摘要
为实现复杂海况下对水柱和靶球目标的高效检测,提出了以YOLOv4网络模型为基础的改进算法。实验设计了4种方案对模型检测效果进行改进:用K-means聚类算法对锚定框进行优化;在YOLOv4骨干网络中嵌入SE注意力模块提高对小目标的检测能力;使用基于灰度共生矩阵的海天线检测算法限定检测范围;采用结构相似性检测算法改善视频流检测效果。检测实验证明4种方法对提高网络检测性能均有效果,综合使用4种方法对YOLOv4网络进行改进,在检靶数据集上mAP_(50)值提升了29.9%。
In complex sea conditions,in order to achieve efficient detection of water-columns and target-balls,the article proposed an improved YOLOv4 network model to detect marine targets.The experiment designed 4 schemes:It used the K-means clustering algorithm to optimize the anchor frame,and embed the SE attention module in the YOLOv4 backbone network to improve the network’s ability to detect small targets,and it used the horizon detection algorithm based on gray level co-occurrence matrix to limit the scope of target detection and used the structural similarity detection algorithm to improve the video stream detection effect.Experiments have proved that the four schemes are effective in improving the performance of network detection.We used the four schemes to improve the YOLOv4 network,and the mAP_(50) value on the target detection data set was increased by 29.9%.
作者
张坤
罗亚松
刘忠
ZHANG Kun;LUO Yasong;LIU Zhong(College of Weapons Engineering, Naval University of Engineering, Wuhan 430033, China)
出处
《兵器装备工程学报》
CSCD
北大核心
2022年第4期211-217,共7页
Journal of Ordnance Equipment Engineering
作者简介
张坤(1993—),男,助理工程师,E-mail:724072076@qq.com;通信作者:罗亚松(1982—),男,博士,副教授,E-mail:yours_baggio@sina.com。
