The conribution of fusion techniques in the recognition systems of radar targets

For several years, different types of classifiers have been developed using several features vector in the automatic target recognition (ATR) field. However, because of measurement conditions and extraction techniques, the individual results performance obtained by different approaches of classification are varied and different. For these reasons and as part of the radar target recognition, we present in this paper a study which deals with a comparison of four fusion techniques: voting majority, Bayesian fusion, belief fusion and possibility fusion. To implement these fusion methods, we used three classifiers: the Support Vector Machines SVM, Neurons Networks and K Nearest Neighbor (KNN) to improve the final decision for ATR. In addition, we present the results obtained from real data performed in the anechoic chamber of ENSTA Bretagne. Thereby demonstrating the contribution, performance and robustness of the approach developed and applied in aid to recognition of radar targets.