New SVM Hierarchical Classifier Fusion

This Phase II proposal is focusing on further development of our new SVRDM-based ATR technologies, developed in Phase I, with the capability of rejecting non-objects. In particular, we will develop segmentation algorithms to omit clutter around the object, develop methods to handle more thermal variations, develop soft decision hierarchical classifiers for the efficient multi-classifier fusion, and develop methods for learning new aspect views of targets during the system operation. In addition to comparison to other classifiers, we will conduct tests on more databases. Our primary objective of Phase II is the product development. We will develop products that include an industrial PC with interface to IR sensors and spectrometers, and a suite of software programs from data acquisition to target classification. During Phase II, we will develop two products: (1) embedded middleware ATR system aimed at an embedded computing platform, and (2) standalone signature-based spectra classification system. These two products will be used in ATR applications for military and in environmental monitoring for process industry. Algorithms developed in Phase I/II will be implemented in C/C++. Effort will be focused on accuracy, autonomous, speed and efficiency of the system. A detailed sales and marketing business plan will be developed.