Aiming at the challenging attributes such as target occlusion and transient disappearance in tracking tasks,a feature game fusion target tracking algorithm combined with memory repair mechanism is proposed.Firstly,under the framework of kernel correlation filter,the FHOG features and SURF features of the target are extracted for channel series,and the Conv4-1 layer depth features extracted by ResNet-50 are fused by adaptive game to jointly construct the target appearance model to avoid the feature singularity of the tracking algorithm.Then,the model update strategy is improved to memorize,recall and repair the feature information through the memory repair neurons and the memory matrix,and the feature loss repair work of the occluded or drifting video frames is completed.The repaired feature template is used to correctly update the kernel correlation filter to improve the recognition of the tracking object.Finally,the proposed algorithm is compared with KCF,DCF_CA,SAMF,M2C2F and EMCF algorithms on OTB100,TC128 and UAV123 experimental benchmark datasets.The experimental results show that the tracking accuracy and robustness of the improved algorithm are improved.