基于CFAR-CNN的轻量级海上目标检测  

Light-Weighted Marine Radar Target Detection Based on CFAR-CNN

作　　者：刘世琦 匡华星 杨昊成 LIU Shiqi;KUANG Huaxing;YANG Haocheng(The 724th Research Institute of China State Shipbuilding Corporation,Nanjing 211153,China;Southeast University,Nanjing 210096,China)

机构地区：[1]中国船舶集团有限公司第七二四研究所,江苏南京211153 [2]东南大学,江苏南京210096

出　　处：《雷达科学与技术》2024年第3期312-320,共9页Radar Science and Technology

摘　　要：针对海防雷达分辨率低、实时性要求高及传统CFAR算法难以满足日益精细的现代化战争需求的问题。本文将传统的CA-CFAR算法融入到计算机视觉的两阶段目标检测框架中,形成了一种轻量、高效的雷达实时目标检测算法。首先使用低门限CFAR(Lo-CFAR)来判断诸多潜在目标的真实位置或虚警位置。然后,根据点迹位置信息和雷达回波距离-方位图做数据切片。最后,采用高性能分类器对数据切片进行训练。实测数值实验表明:与传统CFAR,Faster R-CNN算法相比,所提方法在提高检测概率、抑制虚警和轻量时效性方面有显著优势。Aiming at the problems of low resolution and high real-time requirements of coastal defense radar and the difficulty of traditional CFAR algorithm to meet increasingly sophisticated requirements of modern war.In this paper,we incorporate traditional CA-CFAR algorithm into the two-stage target detection framework of computer vision to form a light-weighted and efficient radar target detection algorithm.Firstly,a low threshold CFAR(Lo-CFAR)algorithm is applied to indicate whether many positions of potential targets are real targets or false alarms.Then,the data slices are made on the radar echo range-azimuth map according to the position of spots.Finally,a high performance classifier is applied to training data slices.The numerical experiments show that the proposed method has significant advantages in improving detection probability,suppressing false alarm and light-weighted timeliness compared with the traditional CFAR and Faster R-CNN algorithm.

关 键 词：雷达目标检测 二阶段 轻量化 神经网络 数据切片 

分 类 号：TN957.51[电子电信—信号与信息处理]