舰船电子工程2026，Vol.46Issue(2)：35-39,83,6.DOI:10.3969/j.issn.1672-9730.2026.02.008
基于改进YOLOv5的钛丝表面缺陷检测算法
A Titanium Wire Surface Defect Detection Algorithm Based on Improved YOLOv5
摘要
Abstract
Titanium wire surface defect detection has important application value in industrial production.Compared with tra-ditional industrial images,titanium wire surface defect has the characteristics of small target,changeable shape and complex back-ground interference.Aiming at the difficulties of feature extraction,lack of detection accuracy,error detection and missing detec-tion of small and medium-sized targets in titanium wire surface defect detection,an improved YOLOv5s based target detection algo-rithm YOLOv5s-SCF is proposed.StarNet architecture is adopted as the backbone network of YOLOv5s,which can effectively re-duce the complexity of the model while maintaining the detection accuracy.The context-anchored attention mechanism(CAA)is in-troduced to supplement the multi-scale local features and enhance the feature extraction ability of the central region of the object.Using Focal-EIoU frame loss function,it focuses on high-quality anchor frames during regression,improves convergence speed and positioning accuracy.The experimental results show that compared with the original YOLOv5s algorithm,the improved model has a 2.1%improvement in mAP@0.5,which is suitable for real-time surface defect detection of industrial titanium wires.关键词
目标检测/YOLOv5s/星操作/上下文锚点注意力机制/损失函数Key words
target detection/YOLOv5s/star operation/contextual anchor attention mechanism/loss function分类
信息技术与安全科学引用本文复制引用
杨林浒,王娟平..基于改进YOLOv5的钛丝表面缺陷检测算法[J].舰船电子工程,2026,46(2):35-39,83,6.基金项目
陕西省秦创原"科学家+工程师"队伍建设项目(编号:2022KXJ-048) （编号:2022KXJ-048）
西安市重点产业链技术攻关项目"人工智能应用场景示范:钛打磨机器人设备研制"(编号:23ZDCYJSGG0029-2023) （编号:23ZDCYJSGG0029-2023）
陕西省技术创新引导专项"钛原材料打磨机器人"(编号:2024ZC-YYDP-85) （编号:2024ZC-YYDP-85）
秦创原总窗口"四链"融合重点项目"人工智能在钛打磨机器人打磨工艺数字产业化的应用研究"(编号:2024PT-ZCK-24) （编号:2024PT-ZCK-24）
陕西省教育厅服务地方专项科学研究计划项目-政企联合资助项目"AI智能钛原材料打磨机器人"(编号:24JB029)资助. （编号:24JB029）
