OPTIMUM BAND SELECTION TECHNIQUES UNDER HYPERSPECTRAL IMAGE ANALYSIS FOR MATERIAL IDENTIFICATION AND CLASSIFICATIONTHE COMPLEXITY INVOLVED IN COLLECTING STORING ANALYSIS AND PROCESSING OF VOLUMINOUS REMOTE SENSING MULTI-DIMENSIONAL DATA FOR AN ARRAY OF "EARTH SYSTEM SCIENCE" APPLICATIONS IS A WELL KNOWN PROBLEM TO NASA. THE RECENT HYPERSPECTRAL IMAGING (HSI) TECHNOLOGY HAVE EVOLVED FROM ITS EARLIER VERSION OF MULTISPECTRAL IMAGING (MSI). IN HSI IMAGES ARE ACQUIRED USING HUNDREDS OF SPECTRAL CHANNELS WHEN COMPARED TO FEWER CHANNELS IN MSI. HOWEVER OVER THE YEARS THERE HAS NOT BEEN MUCH SIGNIFICANT DEVELOPMENT OF PROCESSING ALGORITHMS FOR THESE EVER GROWING (IN THE SPECTRAL DIRECTION) ELECTRO-OPTICAL DATA SETS. THE NEED TO COME UP WITH INTELLIGENT BAND SELECTION TECHNIQUES IS EVEN STRONGER SINCE WE ARE MOVING IN TO THE ERA OF ULTRASPECTRAL IMAGING (USI). THE USI TECHNOLOGY USUALLY CONSISTS OF MEASUREMENTS TAKEN IN THOUSANDS OF SPECTRAL CHANNELS. IN ALL THREE IMAGING TECHNOLOGIES EACH SPATIAL PIXEL IS TREATED AS A COLUMN VECTOR THAT CONTAINS SPECTRAL INFORMATION FROM EACH CHANNEL. DUE TO THE EVER GROWING HIGHER SPECTRAL DIMENSIONALITY OF THE REMOTE SENSING DATA EACH IMAGE PIXEL VECTOR POSSESSES MORE INFORMATION THAN EVER BEFORE. HOWEVER DUE TO NOISE AND OTHER UNAVOIDABLE ATMOSPHERIC INTERFERENCE A PURE SIGNATURE FOR A CLASS KNOWN AS THE END-MEMBER MAY BE REPRESENTED BY A MIXTURE OR COMBINATION OF SPECTRAL SIGNATURE FROM NEIGHBORING PIXELS. TO ENCAPSULATE THE UNCERTAINTY CREATED BY A PIXEL VECTOR DUE TO NOISE AND OTHER INTERFERENCE DEMANDS A STOCHASTIC TREATMENT. THIS PROPOSAL FOCUSES ON DERIVING CLASSIFICATION AND TARGET DETECTION METHODS BASED ON NON-TRANSFORMED DATA. THE INTENT IS THAT WITHOUT THE COST OF THE DECORRELATING TRANSFORM THAT THE DERIVED CLASSIFICATION ALGORITHMS WILL HAVE SUITABLE COMPLEXITY FOR USE IN PRACTICAL CLASSIFICATION SYSTEMS. EARLIER RESEARCH HAS INVESTIGATED USING A REDUCED COLOR SPACE FOR IMPLEMENTING MATERIAL CLASSIFICATION OFHSI DATA. THIS "BEST BAND" SELECTION (BBS) METHOD RELIES SOLELY ON THE DETERMINISTIC DISTANCE MEASURES OF MINIMUM EUCLIDEAN DISTANCE AND SPECTRAL ANGLE MAPPER. THE RESULTS OF THIS RESEARCH WERE ALGORITHMS THAT GAVEENHANCED CLASS SEPARATION USING A REDUCED NUMBER OF SPECTRAL COLOR BANDS. OUR CURRENT RESEARCH EXTENDS THE OPTIMUM BAND SELECTION TECHNIQUES IN TWO SEPARATE DIRECTIONS. THE FIRST IS TO EXTEND THE BBS RESULTS TO STOCHASTIC DISTANCE METRICS SUCH AS HIDDEN MARKOV AND LEIBLER-KULLBACK MEASURES. STUDIES HAVE SHOWN THAT THE STOCHASTIC DISTANCE MEASURES CAN RESULT IN IMPROVED CLASSIFICATION PERFORMANCE RELATIVE TO DETERMINISTIC METRICS. TO DATE THE PROBLEM OF BEST BAND SELECTION USING STOCHASTIC METRICS HAS NOT BEEN ADDRESSED IN THE LITERATURE. THE ISSUE OF STOCHASTIC MEASURES FOR BBS WILL BE THE PRIMARY FOCUS OF THE PROPOSED RESEARCH. THE SECOND DIRECTION FOR EXTENDING BEST BAND SELECTION IS IN INCORPORATING SENSOR MODELS OF HSI DATA COLLECTION INTO THE BBS ALGORITHM. WE WILL SPECIFICALLY EXTEND THE BBS METHOD USING DERIVED ANALYTICAL MODELS OF SPECTRAL PROFILES.