A Novel Approach for Spectral Unmixing and Classification of Chemical and Biological Agents

Besides performing a thorough statistical analysis of the government furnished data sets to determine the features and metrics for spectral unmixing, Intelligent Automation, Inc. and Prof. C. Chang of University of Maryland at Baltimore County also proposean innovative approach to detect and classify chemical and biological agents. There are two major steps. First, we propose to apply a recently developed linear spectral random mixture analysis (LSRMA) method to perform the spectral unmixing operation. Themethod extends commonly used linear spectral unmixing to random spectral unmixing in the sense that the abundance fractions considered in the latter are now modeled by random parameters rather than the unknown constants treated in the former. Thistreatment is more close to reality as agents have deformable shapes. Most importantly, the proposed LSRMA does not require any a priori information and can be fully automated in computers. Second, once the spectral information has been unmixed, we proposeto apply a state-of-the-art technique in pattern recognition to perform automatic detection and classification of agents. The technique is called Support Vector Machine (SVM), which has several remarkable advantages such as global optimal solution, noover-training issue, and better performance than most existing classification schemes. The proposed algorithm will be useful for chemical and biological agent detection and classification. The market for military applications is quite large. Otherpotential applications include law enforcement agency, homeland security, border and coast patrol. The potential market is both domestic and international in scope.