LDDFSF-YOLO11: A Lightweight Insulator Defect Detection Method Focusing on Small-Sized Features

Insulators are essential for maintaining electrical isolation in transmission lines, with the timely and accurate identification of their defects via uncrewed aerial vehicle (UAV) being crucial for ensuring operational stability and safety in high-voltage transmission lines. Due to complex environme...

Full description

Saved in:
Bibliographic Details
Main Authors: Peng Shen, Keyu Mei, Huiqiong Cao, Yongxiang Zhao, Guoqing Zhang
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11003913/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Insulators are essential for maintaining electrical isolation in transmission lines, with the timely and accurate identification of their defects via uncrewed aerial vehicle (UAV) being crucial for ensuring operational stability and safety in high-voltage transmission lines. Due to complex environmental conditions, varying viewing angles, and unstable lighting conditions, existing insulator defect detection algorithms still face challenges, including missed detection, false detection, and inadequate adaptability for detecting small-sized defects in UAV captured images. This study proposes a lightweight insulator defect detection method (LDDFSF-YOLO11) that focuses on small-sized features to efficiently detect and identify insulator defects. In the LDDFSF-YOLO11 model, a focused feature extraction module (MFFEConv) is proposed to enhance the feature extraction capability of the backbone network. In addition, a collaborative attention mechanism and a multi-scale feature fusion module (MDMF) are proposed to enhance the model’s ability to detect small-sized defects in remote and complex backgrounds, effectively solving the problems of missed detections, false detections, and poor adaptability to complex backgrounds. In addition, we propose an RWLoss loss function to improve the accuracy of bounding box matching in the model. And optimize the network structure through channel pruning technology to improve model lightweighting. Finally, experimental validation was conducted using a self built insulator dataset and the China Power Line Insulator Dataset (CPLID). The experimental results showed that compared with the baseline model YOLO11, the LDDFSF-YOLO11 model improved AP50 by 9.8%, 8.2%, and 5.4%, mAP by 7.1%, and reduced model size by 56.1%, accounting for only 4.2MB, in three types of small-sized defects: fractional insulator, drop insulator, and flash insulator, respectively. The LDDFSF-YOLO11 model can quickly and accurately identify small-scale defects in insulators in complex backgrounds, making important contributions to the safe and stable operation of high-voltage transmission lines.
ISSN:2169-3536