2024 | Yanping Chen, Chong Deng, Qiang Sun, Zhize Wu, Le Zou, Guanhong Zhang, Wenbo Li
This article proposes a lightweight Faster R-CNN (Faster R-CNN-tiny) model for detecting self-explosion defects in insulators. Traditional object detection methods for insulator defects face challenges such as excessive parameters, low accuracy, and slow detection speed. To address these issues, the authors replaced the ResNet backbone with EfficientNet, which reduces model parameters while improving detection accuracy. A feature pyramid network was used to generate feature maps at various resolutions for feature fusion, enabling detection of objects at different scales. Depth-wise separable convolutions were introduced to enhance detection speed while slightly reducing accuracy. Transfer learning was also employed to improve the model's ability to detect small target defects. The proposed model was validated using a newly created insulator defect dataset, showing significant improvements in mean average precision (mAP), frames per second (FPS), and parameter count compared to the original Faster R-CNN model. The lightweight Faster R-CNN-tiny model is more suitable for edge deployment and can effectively locate insulator defects. The study also compared the performance of the proposed model with other object detection algorithms, demonstrating its superior accuracy and efficiency in detecting self-explosion defects in insulators. The results show that the proposed model can effectively identify insulator defects in transmission lines and has potential for real-time and efficient detection in power systems.This article proposes a lightweight Faster R-CNN (Faster R-CNN-tiny) model for detecting self-explosion defects in insulators. Traditional object detection methods for insulator defects face challenges such as excessive parameters, low accuracy, and slow detection speed. To address these issues, the authors replaced the ResNet backbone with EfficientNet, which reduces model parameters while improving detection accuracy. A feature pyramid network was used to generate feature maps at various resolutions for feature fusion, enabling detection of objects at different scales. Depth-wise separable convolutions were introduced to enhance detection speed while slightly reducing accuracy. Transfer learning was also employed to improve the model's ability to detect small target defects. The proposed model was validated using a newly created insulator defect dataset, showing significant improvements in mean average precision (mAP), frames per second (FPS), and parameter count compared to the original Faster R-CNN model. The lightweight Faster R-CNN-tiny model is more suitable for edge deployment and can effectively locate insulator defects. The study also compared the performance of the proposed model with other object detection algorithms, demonstrating its superior accuracy and efficiency in detecting self-explosion defects in insulators. The results show that the proposed model can effectively identify insulator defects in transmission lines and has potential for real-time and efficient detection in power systems.