Exploiting Multipath for Efficient Target Classification

Automatic target recognition(ATR) for tactical military targets is a very challenging problem, particularly when the number of observed aspects of a target is subject to the operational constraints of the sensor platform. When the target is situated in the presence of a complicated background medium, such as in urban or hilly settings, the observed signals might be deformed significantly, thus the classification performance suffers. However, it can be proved that the observed signals of a target in the presence of a complicated environment are linear combination of the target’s scattering fields situated in free space. Thus the target is actually observed from multiple aspects for each such measurement. On the other hand, the angle-dependent target responses are typically a smooth function of aspect angles, and therefore are highly compressible. In this Phase I, the concept of in situ compressive sensing (CS) will be employed to exploit the multipath response. ATR can subsequently be performed with a relatively smaller number of observations within compact aspect range. The target RCS/far fields in free space will be simulated via multilevel fast multipole algorithm (MLFMA), and various random projection matrices will be generated to simulate simple and complicated environments, respectively. Sparse Bayesian classifiers will be used to assess the classification performance and efficiency of CS compared with conventional approaches.