基于红外图像检测和改进YOLOv4的高压套管故障识别方法  

Fault Identification Method of High-voltage Bushing Based on Infrared Image Detection and Improved YOLOv4 Algorithm

作　　者：李浩伟 张楚岩 史梓男 孙爱春 刁明光[1] 徐惠勇[1] LI Haowei;ZHANG Chuyan;SHI Zinan;SUN Aichun;DIAO Mingguang;XU Huiyong(School of Information Engineering,China University of Geosciences(Beijing),Beijing 100083,China;Syi Tsing Energy Tech,Beijing 100084,China)

机构地区：[1]中国地质大学(北京)信息工程学院,北京100083 [2]北京西清能源科技有限公司,北京100084

出　　处：《高压电器》2023年第11期24-31,共8页High Voltage Apparatus

基　　金：国家自然科学基金项目(51907178);清华四川能源互联网研究院科技项目(33412022003)。

摘　　要：红外检测是高压外绝缘设备状态在线检测的主要方法之一,为了提高高压套管发热故障红外图像检测的准确率,解决因故障样本较少引起的漏检问题,文中提出了一种基于改进型YOLOv4的故障识别方法,可实现对套管发热区域的高效定位与识别,具有很好的工程应用前景。对YOLOv4算法进行的改进主要包括:首先,将通道注意力机制SE(Squeeze and Excitation)模块插入特征提取网络中的残差模块中,以加强网络对特征信息的提取;其次,分别使用EIoU Loss和Focal Loss取代原模型的边界框回归损失与置信度损失,以提高损失函数的回归精度以及对数据集中困难样本的识别准确率,从而有效减少漏检发生的概率。实验与测试结果表明,所提方法与改进前相比平均精度提高了5.61%,对数据集中更难被识别的故障样本的精确度提升了8.57%。Infrared detection is one of the main methods for online detection of the status of high-voltage external insulation equipment.For improving the accuracy of infrared image detection of heating of high-voltage bushing and solving the problem of missing detection due to the less of fault samples,a kind of fault identification method based on improved YOLOv4 is proposed,which can achieve high efficient location and identification of the bushing heating area and have a good engineering application prospect.The improvement of YOLOv4 algorithm mainly includes:firstly,the channel attention mechanism SE(Squeeze and Extinction)module is inserted into the residual module in the feature extraction network to strengthen the extraction of feature information.Secondly,EIoU Loss and Focal Loss are used respectively to replace the bounding-box regression loss and confidence loss of the original model to improve the regression accuracy of the loss function and the recognition accuracy of the difficult samples in the data set,thus effectively reducing the probability of missing detection.The experimental and test results show that the average accuracy of the proposed method is 5.61%higher than that before the improvement,and the accuracy of the fault samples which are more difficult to be identified in the data set is increased by 8.57%.

关 键 词：套管 故障检测 红外图像 YOLOv4 

分 类 号：TM216.5[一般工业技术—材料科学与工程]