Image Classification of Missing Insulators Based on EfficientNet

Abstract

In this paper, for the detection of missing images of power insulators, the production of image datasets and image classification methods are discussed. Using the drone aerial insulator images from the power grid company, images containing insulators in different scenarios such as high-voltage transmission lines and substations were collected, 2000 insulator images were extracted, and a power insulator database was constructed. Insulator-missing image classification model to verify the robustness of the EfficientNet algorithm. Use EfficientNet to build a transfer learning network, train it, and compare it with the commonly used classifier ResNet-50. By introducing classification evaluation indicators and class activation maps, the experimental results show that EfficientNet-b0 has good transfer ability and can significantly improve the model. Efficiency, better than ResNet-50 for insulator-missing image classification.