【作者】 王娜；

【导师】 匡纲要；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2012， 博士

【摘要】 极化合成孔径雷达（Polarimetric Synthetic Aperture Radar，PolSAR）是目前遥感领域最先进的传感器之一。利用极化信息检测SAR图像中的人造目标是极化SAR图像解译和应用的前沿课题，对于推动极化SAR解译理论研究走向实用化具有重要意义。在上述背景下，本论文立足于极化SAR图像中人造目标与自然杂波的极化信息差异，以满足杂波背景复杂、目标类型多变的实测极化SAR数据自动目标检测技术的应用需求，深入研究了极化SAR图像无监督检测量的构造、检测量杂波的统计建模以及CFAR目标检测等关键技术。所做工作主要包括以下几个方面：（1）理论分析与实验验证了基于多视极化白化滤波（Multi-look PolarimetricWhitening Filter, MPWF）检测量与极化总功率合成检测量（SPAN）的2P-CFAR检测器在均匀度变化较大的杂波场景中检测性能不佳的问题；发展和完善了极化匹配滤波（Polarimetric Matched Filter, PMF）检测量的杂波统计建模与恒虚警（Constant False Alarm Rate, CFAR）目标检测技术。具体的：推导出了地物杂波的雷达散射截面（Radar Cross Section, RCS）服从常数、Gamma分布、逆Gamma分布与广义逆Gauss分布时PMF检测量的统计分布模型；在参数估计环节，基于第二类型统计量（Second Kind Statistics, SKS），将PMF检测量统计分布的SKS参数估计表达式转化为均匀区域检测量统计分布与RCS统计分布的SKS参数估计表达式的求和过程，简化了直接求解PMF检测量统计分布的SKS参数估计的问题；在CFAR检测环节，基于RCS服从逆Gamma分布时推导的PMF检测量统计分布具有更广泛杂波建模能力这一实验验证结论，推导对应统计分布的CFAR检测阈值求解表达式，进而设计了相应的CFAR滑窗检测方案，提高了CFAR检测的准确性与实用性。实验结果表明：所推导的统计分布模型以及参数估计能够确保不同杂波均匀度环境中PMF检测量的精确统计建模；相比基于MPWF与SPAN的2P-CFAR检测器以及经典的PMF检测器，所提基于PMF检测量的CFAR检测方法在均匀度变化较大的杂波背景中具有更优的检测性能；（2）提出了一种基于极化SAR散射机理的无监督目标检测方法。首先在对常见的人造目标与自然杂波散射机理深入剖析的基础上，总结出各自的散射特征及主要差异，并详细分析了不同散射类型的相干矩阵随散射体取向的变化情况；在此基础上，为有效改善人造目标的散射特性，引入Lee等人的方位角补偿方法去除目标取向对其散射特性的影响，进而提取方位角补偿后的极化SAR数据四成份分解结果中偶次散射与螺旋体散射成分功率之和作为检测量；针对新检测量数据的统计分布特征，采用具有广泛建模能力的G0分布，构建了基于数据拟合方式的目标检测流程。实验结果表明：方位角补偿处理能显著地增加目标与杂波的极化散射特性差异，由此得到的检测量在不同类型的地物杂波背景中都具有很高的信杂比（Signal to Clutter Ratio, SCR）；后续基于G0分布的检测量数据拟合及检测流程确保了不同均匀度杂波中目标检测的准确性；（3）提出了一种基于典型散射体特征极化态的无监督CFAR目标检测方法。在对一类典型散射体同极化与正交极化通道极化响应分析的基础上，选取对金属平板具有最小接收功率，同时对零取向的金属二面角具有最大接收功率的极化态，即同极化通道下左右旋圆极化态与正交极化通道下45线极化态，求取上述极化态对应的天线接收功率作为新检测量，并论证了该类检测量与PMF检测量之间存在线性变换关系。利用四成分分解模型，对新检测量的目标增强与杂波抑制效果进行了深入分析，进一步理论分析了新检测量与目标相似性系数、相似性测度的关系，指出了新检测量同时包含目标与杂波散射相似性与功率信息两种差异。在目标检测环节，采用PMF检测量的目标检测流程对新检测量实现CFAR检测。实验结果表明：所提检测量计算简单、运算量小，且在非均匀以及均匀与非均匀混合区域具有目标增强效果显著、检测性能好的优势；（4）提出了一种基于人造目标与自然杂波反射对称性差异的CFAR目标检测方法。在对人造目标与自然杂波反射对称性差异深入分析的基础上，构建了一种同时包含散射媒质方位角分布信息与螺旋性信息的新检测量，即相干矩阵第三副对角线元素（相干矩阵(2,3)项）的样本平均值。对该检测量在均匀区域的统计分布模型进行数学近似，使得近似后的统计分布模型易于进行参数估计与CFAR检测。在简化的均匀区域统计分布的基础上，利用逆Gamma分布描述该检测量在非均匀区域RCS因子的统计分布，推导出非均匀区域该检测量的统计分布，并给出了对应的参数估计表达式和CFAR检测流程。实验结果表明：所提检测量具有计算简单，杂波抑制能力强、稳健性好等优势；简化的检测量统计分布以及由此推导的非均匀与极不均匀区域检测量的统计分布分别能够实现对不同均匀度区域杂波数据较为精确的统计建模，进而基于该分布的CFAR检测能够实现在杂波类型多变的场景中较为准确的自动目标检测。更多还原

【Abstract】 Polarimetric Synthetic Aperture Radar (PolSAR) is one of the most advancedsensors in the present remote sensing field. Using the polarimetric information for theman-made target detection in SAR image is a frontal subject in PolSAR imageinterpretation. It is of theoretical and practical importance in promoting the PolSARsystem for utilization. For the reasons above, grounded on the polarimetric informationdifferences between man made target and natural clutter, and aiming at the requirementof automatic target detection technique for polarimetric SAR image with compoundedclutter background and varies target types, this thesis deeply investigates some keytechniques such as the construct of the unsupervised polarimetric target detectionmetrics, the statistical modeling for the clutter of the metrics and the CFAR targetdetection etc. The main work includes the following aspects.Through theoretical analysis and experimental validation, we present the reason ofthe inefficient detection performance of2P-CFAR detector based on MPWF and SPANmetrics in the compounded background where the degree of the clutter varies vastly.Further, the detection technique of PMF metric including the statistical modeling, theparameter estimation and the CFAR detection are developed and improved. For thePMF metric, the constant, the Gamma distribution, the inverse Gamma distribution andthe inverse Gauss distribution is selected to characterize the distribution of the RCScomponent in the homogenous, heterogeneous, extremely heterogeneous and thecompounded areas, then the distributions of PMF metric are derived for thecorresponding area. In the parameter estimation, based on the SKS method, theparameter estimation expression of the derived distributions are converted into thesummation of the parameter estimation of PMF distribution in the homogenous area andthe parameter estimation of RCS distribution, thus simplifying the parameter estimationin the maximal degree. In the CFAR detection, based on the conclusion that thedistribution of PMF metric with the RCS modeled in inverse Gamma distribution hasmore extensive modeling capacity in various clutter area, the formula of the CFARdetection threshold of the correponding distribution is deduced. Then the CFARdetection with sliding window is designed, which improves the accuracy andpracticability of CFAR detection. The experimental results demonstrate the greatefficiency of the derived distributions and the corresponding parameter estimation ofPMF metric in the areas with different degree of homogeneity. Moreover, comparedwith the2P-CFAR detector based on MPWF and SPAN metrics, the proposed CFARdetector based on the PMF metric has better detection performance in complex clutterenvironment where the homogeneity of terrain varies sharply.(2) An unsupervised PolSAR image target detection method is proposed based on the scattering characteristic difference between man-made target and natural clutter.Through analyzing the polarimetric scattering mechanism of the common man-madetargets and natural clutters, the scattering characteristic and the major differencebetween man-made target and natural clutter is generalized. Then the effect of theorientation on the coherency matrices of the four component model is analyzed in detail.based on the analysis above, the orientation of the PolSAR data is compensated toremove the effect of the orientation to the scattering characteristics. Finally, using theorientation compensated data, the sum of double-bounce and helix scattering powerderived from the four-component model is extracted as the new metric. Afterward,according to the statistical characteristic of the new metric, a data fitting based targetdetection scheme is presented, which utilizes the G0distribution to fit the distribution ofPDEH metric. The experimental results demonstrate the great efficiency of orientationcompensation in increasing the scattering difference between the man made target andnatrual clutter and the derived metric has high SCR. Finally, the target detection schemeusing the G0distribution assures the accurate target detection in clutters with differentdegree of homogeneity.(3) A novel PolSAR unsupervised CFAR target detectionmethod is proposed basedon the characteristic polarization state of the canonical scatterers. Through analyzing thepolarization signature of the canonical scatterers in the co-polarized and cross-polarizedchannels, we select the characteristic polarization states which have the the minimalreceiving power for metallic plane and maximal receiving power for the dihedral in theco-polarized the cross-polarized channels, i. e., the left and the right circularpolarization state in the co-polarized channel and the45degree linear characteristicpolarization state in the cross-polarized channel as the canonical characteristicpolarization states for the target detection. Therefore, we compute the antenna receivingpower in the characteristic polarization states above as the new metrics and validate thelinear relationsip between the new metrics and PMF metric. Utilizing thefour-component decomposition model, we analyze the effect of new metrics on thetarget enhancement and clutter suppression, and the SCR relationship of the antennareceiving power with the polarimetric similarity theoretically. It can be seen the newmetrics include both the scattering similarity and the power information difference, thussuch metrics is provided with distinct target enhancement. In the target detectionsegment, according to the linear relationship between the new metrics and the PMFmetric, we realize the CFAR detection of new metrics using the target detection schemeof PMF metric. The experimental results demonstrate that the proposed metrics aresimple to obtain and have minimal computation load. Moreover, they are provided withthe advantadge of notable target enhancement and effcient detection performance.(4) A new polarimetric SAR CFAR target detector is proposed based on thereflection symmetry. Through analyzing the reflection symmetry difference between most of the man made targets and natural clutters, a new metric, which includes theinformation of the orientation distribution and the helicity of the scattering medium, isconstructed using the magnitude of the third off-diagonal term of the sample averagedcoherency matrix (the (2,3) term of the coherency matrix). In order to realize theparameter estimation and CFAR detection analytically, the statistical model of the newmetric is approximated and simplified in the homogenous area. Then grounded on thesimplified statistical model in the homogenous area, the statistical model of the newmetric is derived in the heterogeneous area by introducing the inverse Gammadistribution to characterize the distribution of RCS component. Based on the statisticalmodels, the parameter estimation and the automatic constant false alarm rate (CFAR)detection scheme are given in detail. The experimental results have demonstrated theproposed metric has the advantage of simple in computing, efficient and robust inclutter suppression and target enhancement. Moreover, the simplified distribution of theproposed metric has reasonable goodness of fit results in the homogenous clutter areasuch as sea, farmland etc, thus validating the rationality of the mathematicalapproximation. In the heterogeneous area, the derived distribution of the proposedmetric can also assure the precise modeling in vegetation area, urban and their mixedarea etc. Further more, the proposed approach can realize the precise CFAR targetdetection in the compounded background with the clutter varies vastly.更多还原