It is an important means to search the suspected shoes in the surveillance video at the scene of crime. Aiming at the problems of low automation and manual screening, this paper proposes a shoe recognition algorithm based on improved deep residual network and data augmentation. In order to enhance the ability of network feature extraction, the deep residual network is studied. The bottleneck structure is improved without adding any parameters to improve the accuracy of the algorithm; Aiming at the problem of down sampling operation in bottleneck structure, the down sampling module is improved to alleviate the problem of information loss in network down sampling; Mixup and optical transform data augmentation algorithm are introduced to establish the linear relationship between data, enrich the diversity of data, and enhance the robustness of network model; Finally, the combined training method of center loss function and softmax loss function is adopted to make the training data achieve better clustering effect. In order to verify the effectiveness and practicability of the proposed algorithm, the proposed algorithm is tested on multi background shoe data sets. The test results show that the accuracy of map and rank-1 of the proposed algorithm is 66.83% and 86.77% respectively, which can effectively improve the accuracy of network recognition.