Physics-Based Automatic Recognition of Small Features Located in Highly Similar Structures With Electromagnetic Scattering Data

Abstract

A physics-based automatic target recognition (ATR) technique is developed to accurately identify small features located in highly similar structures with electromagnetic (EM) scattering data. Automatic target recognition is important due to its widely practical applications. The traditional ATR is usually based on images produced from EM scattering data and sophisticated algorithms. Wideband angular and frequency sweeps are necessary to generate sufficient EM scattering data to produce images with high resolution for the imagery-based ATR to obtain correct recognition results, especially for multiscale structures with small local features. These seriously limit the efficiency of the imagery-based ATR and its practicability. To implement ATR more efficiently, we turn to the physics-based ATR and employ principal component analysis (PCA). The physics-based ATR with PCA can exactly classify objects of different types with one-frequency scattering data and avoid expensive frequency sweeps. However, the pre-existing average feature center (AFC) criterion model for PCA in the literature can only distinguish objects with significant differences and fails to recognize small features located in highly similar structures. Hence, an improved classification criterion for PCA is proposed to precisely identify highly similar structures with different small features. Some numerical examples illustrate the satisfactory performance of the proposed technique.