Multi-label image recognition for electric power equipment inspection based on multi-scale dynamic graph convolution network

Abstract

There exists many types of power equipments in power system operation and inspection scenarios. Among them, different types of equipments often have common representations, thus these various types of power equipments with similar characteristics would bring certain challenges to power equipment defect recognition. The traditional multi-label image recognition is usually with low recognition accuracy. The main reason is that the relationship between each label in the image is ignored. Furthermore, graph convolutional neural networks rely on the modeling ability of graphs to further improve the accuracy of recognition. To this end, we propose a multi-label image recognition model for electric power equipment inspection based on multi-scale dynamic graph convolutional network. In our model, multi-scale image features are extracted through a multi-scale feature extraction network firstly, and then the label relevance of a specific image is adaptively learned through combining the dynamic graph convolutional network. Finally, a self-built dataset of electrical equipment defects is used to conduct experimental results for comparison and validation. According to the analysis of our experimental results, our proposed model shows a better performance. The average accuracy of our model can reach 88.1%, which increases by 0.8% and 31.8% compared with the original model and the baseline model, respectively, showing the effectiveness and superiority of the proposed method.

Keywords

• Electric power equipment inspection
• Multi-label image
• Graph convolution network
• Multi-scale