Automatic Target Recognition Using a Novel Adaptive Technique for Rayleigh Quotient Quadratic Correlation Filters

摘要

A recent approach for Automatic Target Recognition (ATR) uses Quadratic Correlation Filters (QCF). This technique, called Rayleigh Quotient Quadratic Correlation Filter (RQQCF) technique, optimizes the QCF by explicitly maximizing a class separation metric. In the process, the filter coefficients are obtained from the Eigenvectors of a large matrix derived from the autocorrelation matrices of target and clutter training images. Whenever a few new images need to be incorporated, the RQQCF has to be retrained. This can be computationally expensive because the Eigenvalue Decomposition (EVD) has to be performed repeatedly. In this contribution, a novel method is proposed to obtain the new EVD adaptively from the old EVD, in a computationally efficient way relative to existing methods. Using synthetic datasets, sample results are obtained that illustrate the excellent performance of the proposed technique.