一种基于改进YOLOv4的舰炮弹着点水柱检测方法  

A method for detecting impact point water column of shipboard artillery based on improved YOLOv4

作　　者：王智[1] 石章松[1] 吴鹏飞[1] 吴中红 祁江鑫 WANG Zhi;SHI Zhang-song;WU Peng-fei;WU Zhong-hong;QI Jiang-xin(College of Weaponry Engineering,Naval Univ.of Engineering,Wuhan 430033,China)

机构地区：[1]海军工程大学兵器工程学院,武汉430033

出　　处：《海军工程大学学报》2022年第6期35-40,共6页Journal of Naval University of Engineering

基　　金：国家自然科学基金资助项目(61773395)。

摘　　要：为解决弹着点水柱目标准确且快速检测的问题,充分平衡检测精度和实时性要求,首先通过将轻量级深度卷积神经网络MobileNetv3与YOLOv4算法结合,并利用3×3的深度可分离卷积代替PANet中的普通卷积,构建了用于水柱检测的网络模型M-YOLOv4;然后,从检测精度、模型容量和运行速度等方面将M-YOLOv4与YOLOv3、YOLOv4和YOLOv4-tiny等进行比较。研究结果表明:M-YOLOv4对水柱目标具有良好的检测效果,能够达到与YOLOv4相当的检测精度,并且参数量显著减少、运行速度更快。In order to solve the problem of accurate and rapid detection of water column targets at impact points and fully balance detection accuracy and real-time requirements,MobileNet v3,a lightweight deep convolutional neural network,was combined with YOLOv4 algorithm,and the standard convolution in PANet was replaced by 3×3 deep separable convolution.On this basis,M-YOLOv4 was built for water column detection.Next,this model was compared with YOLOv3,YOLOv4 and YOLOv4-tiny network models in the aspect of detection precision,model capacity and running speed.The results show that M-YOLOv4 is of favorable detection effect of water columns,reaching the detection precision equivalent to that of YOLOv4.Besides,the parameter quantity is significantly reduced with higher operating speed.

关 键 词：水柱检测 YOLOv4 深度可分离卷积 MobileNetv3 K-MEANS聚类算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]