Database Expansion Issues

Software tools will be developed that define the underlying phenomenology of greatest importance to radar CID performance, allowing a trace back to the responsible target scattering physics. The proposed research entails development of algorithms that identify the target-sensor orientations (poses) of most relevance for classifying targets of interest to AFRL. These target-sensor poses will then be traced back to the associated phenomenology, to gain an understanding of what poses (and classes of scattering physics) must be modeled accurately for optimal classification. Algorithms will also be developed that address selection of the most-relevant target-sensor poses while simultaneously defining the most-relevant features of these signatures; relevancy will be quantified in terms of CID classification performance. The sensitivity of a CID classifier will then be examined as a function of errors in rendering the most-relevant signatures and signature components. In this manner model accuracy on the most-relevant phenomenology is mapped to anticipated CID performance. The use of signature features will also be investigated, for classifier design, and for development of simplified scattering models. The linkage between most-relevant features and the accuracy of the data needed for their computation will provide a mapping between discriminative features and required model fidelity.