Cross-modality person re-identification aims to match the same person captured by different sensor cameras, but heterogeneous images exhibit significant modality differences. To address the problem of insufficient correlation between different network layers and inconsistent discriminative feature regions in cross-modality feature extraction, this paper proposes a Dual-Path Cross-Level Feature Correlation Network (DPC-FCNet) that enhances multi-level feature correlation. Under the synergistic effect of shallow independent modeling and deep-level fusion, the network significantly improves the discriminative power and correlation of cross-modality features. A dual-modal collaborative enhancement strategy was adopted to improve the adaptability of the model to cross-modal color shifts. Regarding the network design, independent convolution kernels in the dual-path architecture extract modality-specific features at shallow levels, retaining differentiated information. At deeper levels, a channel-space dual attention module was introduced to strengthen cross-modality semantic correlations in a hierarchical manner. Simultaneously, learnable gating parameters dynamically balance the original features and attention features, with residual connections ensuring a stable gradient propagation. The experimental results show significant improvements in the rank and mAP values on the benchmark datasets SYSU-MM01 and RegDB. This design provides a new approach for cross-modality interaction and opens new research directions in the field of multi-modal joint representation, with its dual-path associative learning and dynamic fusion strategy.