Research on Position and Recognition Algorithms for Insulators Based on EfficientDet

Abstract

Insulators, as one of the most used power equipment on transmission lines, are essential for maintaining the safety and stability of the power system. The positioning and recognition of insulators is the prerequisite for fault diagnosis. Although deep learning methods can improve the inspection efficiency, there remain problems such as low accuracy and poor real-time performance for the existing insulator positioning and recognition methods. To solve this problem, this paper proposes an insulator positioning and recognition method based on EfficientDet. The EfficientDet algorithm uses EfficientNet as the backbone and BiFPN as a feature network layer, repeatedly stacking BiFPN to obtain a more advanced feature fusion method. The EfficientDet algorithm performs feature extraction on the collected insulator image data to realize insulator positioning and recognition. This article chooses the open source Pytorch as a tool, combined with tensorboard to visualize the loss curve during the training process. The results show that using the EfficientDet algorithm can achieve rapid model convergence. The average detection time is 0.033s and the average processing speed reaches 29.35 FPS. The method proposed in this paper achieves a balance between model size and detection speed, which verifies that the model has excellent performance in insulator detection, and provides a new idea for intelligent positioning and recognition of power equipment.