Technological advances in sensors and imaging systems for tactical applications have paradoxically created new problems in the process of solving old ones. For example, the development of high-resolution multispectral and hyperspectral sensors, capable of detecting objects at the pixel level, has simultaneously increased the sheer volume of data to be processed on each scene under consideration. Unfortunately, this data dimensionality explosion has not been matched with a proportionate increase in practically useful information for automatic target recognition tasks. This critical problem is exacerbated when surveillance, reconnaissance, or theater combat operations must fuse high-dimensional information obtained from multiple sensor system modalities in near real time. Confidence in automatic target recognition (ATR) decisions is improved by synthesizing a variety of digital image representations, each of which contains information and identification clues regarding target physics unique to a particular region of the electromagnetic spectrum. The objective is to propose image processing correlation techniques capable of multiple-sensor "smart systems" that can enhance identification and provide location coordinates for sensor-to-shooter systems. The United States Air Force is seeking a systematic and principled analytical means for maximizing the information to data ratio in multisensor ATR processing. It is not just enough to rely on the processing of a particular sensor to provide all scene information. Data from sensors must be correlated. Correlating this data across sensor suites and finding the optimal set of salient target features producing rapid and unequivocal automatic target recognition for potentially life-threatening situations is a critical need.
Keywords: IMAGE PROCESSING, ATR, FUSION, HYPERSPECTRAL, INFRARED, NEURAL NETWORKS, SAR
