Research On Registration Method Of UAV Optical Remote Sensing Image

With the rapid development of UAV technology and sensor technology,UAV remote sensing technology is gradually emerging.However,due to the limitation of the perspective of UAV remote sensing,researchers need to splice UAV images with spatial differences acquired in the same scene in order to obtain a complete scene image for further research.As the basis of image stitching technology,image registration technology plays a key role in the practical application of UAV remote sensing technology.Therefore,the registration technology of UAV optical remote sensing images is studied in this paper.The main work is as follows.(1)It cannot take into account the processing speed and accuracy using image registration methods based on point features to process UAV visible light images.Aiming at this problem,this paper proposes to add two bilateral matching methods based on the ORB(Oriented FAST and Rotated BRIEF)algorithm to process the matching feature points and combines with progressive sampling consistency algorithm to remove mismatch points.It is proved that this method has faster processing speed and higher processing accuracy than traditional algorithms by comparing with SIFT(Scale Invariant Feature Transform)and other five traditional image registration algorithms.The UAV images in five common scenes of city,road,building,farmland and forest are selected for registration experiment.The experimental results show that the accuracy of this method can be more significantly improved than the source algorithm when processing scene images with more obvious features such as cities and roads,and the average increase rate is 21.61%.(2)To solve the problem that registration methods based on feature points cannot effectively deal with the registration of UAV visible light images and infrared images,a multi-feature combination image registration method is adopted in this paper.The method extracts the contour image of the input image by edge features,and then registers the contour image by the feature point registration methods.The experimental results show that this method is more suitable for the registration of UAV visible light images and infrared images than the feature point registration methods.And the Scharr operator combined with feature point registration methods has a better and more stable registration effect.(3)Aiming at the problem that traditional image registration methods only use low-level features for registration,this paper adopts an image registration method based on deep convolutional neural network features and proposes improvement measures.Based on the original algorithm,the method is improved by blocking the image,extracting features by using the ResNet-50(Deep Residual Network)network with stronger feature extraction ability,and constructing feature description vectors by feature fusion.The experimental results show that the improved algorithm has higher registration accuracy than the original algorithm and the traditional image registration algorithms,and the improved algorithm can also achieve good registration results for images that cannot be effectively registered by the traditional algorithm.(4)Based on the QT software development platform,a remote sensing image registration software is designed and implemented in this paper.This software integrates the relevant extended applications of the UAV optical remote sensing image registration method and image registration technology in this paper,realizing the automation and visualization of related processing.