Mamba-Driven Dynamic Fusion Strategy for Hyperspectral and LiDAR Image Classification

Advancements in hyperspectral remote sensing technology have significantly enhanced the capability for land cover classification. Both convolutional neural networks and transformer-based approaches have been widely adopted in hyperspectral image (HSI) classification tasks due to their powerful feature extraction capabilities. However, practical applications often face constraints due to the high cost of labeling, resulting in limited availability of accurately labeled data. Moreover, hyperspectral data requires simultaneous modeling of spatial and spectral features, and conventional approaches typically process these features types separately, which may lead to information loss during feature fusion. To tackle these challenges, we introduce an innovative Mamba-driven dynamic fusion strategy (MambaDFS) for hyperspectral and light detection and ranging (LiDAR) image classification. The proposed method leverages the dynamic weighting mechanism of Mamba to adaptively adjust the fusion weights of multimodal features, enhancing discriminative feature representation while maintaining global contextual modeling, thereby achieving high-precision classification with limited samples. Experimental results on benchmark datasets reveal that the proposed MambaDFS achieves superior results in overall accuracy, average accuracy, and Kappa coefficient compared to nine other existing methods.