基于区域划分的极化SAR图像分类方法研究

发布时间：2018-12-15 23:26

【摘要】：极化合成孔径雷达((Polarimetric Synthetic Aperture Radar,POLSAR)极化由于能从多个极化通道获取数据,相较与普通SAR可以获取到更加丰富的地物特征而受到瞩目。极化SAR在军用及民用方面都有着重要的发展前景。通过极化SAR对目标进行识别可以有效帮助部队在战争中着重打击敌军的重要部位。不仅如此,极化SAR数据为地质灾害的检测评估、海冰厚度探测、森林火灾检测等提供了数据基础。目前关于极化SAR的应用研究是一大热点,充分利用极化SAR数据获取信息具有重要的研究意义。作为极化SAR图像解译的重要组成部分,极化SAR图像分类也受到国际遥感领域的重视,成为重要的研究方向。1.本文提出了一种基于近邻传播聚类与区域增长的极化SAR图像分类方法。该算法主要是通过特征提取与分水岭算法得到区域过分割结果,然后利用基于区域的K-means算法进行初始区域划分,减少过分割区域的数量,接着利用基于区域的近邻传播聚类进行分类,充分考虑图像的空间相关性,使用区域增长方法来提高分类准确率,最终通过对边界点的Wishart分类得到分类结果。该方法通过将过分割后得到的匀质区域作为分类单元,有效降低了极化数据中相干斑的影响,提高分类精度。2.本文提出了一种改进的基于分水岭的区域划分方法。将分水岭得到的过分割区域作为分析单元,充分利用区域的空间信息,获取每一个区域的邻接信息,并结合边缘惩罚,计算相邻区域间的区域合并评价值,合并相互都为最适合合并的相邻区域,并得到区域划分结果,大大降低了过分割后的区域数量,有效地将匀质区域内相邻的具有相同地物的区域合并起来,并且区域边缘保持良好。3.本文提出了一种基于区域划分的无监督极化SAR图像分类方法。通过一种新的极化特征提取及边缘强度计算方法,运用分水岭算法得到过分割结果,然后利用改进的基于分水岭的区域划分方法,将过分割后得到的小块区域划分为较大区域,最后利用基于区域的近邻传播聚类与一种考虑空间相关性的Wishart分类器进行分类,得到最终分类结果。该分类方法能够得到较好的分类结果。
[Abstract]:Polarimetric synthetic Aperture Radar (Polarimetric Synthetic Aperture Radar,POLSAR) polarization has attracted much attention due to its ability to obtain data from multiple polarimetric channels, which is more abundant than that of ordinary SAR. Polarized SAR has important prospects in both military and civil fields. The identification of targets by polarized SAR can effectively help the troops to focus on the important positions of the enemy in the war. Moreover, polarized SAR data provide data basis for geological hazard detection and assessment, sea ice thickness detection, forest fire detection and so on. At present, the research on the application of polarized SAR is a hot spot. It is of great significance to make full use of polarized SAR data to obtain information. As an important part of polarimetric SAR image interpretation, polarimetric SAR image classification has also been paid attention to in the field of international remote sensing, and has become an important research direction. 1. A polarimetric SAR image classification method based on nearest neighbor propagation clustering and regional growth is proposed in this paper. The algorithm is mainly based on feature extraction and watershed algorithm to get the results of region over-segmentation, and then the region based K-means algorithm is used to divide the initial region to reduce the number of over-segmented regions. Then the region based nearest neighbor propagation clustering is used to classify the image, and the spatial correlation of the image is fully considered, and the region growth method is used to improve the classification accuracy. Finally, the classification results are obtained by the Wishart classification of the boundary points. By using the homogeneous region obtained by over-segmentation as the classification unit, the effect of speckle in polarimetric data is effectively reduced and the classification accuracy is improved by 2.2. In this paper, an improved watershed based region division method is proposed. The over-segmented region obtained from the watershed is taken as the analysis unit, the spatial information of the region is fully utilized, the adjacent information of each region is obtained, and the combined evaluation value of the adjacent region is calculated by combining with the edge punishment. Merging each other is the most suitable adjacent region for merging, and obtains the result of regional division, which greatly reduces the number of over-segmented regions and effectively combines the adjacent regions with the same features in homogeneous regions. And the edge of the region remains good. 3. In this paper, an unsupervised polarimetric SAR image classification method based on region partition is proposed. Through a new method of polarization feature extraction and edge strength calculation, the watershed algorithm is used to obtain the over-segmentation results, and then the improved watershed based region partition method is used to divide the over-segmented small area into larger regions. Finally, the region based nearest neighbor propagation clustering and a Wishart classifier considering spatial correlation are used to classify, and the final classification results are obtained. The classification method can get better classification results.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【相似文献】