<p>Visible-infrared person re-identification (VI-ReID) aims to retrieve images of the same individual across visible and infrared spectrums, a critical task in intelligent security and surveillance. The challenge lies in bridging the modality gap caused by differences in imaging conditions. To address this, we propose an Adaptive Channel Enhancement Joint Learning (ACEJL) framework that integrates adaptive channel grayscale enhancement for spectral feature alignment and adaptive channel enhancement for discriminative feature learning. We introduce Cutout-based random data augmentation to strengthen modality-invariant representation learning and employ an Improved Kullback–Leibler (IKL) Divergence Loss that incorporates class-wise global information. This approach enhances intra-class consistency and mitigates bias from sample noise, leading to more accurate and robust image feature representations. Extensive experiments on the SYSU-MM01 and RegDB benchmarks demonstrate significant performance improvements over existing methods, with a Rank-1 accuracy of 71.21% and mAP of 67.53% on SYSU-MM01, and 92.10% Rank-1 accuracy, and 86.96% mAP on RegDB. These results validate the effectiveness of our proposed method in the VI-ReID task. The implementation relies on high-performance computing (HPC) resources to manage computationally heavy model training and support real-time inference in multi-camera surveillance environments, underscoring the critical role of supercomputing in advancing robust VI-ReID systems.</p>