摘要
输电线路的绝缘子损坏给电力系统带来诸多安全隐患,引发电弧、火灾等危害。实时高效的绝缘子损害识别技术成为解决该问题的关键。在大量实验的基础上,提出一种基于SA-YOLOv5s的输电线路绝缘子损害识别方法,该方法在YOLOv5s模型的卷积模块中引入CBAM注意力机制,提高模型特征提取能力;使用GhostC3模块替代主干网络的C3模块,降低模型复杂度;使用C2f残差模块替代颈部网络的C3模块,提高检测准确性;使用MPDIoU损失函数替代CIoU定位损失函数,提高检测精度;融合改进多尺度的SAHI切片超推理,提高预测结果的精度与准确度。实验结果表明,改进SA-YOLOv5s模型在数据集上检测的P_(mA0.5)值为95.2%,P_(mA0.5:0.95)值为61.9%,检测速度为98帧/s,且绝缘子、绝缘子破裂、表面闪络损坏的预测准确度分别达到99.2%、100%与100%。改进模型满足复杂环境下对小目标及密集目标的检测需求。
Insulator damage in transmission lines brings many safety hazards to the power system,triggering arcs,fires and other dangers.Real-time and efficient insulator damage identification technology becomes the key to solve this problem.Based on a large number of experiments,a transmission line insulator damage identification method based on SA-YOLOv5s is proposed,which introduces the CBAM attention mechanism into the convolution module of the YOLOv5s model to improve the feature extraction capability of the model;using the GhostC3 module to replace the C3 module of the backbone network to reduce the complexity of the model;using the C2f residual module to replace the neck network's C3 module to improve detection accuracy;using MPDIoU loss function instead of CIoU localisation loss function to improve detection accuracy;and fusing the improved multi-scale SAHI slicing hyper-inference to improve the precision and accuracy of prediction results.The experimental results show that the improved SA-YOLOv5s model detects 95.2%of the P_(mA0.5) value,61.9%of the P_(mA0.5:0.95) value,and 98 frames/s of detection speed on the dataset,and the prediction accuracies of the insulator,insulator rupture,and surface flashover damage reach 99.2%,100%,and 100%,respectively.The improved model meets the detection needs of small and dense targets in complex environments.
作者
杨云皓
韩国政
朱国防
YANG Yunhao;HAN Guozheng;ZHU Guofang(School of Information and Automation,Qilu University of Technology(Shandong Academy of Sciences),Jinan 250353,China;School of Electrical Engineering,Shandong University,Jinan 250100,China)
出处
《齐鲁工业大学学报》
2025年第5期20-29,共10页
Journal of Qilu University of Technology
基金
国家自然科学基金(U23B20122)。
关键词
输电线路
残差模块
注意力机制
平均精度均值
预测准确度
transmission line
residual module
attention mechanism
average precision mean
prediction accuracy
