Small object defect detection of insulators based on YOLO-insulator model in complex background

Authors

Penglin DONG, Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming 650504, China
Jiupeng CHEN, Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming 650504, China; Key Laboratory of Advanced Equipment Intelligent Manufacturing Technology of Yunnan Province, Kunming 650500, China
Sen WANG, Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming 650504, China; Key Laboratory of Advanced Equipment Intelligent Manufacturing Technology of Yunnan Province, Kunming 650500, China
Hongjun SAN, Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming 650504, China; Key Laboratory of Advanced Equipment Intelligent Manufacturing Technology of Yunnan Province, Kunming 650500, ChinaFollow
Hongwei HU, College of Mechanical and Vehicle Engineering, Changsha University of Science & Technology, Changsha 410114, China

Keywords

insulator defect detection; complex background; attention mechanism; squeeze-and-excitation; smallobject detection layer

Abstract

Computer vision-based methods for insulator defect detection from aerial images are widely used in power inspection. To address missed and false detections caused by complex backgrounds and small target scales, a YOLO-insulator defect detection model is proposed to improve detection accuracy. First, the reparameterized convolution based on channel shuffle-one-shot aggregation (RCS-OSA) is introduced to replace the traditional two-dimensional convolution C2f, thus enhancing the network's feature extraction capability. In the neck network, the RCS-OSA module is used to replace some of the C2f convolutions, and the squeeze-and-excitation network (SENet) is introduced to enhance the model's ability to capture inter-channel relationships and express overall features. Finally, to address the difficulty in detecting multiple defect regions due to their small size, a small object detection layer method is proposed. This layer contains more detailed defect information and is more conducive to defect detection. Experimental results on a self-made insulator dataset demonstrate that, compared with the baseline YOLOv 8n, the YOLO-insulator model achieves higher precision, recall, and mean average precision, improving overall model performance.

DOI

10.19781/j.issn.1673-9140.2026.01.025

First Page

262

Last Page

276

Recommended Citation

DONG, Penglin; CHEN, Jiupeng; WANG, Sen; SAN, Hongjun; and HU, Hongwei (2026) "Small object defect detection of insulators based on YOLO-insulator model in complex background," Journal of Electric Power Science and Technology: Vol. 41: Iss. 1, Article 25.
DOI: 10.19781/j.issn.1673-9140.2026.01.025
Available at: https://jepst.researchcommons.org/journal/vol41/iss1/25