基于YOLOv5的电力巡检图像缺陷识别研究

针对输电线路巡检中可能出现的裂化、老化、腐蚀、破损等诸多缺陷的情况,为保证输电线路运行的安全和稳定,文中开展了基于YOLOv5的电力巡检图像缺陷识别研究。在YOLOv5算法的基础上,结合电力巡检图像特点,采用CIOU_Loss作为Bounding box的损失函数,使其具有更快、更好的收敛效果;选用DIOU_NMS用于NMS处理,提高对遮挡重叠目标的识别精度;同时,在对数据集进行分类处理后,采用"分别训练、统一推断"的方法,冻结部分网络层权重来训练网络模型。实验结果显示,基于YOLOv5算法模型可以有效地识别电力巡检图像缺陷情况。

Research on Defect Recognition of Power Patrol Images Based on YOLOv5

In order to ensure the safety and stability of transmission line operation, image defect recognition research based on YOLOv5 is carried out in view of many defects such as cracking, aging, corrosion and breakage that may occur in line inspection. Based on YOLOv5 algorithm and combined with the characteristics of power patrol images, CIOU_Loss is used as the loss function of Bounding box to make it converge faster and better. DIOU_ NMS is selected for NMS processing to improve the recognition accuracy of occluded overlapping targets. At the same time, after classifying the dataset, the network model is trained by freezing some of the network layer weights using the method of "training separately and inferring uniformly". The experimental results show that the YOLOv5 algorithm model can effectively identify the defects of power patrol images.