Joint Information Theoretic and Differential Geometrical Approach for Robust Automated Target Recognition.

摘要

The overall objective of this project is to develop transformative theory and algorithms for robust Automated Target Recognition (ATR). This project addressed the following challenging problems in ATR: modeling uncertainty, small sample size, high dimensional data, irrelevant features/dimensions, heterogeneous data, and outliers. In this project, the PI proposed and developed the following new techniques: 1) kernel local feature extraction (KLFE) for ATR applications, 2) technique for identifying network dynamics under sparsity and stationarity constraints, 3) self-organized-queue- based (SOQ) clustering scheme, 4) robust principal component analysis (RPCA) based on manifold optimization, outlier detection, and subspace decomposition.