Super-resolution reconstruction method applied to MRI (Magnetic Resonance Imaging) medical image

The invention relates to a super-resolution reconstruction method applied to an MRI medical image, and the method comprises the steps: obtaining an undersampled MRI medical image with an artifact, and constructing a mixed dense knowledge distillation model which comprises a teacher model and a student model; and training the mixed dense knowledge distillation model, inputting the undersampled MRI medical image with the artifact into the trained mixed dense knowledge distillation model, and obtaining a full-sampled MRI medical image without the artifact. Through analysis, it is found that the mixed dense network can effectively analyze a plurality of components existing in network output, a multi-valued mapping model is constructed based on the plurality of components to reconstruct a sharpening boundary, and meanwhile, the problem of discomfort in a super-resolution task is effectively solved through regularization constraints obtained from knowledge distillation; and on the basis of analyzing the relation an