LIDAR-GUIDED FAST MULTIMODAL HYPERSPECTRAL UNMIXING

Due to the variations in imaging conditions such as illumination, spectral variability in single-modal hyperspectral unmixing algorithms is unavoidable, leading to inaccuracies in abundance estimation. Considering that LiDAR data is largely unaffected by external imaging condition changes, multimodal hyperspectral unmixing has emerged as a promising solution. By incorporating information that is more robust to illumination variations, multimodal hyperspectral unmixing can achieve enhanced unmixing performance when used synergistically. Existing multimodal unmixing algorithms have successfully addressed shadow removal and strengthened spatial feature extraction. However, they frequently overlook the interplay between geometric and illumination information. Additionally, abundance constraints require prior knowledge and parameter tuning, which necessitates fine-tuning for different datasets. This paper proposes a LiDAR-guided fast multimodal hyperspectral unmixing algorithm, which provides a physical imaging perspective to explain the generation of shadows and enables more accurate abundance estimation by mitigating the effects of illumination variations. Experimental results validate the effectiveness of the proposed method.