Infrared Image Recognition of Substation Equipment Based on Yolov5-GDE

With the development of infrared detection technology and the increased demand for intelligent identification of substation equipment, infrared substation equipment detection networks with lightweight parameters and high detection accuracy have been the focus of research. To meet this demand, we pro...

Full description

Saved in:
Bibliographic Details
Published in2024 7th International Conference on Advanced Algorithms and Control Engineering (ICAACE) pp. 266 - 272
Main Authors Li, Dahua, Zheng, Wenpeng, Yu, Xiao, Gao, Qiang
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.03.2024
Subjects
Analytical models
Attention mechanism
component
Diverse Branch Block
Fault diagnosis
Image recognition
Infrared Imaging
Lightweight
Substations
Target recognition
Training
YOLO
YOLOv5
Online AccessGet full text

Cover

More Information
Summary:With the development of infrared detection technology and the increased demand for intelligent identification of substation equipment, infrared substation equipment detection networks with lightweight parameters and high detection accuracy have been the focus of research. To meet this demand, we propose an infrared object detection network called YOLOv5-GDE. To ensure the speed of model detection, we choose You Only Look Once v5s (YOLOv5s) as the basic framework and design a backbone network of diverse branches for enhancing the feature extraction capability of the network and deploying it efficiently without increasing the number of parameters and model complexity. In addition, we introduce efficient multi-scale attention to increase the receptive field using multi-scale parallel branching, and enhance the global and local feature interactions through cross-space information aggregation methods to improve the stability and accuracy of target detection. Moreover, we redesigned the ordinary convolutional module in the model and the C3 module of the neck network to reduce the number and complexity of model parameters. Furthermore, we improve the loss function so that the network converges quickly during the training process. The experimental results show that the proposed YOLOv5-GDE network reaches 95.8% mean Average Precision (mAP0.5) in the infrared substation equipment dataset. Compared with the original model, the number of parameters decreases by 36.9%. The proposed architecture is able to meet the accuracy and real-time requirements of substation equipment identification, and provides conditions for subsequent fault diagnosis of substation equipment.
DOI:10.1109/ICAACE61206.2024.10548138