摘要
配电线路中绝缘子与避雷器的故障影响供电系统的可靠性,为了实现航拍图像中绝缘子与避雷器目标的快速、精准检测,笔者提出一种基于改进YOLOv4的配电线路绝缘子与避雷器快速检测方法。针对无人机航拍的配电线路图像中各目标数目不均衡的问题,采用基于GraphCut与图像融合的样本扩充方法,建立含真实和虚拟的配电线路图像样本数据库。对于目标检测网络参数多、检测速度慢导致应用性能差的问题,通过迁移轻量级ShuffleNet网络对YOLOv4主干网络进行替换,利用深度可分离卷积原理对PANet模块进行轻量化调整,并引入Complete-IoU损失函数实现模型的轻量化改进。结合学习率衰减和网络层冻结方式对改进模型进行训练与测试,并与常用的目标检测算法进行对比。结果表明,本研究模型的检测速度可达48.7张/秒,对绝缘子和避雷器目标的检测精度均值可达91.6%,优于其它模型。研究可为配电线路绝缘子与避雷器的定位与缺陷检测提供技术参考。
Defective insulators and arresters on distribution lines affect the reliability of power supply systems.In order to realize the rapid and accurate detection of insulator and arrester in aerial images,the author proposes a rapid detection method based on improved YOLOv4.Aiming at the unbalanced number of objects among image data of distribution line photographed by UAVs,the image augmentation method based on GraphCut and image fusion was adopted to establish an image data set containing real and virtual images of the distribution line.In order to solve the limited application due to large number of parameters and slow detection speed of model,the author transfers the light-weight ShuffleNet to replace the backbone network of YOLOv4,and adjusts the PANet modules by using the depthwise separable convolution method,and also introduces the Complete-IoU loss function to achieve the improvement of YOLOv4 model.Combined with the decay of learning rate and network layer freezing method,the improved model is trained,tested and compared with the common object detection models.The result indicates that the detection speed of the proposed model can reach 48.7 frames per second,and the mean average precision of insulator and arrester reach 91.6%,which is better than other models.This study provides a technical reference for the location and defect detection of insulators and arresters in distribution lines.
作者
王佰川
王聪
WANG Baichuan;WANG Cong(School of Electrical and Information Engineering,Panzhihua University,Panzhihua 617000,China)
出处
《电瓷避雷器》
CAS
北大核心
2023年第3期166-174,共9页
Insulators and Surge Arresters
基金
太阳能技术集成及应用推广四川省高等学校重点实验室基金(编号:TYNSYS-2018-Z-04)
攀枝花市指导性科技技术项目基金(编号:2020ZD-G-2)
攀枝花学院博士科研启动项目(编号:2020DOCO001)。
