基于改进YOLOv7的近岸目标船舶检测算法  

A ship detection algorithm for nearshore targets based on improved YOLOv7

作　　者：李毓滦 胡秀波 李鑫军 曹睿 万占鸿[2] 韩冰[1] LI Yuluan;HU Xiubo;LI Xinjun;CAO Rui;WAN Zhanhong;HAN Bing(School of Mechanical Engineering and Automation,University of Science and Technology Liaoning,Anshan,114051,China;College of Oceanography,Zhejiang University,Zhoushan 316000,China)

机构地区：[1]辽宁科技大学机械工程与自动化学院,辽宁鞍山114051 [2]浙江大学海洋学院,浙江舟山316004

出　　处：《辽宁科技大学学报》2024年第3期204-212,共9页Journal of University of Science and Technology Liaoning

基　　金：国家重点研发计划(2017YFC1403306、2016YFC1401603);浙江省重点科技创新团队项目(2010R50036)。

摘　　要：为了提高近岸小目标船舶检测精度,本文提出一种基于YOLOv7网络模型的YOLO-ConSwin船舶目标检测算法,在主干网络中融合ConvNext与Swin-Transformer模块,增强模型在多尺度上捕捉特征的能力。在特征金字塔网络结构中引入SimAM无参数注意力机制,强化对重要通道特征的敏感性,增强船舶目标的权重,抑制背景噪声。实验结果表明,与YOLOv7s相比,船舶识别精确率提升11个百分点,证明YOLO-ConSwin算法满足小目标船舶检测要求。In order to improve the detection accuracy of near-shore small target ships,a YOLO-ConSwin ship target detection algorithm based on YOLOv7 network model is proposed in this paper.ConvNext and Swin-Transformer modules are integrated in the trunk network to enhance the model's ability to capture features on a multi-scale.Introducing SimAM,a parameter-free attention mechanism,into the feature pyramid network structure enhances the sensitivity to important channel features,strengthens the weight of ship targets,and suppresses background noise.Experimental results show that the ship identification accuracy is improved by 11%compared to YOLOv7s,proving that the YOLO-ConSwin algorithm meets the requirements for detecting small ship targets.

关 键 词：船舶检测 目标检测 YOLOv7 注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] U665.261[自动化与计算机技术—计算机科学与技术]