SAR Automatic Target Recognition Based on Classifiers Fusion

Synthetic aperture radar automatic target recognition (SAR ATR) remains a challenging problem in military and civil field. Much work has been done to improve the performance of SAR ATR systems, both in feature extraction and classifier designing. This paper designs a multiple classifier system to solve the target classification problem in the area of SAR ATR. The proposed multiple classifier system trains three classifiers on different feature sets using three leaning algorithms. The outputs of the three classifiers are combined through evidence combination rule and discounting operation of Dempster-Shafer theory of evidence. Experiments on MSTAR public data set demonstrate that the proposed multiple classifier system significantly outperforms single classifiers and also excels adaptive boosting with RBF network as base learner.