基于改进YOLOx的绝缘子自爆缺陷检测方法

Detection method of self-explosion defects of insulators based on YOLOx

针对采集到的不同种类绝缘子图像背景复杂的特点,提出一种改进YOLOx的绝缘子自爆缺陷检测方法。该算法在主干网络引入可以增强网络感知力的CBAM注意力机制,并且对损失函数的IoU损失计算方式优化使用EIoU。将改进算法与YOLOv3算法、YOLOv4算法、YOLOv5算法和YOLOx算法进行对比。试验结果表明,改进后的算法在测试集上的检测精度达到97.26%,能够区分正常的绝缘子和有缺陷的绝缘子,可初步应用于输电线路的绝缘子巡检作业。

An improved YOLOx method for insulator self-explosion defect detection is proposed to address the complex background of the different types of insulator images collected.The algorithm introduces a CBAM attention mechanism in the backbone network that can enhance network perception and optimize the IoU loss calculation of the loss function to use EIoU.The improved algorithm is compared with the YOLOv3 algorithm,YOLOv4 algorithm,YOLOv5 algorithm and YOLOx algorithm.The experimental results show that the improved algorithm achieves 97.26%detection accuracy on the test set,which is able to distinguish normal insulators from defective ones and can be initially applied to insulator inspection operations on transmission lines.