Hyperspectral unmixing with LiDAR-Based total variation regularized non-negative tensor factorization

Spectral unmixing is one of the main research areas of hyperspectral image analysis. In recent years, Non-Negative Tensor Factorization based approaches have gained great importance in remote sensing as they do not lose information and can better represent hyperspectral images. The Total Variation approach preserves the edge information while providing piece-wise smoothness. On the other hand, the Light Detection and Ranging sensor provides Digital Surface Model information that gives height information about the observed scene. In this study, hyperspectral unmixing based on tensor factorization is performed to increase the spatial resolution of hyperspectral images by combining LiDAR Digital Surface Model information with Total Variation constrained. Experimental studies are carried out on simulation and real data sets and high spatial resolution hyperspectral images is obtained. The obtained results is compared with the state of the art Total Variation constrained Matrix-Vector Non-Negative Tensor Factorization approach and it is observed that the proposed method obtain better performance.