MHCAF-Net:The Mamba-Hierarchical Channel-Attention Fusion Network for Multi-Source Remote Sensing Image Classification

To address issues such as feature discrepancies across modalities, long-range channel dependency modeling, local detail loss, and to enhance the generalization ability of the model, this paper proposes the HMCFA-Net for joint classification of HSI and LiDAR. Specifically, the Mambabased Heterogeneous Channel-Attention Merge Module (HCMMamba) is introduced, which dynamically adjusts the weights of multi-modal features and facilitates their efficient fusion, thereby enhancing complementary feature representation, improving channel-wise feature extraction, and mitigating detail loss. To enhance the generalization ability of the network, this paper designs a Hierarchical Multi-Modal Loss (HMML) function, aiming to jointly achieve global semantic modeling and local cross-modal feature optimization. Experiments were designed to show the effectiveness of HMCFA-Net.