基于张量信号处理的极化SAR分类算法研究

发布时间：2018-06-14 07:59

本文选题：张量 + 结构张量　；参考：《电子科技大学》2017年硕士论文

【摘要】：由于张量能够很好地表征高维数据,能够有效保留多维数据之间的结构信息,所以张量分析技术在多维遥感图像分类算法研究中具有重要意义。极化合成孔径雷达(Synthetic Aperture Radar,SAR)是一种具有代表性的多维遥感成像技术。为此,本文研究了基于张量分析的极化SAR边缘检测和分类方法。主要完成以下几个工作:提出了基于加权结构张量的极化SAR边缘检测方法:传统的极化SAR边缘检测方法通常沿袭于SAR边缘检测,在检测效果上并不理想。本文首先研究了基于平均加权结构张量进行极化SAR边缘检测的方法。该方法通过平均加权的方法融合各图像通道的结构张量,其本质是认为每个通道提供等量的边缘信息量。进而,本文提出一种基于特征值度量的加权结构张量的极化SAR边缘检测方法。该方法利用特征值的方式充分度量了每个图像通道的边缘信息量,以此融合得到最终的多通道结构张量。实际数据处理结果表明:相比于传统的极化SAR边缘检测方法,本文提出的基于加权结构张量的极化SAR边缘检测方法可以获得更好的边缘检测结果。提出了极化SAR样本点的高阶表示形式并将秩一支持张量机应用于极化SAR图像分类:极化SAR图像分类的特征量通常由目标极化分解得到。同一分解机制往往能够得到若干特征量,因此所有分解方法得到的特征量集合存在结构关系。传统的极化SAR分类方法将这些特征量排列成向量的形式,这必然破坏了特征量之间的结构关系,导致分类性能较差。为了避免这一问题,本文研究了极化SAR数据的高阶表示形式。基于以上分析,本文还将秩一支持张量机应用于极化SAR图像分类,实验结果表明,本文提出的极化SAR高阶表示形式的确能够利用结构信息,取得更好的分类结果。
[Abstract]:Because Zhang Liang can well represent high-dimensional data and effectively retain the structural information between multidimensional data, Zhang Liang analysis technology is of great significance in the research of multi-dimensional remote sensing image classification algorithm. Polarimetric synthetic Aperture radar (SAR) is a representative multi-dimensional remote sensing imaging technology. Therefore, the edge detection and classification method based on Zhang Liang analysis is studied in this paper. The main work is as follows: a new edge detection method based on weighted structure Zhang Liang is proposed: the traditional edge detection method is usually based on the Zhang Liang edge detection, but the detection effect is not satisfactory. In this paper, the edge detection method of polarimetric Zhang Liang based on average weighted structure is studied. In this method, the structure Zhang Liang of each image channel is fused by the average weighted method, and the essence of the method is that each channel provides the same amount of edge information. Furthermore, an edge detection method based on eigenvalue metric for weighted Zhang Liang is proposed in this paper. In this method, the edge information of each image channel is fully measured by the eigenvalue method, and the final multi-channel structure Zhang Liang is obtained by fusion. The actual data processing results show that the proposed edge detection method based on weighted structure Zhang Liang can obtain better edge detection results than the traditional polarimetric SAR edge detection method. The high order representation of the polarimetric SAR sample points is proposed and the rank one support Zhang Liang machine is applied to the polarimetric SAR image classification. The characteristic quantity of the polarimetric SAR image classification is usually obtained by the target polarization decomposition. In the same decomposition mechanism, some eigenvalues can be obtained, so the set of eigenvalues obtained by all decomposition methods has a structural relationship. The traditional polarimetric SAR classification method arranges these features into vector form, which inevitably destroys the structural relationship between the features and results in poor classification performance. In order to avoid this problem, the high order representation of polarimetric SAR data is studied in this paper. Based on the above analysis, the rank one support Zhang Liang machine is also applied to the classification of polarimetric Zhang Liang images. The experimental results show that the proposed higher-order representation of polarimetric Zhang Liang can obtain better classification results by using the structure information.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52

【相似文献】