- Conference Article
- 10.1109/aecspe66597.2025.00089
Research on Foreign Body Image Recognition Technology of High Voltage Overhead Transmission Line Based on Faster R-CNN Algorithm
- Sep 04, 2025
- Yige Chen + 3 more +3
This study focuses on the image recognition of foreign bodies in high voltage overhead transmission lines, and proposes a technology based on Faster R-CNN algorithm. High-voltage overhead transmission lines are easily invaded by foreign bodies such as bird nesting and kite winding, which leads to accidents. Traditional detection methods have many limitations. Faster R-CNN has become an ideal detection algorithm because of its high accuracy and real-time performance. In this study, 10,000+UAV inspection images covering various scenes and foreign body types were collected to build a data set, and data enhancement methods were used to improve the robustness of the model. In the model architecture, ResNet-50-FPN is selected as the backbone network, and the detection performance of foreign objects is enhanced by introducing FPN structure, optimizing anchor frame parameters and adopting OHEM mechanism. In training, reasonable loss function is designed, Focal Loss is introduced to alleviate category imbalance, and various optimization strategies are applied to improve training effect. The experimental results show that the improved model achieves 89.7% in the test set, the recall rate of plastic bags is 89.6%, the accuracy rate of bird's nest is 94.2%, the model size is compressed to 72MB, and it is stable in different background complexity scenes, which is significantly superior to the original Faster R-CNN and other mainstream detection algorithms, effectively improving the accuracy and real-time performance of foreign body detection for high-voltage overhead transmission lines, and providing an efficient solution for the field of power inspection.
Read more