A many original contributions to an automatic target recognition (ATR) system, that is applied to underwater mine classification. The classification of real-world empirical targets using sensed imagination into different perceptual categories is one in all the most difficult algorithmic elements of radar systems. The contributions think about feature selection and object classification. First, a complicated filter technique is designed for the feature selection. This filter technique utilizes a unique feature relevance measure, the composite relevance measure (CRM). The contributions consider feature selection and object classification. The design of one classification system that was optimized in 2 elementary aspects: the selection of the classification system and also the selection of the optimal feature subset. ATR system is applied to real synthetic aperture sonar imagination to evaluate its performance