基于粒子滤波的雷达弱小目标检测前跟踪算法研究

发布时间：2018-04-20 06:41

本文选题：粒子滤波 + 检测前跟踪　；参考：《江苏科技大学》2017年硕士论文

【摘要】：弱小目标的检测和跟踪是现代雷达必须要面对的关键问题之一。当目标的回波信号能量很低或者背景干扰很强时,雷达传感器接收到的目标回波信噪比(SNR)就很低,此时基于单帧门限判决的传统先检测后跟踪(TAD)方法已很难完成检测和跟踪的任务。检测前跟踪(TBD)方法为此类问题提供了一条有效的解决途径。该方法集检测和跟踪于一体,不对单帧回波数据设置门限,沿着目标可能的路径进行能量积累,能有效地对信噪比很低的弱小目标进行检测和跟踪。在诸多的TBD方法实现算法中,基于贝叶斯递推估计理论的粒子滤波TBD算法(PF-TBD)性能优越,是本文研究的重点。首先,本文研究了标准PF算法及其免重采样改进的高斯粒子滤波(GPF)算法的基本原理,针对将拟蒙特卡罗(QMC)方法应用于GPF算法中虽提高了性能但同时增加了复杂度与运算量的问题,用对基本粒子集的线性变换简化原算法中QMC采样过程,提出了SQMC-GPF算法,该算法具有更低的复杂度和运算量。仿真实验表明SQMC-GPF算法与QMC-GPF算法具有相近的滤波性能但拥有更高的运算速度。其次,针对基于标准PF的TBD算法因为存在重采样步骤而导致粒子失去多样性和并行性减弱的问题,将GPF算法与RPF-TBD算法相结合,提出RGPF-TBD算法,并给出了该算法详细的推导过程。该新算法继承了GPF算法的优点,无需重采样步骤,具有较高的并行性,且粒子的多样性得到了保证,从而具有更好的检测和跟踪性能。再次,本文用QMC方法取代RGPF-TBD算法中的蒙特卡罗采样(MC)方法,同时用超均匀序列取代RGPF-TBD算法中新生后验概率估计中的伪随机序列,提出了QMC-RGPF-TBD算法。该算法可有效提高粒子的多样性和利用率。仿真实验表明,相对于RPF-TBD和RGPF-TBD算法,该算法具有不错的检测和跟踪性能,但因为引入了QMC方法,所以该算法具有较高的复杂度和运算量,针对这一问题,本文又提出了SQMC-RGPF-TBD算法,用线性变换简化连续后验概率分布估计中QMC采样,同时用一次拟随机采样代替新生后验概率分布的估计中的多次拟随机采样,仿真实验表明,该算法在具有不错的检测和跟踪性能的同时具有更快的运算速度。最后,本文对目标运动和雷达量测系统进行建模,将RPF-TBD、RGPF-TBD、QMC-RGPF-TBD和SQMC-RGPF-TBD算法应用到弱小目标的检测中,比较并分析了四种算法在不同信噪比和粒子数量下的检测和跟踪性能,最终得到如下结论:当不考虑运算量时,QMC-RGPF-TBD算法具有最好的检测和跟踪性能;当要求实时性和性能兼顾时,SQMC-RGPF-TBD算法是首选。
[Abstract]:The detection and tracking of small and weak targets is one of the key problems which must be faced by modern radar. When the echo signal energy of the target is very low or the background interference is very strong, the signal to noise ratio (SNR) of the target echo received by the radar sensor is very low. At this time, the traditional detection and tracking algorithm based on single frame threshold decision is difficult to complete the task of detection and tracking. This method provides an effective way to solve this problem. This method integrates detection and tracking, does not set a threshold for single frame echo data, accumulates energy along the possible path of the target, and can effectively detect and track small and weak targets with very low signal-to-noise ratio (SNR). Among the implementation algorithms of TBD, the particle filter TBD algorithm based on Bayesian recursive estimation theory has excellent performance, which is the focus of this paper. Firstly, the basic principles of the standard PF algorithm and the improved Gao Si particle filter (GPF) algorithm without resampling are studied. In order to solve the problem that quasi Monte Carlo QMC (QMC) method can improve the performance but increase the complexity and computational complexity of the GPF algorithm, a SQMC-GPF algorithm is proposed to simplify the QMC sampling process in the original algorithm by linear transformation of the basic particle set. The algorithm has lower complexity and computational complexity. The simulation results show that the SQMC-GPF algorithm and the QMC-GPF algorithm have similar filtering performance but higher computing speed. Secondly, aiming at the problem that the TBD algorithm based on standard PF leads to the loss of diversity of particles and the weakening of parallelism due to the existence of resampling steps, the RGPF-TBD algorithm is proposed by combining the GPF algorithm with the RPF-TBD algorithm, and the detailed derivation process of the algorithm is given. The new algorithm inherits the advantages of GPF algorithm, and it does not need resampling steps. It has high parallelism, and the diversity of particles is guaranteed, so it has better detection and tracking performance. Thirdly, this paper uses QMC method to replace Monte Carlo sampling method in RGPF-TBD algorithm and superuniform sequence to replace pseudorandom sequence in RGPF-TBD algorithm. A new QMC-RGPF-TBD algorithm is proposed. This algorithm can effectively improve the diversity and utilization of particles. The simulation results show that the algorithm has good detection and tracking performance compared with RPF-TBD and RGPF-TBD algorithms, but because of the introduction of QMC method, the algorithm has a high complexity and computational complexity. In order to solve this problem, SQMC-RGPF-TBD algorithm is proposed in this paper. The linear transformation is used to simplify the QMC sampling in the estimation of the continuous posterior probability distribution. At the same time, the quasi random sampling in the estimation of the new posteriori probability distribution is replaced by one quasi random sampling. The simulation results show that, The algorithm not only has good detection and tracking performance, but also has faster operation speed. Finally, the target motion and radar measurement systems are modeled. The RPF-TBDN RGPF-TBD-QMC-RGPF-TBD and SQMC-RGPF-TBD algorithms are applied to the detection of small and weak targets. The detection and tracking performances of the four algorithms under different SNR and particle number are compared and analyzed. The conclusion is as follows: QMC-RGPF-TBD algorithm has the best detection and tracking performance when the computational complexity is not considered, and SQMC-RGPF-TBD algorithm is the first choice when both real-time and performance are required.
【学位授予单位】：江苏科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN713;TN95

【参考文献】