Research On Cross-view Remote Sensing Image Matching Algorithm For Aerial Target Positionin

In recent years,UAVs have been widely used in many fields,and ground target localization technology for UAVs is the basis of UAV applications.Compared with traditional target localization methods,location methods based on image matching do not rely on GPS and have the advantages of good concealment and strong anti-interference ability.The target localization method based on image matching is to match the UAV aerial images with the satellite remote sensing images carrying prior geographic information to obtain the geographical information of the target in the aerial images,in which how to overcome the huge viewpoint difference between the UAV aerial images and the satellite remote sensing images is a difficult research point.This paper is dedicated to applying cross-view image matching technology to the aerial photography target localization problem,and the main research contents are as follows.To address the problem that the matching performance of traditional image matching methods degrades or even fails to match under the huge viewpoint difference,this paper proposes a cross-view image matching algorithm based on generative adversarial network.The algorithm adopts the strategy of extracting the target region first and then matching the target region.Firstly,the target regions in aerial photography images and satellite remote sensing images are extracted using Faster R-CNN network,and the image block dataset is constructed.In the matching step,the target regions in the aerial images are used to generate the satellite viewpoint images by Pix2 pix network,and the image blocks are matched according to the image similarity.The experiments show that the viewpoint generation can effectively improve the matching accuracy.For the multi-target matching problem,the algorithm introduces the dominant set theory,solves the conflicts in multi-target matching by attaching global spatial constraints,and optimizes the matching results.Experiments show that the algorithm has a high matching accuracy.To address the problem of high mis-matching rate of feature points in cross-view image registration task,this paper proposes an improved RFM-SCAN algorithm to remove the mismatching items in the coarse matching results of feature points.To address the problem of data redundancy in the feature vector used for clustering in the RFM-SCAN algorithm,the improved algorithm uses principal component analysis to reduce the dimensionality of the feature vector,which not only reduces the data redundancy but also improves the clustering performance.To address the problem that the RFM-SCAN algorithm is sensitive to the spatial location of feature points and easily classifies isolated positive matches as mis-matches,the improved algorithm uses the consistency of the motion vector of isolated positive matches with the average motion vector of positive sample clusters to recall isolated positive matches,which effectively improves the recall rate of the algorithm.Qualitative experiments show that the improved algorithm has better rotation invariance,scale invariance and viewpoint invariance.Quantitative experiments show that the improved algorithm outperforms the comparison algorithm in terms of accuracy,recall,and score.