Research On Matching Algorithm For Three-line Array CCD Satellite Images

Stereo matching of three-line array CCD satellite images is a key aspect of remote sensing mapping processing,and its accuracy directly affects the accuracy of aerial triangulation and Digital Surface Model(DSM)model generation.At present,remote sensing image sparse matching algorithms use feature points and least-squares matching as the basic method,while dense matching algorithm such as SGM in stereo matching is commonly used to obtain homonymous points pixel-by-pixel.However,the existing feature matching algorithms have low accuracy and the number of matched point pairs obtained by least-squares matching is small,which cannot fully meet the computational requirements of aerial triangulation.In addition,SGM and other dense matching algorithms still have large development space in matching accuracy because they are not adapted to the data characteristics of remote sensing images.To address the above problems,this article conducts research in sparse matching,dense sparse matching,and dense matching algorithms,to design algorithms for remote sensing image data characteristics,improve image matching accuracy,and provide data support for subsequent aerial triangulation and DSM generation.The main research of this article is as follows.(1)Aiming at the problems of uneven brightness of remote sensing images and inconsistent brightness between images,this article proposes a brightness equalization Gamma image dodging algorithm.This article adds the brightness equalization process after the correction of the Gamma function to improve the effect of different scene images in the evaluation criteria of brightness mean,standard deviation,and information entropy,which makes the image brightness change more uniformly and the image details and visual effects better.(2)To address the problem that least-squares local matching is not global,this article proposes a feature fine matching algorithm based on the minimum spanning tree.In the coarseto-fine matching strategy,the algorithm in this article determines the search window of the fine matching process based on the SURF matching and affine transformation,and then calculates the correlation coefficient between the matching windows using the minimum spanning tree as the measure.The experimental result shows that using the image minimum spanning tree as the least-squares fine matching weights can improve the global nature of the matching window,resulting in a 15.38% improvement in matching accuracy.(3)To solve the problems of the small number of sparse matching points and the uneven number of matching in low-texture regions,a sparse matching fusion algorithm for remote sensing images is proposed in this article.The algorithm fuses the Super Point network and Super Glue network to generate dense pairs of sparse feature matching points.In the remote sensing image experiment,the matching points output by this algorithm are relatively more uniform and dense,and the matching effect is stable in each terrain area.(4)In order to solve the problems such as the existence of low-texture regions and the low matching accuracy in regions with repeated texture information in remote sensing images,this article proposes a dense matching algorithm based on texture information constraint.This algorithm reduces the matching computation volume and matching time by epipolar line constraint,and improves the matching accuracy by constraining the energy function of matching cost aggregation with texture features of superpixel segmentation.The DSM model generated by the three-line array CCD remote sensing image is verified,and the algorithm in this article extracts more accurate information of streets,bridges,and other features,and the outlines of the constructed buildings are clearer.In summary,the algorithm proposed in this article for the pre-processing,sparse matching,and dense matching processes of three-line array CCD remote sensing images not only reduces the time of image matching,but also improves the image matching accuracy.This article develops from three aspects of sparse matching,dense sparse matching,and dense matching to achieve accurate matching of three-line array CCD remote sensing images,which can provide strong theoretical support for high-precision spatial triangulation of three-line array CCD remote sensing images and accurate reconstruction of three-dimensional information on the ground surface.