Multilevel Fusion Network Based on Mix Hybrid Attention for Hyperspectral and LiDAR Image Classification

With the rapid development of remote sensing technology, the fusion of hyperspectral image (HSI) and light detection and ranging (LiDAR) data has shown great potential for improving land-cover classification accuracy. However, effectively leveraging the complementary information of HSI and LiDAR remains challenging due to disparities in data representation, cross-modal redundancy, and inadequate fusion of multi-scale features. To address these issues, we propose a novel multi-level fusion network, termed MFNet, built upon attention mechanisms. The model incorporates multi-scale feature extraction to capture hierarchical contextual information from both modalities. A weight-sharing strategy is designed to enhance cross-modal interactions, while a hybrid attention module adaptively weights and selects salient features across different scales. In addilition, a new multi-scale fusion module is proposed to facilitate cross-scale feature integration. By hierarchically combining HSI and LiDAR features, MFNet effectively exploits their complementary strengths while minimizing information conflicts. Extensive experiments conducted on three publicly available benchmark datasets demonstrate that the proposed model achieves state-of-the-art classification performance compared to several existing models.