Blind super-division method and device for multi-degradation category recovery

The invention discloses a blind super-resolution method for multi-degradation category recovery. The blind super-resolution method comprises the following steps: using a random degradation kernel as a real degradation kernel, and degrading a high-resolution image to obtain a low-resolution image; respectively compensating information loss caused by different degeneration by using three recovery modules; the noise feature suppression module suppresses high-frequency information caused by noise in features through an attention mechanism, aims to learn original image information loss caused by noise in a low-resolution image, and compensates the noise features through feature residual errors; the texture feature enhancement module adopts a network structure similar to dense connection to extract fuzzy residual errors of the image, and the fuzzy residual errors are used for learning information loss caused by different fuzzy kernels in degradation and compensating the information loss; and the sampling recovery m