Shoe Print Retrieval Algorithm Based on Improved EfficientnetV2

The retrieval of incomplete and fuzzy shoe prints is a difficult problem in shoe print retrieval algorithm. In order to solve this problem. We improve the feature extraction network by using a new module called Dilated-MBConv on efficientV2. The new module is constructed by introducing multi-scale dilated convolution, and the original MBConv is replaced by Dilated-MBConv in stage 5–7 of the network; At the same time, in order to make the feature extraction network pay attention to the location information of shoe prints, this paper uses coordinate attention to replace the original squeeze and excitation module of the network. The improved network has achieved a better retrieve results in CSS-200 and CS-Database.