Research On Key Technologies Of Image Registration For Multi-Source Remote Sensing Data

In recent years,due to the extensive application of multi-sensor vision systems,multimodal image acquisition technology has been developing.Single image and single modal image cannot meet the needs of image analysis,because they cannot contain all the required information.So it is necessary to register images of different modes or stitch images of the same mode to obtain images of different modes or stitched images of a large view field for better image analysis.However,in the process of multi-source image acquisition,due to the different imaging principles of different modes of images,or the shooting at different times in the same place,these differences lead to great differences in the captured images,which brings difficulties to multi-source remote sensing image registration.At present,most remote sensing image registration methods are only applicable to two modes,and the image data of multiple modes cannot be registered uniformly.Commercial stitching software has a slow analysis speed and is prone to losing image edge information,which affects the quality of splicing images.Therefore,the main research work of this article includes:(1)A multi-source remote sensing image registration algorithm based on adaptive multi-scale partial intensity invariant feature descriptor AMPIIFD is proposed in this dissertation.The algorithm makes use of KAZE features extracted from a scale space constructed by non-linear diffusion filtering,and effectively retains edge feature information while filtering out noise.The proposed AM-PIIFD feature descriptor is used to describe multi-scale features.The mismatches are removed by the consistency of the main directions of feature and RANSAC method to achieve multisource remote sensing image registration.The comparative experimental results show that the algorithm proposed in this paper can register multiple modal images.The CMR is improved by 80%,and the RMSE is reduced by 40%.The registration results of the algorithm proposed in this paper are more accurate(2)A fast image stitching algorithm based on prior geographic information for unmanned aerial vehicle(UAV)cattle farming method is proposed.Using the geographic prior information of drones,first process the collected images of the boustrophedon method,group the images based on geographic coordinate information and determining their relative routes and positions.Search for images to be registered based on parameters such as drone position and flight altitude,thereby reducing the time required for image matching.Finally,the algorithm use feature algorithms to complete image stitching.Experimental results indicate that the proposed algorithm can quickly complete the stitching of unmanned aerial vehicle boustrophedon method images in different backgrounds.And its efficiency is three to four times higher than commercial stitching software such as Pix4Dmapper.Notably,this algorithm preserves image edge information better.