When using traditional Siam network tracking algorithms for target tracking,if the target position is far from the target center point on the feature graph,the tracking performance indexes will be lowered and the tracking effects will be poor.To solve this problem,an improved Siamese Box Adaptive Network (SiamBAN) algorithm is proposed.The center prediction network is introduced to determine the distance between the target position and the target center point on the feature graph.When the distance between the two positions is larger,the centrality score is lower and the algorithms ability to remove outliers in responses is better.The simulation results on datasets of OTB100,VOT2016 and VOT2018 show that the improved SiamBAN algorithm can effectively reject the outliers on the response graph and significantly enhance the tracking performance indexes in comparison with mainstream tracking algorithms.