyperspectral bands is derived using maximum likelihood techniques. The algorithm is a generalization of the weighted spectral difference algorithm, and reduces to that form if the image data is preprocessed to make it spatially white. It is also shown that the algorithm is optimum for non- Gaussian noise when the criterion is to minimize the mean square error between the two groups of bands. The algorithm is applied to TIMS multispectral and SMIFTS hyperspectral data to illustrate the algorithm performance.
The use of hyperspectral visible and infrared sensors is being explored under an ARPA program to provide a means for the detection of buried mines. The purpose of this paper is to summarize the status of the phenomenology of the detection of buried mines using hyperspectral IR detection mechanisms. Both spectral and temperature phenomena related to buried mines will be investigated in the paper. Concepts using the midwave IR (3 to 5 micrometers ), the longwave IR (8 to 12 micrometers ) and the reflection IR (from 1.1 to 2.5 micrometers ) are emphasized in this current effort, although the full IR and visible spectra is considered. Thermally dominated IR is emphasized because of the desire for day/night operation. The program is initially focusing on nonimaging spectrometer measurements of top layers of soil and subsoil, to determine the presence of spectral differences that can be an indicator of mine placement. These spectrometer measurements will be followed by measurements with hyperspectral imaging sensors. While many broad measurements have been made in the MWIR and LWIR, few measurements have been made with an imaging spectrometer. The ARPA/University of Hawaii Spatially Modulated Imaging Fourier Tranform Spectrometer (SMIFTS) can provide such data in the 1.1 to 5.0 micrometers band and the Lawrence Livermore National Laboratory's Livermore Imaging Fourier Transform Infrared Spectrometer (LIFTIRS) will cover the 8-12 micrometers region. The sensors will be deployed in the field from an elevated platform to acquire data in support of both the phenomenology verification and the development of algorithms.
Rail SAR data was analyzed to determine the statistical correlation of background clutter between different UHF frequencies and between passes taken on different days. The purpose of this study was to determine the potential gain from change detection algorithms and multifrequency clutter cancellation algorithms using a high-quality near-ideal SAR data set. The polarimetric radar data used for this study spans a frequency range from 300 to 1200 MHz and had been processed into separate channels with 400 MHz bandwidth. The data is very low noise because of time integration to reduce radio frequency interference (RFI). In this paper, the rail SAR correlation coefficients calculated on data taken on two separate days at the same center frequencies and bandwidths. Secondly, the data was used to explore frequency-to-frequency correlations with correlations calculated for all possible frequency combinations of data taken simultaneously. Both overlapping and nonoverlapping frequency bands were processed. Finally, subtraction studies were performed whereby data from one pass is subtracted from a second pass at the same frequency and data between two frequencies were subtracted using weighted differencing processing. Generally, pass-to-pass correlations on successive days are highest at low frequencies. Correlations approaching .96 between two passes for the lowest frequency band were achieved. These high correlations mean that change detection algorithms can be used and that differencing will result in processing gain. Registration problems will exist with airborne acquired data, that are simplified with rail SAR data. The individual HH and VV polarizations are more highly correlated than the cross polarizations, probably because of higher signal levels relative to noise. Significant clutter reduction (over 10 db) and whitening were demonstrated on data taken in two passes two days apart. The frequency-to-frequency correlations were found to be lower than the pass-to-pass correlations with the level of correlation decreasing with increasing frequecny separation. Some minor whitening was demonstrated for differencing of registered SAR images in nonoverlapping frequencies but no overall clutter reduction was seen.
During 1993, a series of experiments were performed under the Advanced Research Projects Agency (ARPA) sponsorship using the SRI ultra-wide band UHF synthetic aperture radar (SAR). These experiments were performed over a variety of clutter backgrounds to assess the foliage penetration capability of the technology and to investigate target detection in clutter. Experiments were conducted observing tropical rain forest backgrounds in Panama, several different desert backgrounds in the Yuma vicinity, and the mid-latitude temperate forest of Maine. SAR images were formed from the raw data using Differential GPS to aid in the focusing. The three locations represent different levels of foliage cover, ranging from the sparsely vegetated desert sites to the triple canopied rain forest. The characteristics of each site are discussed first through a presentation of photography and SAR imagery. The clutter characteristics are studied through a comparison of the cumulative distributions, which are plotted using a variety of conventions. For each case, at least one reference target is included in the test scene. The signal of that target as processed by a common algorithm will be compared to the processed clutter distribution.
Multi-spectral IR, coupled with advanced image processing procedures, offers the possibility of wide area surveillance and the detection of low concentrations of chemical vapors. A field experiment was performed using the ARPA Multi-Spectral IR Camera sensor mounted on an aircraft. The sensor aircraft flew over a controlled diethyl ether release in a tropical rain forest acquiring image data from both 10000 ft and 22000 ft altitude. The data was processed using multi-spectral algorithms and the vapor was detected over both an open area and the rain forest canopy. This detection was made possible by the removal of most of the background scene by multi-spectral processing.
The DARPA Multi-Spectral Infrared Camera (MUSIC) sensor system was deployed to Australia in July 1991. A series of measurements on targets and backgrounds were performed over a six week period. Two experiments were performed to investigate the use of multi- spectral techniques to enhance extended targets. The first of these was an experiment to detect ship wakes in the tropical ocean. For this experiment data was taken along a ship wake in both an MWIR band and an LWIR band. The thermal structure of the ocean surface showed a high level of spectral correlation even at the 10 millikelvin sensitivity level of the data. Several ship wakes were detected by processing the data spectrally to remove the background. A second extended target experiment was conducted to detect chemical vapor plumes over a cluttered terrain background. In this experiment, two neighboring infrared bands were chosen, one centered on a chemical absorption band and the second placed away from it. The background clutter proved to be highly correlated between these bands allowing its removal using multi- spectral processing. The enhanced chemical plume was then detected. Results from the processing of the data from both of these experiments will be presented along with a description of the algorithm used.
The DARPA multi-spectral infrared camera (MUSIC) was used for a series of experiments in Australia and Maui, Hawaii in 1991. The Maui experiments, conducted from a high mountain, concentrated on the detection of aircraft. The detection of air vehicles without the use of temporal motion (such as the case of a head-on approaching air vehicle) is a challenging problem when background clutter is present. The technique investigated was not dependent upon either the angular motion or the spectral signature of the aircraft. This approach exploits the differential transmission of the atmosphere in neighboring long wave infrared bands. This differential transmission between the target and background `colors' the background relative to the target and allows its removal. This technique was demonstrated on many examples of MUSIC data collected in Maui, Hawaii. Targets approaching the sensor head-on were successfully detected against clouds and other backgrounds using spectral along with spatial techniques. Several different algorithms were investigated and results are compared.
This paper describes adaptive signal processing techniques that utilize spectral and temporal information provided by passive infrared imaging sensors to enhance the detectability of sub- pixel targets in clutter. The approaches are directly applicable to advanced sensors like the DARPA-sponsored MUSIC instrument, which are capable of collecting multi-spectral frame sequences in the thermal infrared region. The performance of several algorithm concepts is demonstrated by processing dual-band frame sequence data taken by the MUSIC sensor. The examples also demonstrate the importance of accurate frame registration prior to multiple- image signal processing.