Multiscale Feature Extraction-Based Remote Sensing Image Matching Network

This paper addresses the challenges of feature extraction and real-time matching in heterogeneous remote sensing images caused by the differing imaging principles of various sensors. We propose a remote sensing image matching network based on Multiscale feature extraction. The network establishes a matching framework for aerial and satellite images, employing a Multiscale Feature Enhancement Module (MSFEM) to extract scale-invariant features and perform salient feature point detection. This approach effectively captures local features and contextual information from heterogeneous images. Subsequently, the feature points and descriptors are decoded and fed into the SuperGlue feature matching network for alignment. Experimental results demonstrate that our proposed matching method improves matching accuracy by 13.79% compared to D2Net and by 2.86% compared to LoFTR, validating the network's potential for feature extraction and matching tasks in heterogeneous remote sensing imagery.