基于融合和特征提取的遥感图像变化检测

发布时间：2018-01-24 00:56

本文关键词： 遥感图像变化检测融合特征提取 PCA Treelet变换核模糊聚类加权系数　出处：《西安电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：遥感图像变化检测是指通过对同一地区不同时期的两幅或多幅遥感图像进行比较分析，根据图像之间的差异来获取地物的变化信息。遥感图像变化检测技术已成功地应用于众多领域，如环境监测、土地利用和土地覆盖的动态监测、森林或植被的变化分析、灾害评估、农业调查、城镇变化研究及在军事中的人造目标监测和地面武装部署分析。本文介绍了遥感图像变化检测的研究背景以及存在的问题，对已有变化检测技术进行了总结，并以差异图融合和特征提取为主要研究内容，针对两时相遥感图像的变化检测问题进行了研究。（1）提出了一种基于图像融合和PCA-核模糊聚类的遥感图像变化检测方法。该方法首先用差值法、对数比值法和均值比法构造三种不同的差异图，，然后对差异图进行融合，对融合后的图像进行PCA（Principal Component Analysis）特征提取，然后用基于核的模糊聚类将特征聚为两类。该方法采用图像融合的方法构造差异图，对不同类型的遥感图像均可获得较好的检测结果，解决了单一类型差异图检测精度低、适用范围窄的问题，具有较好的鲁棒性。该方法对PCA提取的特征采用基于核的模糊聚类方法，将原始数据映射到高维特征空间再进行聚类，实现更为准确的聚类，进一步降低了变化检测的错误率。（2）提出了一种基于Treelet特征融合的遥感图像变化检测方法，首先用差值法、对数比值法和均值比法构造三种不同的差异图，然后用Treelet变换对三幅不同的差异图进行特征融合。该方法由于采用Treelet变换进行特征提取，因而操作简单，正确率高，抗噪性能好；该方法由于利用了不同差异图的有效信息和空间邻域信息进行变化检测，进一步提高了抗噪性能和变化检测精度；此外，该方法对合成孔径雷达（Synthetic Aperture Radar，SAR）图像和光谱图像都可以得到满意的变化检测结果，鲁棒性好。（3）提出了一种基于加权系数和非下采样Contourlet变换（NonsubsampledContourlet Transform, NSCT）特征融合的遥感图像变化检测方法，首先用比值法和差值法构造两种不同的差异图，然后对差异图乘以加权系数，然后用NSCT变换分解带有权值的差异图以获得方向特征，对方向特征和分解前差异图的原始灰度特征进行聚类得到变化检测结果。该方法有效结合不同差异图的信息，并且利用方向特征表达邻域信息，具有一定的抗噪能力，克服了单一类型差异图检测效果不好的弊端，提高了变化检测准确度。
[Abstract]:Remote sensing image change detection refers to the comparison and analysis of two or more remote sensing images in different periods of the same area. Remote sensing image change detection technology has been successfully used in many fields, such as environmental monitoring, land use and land cover dynamic monitoring. Analysis of changes in forests or vegetation, disaster assessment, agricultural surveys, urban change studies and surveillance of man-made targets in the military and analysis of armed deployment on the ground. This paper introduces the research background and existing problems of remote sensing image change detection, summarizes the existing change detection technology, and takes the difference image fusion and feature extraction as the main research content. The change detection of 2:00 remote sensing image is studied. In this paper, a method of remote sensing image change detection based on image fusion and PCA-kernel fuzzy clustering is proposed. The difference method, logarithmic ratio method and mean ratio method are used to construct three different images. Then the difference map is fused and the fused image is extracted by PCA(Principal Component Analysis. Then the features are clustered into two categories by kernel based fuzzy clustering. This method uses image fusion method to construct difference map, and better detection results can be obtained for different types of remote sensing images. The method solves the problem of low detection precision and narrow range of application of single type difference map, and has good robustness. The kernel based fuzzy clustering method is used to extract the features of PCA. The original data is mapped to the high-dimensional feature space and then clustered to achieve more accurate clustering, which further reduces the error rate of change detection. In this paper, a method of remote sensing image change detection based on Treelet feature fusion is proposed. Firstly, three different difference maps are constructed by using difference method, logarithmic ratio method and mean ratio method. Then the Treelet transform is used to fuse the features of three different images. Because of the feature extraction using Treelet transform, the method is easy to operate, has high accuracy and good anti-noise performance. Because the effective information of different difference map and the spatial neighborhood information are used to detect the change, the anti-noise performance and the accuracy of the change detection are further improved. In addition, the method can obtain satisfactory change detection results for synthetic Aperture radar synthetic Aperture radar (SAR) images and spectral images. Good robustness. (3) A nonsubsampled Contour Transform based on weighted coefficient and non-down-sampled Contourlet transform is proposed. NSCT feature fusion remote sensing image change detection method, first using the ratio method and difference method to construct two different difference map, and then the difference map multiplied by the weighting coefficient. Then NSCT transform is used to decompose the difference graph with weight to obtain the direction feature. The change detection results are obtained by clustering the original gray features of the difference map and the orientation feature. The method combines the information of the different difference map effectively and expresses the neighborhood information by using the direction feature. It has a certain ability to resist noise, overcomes the shortcoming of single type difference map detection, and improves the accuracy of change detection.
【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】：TP751

【参考文献】