A Light-Weight Network for Small Insulator and Defect Detection Using UAV Imaging Based on Improved YOLOv5

Insulator defect detection is of great significance to compromise the stability of the power transmission line. The state-of-the-art object detection network, YOLOv5, has been widely used in insulator and defect detection. However, the YOLOv5 network has limitations such as poor detection rate and h...

Full description

Saved in:
Bibliographic Details
Published in:Sensors (Basel, Switzerland) Switzerland), 2023-06, Vol.23 (11), p.5249
Main Authors: Zhang, Tong, Zhang, Yinan, Xin, Min, Liao, Jiashe, Xie, Qingfeng
Format: Article
Language:English
Subjects:
Accuracy
Algorithms
Boxes
convolutional block attention module
Defects
Drone aircraft
Eigenvalues
ghost module
insulator and defect detection
keyword light-weight
Localization
Methods
Neural networks
Power lines
Semantics
Sensors
Unmanned aerial vehicles
Weight reduction
YOLOv5
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Insulator defect detection is of great significance to compromise the stability of the power transmission line. The state-of-the-art object detection network, YOLOv5, has been widely used in insulator and defect detection. However, the YOLOv5 network has limitations such as poor detection rate and high computational loads in detecting small insulator defects. To solve these problems, we proposed a light-weight network for insulator and defect detection. In this network, we introduced the Ghost module into the YOLOv5 backbone and neck to reduce the parameters and model size to enhance the performance of unmanned aerial vehicles (UAVs). Besides, we added small object detection anchors and layers for small defect detection. In addition, we optimized the backbone of YOLOv5 by applying convolutional block attention modules (CBAM) to focus on critical information for insulator and defect detection and suppress uncritical information. The experiment result shows the mean average precision (mAP) is set to 0.5, and the mAP is set from 0.5 to 0.95 of our model and can reach 99.4% and 91.7%; the parameters and model size were reduced to 3,807,372 and 8.79 M, which can be easily deployed to embedded devices such as UAVs. Moreover, the speed of detection can reach 10.9 ms/image, which can meet the real-time detection requirement.
ISSN:1424-8220
1424-8220
DOI:10.3390/s23115249