Super-Resolution Reconstruction of Remote Sensing Images Based on Improved Residual Dense Network

An improved residual dense network (IRDN)-based remote sensing images super-resolution reconstruction technique is suggested in order to address the issues with single feature extraction and equalization of feature processing in the current remote sensing images super-resolution reconstruction algorithm. By strengthening the structure of the original residual dense model, adding a branch to extract more varied and rich features, and establishing the channel and position relationship of features by Coordinate Attention (CA), the algorithm enhances the performance of the model's feature extraction and representation. The strategy can effectively increase the reconstruction performance while significantly lowering the number of model parameters, according to experiments using the publicly accessible remote sensing datasets UC MERCED and NWPU-RESISC45. On the UC MERCED dataset with a scale factor of 4, the objective performance evaluation indexes of peak signal to noise ratio and structure similarity achieve 29.3126dB and 0.7987, respectively, and good visual effects are obtained. It offers a solid foundation for the future remote sensing images' additional analysis and processing.