A Linear NMF-Unmixing-Based Approach Addressing Spectral Variability For Shadow Compensation In Hyperspectral Remote Sensing Data

Currently, with the improvement of the spatial resolution of hyperspectral sensors, the influence of shadows on data gathered from these sensors increases, which makes these data hard to interpret automatically. Therefore, the shadow compensation process of such data is necessary. In this investigation, a linear pixel-by-pixel hyperspectral unmixing-based approach, addressing the spectral variability phenomenon, is used to compensate for the shadow effect in the considered data. Indeed, in this approach, which is based on nonnegative matrix factorization, this effect is considered as a spectral variability phenomenon, and thus spectral variables involved in shaded areas are considered as variants of the same variables in unshaded regions contained in the same imaged scene. Experiments are conducted on a real hyperspectral image to evaluate the performance of the proposed approach. These experiments show that the used approach yields a satisfactory de-shadowing result and outperforms other tested methods from the literature.