雷达图像目标特征提取方法研究

发布时间：2018-06-29 20:31

本文选题：雷达目标识别 + 特征提取　；参考：《西安电子科技大学》2014年博士论文

【摘要】：雷达成像技术具有全天候、全天时、远距离观测能力，有效提高了雷达的信息获取能力，具有重要的军用和民用应用价值。随着雷达成像技术的高速发展，雷达图像收集能力越来越强。从大量雷达图像中获取目标信息进行目标检测、分类与识别则是雷达成像的目的，其中基于雷达图像的目标识别受到越来越广泛的关注。传统的基于数据驱动的目标识别方法依赖于数据本身所反映出来的目标信息，而在实际环境中数据反映的目标信息随环境变化而变化，这给基于数据驱动的目标识别方法带来挑战。目标的一些物理特征（如目标几何尺寸，物理结构等）受环境因素影响较小，而且雷达成像机理在一定程度上可以反映目标物理特征。本文以此为出发点，研究了基于目标参数化回波模型雷达图像中目标物理特征提取技术。论文分别从干涉ISAR（InISAR）横向定标、目标微动特征提取以及目标电磁特征（属性散射中心特征与极化特征）三个方面探讨和研究目标物理特征提取技术，主要工作概括如下： 1.针对干涉ISAR（InISAR）横向定标中相位缠绕问题，提出一种基于随机霍夫变换（Randomized Hough Transform, RHT）的InISAR横向定标算法。该算法利用ISAR图像中的特显点的模糊横距（干涉相位主值得到的横距）与多普勒频率的线性关系，估计真实横距与多普勒频率之间的尺度因子，实现对目标ISAR图像横向定标，从而避免了繁琐的相位解缠绕过程。仿真实验结果表明该算法可以实现对ISAR图像的定标，并具有一定的抗噪声性能，由定标后的目标ISAR图像可以提取目标的几何尺寸特征。 2.研究了目标微动特征提取方法，主要包含两部分：（1）在分析微动目标窄带回波信号基础上，指出窄带目标的微动特征提取等价于多分量非平稳信号的瞬时频率估计问题，我们提出一种基于曲线跟踪（Curve track, CT）算法的目标微动特征提取方法。该算法在时频域通过最近邻数据关联（Nearest Neighbor DataAssociation, NNDA）算法对各分量信号的时频曲线进行关联与分离，然后采用扩展卡尔曼（Kalman）滤波器对分离的时频曲线进行平滑滤波，并基于平滑后的时频曲线估计目标微动参数。（2）通过分析微动目标的宽带回波，指出目标回波在距离包络域具有微距特征，以及时频域具有微多普勒特征。对于微距特征提取，利用幅度相位估计方法（Amplitude and phase estimation, APES）得到目标距离包络信号的超分辨估计，在此基础上由CT算法实现目标微距特征曲线的分离与提取；针对微多普勒特征提取，利用多距离单元信号进行联合时频分析得到目标完整的多普勒谱，并由CT算法提取目标微多普勒特征。最后基于电磁计算数据的仿真结果验证所提算法的有效性。 3.研究了属性散射中心提取方法，包含两部分：（1）考虑目标频率-方位二维观测数据在属性散射中心模型参数空间上的稀疏性，，我们提出一种基于字典缩放的属性散射中心提取与参数估计方法。由于模型参数维数较高，构造的高维联合字典将消耗较多系统资源。针对该问题，所提算法采用交替优化与字典缩放实现了对参数化字典的降维。为了减小邻近散射中心之间的相互干扰，采用正交匹配追踪（OMP）-RELAX联合算法求解稀疏信号恢复问题，实现在频率-方位角域提取属性散射中心。（2）我们提出一种基于距离特性与方位特性解耦合的属性散射中心提取算法，进一步降低对系统资源的要求。该方法通过分别构建包含位置信息与方位特性信息的两个低维字典代替高维的联合字典实现距离特性与方位特性的解耦合，并得到散射中心参数估值。根据提取的属性散射中心可以有效地估计目标或目标重要部件的几何尺寸。基于电磁计算数据和实测数据的实验结果验证了上述算法的有效性。 4.研究了全极化属性散射中心提取方法：（1）考虑目标全极化观测数据在属性散射中心模型参数空间上的联合稀疏特性，利用联合稀疏表示技术提取属性散射中心，并对估计的极化散射矩阵进行极化分解提取目标极化特征，联合干涉测高可以得到目标三维姿态信息。该方法采用基于字典缩放属性散射中心提取算法思想实现参数化字典的降维，对稀疏系数矩阵施加行稀疏约束，通过SOMP（Simultaneous Orthogonal Matching Pursuit）算法求解联合稀疏优化问题并提取属性散射中心。（2）针对散射中心重叠或者同一分辨单元内包含不止一种散射体的情况，依据目标全极化观测在属性特征域（属性参数以及散射类型）的稀疏特性，对目标极化分解系数矩阵施加行稀疏约束与矩阵稀疏约束，该算法利用坐标轮回下降法估计目标极化分解系数矩阵与极化散射机理字典，同时提取目标全极化属性散射中心及其极化特征。基于电磁计算数据的实验结果验证了上述算法的有效性。
[Abstract]:Radar imaging technology has all weather, all day time, long distance observation ability, effectively improve the radar information acquisition ability, has important military and civil application value. With the rapid development of radar imaging technology, radar image collection ability is more and more strong. Target information acquisition from a large number of radar images for target detection, classification and Recognition is the purpose of radar imaging, in which the target recognition based on radar image is becoming more and more widely concerned. The traditional data driven target recognition method relies on the target information reflected by the data itself, and the target information reflected in the actual environment changes with the environment, which is based on the data drive. The moving target recognition method brings challenges. Some physical features of the target (such as the target geometry, physical structure, etc.) are less affected by the environmental factors, and the radar imaging mechanism can reflect the physical characteristics of the target to a certain extent. This paper takes this as the starting point, and studies the target physics based on the target parameterized echo model radar image. Feature extraction technology. This paper discusses and studies the target physical feature extraction technology from three aspects: interference ISAR (InISAR) lateral calibration, target microdynamic feature extraction and target electromagnetic characteristics (attribute scattering center characteristics and polarization characteristics). The main work is summarized as follows:
1. for the phase winding problem in the interference ISAR (InISAR) lateral calibration, a InISAR lateral calibration algorithm based on random Hof transform (Randomized Hough Transform, RHT) is proposed. The algorithm uses the linear relationship between the blurred transverse distance of the explicit point of the ISAR image and the Doppler frequency, and estimates the true transverse distance. The scale factor between the Doppler frequency and the Doppler frequency can realize the horizontal calibration of the target image, thus avoiding the tedious phase winding process. The simulation results show that the algorithm can achieve the calibration of the ISAR image, and has certain anti noise performance. The target ISAR image can extract the geometric feature of the target by the target ISAR image.
2. the extraction method of target micromotion features is studied. It mainly includes two parts: (1) on the basis of the analysis of the narrow band echo signal of the moving target, it points out that the micro motion feature of the narrowband target is extracted from the instantaneous frequency estimation equivalent to the multicomponent nonstationary signal, and we propose a target fretting feature based on the Curve track (CT) algorithm. The algorithm uses the nearest neighbor data association (Nearest Neighbor DataAssociation, NNDA) algorithm to correlate and separate the time-frequency curves of each component signal in time and frequency domain, and then uses an extended Calman (Kalman) filter to smooth the separated time frequency curves and estimate the target micro based on the smooth time frequency curve. (2) by analyzing the wide-band echo of the moving target, it is pointed out that the target echo has a microdistance feature in the range envelope domain and the time frequency domain has the characteristics of micro Doppler. For the extraction of the microdistance feature, the range phase estimation method (Amplitude and phase estimation, APES) is used to obtain the super-resolution estimation of the target distance envelope signal. The CT algorithm is used to separate and extract the target micro feature curve. According to the feature extraction of micro Doppler, the integrated time frequency analysis is used to obtain the complete target Doppler spectrum by the multi distance unit signal, and the target micro Doppler feature is extracted by the CT algorithm. Finally, the simulation results of the electromagnetic calculation data verify the proposed algorithm. Efficiency.
3. the extraction method of attribute scattering center is studied, including two parts: (1) considering the sparsity of the target frequency azimuth observation data in the parameter space of the attribute scattering center model, we propose a method of extracting and estimating the attribute scattering center based on the dictionary scaling. The high dimension joint of the model is constructed because of the high dimension of the model parameter. The proposed algorithm uses alternate optimization and dictionary scaling to reduce the dimension of the parameterized dictionary. In order to reduce the interference between the adjacent scattering centers, the orthogonal matching tracking (OMP) -RELAX algorithm is used to solve the sparse signal recovery problem, and the frequency azimuthal domain extraction is realized. (2) we propose an attribute scattering center extraction algorithm based on the decoupling of distance and azimuth characteristics, which can further reduce the demand for system resources. By constructing two low dimensional dictionaries, which include location information and azimuth characteristics, the distance characteristics and orientation are replaced by high dimensional joint dictionaries. The parameter estimation of the scattering center is obtained. The geometric size of the target or the important part of the target can be estimated effectively according to the extracted attribute scattering center. The experimental results based on the electromagnetic calculation data and the measured data verify the effectiveness of the proposed algorithm.
4. (1) considering the joint sparse characteristic of the target fully polarized observation data in the parameter space of the attribute scattering center model, the joint sparse representation technique is used to extract the attribute scattering center, and the polarization decomposition of the estimated polarization scattering matrix is used to extract the polarization characteristics of the target, and the joint interferometry is used. In this method, the three-dimensional attitude information of the target can be obtained. This method uses the idea of the dictionary scaling attribute scattering center extraction algorithm to reduce the dimension of the parameterized dictionary, applies the sparse constraint on the sparse coefficient matrix, and solves the joint sparse optimization problem by SOMP (Simultaneous Orthogonal Matching Pursuit) algorithm and extracts the attribute scattering center. (2) in view of the overlapping of the scattering centers or the case of more than one scatterer in the same resolution unit, based on the sparse characteristic of the target fully polarized observation in the attribute domain (attribute parameters and scattering types), the line sparse constraint and the sparse constraint are applied to the polarization decomposition coefficient matrix of the target, and the algorithm is estimated by the coordinate rotation descent method. The target polarization decomposition coefficient matrix and the polarization scattering mechanism dictionary are used to extract the full polarization attribute scattering center of the target and its polarization characteristics. The experimental results based on the electromagnetic calculation data verify the effectiveness of the proposed algorithm.
【学位授予单位】：西安电子科技大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TN957.52

【参考文献】