Research On SAR Images Matching Based On An Improved SIFT-Like Algorithm

With the development of Synthetic Aperture Radar(SAR) technology, the research on SAR application domains are becoming more and more popular. In which, the method of SAR images matching has gained growing attention due to its high values in military and civilian. Especially in SAR detection guidance system, it is a key technology to achieve high precision positioning of missile by using SAR image matching.There are shortcomings that Inertial Navigation System(INS) errors gradually increase with accumulation of time in Inertial Navigation System, which cannot meet the requirement for long-time and high-precision positioning of missile. Therefore, it is necessary to propose a new method to breakthrough this bottleneck. Research shows that the method based on images matching can achieve the goal of correction trajectory with all-time and high precision. This method includes two categories: one is based on gray value of matching images, and the other is on feature of images. The former one has higher quality requirements for SAR images, it may unsuitable if gray difference or geometrical deformation exists in SAR images. In addition, the calculation of gray correlation method is relatively large; however, the latter can solve the problems which exist in above matching algorithm under the premise of matching accuracy with invariance of gray and affine transformation. Therefore, the matching technology based on image features is playing an increasingly important role in SAR image application domain.Based on all above, this dissertation makes a study on algorithms of SAR images matching which are suitable for geometrical deformation and speckle noise.Firstly, an introduction is done about basic characteristics of SAR image, and an analysis is done about the influence of geometrical deformation and speckle noise on SAR image features. Then, differences between SAR image and optical image are compared from perspective of imaging principle. Furthermore, shortcomings of Scale Invariant Feature Transformation(SIFT) algorithm which applied in SAR image are pointed out. In this paper, an improved method is proposed for SAR speckle noise and the inadaptability of the original SIFT algorithm.Secondly, a research of SIFT algorithm and its implementation procedures is conducted. Firstly, basic conceptions are introduced, including multi-scale space, gauss pyramid structure of image, and method of Harris Corner Detection, which are the basis of SIFT algorithm. Secondly, a study on SIFT algorithm is performed, such as feature extraction and feature description in gauss pyramid space is presented, then a matching method of the nearest neighbor ratio and a refinement method of random sample consensus are researched in detail. Finally, experiment results prove that the matching accuracy meet our demands and show the invariance of SIFT algorithm to rotation, scale expansion and affine transformation of SAR images.Thirdly, considering that SIFT algorithm is unsuitable for speckle noise of SAR image, an improved method of SIFT-Like is proposed to enhance the robustness of matching algorithm to noise. In order to reduce calculation and increase time efficiency, an operator named the ratio of exponentially weighted averages(ROEWA) is proposed to suppress speckle noise of SAR image, and a SAR-Harris multi-scale space is structured to extract corner information of image. Next, the improved main orientation and description vector are obtained by using ROEWA operator, which enhances the invariance of description vector to speckle noise. At last, experiments confirm the effectiveness of SIFT-Like algorithm in suppression of noise, correct detection rate of image features and time efficiency.