.png){kind=icon}
.png){kind=icon}
Return
Graph neural network method for insulator surface defects classification
.png){kind=icon}
PRE
AI
N
M
S
DOI：10.1177/00202940261432137.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This study aims to develop an effective solution for detecting insulation faults in power transmission systems, which helps to ensure a stable power supply and minimize power supply disruptions and financial losses due to faults. The proposed research method combines deep learning with object detection models and Graph Neural Networks (GNNs). This method applies transfer learning to optimize the detection process, and GNN is used for multi-object tracking (MOT), detecting and linking data from Unmanned Aerial Vehicle (UAV) images. The research results show that the method combining the You Only Look Once (YOLO-v10) model with GNN gives optimal results in detecting insulation images in forest environments, with the following achieved indices: Accuracy 0.62, MOTP 0.61, MOTA 0.73, and IDF1 0.67. The SSD combined with GNN and Particle Swarm Optimization (PSO) combined with GNN methods gave lower results, respectively, Accuracy 0.56 and 0.53, MOTP 0.57 and 0.53, MOTA 0.69 and 0.65, and IDF1 0.61 and 0.57. The value of the study is to provide a robust and accurate solution for detecting and monitoring insulation wire faults in complex background environments, ensuring reliable detection even under difficult conditions.
Keywords:
insulator
power quality
YOLOv10
graph neural network
GNN
Journal
M
IF:
2
Papers:
52
Citations:
0
Organization
Cited Papers