We have developed a method that uses genetic algorithms (GAs) to optimize rules for categorizing the terrain in Landsat data. A rule has two parts: a left side (the 'if' clause) and a right side (the 'then' clause). When the 'if' clause is true, the functions in the 'then' clause are executed to process the Landsat data. Examples of functions for processing the data include pixel by pixel threshold and a linear combination of six bands. Optimized rules are used to identify different terrain categories within Landsat data. Optimization is performed by comparing the results of the rules with ground truth using an objective function which minimizes the number of false positive and false negative pixel labels. Those rules that generate results close to the ground truth (those rules that return a small number of false positive and false negative pixel identifications) are highly rewarded and are used to create the next generation of rules. High altitude photographs were used as ground truth. The GA produced promising results for terrain categorization when compared with results from a maximum likelihood classifier. More work in the area of terrain categroization is planned to build on these promising results.

This paper introduces a model based morphology technique for detecting linear segments in Landsat TM imagery. Linear segment detection has been identified as the initial step in many automated processes -- multiple image registration using linear features, lines of communication (LOC) detection, and target detection algorithms. Even though there have been many research studies aimed at the detection of linear segments in single band imagery, little research has been done on multispectral data which considers all the bands simultaneously. The fundamental idea behind this paper is to apply normal gray level image operations to multispectral imagery. Simply applying a single band edge detection algorithm to each band separately ignores the radiometric response of the edge detector. Applying morphological based operations to each band and accounting for the radiometric response of the target (i.e., LOC) pixels, one may develop a measure for the radiometric fit for each band and combine the individual band responses into a single useful output. Application of the model-based morphological filter designed in accordance with the radiometric responses of the linear segments and fusion of the results from all the bands improves linear segment detection.

In this paper we present a new method for the extraction of information about lines of communciation (LOC) from Landsat Thematic Mapper (TM) imagery. While other techniques have demonstrated some success in the extraction of LOC information, they generally suffer from the inability to handle a wide variety of LOC shapes or they lack the capability to globally integrate information in order to determine the location of the LOCs. In this paper we present the development of an alternative to previously tried methods which promises to do a better job of both local and global extraction of LOC information.

Image analysis is especially labor-intensive in the multispectral domain due to the non-intuitive nature of the data. The data is frequently separated by frequency range into a set of image planes. For image analysts (IAs) and photo interpreters to derive meaningful conclusions from the imagery, the data needs to be fused into a form that is coherent to the human mind. The lines of communication extraction (LOCX) system was designed to merge both spectral and spatial image processing technologies into a user-friendly package that exploits the multispectral aspect of Landsat TM imagery without overwhelming the user with the complexities of the data. This paper discusses in detail the system components, the operations concept and user interface, and preliminary processing results.

A new technique for the extraction of lines of communication (LOCs) from Landsat Thematic Mapper (TM) data is presented. A multi-stage approach is taken. First, LOC segments are detected. Next, gaps between segments are filled by a segment connection routine. Finally, the connected segments are identified. In the segment detection stage a distinction is made between wide LOC and narrow LOC segments. Wide LOC segments are detected by a neural network based segmentation algorithm. Input to the network are spatial and spectral features computed about each pixel. Narrow LOC segments are detected by a cost minimization algorithm designed to work on the multispectral data. Segment connection is performed by connecting those LOC segments that have consistent alignment, position, and spectral signatures. The identification stage consists of a neural network with inputs of shape and spectral features about each connected LOC. While the system is still being refined, most critical pieces have been prototyped and tested. Initial results are encouraging.

Multispectral imaging instruments are often built in a pushbroom scanner configuration with focal planes consisting of large linear arrays containing up to several thousand detectors. Design considerations of such arrays often favor arrangements such that all even numbered detector elements are processed along a path that differs from all odd numbered detector elements. this can cause instrument-induced artifacts in the resulting imagery. Two techniques have been developed that attempt to attenuate this type of artifact. One is based on modeling the artifact as differences in detector bias. These differences are estimated in the spatial domain and corrections are applied. The second approach takes advantange of a convenient wavelet decomposition of the iamge that effectively isolates the artifact from image information. After appropriate filtering in wavelet space, artifacts in the reconstructed image are significantly reduces. These techniques have been applied to SPOT imagery with encouraging results.

Multispectral image sharpening involves enhancing the spatial characteristics of source multispectral imagery (MSI) acquired at low-resolution using a coregistered reference image acquired at a higher spatial resolution. As analysts become better trained in interpreting MSI and rely on spectral information for interpretation, it will be crucial that the sharpened products preserve the spectral information resident in the source MSI. We present a novel approach to sharpening which is explicitly designed to yield results which are consistent with the spectral information in the source MSI, i.e., when the sharpened MSI is filtered and decimated, the source MSI is reconstructed. Our approach involves developing explicit models that embody the assumed relationships among the source, reference and desired sharpened imagery. A sharpening algorithm is then posed as the solution to a constrained model-fitting problem. In this paper we discuss the general model-based image sharpening approach, and discuss a variety of possible models relating the reference and MSI datasets, and the resulting sharpening algorithms.

Image sharpening is defined as a process in which high-resolution information from one image is used to increase the spatial resolution of a second coregistered image of lower resolution. Image sharpening is of interest for applicaiton to multispectral and multisensor imagery. We present an adaptive technique to perform image sharpening based on the use of multiresolution image representations and neural networks. This technique is easily applied to provide resolution enhancement by powers of two and promises to accurately maintain the spectral characteristics of the original low-resolution image.

Many techniques have been developed and demonstrated to combine low spatial resolution multispectral imagery with relatively high resolution panchromatic imagery to accomplish multispectral 'sharpening'. We investigate here a particular approach, high frequency modulation, and show that care must be exercised in the design of spatial weighting functions in order to optimize the resulting fused image. Mathematical relationships are derived showing that, under reasonable assumptions, the modulation equation leads to a specific and intuitively satisfying form of the weighting function. Examples of sharpened multispectral images are presented demonstrating the predicted improvement when attention is given to the spatial weighting function.

Effective change detection techniques for the automated detection of changes of interest have been an elusive goal for many years. The problem has never been one of detecting changes but, rather one of finding the changes of most interest among all the spurious changes. Indeed, changes of interest for one application can be completely different than the changes of interest for another application. In this paper we present a brief overview of techniques to suppress changes between scenes due to different collection conditions. Techniques for sorting the detected multispectral changes according to the intended application and relating them to actual changes on the ground are presented. Our approach is to use multispectral data transformations designed by the use of a visualization tool called Projection Pursuit. This tool allows the user to design a projection of the data into a vector space specifically designed to accentuate the visibility of the changes of interest. Hence, for change detection interactive analysis, projection pursuit offers the important advantage of being able to find the optimum projection without requiring a priori information from the image analyst and requiring little human intervention. This algorithm is complemented by canonical and principal component transformations tailored for specific exploitation requirements. The approach allows design of custom change detection products for a wide variety of applications including: military, economic, and environmental. This capability reduces the burden of data manipulation decisions required of the analyst, while still providing the flexibility required for the demands of exploitation.

Trophy and morphology of rivers, lakes and coastal waters are the most important factors to estimate the status of the waters in respect to biological and ecological situations. Additionally trophy of freshinland waters is a criterion for water quality such as drinking water. Usually the chlorophyll concentration produced by phyto-', meso- and zoo—plankton indicates the status of waters, but it does not regard the dynamic aspects of trophy. Therefore the primary productivity, as the dynamic part of trophy is necessary to xnonitore . This needs much• more complex methods and algorithms than just a correlation of chlorophyll with any remote sensing recorded radiances. The spectral signatures of the primary 4 trophical situations and a concept for the determination of primary productivity from airborne hyperspectral data are described.

A combination of towed and stationary shipboard measurements, bio-optical mooring data, and a series of AVIRIS images were acquired near an offshore sewage outfall which services the Los Angeles metropolitan area. The image containing the outfalls and in-situ measurements was examined using statistical techniques to derive the spectral components responsible for salient variations. These computed components agree quite well with spectra which were hand selected to qualitatively represent the scene. Suspended sediment and aquatic vegetation detection were investigated using spectral derivative techniques, which were quite successful in showing the extent of offshore kelp forests and storm-induced resuspension of previously deposited outfall effluent, or the effluent itself. The in-situ and remote sensing measurements were used to constrain a numerical model of the underwater light field, and the scattering properties along a surface transect were investigated using two phase functions. Using a one-term Henyey-Greenstein function with a backscattering ratio of nearly 30%, we were able to reproduce the remotely- sensed radiance, suggesting that the particulates had a size distribution skewed toward small. These investigations show the utility of high spectral resolution in determining the extent and character of several important natural and anthropogenic components of coastal areas, as well as parameterizing models of the inherent optical properties of the underwater light field.

Synthetic hyperspectral signatures representing an airborne target engine radiation, a decoy flare, and the engine plume radiation are used to demonstrate computational techniques for the discrimination between such objects. Excellent discrimination is achieved for a `single look' at SNR of -10 dB. Since the atmospheric transmittance perturbs the signature of all objects in an identical fashion, the transmittance is equivalent to a modulation of the target radiance (in the spectral domain). The proper spectral signal decomposition may, therefore, recover the original unperturbed signature accurately enough to allow discrimination. The algorithms described here, and in two accompanying papers, have been tested over the spectral range that includes the VNIR and MWIR and are most appropriate for an intelligent, autonomous, air-to-air or surface-to-air guided munitions. With additional enhancements, the techniques apply to ground targets and other dual-use applications.

A system is presented for compression of hyperspectral imagery which utilizes trellis coded quantization (TCQ). Specifically, TCQ is used to encode transform coefficients resulting from the application of an 8X8X8 discrete cosine transform. Side information and rate allocation strategies are discussed. Entropy-constrained codebooks are designed using a modified version of the generalized Lloyd algorithm. This entropy constrained system achieves a compression ratio of greater than 70:1 with an average PSNR of the coded hyperspectral sequence exceeding 40.5 dB.

An approach to the invariant processing system design is proposed, based on the application of the Hermit polynomials. This approach rests on the usage of the expansion of a signal by special polynomials of discrete argument--the Krawtchouk polynomials. The properties of the Krawtchouk polynomials make it possible to select a complete system of informative features of signals, which are invariant with respect to linear and some nonlinear transformations. The algorithm of selection of informative features of a signal provides a basis for the software intended for signal compression and restoration as well as the recognition and classification of signals. We also describe some applications of the above software to the processing of some biological and medical signals and images.

This paper reports results from an analysis of polarimetric hyperspectral imagery collected using a prototype acousto-optic tunable filter instrument in an outdoor environment. Spectra, derivative spectra, and polarization spectra in the image cube form were studied. The issue concerns spectral and polarization signatures of vegetation, contributions due to aerosol, and man- made object detection. The result illustrates potentials of the technology for a variety of remote sensing applications.

Analysis of data collected during a ground-based experiment of an acousto-optic tunable filter (AOTF) hyperspectral imaging system shows potential for utilizing this technology for feature identification. The unique capability of an AOTF system to simultaneously acquire two orthogonally poiarized images allows both spectral and polarimetric characterizafion of ground features. The AOTF sensor used in the experiment operated over a wavelength range from 0.5 1 to 0.77 microns, collecting two differently polarized images for each of 33 bands. The experiment data images selected for analysis contain camouflaged military equipment deployed in a desert background. The Spectral Image Processing System (SIPS) software package was used for data correction and spectral analysis. Processing the AOTF data required geometric correction in addition to removing the solar and atmospheric effects. Two methods of spectral analysis are addressed in this research. The first method assesses the spectral analysis accomplished using the Spectral Angle Mapper utility in SIPS. The other method illustrates the spectral information gained through band ratioing a combination of carefully selected bands which highlights a particular target within a scene. Comparison of images created by the difference between polarizations for each band provides the basis for polarimetric analysis of the data. Finally, an algorithm developed to combine the information provided by spectral and polarimetric analysis shows how features within a scene can be distinguished from the background. Results show that AOTF hyperspectral technology has potential to enhance current military intelligence collection capability.

To achieve the research objectives of the Airborne Visible/IR Imaging Spectrometer (AVIRIS), the sensor calibration must be valid while AVIRIS is acquiring data from the airborne platform. The operational environment inside the aircraft differs significantly from that in the AVIRIS laboratory environment where the sensor is calibrated prior to and following each flight season. To independently validate the calibration of AVIRIS in the flight environment an inflight calibration experiment is conducted at least twice each flight season. Results for a calibration experiment held on the 26th of September 1993 are presented.

In this paper, we describe an experiment to calibrate the Optical Sensor on board the Japanese Earth Resources Satellite-1 with data acquired by the Airborne Visible/IR Imaging Spectrometer. 27 August 1992 both OPS and AVIRIS acquired data concurrently over a calibration target on the surface of Rogers Dry Lake, Nevada. The high spectral resolution measurements of AVIRIS were convolved to the spectral response curves of the OPS. These data, in conjunction with the corresponding OPS digitized numbers, were used to generate the radiometric calibration coefficients for the eight OPS bands. This experiment establishes the suitability of AVIRIS for the calibration of space-borne sensors in the 400-2500 nm spectral region.

Detecting changes in the Earth's environment using satellite images of ocean and land surfaces must take into account atmospheric effects. As a result, major programs are underway to develop algorithms for image retrieval of atmospheric aerosol properties and atmospheric correction. However, because of the temporal and spatial variability of atmospheric transmittance, it is very difficult to model atmospheric effects and implement models in an operational mode. For this reason, simultaneous in situ ground measurements of atmospheric optical properties are vital to the development of accurate atmospheric correction techniques. Presented in this paper is a spectroradiometer system that provides an optimized set of surface measurements for the calibration and validation of atmospheric correction algorithms. The portable ground-based atmospheric monitoring system (PGAMS) obtains a comprehensive series of in situ irradiance, radiance, and reflectance measurements for the calibration of atmospheric correction algorithms applied to multispectral and hypserspectral images. The observations include: total downwelling irradiance, diffuse sky irradiance, direct solar irradiance, path radiance in the direction of the north celestial poles, path radiance in the direction of the overflying satellite, almucantar scans of path radiance, full sky radiance maps, and surface reflectance. Each of these parameters are recorded over a wavelength range from 350 to 1050 nm in 512 channels. The system is fast, with the potential to acquire the complete set of observations in only 8 to 10 minutes depending on the selected spatial resolution of the sky path radiance measurements.