Pedestrian re-identification method based on channel feature aggregation

The invention discloses a pedestrian re-identification method based on channel feature aggregation, which has the following main innovation points: firstly, an SE Block structure is connected at the tail of a residual block, so that the channel feature extraction capability of a network is improved; secondly, two new activation functions WReLU and LWReLU are provided by combining the ReLU and a weight bias item, so that the characterization capability of the neural network on global features is effectively improved; thirdly, the LWReLU is applied in the Splitt-Attention and SE Block, and the weight capability of the Splitt-Attention for calculating each group of feature maps and the extraction capability of the SE Block for channel features are improved, fourthly, a novel loss function Circle Loss is applied, so that the model can be more flexibly optimized, a more clear convergence target can be realized, and the performance of the model is further improved. 本发明公开了一种基于通道特征聚合的行人重识别方法，主要创新点有：一是在残差块的末尾再接一个SE Bl