Self-adaptive memory unsupervised pedestrian re-identification method based on grouping perception label

The invention provides a self-adaptive memory unsupervised pedestrian re-identification method based on grouping perception labels. The method comprises the following steps: A, obtaining a pedestrian data set with multiple groups of labels; b, calculating the distance between the clustering center of the pedestrian data set and the samples in the same cluster, and constructing a weight dictionary; c, constructing a teacher-student network based on an adaptive memory storage structure, and updating the memory storage structure corresponding to the sample by using the weight dictionary; d, in the training process, smoothing the influence of noise on three pedestrian data sets which are generated by setting different clustering conditions and have multiple groups of labels; e, circularly operating the step A to the step D according to a preset number of iterations, and training the teacher network and the student network, and F, inputting test set data after training is completed, and testing. According to the m