【作者】 刘红超；

【导师】 保铮；

【作者基本信息】 西安电子科技大学 ， 信号与信息处理， 2014， 博士

【摘要】 逆合成孔径雷达(Inverse Synthetic Aperture Radar, ISAR)具有全天时、全天候的特点，并且能够在远距离的情况下得到目标的ISAR图像，对雷达获取目标形状信息具有重大意义，因此在军事和民用中具有重大的应用价值。随着对ISAR的实时成像、超分辨、方位定标的不断需求，ISAR成像技术的研究不断深入。为了提高雷达的成像能力，更有利于后续基于ISAR图像的目标识别工作的展开，本文对有限脉冲、快速、自适应的ISAR成像算法做了一些研究。本文的主要内容概括如下：（1）基于稀疏贝叶斯学习的超分辨ISAR成像技术本文第三章首先介绍了刚体和微动目标ISAR成像模型，然后通过理论和公式的推导，提出了基于稀疏贝叶斯学习的超分辨ISAR成像算法。针对快速成像和自适应ISAR成像开展了以下研究工作：提出了一种基于稀疏贝叶斯学习(Sparse Bayesian Learning, SBL)ISAR超分辨成像算法。近年来,压缩感知已经成功应用到ISAR成像中。由于压缩感知方法将稀疏约束L0松弛到L1范数，导致最终解的稀疏度下降。另外，正则化参数需要手动调节，限制了其在实际ISAR成像的应用。由于SBL算法采用独立高斯分布，更能表征最终解的稀疏度。此外，利用该算法还可以通过置信最大化程序的得到相应的参数，不用人为干预，提高了算法的实际应用价值。飞机和舰船实测数据处理实验验证了基于稀疏贝叶斯学习ISAR超分辨成像算法的有效性。提出了一种基于多稀疏贝叶斯学习的微动目标ISAR成像算法。首先对微动目标的回波信号模型进行分析，得到目标图像具有不规则图形(微动成份)和直线的规则图形(主体成分)。由于主体部分具有散射点位置不变和幅度有起伏的特性，通过多稀疏贝叶斯学习(Mutiple Sparse Bayesian learning, MSBL)得到目标主体成分图像。进而通过减去主体的ISAR图像得到微动成分的图像。最终得到清晰的主体ISAR图像和微动参数。仿真和实测数据处理验证了该算法的有效性。（2）基于置信框架的自适应ISAR超分辨成像算法本文第四章提出了一种基于置信框架(Evidence Framework)的自适应ISAR超分辨成像算法。在压缩感知ISAR成像模型基础上，通过对稀疏约束L1范数的近似，利用置信框架推导得到相应参数的近似闭式解。本算法在稀疏编码和参数求解交替迭代，直到算法收敛到指定的步数，提高了ISAR成像算法的自适应性。仿真和实测数据处理验证了该算法的有效性。（3）基于自适应字典的压缩感知的ISAR超分辨成像及定标一体化技术本文第五章主要针对ISAR图像的方位定标进行了研究。首先建立了调频率-压缩感知的超分辨ISAR成像模型，通过构造自适应字典得到目标代价函数。由于此信号模型多了一个调频率的干扰矩阵，原压缩感知重构方法不再适用。通过理论推导和实验验证，得出在一定条件下可以通过交替迭代算法高概率的恢复出稀疏的目标信号。迭代算法分为两步：1)固定调频率的值，通过稀疏重构算法得到目标的超分辨ISAR图像；2)固定超分辨ISAR图像，通过梯度算法得到调频率的值。最后，利用得到的调频率的值，对目标的旋转速度进行最小二乘估计，利用得到的目标的转速值对超分辨ISAR图像进行方位定标，最终得到目标的距离-方位距离超分辨ISAR图像。实测数据验证了该算法的有效性。（4）一种匀加速空间目标一维距离像补偿算法本文第六章从宽带线性调频雷达信号目标回波的模型出发，分析了处于匀加速状态的空间目标回波的特性，得到回波为立方相位信号，通过推导得到速度和加速度估计的克拉美-罗界，提出基于立方相位函数参数估计的运动补偿方法。该方法可以在较低信噪比下实现速度和加速度的估计以及高分辨一维距离像的运动补偿，更有利于后续的成像和目标识别，仿真实验表明，该算法可以有效的实现匀加速目标的高分辨一维距离像的运动补偿。更多还原

【Abstract】 Inverse Synthetic Aperture Radar (ISAR) can acquire ISAR image ofnon-cooperative target under the condition of all weather, all day and far distance,which enhances radar’s information inquisition capability dramatically. Therefore, inboth military and civilian application, ISAR imaging method has great importance.Recently, the problem of real-time, super-resolution and cross-range scaling in ISARimaging has been studied deeply. In order to improve the imaging capability of ISAR,which is more suitable for radar automatic target recognition base on ISAR image, somenew fast and adaptive imaging ISAR imanging algorithms with limited pulses arestudied in this dissertation. The main work in this dissertation can be shown as follows.(1) Superresolution ISAR imaging based on sparse Bayesian learningIn Chapter3, ISAR imaging models of ridge target and micro-motion target havebeen analyzed, and the process of superresolution ISAR imaging based on sparse signalprocessing is presented. The following works have been done to improve the ISARimage quality for real-time and adaptive ISAR imaging.An algorithm for superresolution ISAR imaging based on sparse Bayesianlearning (SBL) is proposed. Recently, compressive sensing (CS) has been successfullyused in ISAR imaging. Since the exact sparse reconstruction, i.e., L0-norm constraint, isNP hard, L1-norm relaxation is widely used at the cost of performance degradation inthe sparseness of the solution. Furthermore, the regularized factor in CS-based ISARimaging algorithms should be adjusted manually. This makes the existing algorithmsinconvenient to be used in practice. SBL adopts individual Gaussian prior, which retainsa preferable property of diversity measure and can give more sparse solution. Moreover,all the necessary parameters can be estimated using an efficient evidence maximizationprocedure, which can be easily used in practice. The validity of the superresolutionbased on SBL has been verified by measured data of airplane and ship.A novel ISAR imaging algorithm for micromotion target based on multiplesparse Bayesian learning (MSBL) is proposed. The signal model of micromotion targetis analyzed. The ISAR image can be divided into irregular image (rotating part) andstraight lines (main body). Since the signal of the main body has the property ofcommon profiles, MSBL can be used to obtain the signal of main body effectively. Theimage of micromotion parts can be obtained by substracting the image of main body.Finally, the clear ISAR image of main body and the micromotion parameter of rotating part can be obtained. The validity of the proposed ISAR imaging algorithm based onMSBL has been testified by simulated and measured data.(2) An adaptive ISAR imaging algorithm based on evidence frameworkIn Chapter4，An adaptive ISAR imaging algorithm based on evidenceframework is proposed. Based on the CS ISAR signal model and the approximation ofsparse constraint, i.e., L1-norm, the closed forms of all necessary parameters areobtained using evidence framework. This algorithm iterates between sparse coding andparameter estimation until a fixed number of iterations is reached. This ISAR imagingalgorithm improves the adaptivity of ISAR imaging. Experiments based on simulatedand measured data are demonstrated to show the efficiency of the proposed ISARimaging algorithm.(3) Joint ISAR imaging and cross-range scaling using compressive sensing withadaptive dictionary (CSAD)In Chapter5, ISAR cross-rang scaling is mainly studied. Firstly, the chirprate-CSISAR signal model is presented, and the cost function is obtained by constructing theadaptive dictionary. Since a disturbed matrix with chriprate is added to original CSISAR imaging model, the sparse reconstruction method used in CS method is invalid.An alternative recursion algorithm can be used to reconstruct the sparse signal with highprobability and this has been confirmed by theories analysis and data demonstration.The recursion algorithm can be divided two stages:1) fix the value of chirprate, get thesuperresolution ISAR image by the algorithm of sparse reconstruction;2) fix thesuperresolution ISAR, get the value of chirprate by gradient algorithm. Finally, therotation rate can be obtained by least square method, and the range and corss-rangeISAR image can be obtained by using the rotation rate to scale the ISAR image. Thevalidity of the proposed ISAR imaging algorithm based on CSAD algorithm has beentestified by measured data.(4) A range profile compensation algorithm for space target with uniformaccelerationIn Chapter6，Based on the model of wideband linear frequency modulation signal,the property of space target echo with uniform acceleration is analyzed. The target echocan be modeled as cubic phase signal, the Cramer-Rao bound of velocity andacceleration is deduced. A compensation algorithm based on cubic phase function isproposed, which can estimate the velocity and acceleration of target and compensate thehigh resolution range profile under lower signal to noise ratio, then it is preferable to the following imaging and target recognition. The results of simulation data show that theproposed algorithm can effective compensate the motion of space target with uniformacceleration.更多还原