Super-resolution reconstruction of remote sensing images based on multi-scale reparameterized pyramids

In order to solve the problems of loss of structural details and insufficient attention to edge information in the existing lightweight remote sensing super-resolution reconstruction methods, a multi-scale reparameterized pyramid network (MRPN) method for remote sensing image super-resolution reconstruction based on multi-scale reparameterized pyramids was proposed. The model is based on the deep residual network structure, and multiple cascading multi-scale reparameterized pyramid modules are designed in the network for feature extraction. In the module, a multi-scale separation convolution and fusion module is designed to help the network refine image information, and the unglobal and global feature expressions are enhanced, and the link-enhanced spatial attention module is used to improve the attention to high-frequency feature information. Finally, the jumping connection method is used to fuse multi-level features to achieve effective spatial information fusion. The model is verified by 2x, 3x and 4x super-resolution reconstruction on the NWPU-RESISC45 dataset and the AID dataset, and the experimental results show that the peak signal-to-noise ratio and structural similarity are improved, and the method has good performance in improving image quality and preserving details.