Abstract: In the research field of cross-modal person re-identification (ReID), the modality difference between visible light and infrared modalities is a key issue that increases the difficulty of extracting shared features. To mitigate the differences between these two modalities and enhance the performance of person re-identification, a progressive difference reduction network (PDRNet) is proposed. During the retrieval and recognition stages from visible light targets to infrared image sets, a specific factual intervention module is designed based on causal reasoning theory. It intervenes with visible light images through channel transformation to generate intervened images, suppressing color information interference in visible light images. In the recognition stage from infrared targets to visible light image sets, a channel coordination module is designed to transform multi-channel feature extraction into a single-channel manner, focusing the internet on learning the channel correlation between visible light and infrared images. Finally, a modality balance loss method is proposed for mutual retrieval and recogni tion of visible light and infrared target images. It achieves balanced learning across multiple modalities by utilizing intervened images, visible light images, and infrared images, further suppressing color features and compensating for the discriminative loss features in the specific factual intervention process of visible light images. Simulation experimental results demonstrate that the proposed network outperforms existing mainstream cross-modal person re-identification methods on the SYSU-MM01 and RegDB standard datasets, with rank1 and mAP improving by over 2%. Internet source: https://cstr.cn/31253.11.sciencedb.27692.