无线电工程2025，Vol.55Issue(2)：281-290,10.DOI:10.3969/j.issn.1003-3106.2025.02.007
改进YOLOv8的绝缘子及其自爆缺陷目标检测算法
Improved YOLOv8 Algorithm for Insulator and Its Self-explosion Defect Detection
摘要
Abstract
To deal with the problems of false detection,missed detection and low detection accuracy in the current insulator defect object detection algorithms,an improved YOLOv8 algorithm for insulator and its self-explosion defect detection is proposed.Firstly,a small object detection head is added to the original structure of YOLOv8 model,which makes the network model more focused on small objects.Secondly,Deformable ConvNets v2(DCNv2)is used to replace the ordinary convolution in the C2f module,so that the model can better capture and perceive the characteristics of the detected object.Thirdly,the Inverted Residual Mobile Block(iRMB)is introduced into the backbone network to enable the model to capture the relationship between local features and complex spaces,so that it can focus more on the detection of object feature information.Finally,Wise Intersection over Union(WIoU)is applied as the loss function,which makes the model achieve better results in training and improve the detection accuracy.On the transmission line insulator dataset,the mean Average Precision(mAP)of the algorithm is improved to 98.0%,which is 2.9%higher than that of the original algorithm.Precision and Recall of the algorithm is improved to 92.8%and 94.8%respectively.Compared with the original algorithm,the improved algorithm has a significant improvement,especially for small object defect detection,which proves the feasibility of the improved algorithm.关键词
目标检测/绝缘子检测/可变形卷积/注意力机制Key words
object detection/insulator detection/DCNv2/attention mechanism分类
信息技术与安全科学引用本文复制引用
高英铭,韩顺杰,周波,李东奇..改进YOLOv8的绝缘子及其自爆缺陷目标检测算法[J].无线电工程,2025,55(2):281-290,10.基金项目
吉林省科技厅重点研发项目(20210201025GX)Key Research and Development Project of Jilin Provin-cial Science and Technology Department(20210201025GX) （20210201025GX）
