For the difference between modalities, a cross-modality person re-identification algorithm which based on symmetric network was proposed. The network combined the modal confusion based on probability distribution with adversarial learning, and generated modal-invariant features through symmetric network to achieve modal confusion. To deal with appearance differences and intra-modality differences, the network constructed a mixed-triplet loss using convolution features of different hidden layers, which can improve the characterization capability of the network. Numerous experimental results on the RegDB and SYSU-MM01 datasets demonstrate the effectiveness of the method.