In this paper, the classification algorithm is used to classify and recognize ten types of targets and three types of target variants on the MSTAR data set. Then, according to the prior knowledge in the classification adjustment process, the evidence is corrected, namely the output of the classifier, and the basic confidence function is constructed. The improved combination rule is a conflict measurement method based on the combination of the conflict coefficient K and the Pignistic probability distance, and the conflict evidence is synthesized by the combination rule that distributes the conflict degree in proportion. Without fusion, the highest classification accuracy rate of the three types of target variants is 93.553%. After fusion, the classification and recognition rate of the three types of target variants is 95.092%, which is increased by about 37% of that of the ideal state.