Image flare removal system and method based on cross-channel feature information interaction

The invention discloses an image flare removal system and method based on cross-channel feature information interaction. The system comprises a feature extraction layer, an encoder unit, a long-range information capture unit, a decoder unit and a feature reconstruction layer. The method comprises the following steps: firstly, extracting bottom layer features by using a 3 * 3 convolutional layer to obtain a feature map; secondly, processing the extracted feature map through four encoder modules, capturing information of various feature scales, properly compressing the information, and capturing long-range information of the encoded feature map through a long-range information capturing unit to increase the sensing range of the network; thirdly, gradually recovering details of the feature map through four decoder modules, and fusing the details with features of different scales; and finally, remodeling the recovered feature map into a two-dimensional feature map, and generating a final prediction image by using