基于改进YOLOv5s的绝缘子缺陷检测方法  

Insulator Defect Detection Method Based on Improved YOLOv5s

作　　者：王子玉 陈佳星 白博文 徐爱婷 Wang Ziyu;Chen Jiaxing;Bai Bowen;Xu Aiting(Liaoning Technical University Faculty of Electrical and Control Engineering,Huludao,Liaoning 125105,China)

机构地区：[1]辽宁工程技术大学电气与控制工程学院,辽宁葫芦岛125105

出　　处：《机电工程技术》2024年第8期197-200,共4页Mechanical & Electrical Engineering Technology

摘　　要：绝缘子在高压和超高压架空输电线路中的安全运行至关重要,因此对绝缘子图像的准确识别显得尤为关键。为了提高绝缘子缺陷检测的精度,提出了一种基于改进YOLOv5s模型的绝缘子缺陷检测方法。通过引入scSE注意力机制,以实现通道和空间方面的校准,强化了该模型在对绝缘子缺陷的辨识能力,从而进一步提高对绝缘子图像识别的精度,特别是在遮挡和相似目标干扰的情况下表现更为出色;在此基础上增加了Mosaic数据增强处理,为模型的训练提供更多数据,使模型在多场景下表现优异。经实验证实,该算法准确率达到了92.8%,召回率达到97.5%,mAP达到了98.8%,其检测精度明显优于其他模型,并在复杂场景下展现了出色的鲁棒性,为绝缘子检修工作提供了新的思路。Insulators are crucial for the safe operation of high-voltage and ultra-high-voltage overhead transmission lines,making accurate identification of insulator images particularly important.To improve the accuracy of insulator defect detection,this study proposes an insulator defect detection method based on an improved YOLOv5s model.By introducing the scSE attention mechanism to achieve calibration in both channel and spatial aspects,the model's ability to identify insulator defects is strengthened,thereby further enhancing the accuracy of insulator image recognition,especially performing more outstandingly in cases of obstruction and similar target interference.Additionally,Mosaic data augmentation processing is added on this basis to provide more data for training the model,enabling the model to perform excellently in multiple scenarios.Experimental results show that the algorithm achieves an accuracy of 92.8%,a recall rate of 97.5%,and a mean average precision(mAP)of 98.8%.Its detection accuracy is significantly superior to that of other models,and it demonstrates outstanding robustness in complex scenarios,providing new insights for insulator maintenance work.

关 键 词：绝缘子检测 YOLOv5s模型 scSE注意力机制 图像处理 

分 类 号：TM216[一般工业技术—材料科学与工程] TP391.4[电气工程—电工理论与新技术]