Due to the structure size of the atomic force microscope probe tip, image edge distortion will occur when micro-nano measurement is performed. Thus, a blind restoration method of atomic force microscopy image based on transfer learning is proposed, where the blind restoration for the one-dimensional raster image can be realized by training sourcing model and target model. This method uses the corrosion algorithm of mathematical morphology to generate grid training samples, extracts the characteristic vectors of the convolution effect from the samples by applying the U-Net network source model, where the weight parameters are migrated to the U-Net network target model. Then the source model can conduct supervised learning under adaptive regularization method. The experimental results show that the proposed method can effectively restore the atomic force microscopy measurement image of one-dimensional grid, improve the lateral resolution, and be used in the linewidth detection of nano grid.