Pedestrian re-recognition method based on hierarchical multi-mode attention mechanism

The invention provides a pedestrian re-recognition method based on a hierarchical multi-mode attention mechanism, which mainly comprises a training stage and a re-recognition stage of a network model.the training stage of the network model mainly comprises four parts: 1, high-level significant feature learning based on a channel attention model; 2, extracting middle-layer significant features based on a hybrid attention model; 3, low-level significant feature learning based on a regional attention model; 4, performing adaptive fusion learning based on hierarchical feature similarity; the re-recognition stage mainly comprises two parts: 1, carrying out hierarchical salient feature extraction and hierarchical similarity measurement by using a trained network, and 2, carrying out multilayersimilarity fusion by using a trained weight. According to the pedestrian re-recognition method based on the hierarchical multi-mode attention mechanism, high recognition accuracy can be obtained, andthe pedestrian re-recogniti