A Study On Image Pairs Matching Method For Tracking Analysis Of Optical Remote Sensing Image Time Series Offset Tracking

Frequent disasters and serious hidden dangers are the factual descriptions of geological disasters,which cause great danger to the lives and property safety of our people.Using satellite remote sensing images,inversion analysis of surface deformation is an effective way to predict disaster risks and proactively prevent and mitigate disasters.In this paper,I study the optical and SAR image pixel offset tracking techniques for seismic isochronous field inversion,and propose an image pairing strategy based on the minimum spanning tree algorithm to reduce the cost of pixel offset tracking solution for the existing problems of pixel time series offset analysis.The main research work and results include:(1)For the inversion of the deformation in the co-seismic field of the portal earthquake,the optimized parameter setting scheme of the COSI-Corr software is investigated in this paper.The surface rupture zone deformation information of the portal seismic coseismic field is obtained using Sentinel-2 and Landsat9 images,and the practical accuracy and theoretical accuracy of the two optical platform offset estimation techniques are analyzed and compared for the extraction of small deformation information.Further,the Sentinel-1A downgraded image is used to acquire the surface rupture zone deformation information in the same seismic field of the portal earthquake to verify the offset estimation accuracy of the Sentinel-2 image with similar correlation matching principle but different imaging methods.The final results show that the Sentinel-2 image with 10-m spatial resolution is the best choice for the COSICorr software to implement the optical image pixel offset tracking technique;the actual effective accuracy of the Sentinel-2 image pixel offset estimation is higher than the theoretical effective accuracy of 1/20 th pixel(0.5 m);the overall deformation information acquired by the Sentinel-1A image is similar to that of the Sentinel-1A image.The overall deformation information obtained from Sentinel-1A images is similar to the results of Sentinel-2 images.(2)To address the problems of the number of image matching pairs and the higher computational cost and lower resolution rate of the temporal offsets in various "redundant observation" matching strategies in optical temporal offset analysis,this paper proposes a strategy to obtain image matching pairs by the Minimum Spanning Tree(MST)algorithm of the graph.In this paper,I propose a selection strategy to obtain image matching pairs by the minimum spanning tree(MST)algorithm,which uses three types of information,namely,solar azimuth difference(SAD),solar height difference(SED),and time baseline(TB),as the baseline variables of MST.The analysis compares the deformation information with the cumulative offset based on the first image and the cumulative temporal offset based on the "Baseline Threshold Method" matching strategy.Optical image pixel time series offset inversion analysis of the Siguas landslide using Sentinel-2 imagery shows: Compared with the cumulative offset,the cumulative temporal offset has more realistic deformation information;compared with the "Baseline Threshold" image matching pair selection strategy,the MST image matching pair selection strategy achieves similar deformation information of temporal offset with fewer image matching pairs,the computational cost is lower and the time series offset is solved faster.