【作者】 徐牧；

【导师】 肖顺平；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2008， 博士

【摘要】 利用高分辨成像及极化信息处理技术对SAR图像中的典型人造目标进行提取及几何结构反演,是SAR战场侦察及情报获取的重要任务,也是SAR图像解译的前沿课题,具有重要的理论意义和实用价值。论文立足于高分辨极化SAR雷达体制,以战场侦察为应用背景,以提高极化SAR战场目标情报分析及处理能力为目的,深入开展了极化SAR图像人造目标提取与几何结构反演的研究工作。第一章介绍了极化SAR系统的发展状况,归纳了极化SAR图像解译技术的研究现状,特别指出现有极化SAR图像人造目标提取与几何结构反演存在的问题,阐述了本文的主要研究工作。第二章深入研究了极化SAR图像目标对比增强问题。传统的对比增强处理通常直接基于人工划分训练样本进行增强,对于极化特性非均匀区域的增强效果较差。针对此不足,提出了基于样本像素筛选的极化SAR图像目标对比增强方法,结合“基于极化分解的初步筛选”及“基于统计检验的最终筛选”等处理,提高了传统方法的增强性能。此外,特别针对高分辨极化SAR图像人造目标的对比增强,提出了基于改善极化相似性的增强方法。通过改善同类区域内部像素点的极化相似性,实现了对高分辨极化SAR图像人造目标整体区域的有效增强,增强结果更有利于后续的人造目标检测及提取等处理。第三章深入研究了极化SAR图像人造目标提取问题。首先分析了传统提取方法在提取准确性及完整性等方面的不足。在此基础上,分别引入频谱相干性特征及极化通道相关性特征指导人造目标的提取,提出了两种新的提取方法——基于散射机理分类与频谱相干性分析的方法、基于散射机理分类与方位对称性判决的方法。新方法可弥补传统方法因判别信息量有限而存在的不足,为多特征联合的人造目标提取提供了重要参考。特别地,通过引入频谱相干性特征提取了确定性的人造目标像素点。对于此类像素点,可不经过统计平均而直接基于散射矩阵进行结构类型判别等解译处理,有效避免了邻近像素点对像素点真实属性的影响,提高了人造目标解译的准确性。第四章研究了极化SAR图像人造目标几何结构反演问题。首先针对高分辨极化SAR雷达体制,分析了典型散射结构的空频域极化特性,揭示了空频域极化特征在散射结构类型判别及目标几何结构反演等方面的应用潜力。在此基础上,提出了基于二维CP-GTD模型与STFT处理的全极化散射中心空域极化特性分析方法。最后,利用空频域极化特征在全面揭示目标散射信息等方面的优势,提出一种基于空频域极化特征的人造目标几何结构反演策略。新方法结合极化分类、空域极化特性分析、频域色散特性判别等处理,不仅可克服传统反演方法在散射结构类型判别时存在的模糊性,还可反演特定散射结构的几何参数,反演更为准确和彻底。反演结果可为极化SAR图像解译及目标识别提供直观而有效的判别信息。第五章系统地总结了论文的研究工作和主要创新点,指出了需要进一步研究的问题。更多还原

【Abstract】 The extraction and geometrical structure retrieval of man-made target in SAR imagery play important roles in SAR battlefield reconnaissance and intelligence collection applications, and are also advanced issues of SAR image interpretation with much theoretical and applicable significance. To improve the capability of intelligence analysis and processing by POLSAR sensors, the extraction and geometrical structure retrieval of man-made target in POLSAR imagery are investigated in this thesis.Chapter 1 summarizes the current development of typical POLSAR platforms and various techniques of POLSAR image interpretation. The disadvantages of the current research on extraction and geometrical structure retrieval of man-made target in POLSAR imagery are especially analyzed, and the major research topics of this dissertation are also explained.Chapter 2 studies the problem of target contrast enhancement in POLSAR imagery. Traditional enhancement schemes, which select pixel samples with manual intervention, would be of less effect when dealing with regions of heterogeneous polarimetric characteristics. In order to overcome this disadvantage, a novel scheme based on the strategy of sample selection is proposed. The novel scheme, which consists of two sequential sample selection stages, can effectively improve the performance of traditional method. Moreover, a scheme based on the improvement of polarimetirc characteristics similarity is proposed especially for the enhancement of man-made targets in high resolution POLSAR imagery. The enhancement result makes it much easier for the further detection and extraction process. The validities of the two methods proposed in this chapter are indicated by experimental results with fully polarimetric SAR data sets.In Chapter 3, the extraction of man-made target in POLSAR imagery is discussed. Aiming to remedy the deficiency of traditional extraction methods on the integrality and accuracy, novel extraction methods are proposed with more effective features for extraction introduced. By introducing the spectral correlation property, a novel method based on scattering mechanism identification and spectral correlation property analysis is proposed. By introducing the property of polarimetric correlation between like- and cross-polarized channels, another novel method based on scattering mechanism identification and azimuthal symmetry decision is proposed. The two novel methods can accurately extract man-made targets with different polarimetric characteristics, and provide instructive candidates for the construction of multi-features joint man-made target extraction framework. Moreover, the method based on scattering mechanism identification and spectral correlation property analysis has a unique advantage of extracting point-like coherent scatterers and achieving a direct interpretation of these scatterers, which can improve the accuracy of target interpretation.In Chapter 4, the problem of geometrical structure retrieval of man-made target in POLSAR imagery is discussed. Firstly, the space-frequency polarization characteristics of canonical scatterers such as dihedral corner, trihedral corner, plate, are compared, which reveal the potential of space-frequency polarization features to benefit the geometrical structure retrieval of man-made target. A strategy combined with two-dimensional coherent polarization GTD model (2D CP-GTD) and Short-Time Fourier Transform (STFT) is proposed to extract and analyze the spatial polarization characteristics of scattering centers. A novel retrieval scheme based on space-frequency polarization feature is proposed. Compared with traditional retrieval trials, the novel scheme can obtain a more accurate and complete retrieval result by making fully exploitation of the information embedded in space-frequency polarization features, and can offer great benefit to POLSAR image interpretation and automatic target recognition.更多还原