| China's space science and technology continue to break through over the past decade.The number of on-orbit optical remote sensing satellites has increased dramati-cally,among them the ZY-3 satellite and the GF-1 satellite which launched successively in 2012 and 2013 led China officially into a new era of civilian high-resolution remote sensing satellite application.At present,multi-sources optical remote sensing satellite application has been deep into the information detection and decision-making such as urban planning,forestry monitoring,land use and so on.Behind these applications,multi-sources optical remote sensing satellite image matching,as one of the core issues of optical remote sensing satellite service application system,directly affects the quality and application prospect of the final satellite image products.Due to the limitations of many technical conditions and the traditional image matching ideas,multi-sources optical remote sensing satellite image matching technology is still immature,there are many key issues should to be addressed.Therefore,collating and analysing previous studies on multi-sources optical remote sensing satellite image matching,concluding and summarizing the core problems of multi-sources optical remote sensing psatellite im-age matching,exploring and practising a new automatic multi-sources optical remote sensing satellite image matclhing method,is of great significance for the application of multi-sources and multi-temporal optical remote sensing satellite data.In this thesis,optical remote sensing satellite images from different sensors and different time are the research objects,and obtaining the correct and uniform distri-bution correspondences from multi-sources and multi-temporal images is the research target.Based on the previous research results,an effective multi-sources remote sensing satellite image automatic matching method with multiple constraints is designed.The main work of this thesis is as follows:1.This thesis deduces that the deviation of the image point along the column/row direction,which is caused by the error of roll/pitch angle of the optical remote sensing satellite,is proportional to its elevation difference in object space.Thus a RFM-based(Rational Function Model)affine transformation plus elevation difference compensa-tion model is proposed to overcome the problem that the traditional RFM-based affine transformation compensation model is unstable in dealing with the optical remote sens-ing satellite images with side pendulum.Several experimental results show that the proposed method,compared with the RFM-based affine transformation compensation model,not only has the same accuracy in dealing with flat area but also has higher accuracy when dealing with mountian area.2.Each SIFT(Scale-Invariant Feature Transform)point from the SIFT satellite image matching algorithm is given a weight according to its scale factor and used in three important steps of the entire image matching process:the weight matrix composed of each point correspondence weight is being substituted for the unit weight matrix in the least square method for the transformation model calculation to improve the accuracy of the solution process;The SIFT weight is introduced in the traditional RANSAC(RANdom SAmple Consensus)algorithm to help it optimize the random sample selection process and assist in determining the residual threshold of each SIFT point;The search radius is a variable which determined by each SIFT point's weight rather than a constant value for all SIFT points in guided matching based on geometric constraint,this effectively solves the search radius threshold selection problem;3.An Extended Phase Correlation algorithm based on Log-Gabor filtering(LGEPC)is proposed by fully understanding and analyzing many previous studies on phase cor-relation and Log-Gabor filter.It transforms the original image matching problem into the correlation problem between two multi-scale image sets by using Log-Gabor filter with different wavelengths to build the multi-scale structure set of each image in the frequency domain,and solve the correlation problem by the phase correlation algo-rithm.This method can effectively avoid the failure of the traditional phase correlation algorithm caused by the nonlinear radiation difference and the scaling ratio difference between two input images,and improve the applicability and stability of the existing phase correlation extension algorithm;4.A Patch-based Extended Phase Correlation algorithm(PEPC)based on LGEPC,pyramid searching and patch constraint,is proposed to overcome the high mis-matching rate even matching failure caused by the low overlap ratio or large change between the images.This method firstly determines the pyramid layer number by the initial search radius,secondly splits each pyramid image into patches,and uses LGEPC to obtain the transformation model parameters between each patch and the reference,thirdly quickly locates suspected mis-matching patches from the global and local scales,and finally obtains a number of correct and uniform distribution correspondences between two original images.In this thesis,we summarize the image matching problem in practical applications of multi-sources optical remote sensing satellite imagery,analysis the shortcomings of the existing multi-sources optical remote sensing satellite image matching algorithms,deeply fuse a variety of image matching ideas,and propose a automatic multi-sources optical remote sensing image matching process based on multiple constraints.Multiple experimental data and real multi-sources optical remote sensing satellite data are tested to verify the feasibility of each study content and the overall workflow.The experimental results show that each study content has practical importance and application prospects,and the overall workflow proposed in this thesis can lay a solid foundation for the follow-up image registration and applications. |
PDF Full Text Request