引用本文
  • 李斌,文莉莉,邬满,刘画宁,许贵林.基于注意力机制的SK-YOLOv5海洋目标检测分类算法[J].广西科学,2023,30(1):132-138.    [点击复制]
  • LI Bin,WEN Lili,WU Man,LIU Huaning,XU Guilin.SK-YOLOv5 Ocean Target Detection and Classification Algorithm Based on Attention Mechanism[J].Guangxi Sciences,2023,30(1):132-138.   [点击复制]
【打印本页】 【在线阅读全文】【下载PDF全文】 查看/发表评论下载PDF阅读器关闭

←前一篇|后一篇→

过刊浏览    高级检索

本文已被:浏览 153次   下载 494 本文二维码信息
码上扫一扫!
基于注意力机制的SK-YOLOv5海洋目标检测分类算法
李斌1,2, 文莉莉2,3, 邬满2, 刘画宁4, 许贵林2
0
(1.广西自然资源职业技术学院商贸管理系, 广西崇左 532100;2.广西科学院, 广西近海海洋环境科学重点实验室, 广西人机交互与智能决策重点实验室, 数字孪生新技术研究院, 广西南宁 530007;3.广西壮族自治区药用植物园, 信息产业办, 广西南宁 530023;4.广西壮族自治区机构编制和绩效管理数据中心, 业务部, 广西南宁 530012)
摘要:
基于遥感影像的海洋目标图像具有多尺度、形状变化大、颜色暗淡、目标边界不清、图像模糊等特点,需要在现有的目标检测算法上进行改进,以满足遥感影像海洋目标检测及分类需要。针对这些问题,在You Only Look Once version 5 (YOLOv5)的网络架构中引入Selective Kernel Networks (SKNet)注意力模块,提出一种新的SK-YOLOv5网络,增强网络对多尺度复杂海洋目标的特征提取和自适应能力。经对比实验测试,在相同的海洋目标数据集上,改进后的网络比原网络整体检测及分类准确率提升了约9%。
关键词:  注意力机制|SKNet|YOLOv5|海洋目标检测|特征提取
DOI:10.13656/j.cnki.gxkx.20230308.015
基金项目:广西科技重大专项“空天地一体协同重大灾害应急智慧服务平台研发与应用示范”(桂科AA22068072),国家自然科学基金区域创新发展联合基金重点支持项目“台风影响下北部湾海浪与风暴潮演变特征、预测预报及灾变关系研究”(U20A20105)和自然资源部海洋信息技术创新中心2019年度开放基金项目“多源时空数据海洋目标智能提取与行为分析预警”资助。
SK-YOLOv5 Ocean Target Detection and Classification Algorithm Based on Attention Mechanism
LI Bin1,2, WEN Lili2,3, WU Man2, LIU Huaning4, XU Guilin2
(1.Department of Business Management, Guangxi Vocational and Technical College of Natural Resources, Chongzuo, Guangxi, 532100, China;2.Guangxi Key Laboratory of Marine Environmental Science, Guangxi Key Laboratory of Human-machine Interaction and Intelligent Decision, New Technology Research Institute on Digital Twin, Guangxi Academy of Sciences, Nanning, Guangxi, 530007, China;3.Information Industry Office, Guangxi Botanical Garden of Medicinal Plants, Nanning, Guangxi, 530023, China;4.Business Department, Guangxi Zhuang Autonomous Region Organization Establishment and Performance Management Data Center, Nanning, Guangxi, 530012, China)
Abstract:
The ocean target image based on remote sensing image has the characteristics of multi-scale,large shape change,dim color,unclear target boundary and fuzzy image.Therefore,the existing target detection algorithm needs to be improved to meet the needs of ocean target detection and classification of remote sensing image.Aiming at this problem,the Selective Kernel Networks (SKNet) attention module is introduced into the network architecture of You Only Look Once version 5 (YOLOv5),and a new SKYOLOv5 network is proposed to enhance the feature extraction and adaptive ability of the network to multi-scale complex ocean targets.Through comparative experimental tests,on the same ocean target data set,the overall detection and classification accuracy of the improved network is improved by about 9% compared with the original network.
Key words:  attention mechanism|SKNet|YOLOv5|ocean target detection|feature extraction

用微信扫一扫

用微信扫一扫