Modified eigen-image approach to radar target identification

Abstract

A modified approach to the eigen-image classifier is presented. The criterion for selection of eigenvectors is based on minimizing the ratio of the sum of the intra-class variances to the inter-class variance. On average, the percentage of correct classification has gone up by 3.6% from 89.3% to 92.9% using this modified selection criterion. Correct classification performance is not affected by the modified selection criteria, if the images in that sector behaved according to the ideal case, where the intra-class variance is small and inter-class variance is large.