Title: InsuDet: a lightweight insulator defect detection algorithm based on YOLOv8

Authors: Jinqing Shen; Hong Ye; Chunjun Tang; Yifu Chen; Yan He; Enhui Zheng

Addresses: Jinhua Bada Group Ltd., Jinhua, Zhejiang, China ' Jinhua Bada Group Ltd., Jinhua, Zhejiang, China ' Jinhua Power Supply Company of State Grid Zhejiang Electric Power Co., Jinhua, Zhejiang, China ' China Jiliang University, Hangzhou, Zhejiang, China ' Jinhua Power Supply Company of State Grid Zhejiang Electric Power Co., Jinhua, Zhejiang, China ' China Jiliang University, Hangzhou, Zhejiang, China

Abstract: Failures caused by insulator defects may seriously threaten the normal operation ofpower systems. Hence, how to detect and maintain insulator defects in a timely and accuratemanner is of vital importance. However, the small size of insulator shells coupled with complexbackgrounds make such detection an arduous task. To address the issue of low accuracyin detecting insulator defects, this paper proposes InsuDet, an improved lightweight objectdetection network based on YOLOv8. In order to further reduce the complexity of the model,MobileNetv3 blocks are chosen as the backbone of the model in this paper. Additionally,the bottleneck structure in C2f is replaced with deformable convolution structure and smallobject detection layers are introduced to enhance the feature-extraction performance of themodel, improving its ability to identify dense small objects. Our experimental results show thatcompared to the YOLOv8-s baseline model, our model has increased precision and recall ratesby 2.5% and 1.8%, respectively, reflecting at 96.8% and 95.5%, correspondingly. Furthermore,our proposed model reduces the number of parameters by 21%, and achieves a detection speedof 111.6 frames/second. Our model can accurately and timely detect insulator defects, whileits lightweight structure ensures that it can be effortlessly deployed on mobile devices likeunmanned aerial vehicles.

Keywords: YOLOv8 network; insulator defect detection; aerial inspection images; image processing.

DOI: 10.1504/IJAMECHS.2024.143358

International Journal of Advanced Mechatronic Systems, 2024 Vol.11 No.4, pp.179 - 191

Received: 30 Jan 2024
Accepted: 24 Apr 2024
Published online: 16 Dec 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article