Research On Key Techniques Of Feature Extraction And Matching Of Remote Sensing Images

Nowadays,the satellite remote sensing technology has become the focus research of establishing the space perception system at home and abroad.As a crucial step in remote sensing image processing,the remote sensing image registration has received more and more attention.Feature extraction and feature matching of remote sensing images are the basis of feature based image registration technology,which will directly affect the final results.In this instance,this paper focuses on the key technologies of feature extraction and feature matching in multi-source remote sensing image registration,and mainly did the following researches:According to the characteristics of remote sensing image,firstly,the image preprocessing is analyzed for the noise removal and radiation correction in remote sensing images.Then the basic theory of remote sensing image registration is introduced systematically.For the feature extraction of remote sensing image,this paper introduces the extraction principle of the mutual information feature,SIFT feature and SURF feature.In particular,the idea of information entropy and grid division is added in this paper.The large-scale remote sensing images are divided by two-level grid before the SURF feature extraction.The computation of feature extraction is greatly reduced by the way of extracting feature grid based on information entropy.For the multi-source remote sensing image with large difference of gray level,a shape context feature extraction algorithm based on polynomial fitting is proposed.The feature points are extracted by using the polynomial fitting algorithm on the contour edge,in addition,the shape content circular template is improved,so that it has a good invariance to the rotation,scaling,and translation transform.Experimental results show that the algorithm proposed in this paper has advantages in both speed and robustness.In terms of the feature matching of remote sensing images,different coarse matching algorithms are proposed for the SURF feature descriptor and the improved shape context descriptor.On the basis of this,a precise matching algorithm combining RANSAC algorithm with mutual information is proposed.Experiments show that this method can realize the automatic and accurate registration of multi-source remote sensing images.