Deep Convolutional Neural Networks With Integrated Quadratic Correlation Filters for Automatic Target Recognition

Brian Millikan, Hassan Foroosh, Qiyu Sun

Abstract: Automatic target recognition involves detecting and recognizing potential targets automatically, which is widely used in civilian and military applications today. Quadratic correlation filters were introduced as two-class recognition classifiers for quickly detecting targets in cluttered scene environments. In this paper, we introduce two methods that integrate the discrimination capability of quadratic correlation filters with the multi-class recognition ability of multilayer neural networks. For mid-wave infrared imagery, the proposed methods are demonstrated to be multi-class target recognition classifiers with very high accuracy.