双参数恒虚警率CFAR(Constant False Alarm Rate)是舰船目标检测中的常用算法。近年来,合成孔径雷达(SAR)分辨率不断提高,SAR图像幅宽增大,并且在检测时希望尽量保持舰船轮廓以便后续的舰船目标识别。双参数CFAR算法虽然能满足目标检测需求,但算法运行时间过长,不利于信息的及时处理。传统的MPI(Message Passing Interface)并行化解决方案在分配检测任务给各进程时,没有考虑因陆地掩膜,几何校正等预处理所导致的图像中待检测点分布不均。针对这一问题,本文提出改进的MPI并行化解决方案。与传统的MPI并行化解决方案相比,该方案能较为均衡地为各个进程分配检测任务。在集群计算机上的实验结果表明,改进后并行标准效率提高约43%。为应对机载SAR实时舰船目标检测的需求,在多核PC机上进行实验,结果表明,本文算法在多核PC机上也能有效地缩短检测时间,对实现机载SAR实时舰船目标检测有积极意义。

Ship detection is important in military and civilian applications. Synthetic Aperture Radar(SAR) with all-day, all-weather, and ultra-long-range characteristics has been extensively used. The two-parameter Constant False Alarm Rate(CFAR) method is one of the most well-known methods for target detection. CFAR is an adaptive threshold detection scheme that works efficiently when the background clutter is unevenly distributed. However, in recent years, the resolution of SAR images is significantly improved by the rapid development of the SAR sensor. With the improvement of the resolution, the size of SAR images significantly increased and the ship targets no longer appear as point targets. Instead, the ship targets appear as hard targets. The contour of the targets becomes clearer as well. When the two-parameter CFAR is used to detect ship targets with good contour, a longer computation time is needed. Message Passing Interface(MPI) parallelization is a workable solution used to shorten the computation time of two-parameter CFAR with MPI parallel technique.
The traditional MPI parallelization divides the SAR image horizontally/vertically on average. However, in practical applications, preprocessing methods, such as land masking and geometric correction, are conducted before detection. These preprocessing methods can cause the uneven distribution of the points to be detected. This uneven distribution leads to the unbalanced tasks between the parallel processes. Thus, the efficiency of MPI parallelization is highly influenced. The objective of this study is to eliminate the negative influence caused by the uneven distribution.
In this study, we propose an improved MPI parallel solution of the two-parameter CFAR ship detection method, in which the SAR image is divided in terms of the number of points to be detected. The partitioning strategy includes:First, the total number of points to be detected is calculated. Second, the approximate number of responsible points for each process is computed. Third, the responsible rows of image for each process are identified.
In this manner, the entire detection task is equally divided among the processes. The details of the improved parallel algorithm are listed as below:(1) The first process computes the partitioning strategy and transmits it to the other processes.(2)Each process imports its responsible part of the image.(3)Each process implements two-parameter CFAR detection on its responsible part of the image.(4)The first process obtains the detection results from the other processes.
The numerical experiment is conducted on a cluster computer. When the number of processes is 8, the speedup of the improved parallel algorithm is 7.45, which is better than that of the normal parallel algorithm. The efficiency of parallelization increases by approximately 43%. A similar experiment is conducted on a multicore computer, and a similar result is obtained.
The experimental results show that the improved parallel solution can shorten the detection time and improve the parallel efficiency of the cluster or multicore computer. This study is positively significant for real-time ship detection based on airborne SAR images. Further research is needed to shorten the detection time by using the GPU or Intel MIC architecture.