AUTOMATED DETECTION AND IDENTIFICATION OF MATERIALS IN HYPERSPECTRAL IMAGES

The goal of this program is to develop and implement optimal spectral signature detection techniques which provide automated detection and identification of materials in hyperspectral image sequences. The requirement to detect, identify and discriminate materials is a common theme in many applications of imaging spectrometry. In general, materials of interest are often smaller than the spatial resolution of the sensor. In this case, the signal observed at the sensor is a mixture of the reflected materials. This subpixel mixture characteristic offers a unique signal processing problem which cannot be optimally solved with standard classification techniques or typical detection strategies such as the matched filter and spectral signature matching. In this Phase I effort, an optimal detector for subpixel signatures of interest in the presence of multiple undesired signatures and noise will be developed. Two additional approaches for detecting subpixel materials in cases where the undesired signatures are not known 'a priori will also be developed and implemented. The theoretical performance of the techniques will be rigorously described, and verified by application to available HYDICE and AVIRIS data.