基于SIFT算法的机载SAR影像匹配研究

发布时间：2018-05-15 23:17

本文选题：SIFT算法 + RANSAC算法　；参考：《山东农业大学》2013年硕士论文

【摘要】：由于SAR影像特殊的成像机理，导致SAR影像匹配的成功率、正确率、精度及匹配效率较低，SAR影像匹配成为SAR影像应用的技术难点之一。本文针对机载SAR影像特殊的成像特点，以机载SAR影像匹配算法流程为主线，对机载SAR影像SIFT匹配算法进行研究，并引入了基于2D单应变换的RANSAC剔除误匹配点对算法及基于物方约束的匹配点预测算法，对SIFT匹配算法进行改进、总结；论文重点研究了SIFT匹配算法，并提出了两种改进SIFT匹配算法的方法，第一种是SIFT和粗差剔除算法相结合的匹配方法；第二种是基于物方约束的SIFT匹配方法。前者是利用SIFT算法提取特征稳定的同名点对，并结合基于2D单应变换的RANSAC算法剔除SIFT误匹配点对，以提高匹配点对的精度。而后者对前者进一步改进，基于物方约束的SIFT匹配方法是针对机载SAR影像特殊的成像特点和几何特点，将机载SAR影像的几何约束加入到SIFT算法匹配中，提高SIFT匹配点对的准确率、精度及效率。首先，以机载SAR影像匹配SIFT算法匹配流程为基础，通过机载SAR影像SIFT匹配实验，从选取的机载SAR影像中三组不同地物类别的代表区域角度来分析，验证了SIFT算法能够较准确地匹配到稳定的特征，具有一定的鲁棒性。然后，在机载SAR影像匹配算法流程分析基础上，通过编程，对选取的机载SAR影像中三组不同地物类别的代表区域分别进行了SIFT和粗差剔除算法相结合的匹配实验和基于物方约束的SIFT匹配实验，并对匹配点对结果精度进行了分析和评价。本文研究的内容和创新点如下：（1）本文分别采用了三种不同地物类别的机载SAR数据做实验，在含有人工建筑物、含有自然植被及纹理信息缺乏的机载SAR影像中，，从不同角度说明SIFT算法可有效地提取稳定的匹配点对，其正确率高，即使是在纹理信息缺乏的区域，便于在实际中实现整景SAR影像间重叠区域的匹配。（2）引入了2D单应变换以作为RANSAC算法剔除SIFT误匹配点对的模型，提出了利用RANSAC算法剔除SIFT误匹配点对，在基于2D单应变换的RANSAC算法机载SAR影像误匹配剔除实验中，通过对匹配点对数据的分析，验证了该方法可以有效地剔除SIFT误匹配点对，进而提高了SIFT匹配点对的准确率和精度。（3）分析了合成孔径雷达的成像机理、影像特点和构像模型，阐明了SAR影像匹配必须考虑到其自身影像特点。针对机载SAR影像特殊的成像特点，提出了以R-D模型为基础，POS与DEM数据相结合辅助像点定位，将机载SAR影像的几何约束加入到SIFT算法匹配中，利用物方约束来预测待匹配SAR影像的匹配点，建立以预测匹配点为中心的匹配搜索窗口，利用SIFT算法在此约束范围内进行匹配。（4）基于物方几何约束的SIFT匹配实验，与利用SIFT和粗差剔除相结合的匹配算法相比，大大减少了误匹配点对，进一步提高了匹配点对的准确率和精度；基于物方几何约束的SIFT匹配并且约束了待匹配SAR影像的搜索匹配范围，进而提高了匹配效率。基于物方几何约束的SIFT匹配实验结果表明该方法是一种非常有效的机载SAR影像匹配算法。
[Abstract]:Due to the special imaging mechanism of SAR image, the success rate, accuracy, accuracy and matching efficiency of SAR image matching are low. SAR image matching has become one of the technical difficulties in the application of SAR images. This paper, aiming at the special imaging characteristics of airborne SAR images, takes the process of airborne SAR image matching arithmetic as the main line, and advances the algorithm of SIFT matching for airborne SAR images. In this paper, the algorithm of RANSAC elimination mismatch point pair based on 2D single transformation and the matching point prediction algorithm based on object constraint are introduced, and the SIFT matching algorithm is improved and summed up. The paper focuses on the SIFT matching algorithm, and two methods to improve the SIFT matching algorithm are proposed. The first is the combination of SIFT and the coarse difference elimination algorithm. The second is the SIFT matching method based on the object constraint. The former uses the SIFT algorithm to extract the homonym pairs of the stable feature, and combines the RANSAC algorithm based on the 2D single stress transform to eliminate the SIFT mismatch point pair, in order to improve the accuracy of the matching point pair. The latter improves the former one step and the SIFT matching method based on the object constraint. In view of the special imaging features and geometric features of the airborne SAR image, the geometric constraints of the airborne SAR image are added to the SIFT algorithm matching to improve the accuracy, accuracy and efficiency of the matching point pairs of the SIFT.
First, on the basis of the matching process of airborne SAR image matching SIFT algorithm, through the airborne SAR image SIFT matching experiment, from the representative area angle of three different objects category of the selected airborne SAR image, it is proved that the SIFT algorithm can match the stable feature more accurately and has certain robustness. Then, in the airborne SAR image, the airborne SAR image has a certain robustness. On the basis of the matching algorithm flow analysis, the matching experiment combined with the SIFT and the gross error elimination algorithm and the SIFT matching experiment based on the square constraint are carried out on the selected representative regions of the selected airborne SAR images of three different types of ground objects. The results are analyzed and evaluated by the matching points.
The contents and innovations of this paper are as follows:
(1) in this paper, the airborne SAR data of three different types of ground objects are used to do the experiment. In the airborne SAR images containing artificial buildings and lack of natural vegetation and texture information, the SIFT algorithm can effectively extract the stable matching points from different angles, and the correct rate is high, even in the area lacking texture information. In practice, the matching of overlapping areas between SAR images is achieved.
(2) the 2D single stress transformation was introduced to remove the SIFT mismatch point pair as the RANSAC algorithm, and the RANSAC algorithm was used to eliminate the SIFT mismatch point pair. In the RANSAC algorithm based on the 2D single stress transformation, the airborne SAR image mismatch elimination experiment was tested. By analyzing the matching points to the data, it was proved that the method could effectively eliminate the SIFT mismatch. Matching point pairs further improve the accuracy and accuracy of SIFT matching points.
(3) the imaging mechanism, image feature and image model of synthetic aperture radar (SAR) are analyzed. It is stated that SAR image matching must take into account its own image features. In view of the special imaging features of airborne SAR images, the R-D model is proposed, and POS and DEM data are combined to assist image point positioning, and the geometric constraints of airborne SAR images are added to SIFT. In the algorithm matching, the matching points of the SAR image to be matched are predicted by the object square constraint, and the matching search window is set up to predict the matching point as the center, and the matching is made by using the SIFT algorithm in this constraint range.
(4) the SIFT matching experiment based on geometric constraint, compared with the matching algorithm combined with SIFT and gross error elimination, greatly reduces the mismatch point pair and further improves the accuracy and accuracy of the matching point pair; based on the SIFT matching of the object geometric constraint and constrains the search matching range of the matched SAR image, then the matching range is improved. Matching efficiency. Experimental results of SIFT matching based on object geometry constraint show that the method is a very effective algorithm for airborne SAR image matching.

【学位授予单位】：山东农业大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P237

【参考文献】