基于自监督学习的输电线路螺栓螺母异常检测技术

Fault Detection Technology of Bolts and Nuts based on Self-Supervised Learning in Transmission Line

针对输电线路螺栓螺母异常检测问题,对无标签螺栓数据训练深度学习预训练模型进行了研究,首次采用了自监督学习的方法,使用大量无标签数据集进行学习,首先,使用少量带标签的螺栓目标检测数据集训练目标检测模型,推理获取大量的单图无标签螺栓数据集,然后,采用自监督学习方法通过单图无标签螺栓数据集训练螺栓预训练大模型并获取可视化的螺栓注意力图,最后通过实验对比螺栓预训练大模型在螺栓分类和检索任务中与非预训练模型的效果,实验结果表明无标签数据得到的预训练大模型可以注意到螺杆、螺母、连接件的位置,进一步表明螺栓预训练大模型在分类任务中准确率提升了2%到7%,在螺栓检索任务中平均精度提升了8%。

Aiming at the detection problem of bolts and nuts in transmission lines,an unlabeled bolt data training deep learning pre-training model is studied,a self-supervised learning method is adopted for the first time,a large number of unlabeled data sets are used to learn.Firstly,a small amount of labeled bolt object detection data sets are used to train the object detection model,inference and obtain a large number of single-image unlabeled bolt datasets,and then the self-supervised learning method is used to train the bolt pre-training large model through the single-image unlabeled bolt dataset and obtain the visualized bolt attention map.Finally,the effects of the bolt classification and retrieval tasks on the bolt pre-training large model are compared by the experiment with the non-pre-trained model.The experimental results show that the pre-trained large model obtained from unlabeled data sets can notice the positions of screws,nuts,and connectors,which further shows that the accuracy of the bolt pre-trained large model in the classification task is improved by 2%to 7%,and the average accuracy by 8%in the bolt retrieval task.