Atmospheric Correction And Cloud Removal Of Degraded Optical Remote Sensing Images

The thesis includes a complete set of degraded remote sensing image correction algorithms:atmospheric correction and cloud removal,which provide a powerful algorithm tool for restoring degraded remote sensing images;Atmospheric correction is dedicated to solving the inefficiency problem of traditional algorithms by simplifying the look-up table;Cloud removal is dedicated to solving the problem that traditional algorithms cannot handle non-uniform thick clouds.We propose a simplified look-up table atmospheric correction algorithm based on the 6s model.By analyzing the sensitivity of input parameters to atmospheric correction coefficients,a simplified look-up table is established,which greatly improves the efficiency of table building.The step length and range are set according to the specific research area,which reduces the time for interpolation and table lookup.Using this algorithm to perform atmospheric correction on Sentinel-2B satellite images,and comparing with FLAASH correction results,our method has achieved good results,and has been well verified on the spectral curve of ground features and the NDVI value.We propose a cloud removal algorithm using deep learning based on a coarse-to-fine generative adversarial network.By introducing the pyramid cavity convolution module and the fusion attention module to obtain the multi-scale features of remote sensing images,the problem of non-uniform cloud and fog is solved.The contrast enhancement loss and color fidelity loss for remote sensing images are also designed to improve the visual effect.We conduct ablation experiments on the network structure,loss function and training data.This algorithm has achieved good results in removing clouds and fog from real remote sensing images,especially for the processing of non-uniform clouds and fog.