轻量级YOLOv8的绝缘子缺陷检测算法

打开文本图片集

中图分类号：TP391.41 文献标志码：A 文章编号：1672-1098（2025）02-0010-09

引文格式：.轻量级YOLOv8的绝缘子缺陷检测算法[J].安徽理工大学学报（自然科学版），2025，45（2） ：10-18.

Insulator Defect Detection Algorithm for the Lightweight YOLOv8 ZHOU Mengran，CAI Wentao，LAI Wenhao

ScholfElectricalandInformationEnginering，Anui UniversityofScienceandTechnolog，HuainanAnui32O，China）

Abstract：Objective To address isses such as low detection accuracy，prolonged detection time，and frequent missed or false detections in insulator defect detection methods，an improved detection approach based on YOLOv8 was proposed. Methods First，a lightweight bidirectional feature pyramid lightweight network （BiFPN） was introduced to enhance multi ⋅- scale feature fusion while reducing computational complexity. Second，the detection head was redesigned by replacing the original convolutions with two re-parameterized convolutions，enabling richer and more complex feature extraction，thereby improving training eficiency and model generalization.Finally，the original lossfunction（CIoU）was replaced with MPDIoU to optimize bounding box regression and enhance detectionaccuracy.Results Experimental results demonstrated that compared to theoriginal model，the improved algorithm reduced computational load by 27.2% ，decreased parameters by 41.2% ，increased FPS by 81.8% ，and improved mAP by 3.2% . Conclusion The proposed method successfully enhanced detection accuracy while maintaining model lightweightness，meeting the practical requirements for insulator defect detection.

Key words ： insulator;defect detection;lightweight; detection head ;loss function

在电力系统中，绝缘子作为电气设备的重要组成部分，起着将电力线路导线与支架、横担等接地构件隔离的重要作用[1]。（剩余9085字）