SAR立体影像匹配关键技术研究

发布时间：2018-05-10 07:30

本文选题：SAR立体影像 + 核线影像　；参考：《中国矿业大学》2017年博士论文

【摘要】：雷达立体测量是SAR进行大面积地形测量的主要技术手段之一。雷达立体测量利用不同入射角获取的SAR立体影像间同名点进行地表三维信息提取。然而,由于SAR影像上存在固有的斑点噪声和SAR立体影像间存在较大的几何差异与辐射差异,导致SAR立体影像间的同名点自动提取困难。因此,研究SAR立体影像匹配方法,完成不同角度的SAR影像间同名点自动提取,无论对于提高雷达立体测量的地形测量精度,还是推动雷达立体测量大面积地形测量的自动化程度,均具有重要意义。基于此,本文从SAR立体影像近似核线约束、多特征匹配和顾及匹配窗口信息量的相关匹配三方面开展研究。本文的主要贡献和创新如下:(1)研究遥感影像的核线定义,探讨SAR立体影像间的近似核线。研究了星载SAR影像范围内的近似核线形式,同时引入星载SAR斜地距几何关系,提出星载SAR的近似核线影像生成方法。研究了机载SAR影像范围内的近似核线形式,同时引入机载斜地距几何关系,提出机载SAR的近似核线影像生成方法。实验表明上述方法均能有效消除影像间的上下视差。研究物方空间的投影基准面核线模型,发展一种改进的星载SAR近似核线影像生成方法。该方法利用物方空间不同基准面的核线近似平行特性,在核线生成过程中引入外部DEM。实验结果表明该方法既能保证星载SAR近似核线影像的上下视差精度,又能减弱星载SAR近似核线影像间的几何差异,进一步缩短匹配搜索范围。(2)SAR影像乘性斑点噪声影响特征主方向提取的准确性和特征描述子的分辨能力;特征匹配易受相似性纹理影响,且单一特征获取的匹配点数量较少,匹配成功率低。针对这些问题,提出一种三角形多条件约束的多特征匹配方法。该方法利用比值梯度替代差值梯度,融合最大稳定极值区域(Maximally Stable Extremal Regions,MSER)算法和尺度不变特征变换(Scale Invariant Feature Transform,SIFT)算法获取较多的稳定特征匹配点,然后在稳定特征匹配点构建的三角形特征空间、特征尺度和特征方向多条件约束下,进行剩余特征匹配。以机载SAR立体影像进行实验,实验证明该算法能有效增加匹配点数量,提高匹配成功率。针对SAR影像信噪比低引起的特征提取阶段噪声点多和特征匹配点集中在纹理信息丰富区域的问题,提出一种几何约束的SAR立体影像匹配方法。方法通过在多视处理后的影像上进行SIFT特征匹配,然后在构建的投影变换几何约束和格网控制下进行匹配传播。实验验证了方法在匹配效率、匹配精度以及匹配点的空间分布方面的优越性。(3)针对SAR立体影像间的几何差异的影响,提出一种核线约束的自适应窗口匹配方法。方法利用外部高程信息和影像参数进一步缩短核线约束的搜索范围,采用归一化相关和(Summed Normalized Cross Correlation,SNCC)进行匹配,匹配过程中进行匹配窗口自适应扩展,以减弱SAR立体影像的几何差异影响。采用TerraSAR-X聚束立体影像进行实验,表明本文方法在山地区域、高对比度区域均能增加正确匹配点,提高匹配正确率。提出一种SAR立体影像匹配的视差图融合方法。受影像纹理信息的影响,不同尺寸的匹配窗口进行归一化相关系数(Normalized Cross Correlation,NCC)匹配,获取的视差不同,为了对视差进行优选,充分利用左右一致性约束和信噪比两种置信测度的优势,提出一种新的置信测度,评价视差的置信水平。在此基础上,提出一种视差图融合策略,对不同尺寸匹配窗口获取的视差图进行融合,融合时既考虑视差本身的置信水平,也兼顾其邻域视差的影响。实验表明该方法能有效提高SAR立体影像三维信息提取精度。(4)利用覆盖同一地区的升轨立体影像和降轨立体影像进行DEM提取实验。在多特征匹配研究的基础上,设计一种区域网多航带影像连接点自动选取方法,实验证明了该方法可为区域网平差提供分布均匀的有效连结点。然后结合SAR近似核线影像和SAR相关匹配的研究,设计一种影像自动匹配方法,实验表明该方法提取的DEM中误差约为7米。
[Abstract]:Radar stereo measurement is one of the main technical means for large area topographic measurement by SAR. Radar stereo measurement uses the same name points of SAR stereoscopic images obtained from different incident angles to extract the three-dimensional information of the surface. However, there is a large geometric difference and radiation difference between the intrinsic speckle noise and the SAR stereoscopic image on the SAR image. It is difficult to automatically extract the same name points between the SAR stereoscopic images. Therefore, it is important to study the SAR stereo image matching method and complete the automatic extraction of the homonym points between different SAR images. The main contributions and innovations of this paper are as follows: (1) study the nuclear line definition of remote sensing images, explore the approximate nuclear lines between SAR stereoscopic images, and study the approximation in the range of spaceborne SAR images. In the form of nuclear line, the geometric relation of spaceborne SAR slope is introduced, and an approximate nuclear line image generation method for Spaceborne SAR is proposed. The approximate nuclear line form in the range of airborne SAR image is studied. At the same time, the geometric relation of airborne oblique distance is introduced, and an approximate nuclear line image generation method for airborne SAR is proposed. The experiment shows that the above method can effectively eliminate the image. The projection datum line model of the square space of the object space is studied to develop an improved image generation method for the spaceborne SAR approximation nuclear line. This method uses the nuclear lines of different datum planes in the square space to approximate the parallel characteristics. In the process of generating the nuclear line, the external DEM. experiment results show that the method can not only guarantee the spaceborne SAR approximate nuclear line. The upper and lower parallax accuracy of the image can also reduce the geometric difference between the spaceborne SAR approximate nuclear line images, and further shorten the matching search range. (2) the multiplicative speckle noise of the SAR image affects the accuracy of the feature extraction and the resolution of the feature descriptor; the feature matching is easily affected by the phase resemblance texture and the number of matching points obtained by the single feature. This method uses the ratio gradient to replace the difference gradient, and combines the maximum stable extremal region (Maximally Stable Extremal Regions, MSER) algorithm and the scale invariant feature transform (Scale Invariant Feature Transform, SIFT). The algorithm obtains more stable feature matching points, and then performs the residual feature matching under the constraints of the feature and characteristic directions in the triangular feature space constructed by the stable feature matching points. The experiment is carried out by the airborne SAR stereo image. The experiment shows that the algorithm can increase the number of matching points effectively and improve the matching success rate. For the SAR shadow, the algorithm is proved to be effective. A geometric constraint SAR stereo matching method is proposed in the feature extraction phase, which is caused by low SNR and the feature matching points are concentrated in the rich area of texture information. The method is used to match the SIFT feature on the image after the multi view processing, and then the geometric constraint of the projected transformation and the grid control are used. The superiority of the method in matching efficiency, matching precision and the spatial distribution of matching points is verified experimentally. (3) an adaptive window matching method for nuclear line constraints is proposed in view of the influence of geometric differences between SAR stereoscopic images. The method uses external Gao Cheng information and image parameters to further shorten the search of nuclear line constraints. The cable range is matched by normalized correlation and (Summed Normalized Cross Correlation, SNCC), and the matching window is adaptively extended to reduce the geometric difference influence of the SAR stereoscopic image. The experiment is carried out by the TerraSAR-X bunching stereoscopic image, which shows that this method can increase the positive in the high contrast area in the mountain area. To match points and improve the correct rate of matching, a fusion method of parallax graph for SAR stereo matching is proposed. Under the influence of image texture information, the matching window of different sizes is matched by the normalized correlation coefficient (Normalized Cross Correlation, NCC), and the parallax is different. In order to optimize the parallax and make full use of the left and right conformance With the advantages of two confidence measures, a new confidence measure is proposed to evaluate the confidence level of the parallax. On this basis, a parallax graph fusion strategy is proposed to fuse the parallax graphs obtained from different size matching windows. The fusion of parallax itself is considered and the effect of its neighborhood parallax is taken into account. The method can effectively improve the 3D information extraction accuracy of SAR stereoscopic images. (4) DEM extraction experiments are carried out by using the stereo image of the same area and the stereoscopic image of the rails. On the basis of the multi feature matching research, an automatic selection method is designed for the area network multi flight band image connection point. The experiment proves that this method can be used as a regional network. The adjustment provides an effective continuous node with uniform distribution. Then, an automatic image matching method is designed based on the research of SAR approximation nuclear line image and SAR correlation matching. The experiment shows that the error extracted by this method is about 7 meters.

【学位授予单位】：中国矿业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TN957.52

【相似文献】