Channel attention-based swin-transformer image denoising method and system

A channel attention (CA)-based Swin-Transformer image denoising system, comprising:

• a data preparation module, configured to:
• acquire an original high-resolution picture data set, and preprocess the original high-resolution picture data set to obtain a training data set of a noisy image and a high-resolution image for training a denoising network model;
• a shallow layer feature extraction module, configured to input the noisy image into a shallow layer feature extraction network in the denoising network model to extract feature information, to obtain a shallow layer feature map;
• a deep layer feature extraction module, configured to:
• use the shallow layer feature map as an input of a deep layer feature extraction network in the denoising network model, and perform feature extraction to obtain a deep layer feature map, the deep layer feature extraction network comprising a plurality of CARSTB modules and one convolutional layer, wherein the CARSTB module comprises one residual block and a plurality of channel attention Swin-Transformer layers, the channel attention Swin-Transformer layer is obtained by adding a channel attention module to each MSA module and MLP module in a Swin-Transformer layer, and after the CARSTB module extracts feature information of the shallow layer feature map and differentiates importance on the extracted information, the information passes through the convolutional layer to obtain the deep layer feature map;
• an image reconstruction module, configured to perform feature fusion on the shallow layer feature map and the deep layer feature map to obtain a reconstructed image;
• and
• a training module, configured to:
• constrain a difference between the reconstructed image and the high-resolution image by using a loss function, and continuously adjust parameters of the denoising network model until the denoising network model converges, to complete training of the denoising network model.