The Office of Naval Research (ONR) and the Naval Research Laboratory (NRL) have initiated the Hyperspectral Remote Sensing Technology (HRST) program to demonstrate the utility of a hyperspectral earth-imaging system to support Naval needs for characterization of the littoral regions of the world. One key component of the HRST program is the development of the Naval EarthMap Observer (NEMO) satellite system to provide a large hyperspectral data base. NEMO will carry the Coastal Ocean Imaging Spectrometer (COIS) which will provide images of littoral regions with 210 spectral channels over a bandpass of 0.4 to 2.5 micrometer. Since ocean environments have reflectances typically less than 5%, this system requires a very high signal-to-noise ratio (SNR). COIS will sample over a 30 km swath width with a 60 m Ground Sample Distance (GSD) with the ability to go to a 30 m GSD by utilizing the systems attitude control system to 'nod' (i.e., use ground motion compensation to slow down the ground track of the field of view). Also included in the payload is a co-registered 5 m Panchromatic Imager (PIC) to provide simultaneous high spatial resolution imagery. A sun-synchronous, 97.81 degree inclination, circular orbit of 605 km allows continuous repeat coverage of the whole earth. One unique aspect of NEMO is an on-board processing system, a feature extraction and data compression software package developed by NRL called the Optical Real-Time Spectral Identification System (ORASIS). ORASIS employs a parallel, adaptive hyperspectral method for real time scene characterization, data reduction, background suppression, and target recognition. The use of ORASIS is essential for management of the massive amounts of data expected from the NEMO HSI system, and for developing Naval products under HRST. The combined HSI and panchromatic images will provide critical phenomenology to aid in the operation of Naval systems in the littoral environment. The imagery can also satisfy a number of commercial and science community requirements for moderate spatial and high spectral resolution remote sensing data over land and water. Specific areas of interest for the Navy include bathymetry, water clarity, bottom type, atmospheric visibility, bioluminescence potential, beach characterization, underwater hazards, total column atmospheric water vapor, and detection and mapping of subvisible cirrus. These data support requirements for Joint Strike and Joint Littoral warfare, particularly for environmental characterization of the littoral ocean. Demonstrations of direct downlinking of near real-time data to the warfighter are also being formulated. The NEMO satellite is planned to launch in 2000 followed by an operational period of 3 to 5 years.

The European Space Agency (ESA) is defining candidate missions for Earth Observation. In the class of the Earth Explorer missions, dedicated to research and pre-operational demonstration, the Land Surface Processes and Interactions Mission (LSPIM) will acquire the accurate quantitative measurements needed to improve our understanding of the nature and evolution of biosphere-atmosphere interactions and to contribute significantly to a solution of the scaling problems for energy, water and carbon fluxes at the Earth's surface. The mission is intended to provide detailed observations of the surface of the Earth and to collect data related to ecosystem processes and radiation balance. It is also intended to address a range of issues important for environmental monitoring, renewable resources assessment and climate models. The mission involves a dedicated maneuvering satellite which provides multi-directional observations for systematic measurement of Land Surface BRDF (Bi-Directional Reflectance Distribution Function) of selected sites on Earth. The satellite carries an optical payload: PRISM (Processes Research by an Imaging Space Mission), a multispectral imager providing reasonably high spatial resolution images (50 m over 50 km swath) in the whole optical spectral domain (from 450 nm to 2.35 micrometer with a resolution close to 10 nm, and two thermal bands from 8.1 to 9.1 micrometer). This paper presents the results of the Phase A study awarded by ESA, led by ALCATEL Space Industries and concerning the design of LSPIM.

Airborne imaging spectroscopy has undergone a rapid development over the last years. The number of research groups making use of this technology has increased by an order of a magnitude. To be able to provide reliable imaging spectrometer data to these users not only the operation of a well- characterized imaging spectrometer by an experienced flight crew is essential. Furthermore facilities for the data handling and processing as well as higher-level data correction algorithms have to be established. In this paper the hard- and software components of the imaging spectrometer facility as implemented at the DLR, Institute of Optoelectronics are described.

We present experimental results of dispersive coherence spectrotomography (DCST) with the white-light continuum. The dispersive coherence spectrotomography enables us to extract both range and spectral properties inside a medium. The main feature is that DCST has the high dynamic range in depth and high signal-to-noise ratio making the most of the extreme brightness of the white-light continuum. The principle of DCST is based on a spectral decomposition of the white-light interferograms. The system consists of a Michelson interferometer followed by a spectrometer. The system is illuminated by the white-light continuum through a cylindrical lens and narrow strip lines of the white continuum are formed in the 3-D sample and on the reference mirror surface. Backscattered light from the sample is recombined with the reference light. We superpose spectrally dispersed images of the strip lines distributed along a chromatic axis on the surface of the CCD. By translating the sample along the optical axis, the spectrally decomposed interferograms are obtained from the detector elements. We sum up the spectrally- decomposed interferograms according to spectral filters designed for the spectromography and get the coherence spectrotomogram that contains the spatial and spectral distribution over the 2-D cross-section of the sample volume. The depth and spectral properties of colored layers have successfully been extracted.

Passive standoff detection of chemical warfare (CW) agents is currently achieved by remote sensing infrared spectrometry in the 8 - 12 micrometer atmospheric window with the aid of automatic spectral analysis algorithms. Introducing an imaging capability would allow for rapid wide-area reconnaissance and mapping of vapor clouds, as well as reduce false alarms by exploiting the added spatial information. This paper contains an overview of the CW agent standoff detection problem and the challenges associated with developing imaging LWIR hyperspectral sensors for the detection and quantification of vapor clouds, as well as a discussion of spectral processing techniques which can be used to exploit the added data dimensionality.

An infrared-imaging instrument is being developed to provide in situ qualitative and quantitative assessment of hydrocarbon contaminants on metallic surfaces for cleaning verification. A continuous-wave infrared optical parametric oscillator (OPO), based on the quasi-phasematched material periodically poled lithium niobate (PPLN), is interfaced with an InSb focal plane array camera to perform fast, non-invasive analysis by reflectance spectroscopy. The period range of the designed fan-out PPLN crystal determines the range of the output wavelength of the light source. It is able to scan hundreds of wavenumbers positioned in the range of 2820 - 3250 cm^-1, which is sufficient to detect functional groups of common organic compounds (-CH, -OH, and -NH). The capability of the instrument has been demonstrated in a preliminary investigation of reflectance measurements for hydrocarbon solvents (methanol and d-limonene) on an aluminum surface. A substantial difference in absorption is obtained for the two solvents at two different laser-illumination wavelengths, thus permitting hydrocarbon detection and molecular species differentiation. Preliminary reflectance spectra of a mixture of aliphatic hydrocarbon lubricants and drawing agents on an aluminum panel are also presented. The relative thickness of the hydrocarbon thin film is determined by the intensity ratio of images acquired at two different laser illumination frequencies.

A new, state-of-the-art atmospheric correction algorithm for the solar spectral range has been developed based on the MODTRAN4 code. The primary data products are surface reflectance spectra, column water vapor maps and relative surface elevation maps. In addition, a radiance simulation tool, an automated visibility retrieval algorithm and a spectral 'polishing' algorithm are included. Validations of retrievals have been carried out by analyzing data that encompass a variety of atmospheric and surface conditions. Some results and their implications for atmospheric correction and spectroscopy are discussed.

Multi-channel remote sensing of ocean color from space has a rich history -- from the past CZCS, to the present SeaWiFS, and to the near-future MODIS. The atmospheric correction algorithms for processing remotely sensed data from these sensors were mainly developed by Howard Gordon at University of Miami. The algorithms were primarily designed for retrieving water leaving radiances in the visible spectral region over clear deep ocean areas. The information about atmospheric aerosols is derived from channels between 0.66 and 0.87 micrometer, where the water leaving radiances are close to zero. The derived aerosol information is extrapolated back to the visible when retrieving water leaving radiances from remotely sensed data. For the turbid coastal environment, the water leaving radiances for channels between 0.66 and 0.87 micrometer may not be close to zero because of back scattering by suspended materials in the water. Under these conditions, the channels are no longer useful for deriving information on atmospheric aerosols. As a result, the algorithms developed for applications to clear ocean waters cannot be easily modified to retrieve water leaving radiances from remote sensing data acquired over the costal environments. We have recently developed a fast and fully functional atmospheric correction algorithm for hyperspectral remote sensing of ocean color with the Coastal Ocean Imaging Spectrometer (COIS). Our algorithm uses lookup tables generated with a vector radiative transfer code developed by Ahmad and Fraser (1982) and a spectral matching technique for the retrieval of water leaving radiances. The information on atmospheric aerosols is estimated using dark channels beyond 0.86 micron. Quite reasonable results were obtained when applying the algorithm to process spectral imaging data acquired over Chesapeake Bay with the NASA JPL Airborne Visible Infrared Imaging Spectrometer (AVIRIS).

Only a fraction of the bands in a hyperspectral image are typically needed to discriminate materials of interest for any given application. Many of the bands are in superfluous portions of the spectrum, where no substantive differences exist between the material spectra. Other bands are spread across the widths of the absorption/emission features, causing them to be highly correlated and of minimal added discrimination value. It would be desirable to be able to efficiently identify which bands are the most discriminating for a specific application, and to process only those image planes to detect the material of interest. The volume of data that needs to be processed would be reduced, discrimination sensitivity could be increased through elimination of extraneous bands, and multispectral exploitation algorithms could be used. Hyperspectral Distiller performs this function using only representative spectra of the materials to be discriminated as input. Distiller utilizes the BANDS algorithm to identify the wavelength positions, widths, and strengths of constituent absorption/emission features in the input spectra. Distiller then uses this information to characterize and rank each band according to its discrimination potential. The most discriminating bands (user defined number) are reported as output, and a subset image is created.

Previous papers have described the concept behind the MightySat II.1 program, the satellite's Fourier Transform imaging spectrometer's optical design, the design for the spectral imaging payload, and its initial qualification testing. This paper discusses the on board data processing designed to reduce the amount of downloaded data by an order of magnitude and provide a demonstration of a smart spaceborne spectral imaging sensor. Two custom components, a spectral imager interface 6U VME card that moves data at over 30 MByte/sec, and four TI C-40 processors mounted to a second 6U VME and daughter card, are used to adapt the sensor to the spacecraft and provide the necessary high speed processing. A system architecture that offers both on board real time image processing and high-speed post data collection analysis of the spectral data has been developed. In addition to the on board processing of the raw data into a usable spectral data volume, one feature extraction technique has been incorporated. This algorithm operates on the basic interferometric data. The algorithm is integrated within the data compression process to search for uploadable feature descriptions.

The clear waters of Lake Superior constitute the heart of one of the most significant fresh water ecosystems in the world. Lake Superior is the world's largest lake by surface area (82,100 km2) holding approximately 10% of the earth's freshwater (12,230 km3) that is not locked into glaciers or ice caps. Although Superior is arguably the most significant fresh water ecosystem on earth, questions relating to the lake and its watershed remain unanswered, including the effects of human habitation, exploitation, and economic potential of the region. There is a great diversity of scientific disciplines with a common interest in remote sensing of the Lake Superior ecosystem which have the need for data at all spatial, spectral, and temporal scales-from scales supplied by satellites, ships or aircraft at low spatial, spectral or temporal resolution, to a requirement for synoptic high resolution spatial (approximately 1 meter)/spectral (1 - 10 nm) data. During May and August of 1998, two week-long data collection campaigns were performed using the Kestrel airborne visible hyperspectral imager to acquire hyperspectral data of a broad taxonomy of ecologically significant targets, including forests, urban areas, lakeshore zones and rivers, mining industry tailing basins, and the Lake itself. We will describe the Kestrel airborne hyperspectral sensor, the collection and data reduction methodology, and flight imagery from both campaigns.

Using a spectral signature derived from training data in one image to detect the material of interest in another image is not always successful using image normalization alone. Normalization can correct for scene-to-scene differences in atmospheric, sun angle, and sensor calibration effects, but it does not compensate for scene-to-scene changes in inherent target characteristics. It is generally difficult to predict what kind of change may occur and what its impact on the signature might be. Even when the expected variations in signature can be predicted, expanding a signature's tolerance or creating a family of signatures to accommodate the variance can reduce discrimination and produce unacceptable false alarm rates. Ignoring them can lead to missed detections. ASK has been developed for use with Subpixel Classifier, software that automatically detects these scene-to-scene changes and accordingly adapts the spectral signature to maintain strong detection performance. Subpixel Classifier is used to make target detections in an image, and the spectra of the detections are automatically analyzed to characterize the change and adapt the signature to the target condition(s) in the scene. An interactive functionality can take advantage of scene knowledge when available.

Hyperspectral imaging is the latest advent in imaging technology, providing the potential to extract information about the objects in a scene that is unavailable to panchromatic imagers. This increased utility, however, comes at the cost of tremendously increased data. The ultimate utility of hyperspectral imagery is in the information that can be gleaned from the spectral dimension, rather than in the hyperspectral imagery itself. To have the broadest range of applications, extraction of this information must occur in real-time. Attempting to produce and exploit complete cubes of hyperspectral imagery at video rates, however, presents unique problems for both the imager and the processor, since data rates are scaled by the number of spectral planes in the cube. MIDIS, the Multi-band Identification and Discrimination Imaging Spectroradiometer, allows both real-time collection and processing of hyperspectral imagery over the range of 0.4 micrometer to 12 micrometer. Presented here are the major design challenges and solutions associated with producing high-speed, high-sensitivity hyperspectral imagers operating in the Vis/NIR, SWIR/MWIR and LWIR, and of the electronics capable of handling data rates up to 160 mega-pixels per second, continuously. Beyond design and performance issues associated with producing and processing hyperspectral imagery at such high speeds, this paper also discusses applications of real-time hyperspectral imaging technology. Example imagery includes such problems as buried mine detection, inspecting surfaces, and countering CCD (camouflage, concealment, and deception).

A number of organizations are using the data collected by the HYperspectral Digital Imagery Collection Experiment (HYDICE) airborne sensor to demonstrate the utility of hyperspectral imagery (HSI) for a variety of applications. The interpretation and extrapolation of these results can be influenced by the nature and magnitude of any artifacts introduced by the HYDICE sensor. A short study was undertaken which first reviewed the literature for discussions of the sensor's noise characteristics and then extended those results with additional analyses of HYDICE data. These investigations used unprocessed image data from the onboard Flight Calibration Unit (FCU) lamp and ground scenes taken at three different sensor altitudes and sample integration times. Empirical estimates of the sensor signal-to-noise ratio (SNR) were compared to predictions from a radiometric performance model. The spectral band-to-band correlation structure of the sensor noise was studied. Using an end-to-end system performance model, the impact of various noise sources on subpixel detection was analyzed. The results show that, although a number of sensor artifacts exist, they have little impact on the interpretations of HSI utility derived from analyses of HYDICE data.

Optical systems associated with imaging sensors and instruments typically distort the 'true' or object image, I(x), in a manner usually characterized by their point spread function (PSF). Determining I(x) from the measured image data, M(z), using the convolutional relation with the PSF is called deconvolution. This paper proposes what appears to be a new deconvolution technique by taking advantage of a remarkable coincidence. It is that for most optical systems of interest here the PSF is Gaussian, which is a zeroth order Hermite function. By expressing I(x) in an orthogonal representation using Hermite functions, which are to be distinguished from Hermite polynomials, the convolution integral can be evaluated exactly in analytical form, perhaps for the first time for the general case. This, in turn, leads to simple, precise linear relations between the coefficients of the Hermite representation of I(x) and that of M(x); while avoiding the common problem of division of noisy data by small quantities. The coefficients in those linear equations have precise values obtained from the nature of Hermite function interrelations rather than measured data. These values of I(x) may be more useful than M(x) as the initial iterate in the iteration techniques commonly used for deconvolution.

A calibration method for determining and correcting the optical distortion in the spectral axis of a Fourier Transform spectrometer is proposed. In our method, a narrowband optical calibration source is used, and the resulting interferogram is treated as a sinusoid which is phase-modulated as a result of the optical distortion. A discrete-time quadrature demodulator is used to determine the phase modulation, and the distortion is then characterized using a third-order polynomial fit to this estimated phase. Correction of interferogram data is accomplished by resampling the discrete data onto a uniform grid in the predistortion space. Our method has been applied to calibration data from the Digital Array Scanned Interferometer (DASI), with excellent results.

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.

The Ocean Portable Hyperspectral Imager for Low-Light spectroscopy (Ocean PHILLS), is a new hyperspectral imager specifically designed for imaging the coastal ocean. It uses a thinned, backside illuminated CCD for high sensitivity, and an all-reflective spectrograph with a convex grating in an Offner configuration to produce a distortion free image. Here we describe the instrument design and present the results of laboratory calibration and characterization and example results from a two week field experiment imaging the coastal waters off Lee Stocking, Island, Bahamas.

A complete Stokes imaging spectropolarimeter has been designed, constructed and tested by researchers at the U.S. Army Engineer Research and Development Center (USAERDC) in collaboration with researchers at the University of Arizona's Optical Sciences Center. CTISP is a polarimetric extension to CTIS, developed by the authors associated with the Optical Sciences Center. Currently, CTISP characterizes an object's spectropolarimetric radiance over the 440 to 740 nm range using 20 nm spectral bins and subdividing the FOV with a 32 X 32 resolution. CTISP's output is a four-fold increase in object cube information when compared to spectral radiance alone as CTISP effectively extracts the polarization information from the radiance of each of the 'N' voxels to form 'N' 4 element Stokes vectors, where N equals (# horizontal resolution elements in FOV)*(# vertical resolution elements in FOV)*(# of wavelength bands). Voxel polarization calibration is performed using a fully computer automated spectropolarimetric calibration facility. The facility generates an object whose spatial and spectral dimensions define a voxel and whose radiance is purely polarized. CTISP's response to each generated polarized voxel is recorded and used to calculate a polarization characteristic matrix (PCM) for each voxel. CTISP utilizes four polarization analyzers in an automated rotating filter wheel configuration to acquire four images of the object. The results from the image reconstructions behind four analyzers are utilized with the PCMs to estimate the Stokes vector for each voxel in the object cube. CTISP utilizes a host of software tools to control the calibration facility, perform image acquisition and perform reconstruction and Stokes vector calculation. The order of use and inter-relation of these tools is described. Results will be presented and indicate that CTISP is capable of reconstructing objects containing complex spectral, spatial and polarization content. A spectral comparison is made to a reference spectrometer using a reflectance standard for normalization.

The CHRIS instrument is a space-based imaging spectrometer that will provide 10 nm spectral resolution over the spectral range from 415 nm to 1050 nm. The nominal spatial sampling interval will be 25 m, however, larger sampling intervals are possible. Band selection, spectral bandwidths and the spatial sampling interval will be programmable. The instrument is planned to be launched on an agile small satellite of the 100 kg class. This satellite will operate in a sun-synchronous, high inclination orbit at approximately 830 km. At this altitude the instrument can provide 19 spectral bands with a spatial sampling interval of 25 m at nadir. The field of view of CHRIS is 18.6 km. Attitude control of the platform will allow access to non-nadir targets, multi-angle observations of selected targets and improved radiometric resolution. This paper describes the optical design of the instrument, including the telescope, spectrometer detector and in-flight calibration hardware, as well as critical alignment procedures, with emphasis on spectrometer assembly and stray light control. Results of performance and calibration measurements are presented.

We report results of experimentation with a MWIR non-scanning, high speed imaging spectrometer capable of simultaneously recording spatial and spectral data from a rapidly varying target scene. High speed spectral imaging was demonstrated by collecting spectral and spatial snapshots of filtered blackbodies, combustion products and a coffee cup. The instrument is based on computed tomography concepts and operates in a mid-wave infrared band of 3.0 to 4.6 micrometer. Raw images were recorded at a video frame rate of 30 fps using a 160 X 120 InSb focal plane array. Reconstructions of simple objects are presented.

The concept for an imaging spectral sensor is presented. The starting point is an advanced raster scanning with multiplexing in spatial terms. The essential components are an entrance array, an imaging grating and a detector array -- a double array architecture. A programmable micro slit matrix is used as a two-dimensional HADAMARD-mask. It scans the two- dimensional images to be processed in a one-dimensional spectrometer. A multiplexing in spatial terms and simultaneously in spectral terms is possible. The spectra recorded by a commercial detector array spectrometer. At the present state the multiplexing of more than 100 spatial pixels within the quality of spectrometry is demonstrated. The concept provides a HADAMARD or multiplex advantage resulting from the micro slit matrix (increase of SNR by (root)n/2, n- number of imaging pixels) and a multidetector advantage resulting from the linear detector array of the spectrometer (increase of SNR by (root)N, N-number of detector elements). Experiments reported on a further paper.

We present results from an improved ORASIS (Optical Real-time Adaptive Spectral Identification System) hyperspectral-data compression-algorithm that is being implemented on the Naval EarthMap Observer (NEMO) satellite. The algorithm is shown to produce results that are statistically improved from previous findings. To augment the statistical testing, the re-inflated data are run through analysis programs such as unsupervised classification. ORASIS compression is a series of algorithms. The first algorithm, the exemplar selector process (ESP), is a variation of Learned Vector Quantization (LVQ) that builds up a relatively small set of spectra to represent the full data set. Subsequent algorithms find approximate endmembers for the exemplar set and project the set into the space defined by the endmembers. Both the ESP and the projection process contribute to the compression of the data. The obtainable compression ratios vary with scene content and other factors but ratios between 10:1 and 30:1 are possible. The compressed data format is designed to allow direct access to individual pieces of the data without reinflation of the entire data set. Details of the hardware implementation of the Imagery On-Board Processor (IOBP) of NEMO is discussed, as well as the use of the compressed data on the ground.

The consistent simulation of airborne and spaceborne hyperspectral data is an important task and sometimes the only way for the adaptation and optimization of a sensor and its observing conditions, the choice and test of algorithms for data processing, error estimations and the evaluation of the capabilities of the whole sensor system. The integration of three approaches is suggested for the data simulation of APEX (Airborne Prism Experiment): (1) a spectrally consistent approach (e.g. using AVIRIS data), (2) a geometrically consistent approach (e.g. using CASI data), and (3) an end-to- end simulation of the sensor system. In this paper, the last approach is discussed in detail. Such a technique should be used if there is no simple deterministic relation between input and output parameters. The simulation environment SENSOR (Software Environment for the Simulation of Optical Remote Sensing Systems) presented here includes a full model of the sensor system, the observed object and the atmosphere. The simulator consists of three parts. The first part describes the geometrical relations between object, sun, and sensor using a ray tracing algorithm. The second part of the simulation environment considers the radiometry. It calculates the at-sensor-radiance using a pre-calculated multidimensional lookup-table for the atmospheric boundary conditions and bi- directional reflectances. Part three consists of an optical and an electronic sensor model for the generation of digital images. Application-specific algorithms for data processing must be considered additionally. The benefit of using an end- to-end simulation approach is demonstrated, an example of a simulated APEX data cube is given, and preliminary steps of evaluation of SENSOR are carried out.

Hyperspectral image data presents challenges to current transmission bandwidth and storage capabilities. To overcome these challenges and to retain the radiometric accuracy of the data, there is a need for good hyperspectral lossless compression. The current state-of-the-art lossless compression algorithm is JPEG-LS, which uses a 2-D edge-detecting predictor. Hyperspectral systems sample the electromagnetic spectrum very finely, which results in increased spectral correlation. A predictor that takes into account previous band information can obtain substantial gains in compression ratio. This paper discusses a number of different predictors that take advantage of the significant band-to-band (spectral) correlation within the hyperspectral imagery. A sample set of HYDICE, AVIRIS, and SEBASS imagery was used to evaluate the different predictors. While the JPEG-LS algorithm achieved just greater than 2:1 on most imagery, some of the 3-D prediction techniques achieved greater than 3:1 compression ratio. The characteristics of these test images and results from different predictors are presented in this paper.

We present a method for identifying hyperspectral textures composed of a set of given materials. The algorithm is invariant to illumination and atmospheric conditions as well as the spatial sampling of the texture. Only the spectral reflectance functions for the materials in the texture are required by the algorithm. A texture analysis method based on minimizing the squared error between a pixel spectrum and a synthetic spectral mixture allows pixels to be ranked according to consistency with the texture model. Experimental results using HYDICE imagery demonstrate the use of the method to identify hyperspectral textures under different conditions.

The analysis of hyperspectral data sets requires the determination of certain basis spectra called 'end-members.' Once these spectra are found, the image cube can be 'unmixed' into the fractional abundance of each material in each pixel. There exist several techniques for accomplishing the determination of the end-members, most of which involve the intervention of a trained geologist. Often these-end-members are assumed to be present in the image, in the form of pure, or unmixed, pixels. In this paper a method based upon the geometry of convex sets is proposed to find a unique set of purest pixels in an image. The technique is based on the fact that in N spectral dimensions, the N-volume contained by a simplex formed of the purest pixels is larger than any other volume formed from any other combination of pixels. The algorithm works by 'inflating' a simplex inside the data, beginning with a random set of pixels. For each pixel and each end-member, the end-member is replaced with the spectrum of the pixel and the volume is recalculated. If it increases, the spectrum of the new pixel replaces that end-member. This procedure is repeated until no more replacements are done. This algorithm successfully derives end-members in a synthetic data set, and appears robust with less than perfect data. Spectral end-members have been extracted for the AVIRIS Cuprite data set which closely match reference spectra, and resulting abundance maps match published mineral maps.

Numerous statistical approaches have been developed for small target detection in cluttered environments. Examples include orthogonal background suppression (OBS) where the initial principal components are suppressed, and the clutter matched filter (CMF) where the principal components are weighted by the inverse of the eigenvalues and the latter principal components are discarded. Our research has shown that improved target detection performance can be obtained by combining certain aspects of both OBS and CMF approaches. This is especially true in the presence of limited scene data (finite number of pixels) or an imperfect reference target spectrum. The basis of this idea is to use weighting by the inverse of the eigenvalues (from CMF) for the initial PCs and the uniform weighting for the later PCs (from OBS). Examples of this new technique and comparisons with OBS and CMF will be shown with model data with realistic clutter containing a chemical plume.

In the thermal infrared (TIR), a surface emits radiation based on its temperature and emissivity. TIR imaging spectrometry involves extracting temperature, emissivity, and/or surface composition information, which are useful in a wide variety of studies ranging from climatology to land use analyses. A simple technique of temperature emissivity separation (TES) has been developed to separate the effects of emissivity from temperature within the radiance signal recorded by a sensor. This technique can be employed to map the composition and temperature of a surface. Likewise, spectral mixture analysis (SMA) has been successfully applied in this spectral region to discern spectral features and their temperatures. This paper describes an application of TES and SMA that has been employed to characterize the isothermal combinations of thermal radiance features on a sub-pixel basis. This approach, referred to in this paper as a Temperature Emissivity Separation Spectral Mixture Analysis (TESSMA), uses the relationship between a 'virtual cold' endmember fraction and surface temperature to extract image temperature estimates, which are then used to constrain an isothermal unmixing of pixel endmembers. Work presented includes characterizations of synthetically generated temperature-endmember fraction test images, a discussion of methods used to separate temperature and endmember attributions, and a fraction estimate analysis. This paper also demonstrates the temperature dependence of isothermal SMA on accurate temperature estimates, two TESSMA approaches, and their results.

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.

An invariant material tracking algorithm requires a representation that is unaffected by environmental factors. An airborne hyperspectral sensor measures a radiance spectrum that, in general, will vary with atmospheric conditions and the scene geometry. Material tracking algorithms that directly match hyperspectral sensor measurements are therefore limited under general conditions. In this paper, we develop a low- dimensional subspace representation that can be used for material tracking over wide variation in environmental parameters. The material subspace is constructed using a single radiance spectrum from an unknown target material measured under unknown conditions. We demonstrate the subspace representation with a set of material identification experiments on HYDICE imagery from the Forest Radiance I and Desert Radiance II collections.

The TRW Imaging Spectrometer III (TRWIS III) airborne hyperspectral sensor collects imagery in 384 contiguous spectral channels covering the 400 nm to 2450 nm wavelength range. TRWIS III has been used to gather data for a number of remote sensing applications including classification of desert vegetation, forest species, and man-made materials. Analysis has also been performed on TRWIS III agricultural images. This paper will describe the subspace projection technique that was used to process agricultural imagery. An example will be shown in which data dimensionality is reduced by projecting test data onto the space spanned by training data. A minimum Mahalanobis distance measure from the means of the training data class clusters is used as the pixel classification criterion.

Urban areas are characterized by a high frequency of small sized changes in surface cover types whose spatial patterns strongly influence the environmental conditions in cities. Airborne hyperspectral data yield a new potential for their spectrally-based identification, but also raise new challenges in image analysis caused by high spatial and spectral variability of the data. In this context we present a new linear unmixing approach including a pixel-oriented selection of endmember combinations. This approach was especially developed for analyzing urban conditions using data of airborne DAIS 7915 scanner for the city of Dresden, Germany. Because of the big number of spectrally similar endmembers in the urban environment, application of standard unmixing techniques lead to strong local variations of different endmembers and confusion of surface cover types. In comparison to these standard techniques a new extended mathematical model is used which includes stochastic models for each endmember. Additionally, a procedure for pixel oriented selection of likely endmember candidates is developed based on the assumption that the number of endmembers is limited within a pixel. For this purpose, all possible combinations of different endmembers are defined and stored in a list which forms the basis for spatially and thematically constrained endmember selection. These combinations of endmembers are tested during the linear unmixing process. In the result, sensible endmember combinations could be identified during the unmixing process for the 10 km by 4.5 km study area in Dresden. In comparison with standard image classification techniques our approach shows advantages especially in areas dominated by mixed pixels. Thus, a spatially and thematically precise identification of urban surface cover types could be achieved.

The Multispectral Thermal Imager (MTI) is a research and development project sponsored by the Department of Energy and executed by Sandia and Los Alamos National Laboratories and the Savannah River Technology Center. Other participants include the U.S. Air Force, universities, and many industrial partners. The MTI mission is to demonstrate the efficacy of highly accurate multispectral imaging for passive characterization of industrial facilities and related environmental impacts from space. MTI provides simultaneous data for atmospheric characterization at high spatial resolution. Additionally, MTI has applications to environmental monitoring and other civilian applications. The mission is based in end-to-end modeling of targets, signatures, atmospheric effects, the space sensor, and analysis techniques to form a balanced, self-consistent mission. The MTI satellite nears completion, and is scheduled for launch in late 1999. This paper describes the MTI mission, development of desired system attributes, some trade studies, schedule, and overall plans for data acquisition and analysis. This effort drives the sophisticated payload and advanced calibration systems, which are the overall subject of the first session at this conference, as well as the data processing and some of the analysis tools that will be described in the second segment.

MTI is a comprehensive research and development project that includes up-front modeling and analysis, satellite system design, fabrication, assembly and testing, on-orbit operations, and experimentation and data analysis. The satellite is designed to collect radiometrically calibrated, medium resolution imagery in 15 spectral bands ranging from 0.45 to 10.70 micrometer. The payload portion of the satellite includes the imaging system components, associated electronics boxes, and payload support structure. The imaging system includes a three-mirror anastigmatic off-axis telescope, a single cryogenically cooled focal plane assembly, a mechanical cooler, and an onboard calibration system. Payload electronic subsystems include image digitizers, real-time image compressors, a solid state recorder, calibration source drivers, and cooler temperature and vibration controllers. The payload support structure mechanically integrates all payload components and provides a simple four point interface to the spacecraft bus. All payload components have been fabricated and tested, and integrated.

The Multispectral Thermal Imager Optical Assembly (OA) has been fabricated, assembled, successfully performance tested, and integrated into the flight payload structure with the flight Focal Plane Assembly (FPA) integrated and aligned to it. This represents a major milestone achieved towards completion of this earth observing E-O imaging sensor that is to be operated in low earth orbit. The OA consists of an off- axis three mirror anastigmatic (TMA) telescope with a 36 cm unobscured clear aperture, a wide-field-of-view (WFOV) of 1.82 degrees along the direction of spacecraft motion and 1.38 degree across the direction of spacecraft motion. It also contains a comprehensive on-board radiometric calibration system. The OA is part of a multispectral pushbroom imaging sensor which employs a single mechanically cooled focal plane with 15 spectral bands covering a wavelength range from 0.45 to 10.7 micrometer. The OA achieves near diffraction-limited performance from visible to the long-wave infrared (LWIR) wavelengths. The two major design drivers for the OA are 80% enpixeled energy in the visible bands and radiometric stability. Enpixeled energy in the visible bands also drove the alignment of the FPA detectors to the OA image plane to a requirement of less than plus or minus 20 micrometer over the entire visible detector field of view (FOV). Radiometric stability requirements mandated a cold Lyot stop for stray light rejection and thermal background reduction. The Lyot stop is part of the FPA assembly and acts as the aperture stop for the imaging system. The alignment of the Lyot stop to the OA drove the centering and to some extent the tilt alignment requirements of the FPA to the OA.

The focal plane assembly for the Multispectral Thermal Imager (MTI) consists of sensor chip assemblies, optical filters, and a vacuum enclosure. Sensor chip assemblies, composed of linear detector arrays and readout integrated circuits, provide spatial resolution in the cross-track direction for the pushbroom imager. Optical filters define 15 spectral bands in a range from 0.45 micrometer to 10.7 micrometer. All the detector arrays are mounted on a single focal plane and are designed to operate at 75 K. Three pairs of sensor chip assemblies (SCAs) are required to provide cross-track coverage in all 15 spectral bands. Each pair of SCAs includes detector arrays made from silicon, indium antimonide, and mercury cadmium telluride. Readout integrated circuits multiplex the signals from the detectors to 18 separate video channels. Optical filter assemblies defining the spectral bands are mounted over the linear detector arrays. Each filter assembly consists of several filter strips bonded together side-by- side. The MTI focal plane assembly has been integrated with the rest of the payload and has undergone detailed testing and calibration. This paper includes representative test data for the various spectral bands and the overall performance of the focal plane assembly.

The Multi-spectral Thermal Imager (MTI) will be a satellite- based imaging system that will provide images in fifteen spectral bands covering large portions of the spectrum from 0.45 through 10.7 microns. An important goal of the mission is to provide data with state-of-the-art radiometric calibration. The on-orbit calibration will rely on the pre-launch ground calibration and will be maintained by vicarious calibration campaigns. System drifts before and between the vicarious calibration campaigns will be monitored by several on-board sources that serve as transfer sources in the calibration of external images. These sources can be divided into two groups: a set of sources at an internal aperture, primarily intended to monitor short term drifts in the detectors and associated electronics; and two sources at the external aperture, intended to monitor longer term drifts in the optical train before the internal aperture. The steps needed to transfer calibrations to image products, additional radiometric data quality estimates performed as part of this transfer, and the data products associated with this transfer will all be examined.

The major science goal for the Multispectral Thermal Imager (MTI) project is to measure surface properties such as vegetation health, temperatures, material composition and others for characterization of industrial facilities and environmental applications. To support this goal, this program has several coordinated components, including modeling, comprehensive ground-truth measurements, image acquisition planning, data processing and data interpretation. Algorithms have been developed to retrieve a multitude of physical quantities and these algorithms are integrated in a processing pipeline architecture that emphasizes automation, flexibility and robust operation. In addition, the MTI science team has produced detailed site, system and atmospheric models to aid in system design and data analysis. This paper will provide an introduction to the data processing and science algorithms for the MTI project. Detailed discussions of the retrieval techniques will follow in papers from the balance of this session.

The Multispectral Thermal Imager (MTI) has a number of core science retrievals which will be described. We will concentrate on describing the major Level-2 algorithms which cover land, water and atmospheric products. The land products comprise atmospherically corrected surface reflectances, vegetation health status, material identification, land temperature and emissivities. The water related products are: water mask, water quality and water temperature. The atmospheric products are: cloud mask, cirrus mask and atmospheric water vapor. We will present several of these algorithms and present results from simulated MTI data derived from AVIRIS and MODIS Airborne Simulator (MAS). An interactive analysis tool has been created to visually program and test certain Level-2 retrievals.

The retrieval of scene properties (surface temperature, material type, vegetation health, etc.) from remotely sensed data is the ultimate goal of many earth observing satellites. The algorithms that have been developed for these retrievals are informed by physical models of how the raw data were generated. This includes models of radiation as emitted and/or reflected by the scene, propagated through the atmosphere, collected by the optics, detected by the sensor, and digitized by the electronics. To some extent, the retrieval is the inverse of this 'forward' modeling problem. But in contrast to this forward modeling, the practical task of making inferences about the original scene usually requires some ad hoc assumptions, good physical intuition, and a healthy dose of trial and error. The standard MTI data processing pipeline will employ algorithms developed with this traditional approach. But we will discuss some preliminary research on the use of a genetic programming scheme to 'evolve' retrieval algorithms. Such a scheme cannot compete with the physical intuition of a remote sensing scientist, but it may be able to automate some of the trial and error. In this scenario, a training set is used, which consists of multispectral image data and the associated 'ground truth;' that is, a registered map of the desired retrieval quantity. The genetic programming scheme attempts to combine a core set of image processing primitives to produce an IDL (Interactive Data Language) program which estimates this retrieval quantity from the raw data.

We have been studying the use of spectral imagery to locate targets in spectrally interfering backgrounds. In making performance estimates for various sensors it has become evident that some calculations are unreliable because of overfitting. Hence, we began a thorough study of the problem of overfitting in multivariate classification. In this paper we present some model based results describing the problem. From the model we know the ideal covariance matrix, the ideal discriminant vector, and the ideal classification performance. We then investigate how experimental conditions such as noise, number of bands, and number of samples cause discrepancies from the ideal results. We also suggest ways to discover and alleviate overfitting.

Sandia National Laboratories (SNL), Los Alamos National Laboratory (LANL) and the Savannah River Technology Center (SRTC) have developed a diverse group of algorithms for processing and analyzing the data that will be collected by the Multispectral Thermal Imager (MTI) after launch late in 1999. Each of these algorithms must be verified by comparison to independent surface and atmospheric measurements. SRTC has selected 13 sites in the continental U.S. for ground truth data collections. These sites include a high altitude cold water target (Crater Lake), cooling lakes and towers in the warm, humid southeastern U.S., Department of Energy (DOE) climate research sites, the NASA Stennis satellite Validation and Verification (V&V) target array, waste sites at the Savannah River Site, mining sites in the Four Corners area and dry lake beds in Nevada. SRTC has established mutually beneficial relationships with the organizations that manage these sites to make use of their operating and research data and to install additional instrumentation needed for MTI algorithm V&V.

The amount of information contained in a single hyperspectral image is overwhelming for the human operator. As a result, assessing the spatial and spectral variability of a hyperspectral image is very difficult. The existing techniques mainly rely on different preprocessing algorithms that reduce the high-dimensionality of the hyperspectral data down to a few images that can be visualized using traditional RGB or RGBI combinations. The proposed auto-correlogram approach provides a simple framework for reducing a hyperspectral image cube to a single grayscale image that is easy to interpret and screen for spectral anomalies.

One of the elements of the European Space Agencies new 'Living Planet' Programme for Earth Observation are the Earth Explorer Missions. These are research/demonstration missions with the emphasis on advancing understanding of different processes, which help govern the Earth Systems. One of the four Earth Explorer Missions which was subjected to a Phase A study is the Land Surface Processes and Interactions Mission (LSPIM). The major scientific objective of the LSPIM is to increase our understanding of land-surface processes and interactions. In order to retrieve relevant geo-/biophysical variables needed to feed and adjust the models describing the processes and interactions, detailed information at high spatial, spectral, directional and temporal resolution is required. To fulfill the mission objectives, a hyperspectral imager is proposed as the LSPIM instrument. The mission will provide contiguous spectral coverage in 142 bands within the solar spectral region (0.45 - 2.35 micrometer). The spectral sampling interval will be 10 nm in the VNIR and 15 nm in the SWIR. Further TIR observations will be performed in two wavebands. A spatial resolution of 50 m X 50 m with a swath width of 50 km will be provided. This mission will also have a depointing capability for angular observations along-track and areal access across track. Seven directions from which one is programmable available for angular observation. It is the purpose of this paper to outline the planned spaceborne mission.

Utilization of sub-pixel targets for radiometric calibration of airborne and space-borne imaging sensors involves the uncertainty of their contribution to the pixel-integrated radiance. This contribution depends not only on the target area but also on an unknown location of the sub-pixel target within a sensor pixel. A technique is proposed to retrieve both the target radiance and its sub-pixel location from the target image, taking into account the effects of the sensor point spread function. The technique was used for in-flight calibration of the thermal channels of the airborne imaging spectrometer DAIS-7915.

A coherent fiber bundle for infrared image transmission was prepared by arranging 8400 chalcogenide (AsS) glass fibers. The fiber bundle, 1 m in length, is transmissive in the infrared spectral region of 1 - 6 micrometer. A remote spectroscopic imaging system was constructed with the fiber bundle and an infrared PtSi CCD camera. The system was used for the real-time observation (frame time: 1/60 s) of gas distribution. Infrared light from a SiC heater was delivered to a gas cell through a chalcogenide fiber, and transmitted light was observed through the fiber bundle. A band-pass filter was used for the selection of gas species. A He-Ne laser of 3.4 micrometer wavelength was also used for the observation of hydrocarbon gases. Gases bursting from a nozzle were observed successfully by a remote imaging system.

An imaging variable retardance, Fourier transform spectropolarimeter (VRFTSP) is presented that is capable of creating spectropolarimetric images of scenes with independent characterization of spatial, spectral, and polarimetric information. The device is used to image simple, canonical targets such as spheres and cylinders in a laboratory setup. The resulting images demonstrate the capability of developing systems to collect spectropolarimetric data of field images using the concept of pushbroom scanning and serial collection of polarimetric information, with the possibility of developing a parallelized collection strategy allowing the collection of near-real-time images of real-world targets.

Two simple and compact pushbroom spectrometer forms are described that can satisfy stringent spectral and spatial uniformity requirements in terms of minimizing distortion as well as the variation of the pixel spectral and spatial response functions.