Combining Dark Pixels and Spectral Characteristics for Thin Cloud Removal in High-Resolution Remote Sensing Images

High-resolution optical remote sensing satellites usually refers to Earth observation satellites carrying meter or submeter spatial resolution bands, which are capable of capturing more detailed features of the Earth’s surface. In addition, high-resolution optical satellites have obvious spectral limitations, and the existing thin cloud removal methods are mostly designed for low- and medium-resolution images lacking applicability to high-resolution images. In this article, we propose a method combining dark pixels and spectral characteristics for thin cloud removal in high-resolution remote sensing images, which can adaptively remove thin clouds under different sensors and scenes. For the effective identification of thin cloud information, a new band considering spectral statistical information is synthesized, and an iterative side window minimum filtering (ISWMF) technique is proposed. ISWMF is utilized to construct a thin cloud thickness map (TCTM) containing more thin cloud edge information. To reduce the interference of bright surfaces on the TCTM, the bright surfaces are extracted using interband spectral characteristics and corrected to ensures fidelity of bright surfaces in the results. In addition, the relative aerosol thickness is calculated and compensated using the TCTM within cloud-free vegetation areas. Finally, the linear relationship is combined with the scattering law to estimate the thin cloud reflectance of each band. High-resolution images of various surface types were selected for the experiments, the results show that the proposed method can effectively remove thin clouds and maintain spectral fidelity. The proposed method is effective with various sensor data and large-scale applications and has significant adaptability and universality.