基于改进YOLOv4算法的小型多旋翼无人机目标检测  

Detection of Small Multi-rotor Unmanned Aerial Vehicle Based on Improved YOLOv4 Algorithm

作　　者：王磊[1,2] 张启亮 翁明善[4] WANG Lei;ZHANG Qiliang;WENG Mingshan(Graduate School,Air Force Engineering University,Xi'an 710051,China;Unit 93688 of People's Liberation Army,Tianjin 300202,China;Air Defense and Antimissile School,Air Force Engineering University,Xi'an 710051,China;Unit 93159 of PLA,Dalian 116033,China)

机构地区：[1]空军工程大学研究生院,陕西西安710051 [2]中国人民解放军93688部队,天津300202 [3]空军工程大学防空反导学院,陕西西安710051 [4]中国人民解放军93159部队,辽宁大连116033

出　　处：《探测与控制学报》2022年第5期125-131,共7页Journal of Detection & Control

摘　　要：针对基于传统恒虚警概率检测算法的多输入多输出雷达在强地杂波背景下对于小型多旋翼无人机目标检测能力急剧下降的问题,引入了光学图像处理领域的YOLOv4目标检测算法,并在原算法的基础上加入SE模块,形成SE-YOLOv4算法。通过对雷达一维原始回波信号进行处理,获得目标回波信号在距离多普勒域能量分布的二维数据矩阵,形成特征明显的二维距离多普勒谱图,进行标注后构建数据集,模型训练完成后,在测试集上对模型的检测性能进行评估。实验结果表明SE-YOLOv4算法的检测性能优于传统的CFAR算法。MIMO radar using traditional constant false alarm rate detection algorithm has a sharp decline in target detection ability for small multi-rotor UAV when the ground clutter is relatively strong.To solve this problem,we introduced YOLOv4 target detection algorithm in optical image processing field and add SE module on the basis of YOLOv4 algorithm to form SE-YOLOV4 algorithm.After processing the original one-dimensional radar echo signal,the two-dimensional data matrix of the target echo signal energy distribution in Range-Doppler domain could be obtained,and the two-dimensional R-D spectrum with obvious characteristics could be formed.After labeling,then built the data set.After the model training is completed,the detection performance of the model was evaluated on the test set.The experimental results showed that the detection performance of SE-YOLOv4 algorithm was better than that of traditional constant false alarm rate detection algorithm.

关 键 词：多旋翼无人机 恒虚警概率检测 YOLOv4算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]