基于改进YOLOv8的变压器磁芯缺陷检测算法  

Transformer Core Defect Detection Algorithm Based on Improved YOLOv8

作　　者：胡卫昊 HU Weihao(College of Electrical and Information Engineering,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学电气与信息工程学院,安徽淮南232001

出　　处：《湖北民族大学学报（自然科学版）》2024年第4期539-543,共5页Journal of Hubei Minzu University:Natural Science Edition

基　　金：安徽省研究生创新创业实践项目(2023cxcysj081)。

摘　　要：针对变压器磁芯缺陷对电网系统的安全稳定运行构成潜在风险却缺少检测手段的问题,提出了基于改进你只看一次版本8(you only look once version 8,YOLOv8)的变压器磁芯缺陷检测算法。该算法首先通过引入幽灵网络(ghost network,GhostNet)实现了模型参数量的降低,同时保持了较高的检测精确率;然后,引入了具有较大尺寸特征图的小目标检测层,适用于小目标检测;最后,引入了融合注意力机制的动态检测头(dynamic head,DyHead),增强了检测头的表征能力。在磁芯数据集上进行训练和测试,结果表明,该算法的平均精确率均值较原版YOLOv8算法提升了4.3%,模型参数量减少了51.8%,既达到了较高的精确率,又能满足边缘计算设备的部署要求。该算法能比较准确地检测变压器磁芯缺陷,从而为电网安全性能提升提供重要技术支持。To address the problem that transformer core defects pose a potential risk to the safe and stable operation of power grid systems but lack of detection means,a transformer core defect detection algorithm based on improved you only look once version 8(YOLOv8)was proposed.The algorithm firstly reduced the number of model parameters by introducing a ghost network(GhostNet),while maintaining a high detection accuracy.Secondly,it introduced a small-target detection layer with a large-size feature map,which was suitable for the detection of small-object.Finally,a dynamic detection head(DyHead)incorporating an attention mechanism was introduced to enhance the characterization of the detection head.After training and testing on the magnetic core dataset,the results showed that the algorithm′s mean average accuracy rate was improved by 4.3%compared with the original YOLOv8 algorithm,and the amount of model parameters was reduced by 51.8%,which achieved a high accuracy rate and also met the deployment requirements of edge computing devices.The algorithm can detect transformer core defects more accurately and provide important technical support for the improvement of power grid safety performance.

关 键 词：磁芯缺陷检测 GhostNet DyHead 小目标 YOLOv8 

分 类 号：TM755[电气工程—电力系统及自动化]