基于改进YOLOv5s的趸船钢丝绳缺陷检测算法  

Defect Detection Algorithm for Pontoon Wire Rope Based on Improved YOLOv5s

作　　者：杨宏亮 姜宇 YANG Hongliang;JIANG Yu(CHN Energy Chongqing Electric Power Co.,Ltd.,Chongqing 401147,China)

机构地区：[1]国家能源集团重庆电力有限公司,重庆401147

出　　处：《安全》2024年第11期85-89,共5页Safety & Security

摘　　要：为提高钢丝绳开裂、抽丝图像的识别精度与召回率,本文提出一种基于改进YOLOv5s(you only look once,你只需看一次)的趸船钢丝绳缺陷检测算法。首先在YOLOv5s模型的基础上进行改进,改进方案包括:用特征重组算子(CARAFE)替代最近邻插值进行上采样,以增强特征图的完整性;引入卷积注意模块(CBAM)强化重要特征通道;损失函数由完全交并比损失(CIoU_Loss)替换为扩展交并比损失(EIoU_Loss),以提高边框位置的精度;采用解耦合头减少计算量,提升模型性能与鲁棒性。随后,构建一个专门用于训练和测试的钢丝绳缺陷数据集。通过对比实验结果表明:改进后的YOLOv5s算法在召回率上提高了1.2%,平均精度均值提升了2.2%,呈现出更优的检测效果,并为未来的检测研究提供了理论基础。In order to improve the identification precision and recall rate of the wire rope cracking and strand slippage,in this paper,a defect detection algorithm for pontoon wire ropes was proposed based on YOLOv5s.Firstly,the improvement was implemented based on the YOLOv5s model.The improved scheme includes that the content-aware reassembly of features(CARAFE)operator is used instead of the nearest-neighbor interpolation for upsampling to ensure the feature map integrity;the CBAM module is introduced to enhance important feature channels;the loss function CIoU_Loss is replaced with the EIoU_Loss to improve the position accuracy of the bounding box;and decoupled heads is used to reduce the computation and enhance the performance and robustness of the model.Then,a wire rope defect dataset was constructed for specifically training and testing.The results show that the improved YOLOv5s algorithm increases the recall rate by 1.2%and the mean average precision by 2.2%,demonstrating superior detection performance,and provides a theoretical foundation for the future defect detection in wire ropes.

关 键 词：深度学习 钢丝绳缺陷检测 改进的YOLOv5s算法 注意力机制 特征重组算子(CARAFE) 

分 类 号：X924.3[环境科学与工程—安全科学] TP181[自动化与计算机技术—控制理论与控制工程]