A Cross-View Image Matching Method with Feature Enhancement

Most cross-view image matching algorithms focus on designing network structures with excellent performance, ignoring the content information of image. At same time, there are non-fixed targets such as cars, ships, and pedestrians in ground perspective images aerial images. Differences perspective, direction, scale cause serious interference process. This paper proposes a method feature enhancement, which first transforms empty to generate transformation aligned ground–aerial domain establish preliminary geometric correspondence between ground-space Then, rich deep edge cross-convolution layer used The fusion module enhances tolerance model differences, improving problem transient performance Finally, maximum pooling aggregation strategies adopted aggregate local features obvious distinguishability into global complete accurate experimental results show that proposed has good advance high accuracy CVUSA, is commonly public datasets, reaching 92.23%, 98.47%, 99.74% top 1, 5 10 indicators, respectively, outperforming original dataset limited field view center, better completing cross-perspective task.