Document image deblurring method, system and equipment based on deep learning

The invention discloses a document image deblurring method, system and device based on deep learning, and relates to the field of image restoration and reconstruction. According to the method, firstly, blurred document images with different blurring degrees are generated in batches in a blurring kernel generation mode based on an original clear document image; forming an image data set by the fuzzy document image and the corresponding original clear document image; constructing a convolutional neural network model comprising a residual error backbone network and a feature fusion module; training and testing the convolutional neural network model by using the image data set to obtain a trained convolutional neural network model as a deblurring model of the document image; and inputting a document image to be deblurred into the deblurring model, and directly outputting a corresponding clear document image in an end-to-end manner. The document image subjected to blurring removal through the method has high conte