Unsupervised domain adaptive person re-identification methods have problems with style differences between datasets, and noisy pseudo-labels in the clustering process. In order to make better use of global and local information, this paper proposes an unsupervised domain adaptive method based on label mutual optimization. This network inserts a feature enhancement module into the feature extractor to extract more robust features. At the same time, it utilizes a multi-branch network structure to extract global and local features of a person separately. By calculating its cross-consistency score, the iterative network is optimized to reduce the impact of pseudo-label noise. By applying clustering reliability evaluation standards in clustering algorithms, the performance of the network is further improved. The effectiveness of the method proposed in this paper is verified on several datasets.