Shadow detection and shadow removal: a benefit for land cover classification

In airborne remote sensing data, detecting and removing shadows for accurate land cover classification is important. Shadows can obscure features and distort surface appearance, leading to misclassifications. Eliminating shadows reveals true surface characteristics, enhancing classification accuracy. This study presents a methodology for detecting and removing shadows to enhance the quality of land cover classification results. For shadow detection, specific color bands of LCh and HSI color spaces are selected and processed by two shadow detection methods: the spectral ĤL ratio and Particle Swarm Optimization. Shadow removal is performed using two methods: The first one is based on histogram matching and the second on Distance-based Gamma Correction. Land cover classification is performed on the corrected image data and the results are compared to those achieved with the original data. In both cases, the influence of including 3D data on the accuracy of the land cover classification in comparison with 2D data has been carried out. The data is used to train a Random Forest algorithm. The results indicate that the shadowy regions in test areas are best classified using the combination of Particle Swarm Optimization and Distance-based Gamma Correction for shadow detection and shadow removal, respectively.