A Method for Detecting Feature-Sparse Regions and Matching Enhancement

Image matching is a key research issue in the intelligent processing of remote sensing images. Due to large phase differences or apparent ground features between unmanned aerial vehicle imagery and satellite imagery, as well number sparsely textured areas, image two types very difficult. Tackling difficult problem feature sparse region detection enhancement algorithm (SD-ME) proposed this study. First, SuperGlue was used initially match images, feature-sparse performed with help initial results, detected areas stored linked list one by one. Then, according order storage, re-extraction on individually, an adaptive threshold screening filter screen re-extracted features. This retains only high-confidence improves reliability results. Finally, local high scores that were aggregated input network for matching, thus, reliable results obtained. The experiment selected four pairs un-manned compared SD-ME SIFT, ContextDesc, algorithms. revealed far superior other algorithms terms correct points, accuracy uniformity distribution points. correctly matched points each pair increased average 95.52% SuperGlue. can effectively improve quality has practical value fields registration change detection.