Remote Sensing Image Super-Resolution via Efficient Non-Local Feature Extraction Strategy
Jinpeng Shi, Dong Liang, S. S. Weng · 2023
Remote sensing (RS) images typically exhibit com plex spatial distributions, making non-local features critical for achieving high-quality super-resolution (SR). Most existing SR networks extract local and non-local features alternately, making the non-local feature to be explored at high spatial resolution. This leads to substantial computational costs and limits the performance of these networks. In this paper, we propose an efficient non-local feature extraction strategy to solve this problem. Specifically, we propose a dual branch super-resolution network (DBSRN) with different branches focusing on local and non-local feature extraction. For the local feature extraction branch (LFEBranch), we design an adaptive feature enhancement block (AFEB) to optimize its processing of local features. For the non-local feature extraction branch (NFEBranch), we propose a non-local feature aggregation block (NFAB) to extract non-local features more efficiently by continuously reducing the spatial resolution of the input. Extensive experiments have demonstrated that the proposed DBSRN can effectively leverage the non-local features of RS images, resulting in superior SR performance compared to state-of-the-art networks.