Summary
To address the recognition problem of infrared images for substation equipment faults, a weakly supervised recognition method is proposed. First, based on the faster region convolutional neural network(RCNN) model, substation equipment is identified, and the accuracy of equipment detection is further improved by adjusting the model's network structure and parameters. Second, considering the inherent physical characteristics of various operational equipment in substations, an image feature extraction method is proposed based on kernel density estimation of temperature probability distributions. Finally, to tackle challenges such as difficulties in data collection in practical applications, weakly supervised learning is introduced. Labeled data are used to compute prototype vectors representing equipment states, and unlabeled samples are utilized to update these prototype vectors.
Read more Read lessAdditionally, a confidence parameter is incorporated to effectively enhance the model's generalization performance. The experimental results demonstrate that the proposed algorithm significantly improves the recognition accuracy for different types of equipment faults.
Compared with existing methods, the proposed model achieves a 6.7% improvement in overall average recognition accuracy, with significant accuracy increases observed across all equipment types.
Additional informations
| Publication type | Session Materials |
|---|---|
| Reference | A2_11495_2026 |
| Publication year | |
| Publisher | CIGRE |
| Country | China, People's Republic of |
| Study committees | |
| File size | 542 KB |
| Price for non member | 30 € |
| Price for member | 30 € |
Authors
XIE Zhicheng - China Southern Power Grid; DENG Jun - China Southern Power Grid; ZHOU Haibin - China Southern Power Grid; PAN Zhicheng - China Southern Power Grid
Keywords
Weakly-supervised, Faster RCNN, Infrared Image Recognition, Fault Detection
