Research on defog algorithm of single remote sensing image based on deep learning

Abstract:
Since deep learning has shown superior performance in the field of natural image defogging, this paper draws on the network structure and ideas of natural image defogging to design and implement a gated context aggregation network for defogging of a single remote sensing image. The network consists of an encoder-decoder structure, smooth expansion blocks, and gated fusion subsystems that enable it to directly learn the mapping between the input image and the corresponding fog-free image, without relying on traditional atmospheric scattering models. At the same time, this paper establishes a large-scale cloud remote sensing dataset for training and testing the methods used, containing uniform and non-uniform synthetic cloud and fog remote sensing images. Experimental results on the established dataset show that the designed defogging method has made significant progress compared with the traditional defogging method.