HRR TOS/TOM Features and Classifiers Using Boundary Methods

摘要

This project is focused on the development of features and classifiers for TOS/TOM classifiers, specifically for HRR applications. The technical approach to this problem is broken into two parts: 1. feature generation and feature set evaluation, and 2. classifier design. Both data- driven and physics-based models to produce features are evaluated to determine which set(s) of features are robust to the differences between measured and synthetic data. Feature Set Evaluation (FSE) is accomplished using both conventional techniques (e.g., kNN) and using a technique called Boundary Methods. New classifier designs are being developed that use these features to construct classifiers and verify that the developed classifiers perform well when trained with synthetic data and tested on measured data. Moreover, the classifiers make efficient use of a limited set of stored templates in order to mitigate computational problems.