双极化SAR影像分类研究与应用

发布时间：2018-10-30 16:24

【摘要】：论文通过对ALOSPALSAR双极化SAR数据特征及其地物散射机理研究，利用极化目标分解方法提取的特征参数进行双极化SAR影像分类处理，以达到提高极化SAR影像分类精度的目的。较光学数据和传统雷达数据而言，极化SAR数据不仅包括了幅度信息，也带有相位信息，因此该数据以每个分辨率单元不同极化组合状态记录了丰富的地物后向散射信息。基于极化雷达具有不受昼夜云层影响、能穿透植被和浅层地表、多波段和多极化、高分辨率有源成像的特点，极化SAR雷达在城市规划与变迁、农作物生长、地质体与地质现象(隐伏)、地质灾害等监测与制图方面具有独特的优势。由于地物目标所在的位置、表面几何结构和介电性不同，极化SAR接收的回波具有复杂的散射过程，在分析极化SAR成像机理时,一些表征地物属性的参数必须从这些复杂的散射回波中提取出来，目标分解方法由此应运而生。论文着眼于揭示极化SAR目标分解提取各参数所代表的地物散射机理，以及提高双极化SAR图像分类精度两方面，对长白山地区进行地物分类研究，取得了以下成果： 1．本研究选用的双极化SAR影像存在数据压缩和斑点噪声现象，为保证地物信息提取的精度，对该影像数据进行了一系列预处理。通过分析相干斑统计特性和噪声模型，，结合双极化SAR影像特点，首先对ALOSPALSAR双极化数据进行多视处理，提高了极化SAR图像的辐射分辨率；然后分别采用Boxcar、Lee-sigma和增强型Lee三种滤波算法对多视处理后的图像做降噪对比分析，各滤波算法均能起到降噪作用，而增强型Lee滤波最能有效抑制斑点噪声、保持空间分辨率和极化信息，是一种高性能、优质的滤波方法。 2．通过对常规雷达数据分别采用ML和SVM两种具有代表性的算法进行影像分类对比研究。与ML的分类算法相比，SVM法提高了地物分类的准确率，验证了分类器的选择直接影响着极化SAR图像分类质量。 3．极化特征参数提取。通过对双极化SAR数据的相干矩阵进行Cloude目标分解，提取了反映目标散射机理的4个特征参数，分析表明4个参数表征了不同散射机制下的地物散射信息和物理意义，为后续基于目标分解的极化SAR影像分类提供了有效的特征参数集。这是本论文的特色。 4．基于Cloude目标分解双极化SAR影像分类算法的实现。由于目标分解后得到的特征参数具有较为明确的物理意义，将目标分解技术应用到极化SAR图像分类研究具有高效、可行性强的优势。论文利用目标分解后得到的特征参数与高性能的SVM分类器相结合，对双极化SAR影像进行地物分类算法的实现。结果表明，较常规雷达数据的图像分类而言，基于目标分解的双极化SAR影像分类精度更高，其各类地物均能准确的分离出来。极化目标分解方法可成为双极化SAR图像分类的一种有效的技术手段。
[Abstract]:Based on the study of the characteristics of ALOSPALSAR bipolar SAR data and the scattering mechanism of ground objects, the feature parameters extracted by the polarimetric target decomposition method are used to classify and process the dual-polarized SAR images in order to improve the classification accuracy of the polarimetric SAR images. Compared with optical data and traditional radar data, polarimetric SAR data not only includes amplitude information, but also has phase information, so the data record abundant backscattering information in different polarimetric states of each resolution unit. Based on the characteristics of polarimetric radar, which is not affected by day and night clouds, can penetrate vegetation and shallow surface, multi-band and multi-polarization, high-resolution active imaging, polarimetric SAR radar in urban planning and change, crop growth, Geological bodies and geological phenomena (hidden), geological hazards and other aspects of monitoring and mapping have unique advantages. Due to the location of the object, the surface geometry and dielectric properties, the echo received by polarized SAR has a complex scattering process. When analyzing the imaging mechanism of polarimetric SAR, Some parameters representing the properties of objects must be extracted from these complex scattering echoes, and the target decomposition method emerges as the times require. This paper focuses on revealing the scattering mechanism of ground objects represented by the extraction parameters of polarimetric SAR targets and improving the classification accuracy of dual-polarized SAR images, and studies the ground objects classification in Changbai Mountain area. The results are as follows: 1. The dual-polarization SAR image in this study has the phenomenon of data compression and speckle noise. In order to ensure the accuracy of information extraction, a series of preprocessing of the image data is carried out. By analyzing the statistical characteristics of speckle and the noise model, combining the characteristics of dual-polarized SAR images, the multi-view processing of ALOSPALSAR dual-polarization data is carried out, which improves the radiative resolution of polarized SAR images. Then, three filtering algorithms, Boxcar,Lee-sigma and enhanced Lee, are used to compare and analyze the noise reduction of the multi-view image. Each filtering algorithm can reduce the noise, and the enhanced Lee filter is the most effective to suppress speckle noise. Keeping spatial resolution and polarization information is a high performance and high quality filtering method. 2. The conventional radar data are classified by ML and SVM respectively. Compared with ML's classification algorithm, SVM improves the accuracy of ground object classification, and verifies that the choice of classifier directly affects the classification quality of polarimetric SAR images. 3. Extraction of polarization characteristic parameters. Through the Cloude target decomposition of the coherent matrix of dual-polarized SAR data, four characteristic parameters reflecting the scattering mechanism of the target are extracted. The analysis shows that the four parameters represent the scattering information and physical significance of the ground objects under different scattering mechanisms. It provides an effective feature parameter set for polarimetric SAR image classification based on target decomposition. This is the characteristic of this thesis. 4. Realization of dual-polarization SAR image classification algorithm based on Cloude target decomposition. Because the characteristic parameters obtained by target decomposition have definite physical significance, it is effective and feasible to apply target decomposition technology to the classification of polarimetric SAR images. In this paper, the feature parameters obtained from the target decomposition are combined with the high performance SVM classifier to realize the ground object classification algorithm for the dual-polarization SAR images. The results show that the classification accuracy of dual-polarization SAR images based on target decomposition is higher than that of conventional radar data, and all kinds of ground objects can be separated accurately. The polarimetric target decomposition method can be used as an effective technique for the classification of bipolar SAR images.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P225.1;TP391.41

【参考文献】