Power Equipment Defect Detection Based On Hierarchical Multi-tag Classification

Nowadays,the demand for electricity in industrial production and people’s daily lives in China is increasing,and the power system is facing huge challenges.The detection and maintenance of the operation status of power equipment are related to the safety and stability of the entire power system.Traditional defect detection of power equipment is mainly achieved through manual inspection or data collection by cameras.The manual detection method requires a large workload and requires high professionalism from the staff.Moreover,with the increasing number and variety of power equipment in substations,the detection efficiency of camera acquisition technology cannot meet the requirements.At present,deep learning and object detection techniques are increasingly being applied to defect detection of power equipment,which has improved the efficiency of defect detection to some extent.However,in methods based on deep learning and object detection,some methods only perform multi-classification detection for defects generated by a certain type of power equipment,while some defect detection methods that perform multi-label classification do not fully explore the relationship between device defect labels.Therefore,these defect detection methods cannot meet the needs of comprehensive and accurate detection of power equipment defects in substations.Faced with the current situation of complex types of defects in power equipment in substations,traditional defect detection methods and multi-classification object detection methods cannot accurately and comprehensively detect power equipment defects.A power equipment defect detection method based on hierarchical multi-label classification is proposed to address this issue.Firstly,defect image collection was conducted on power equipment from multiple substations in different regions,and the relationship between equipment defect labels was fully explored to construct a power equipment defect image dataset with hierarchical label structure.Then,a hierarchical classification model based on cross-modal feature fusion is proposed.This model utilizes multimodal feature fusion technology to enhance the feature representation of defect targets,and adopts a hierarchical classification method to achieve multi-label classification detection of defects.The model first extracts image features through the Resnet50 network;Then,a region proposal network is used to predict the target position,foreground,and background probabilities,and the ROI Align region feature aggregation method is used to generate more accurate position coordinates;Finally,hierarchical classification is used to embed the parent category label into the target feature representation of the current layer for layer by layer defect classification.The final layer obtains the final defect detection result.To verify the effectiveness of the method,comparative experiments were conducted on the power equipment defect dataset and benchmark dataset with popular multi-label classification defect detection methods and commonly used object detection algorithms.The experimental results showed that the model had the best detection accuracy for the vast majority of equipment defect categories,with an average detection accuracy of 83.4%.Compared with the second best performing model,the accuracy improved by 2.1%,And the average detection accuracy on the benchmark dataset has also improved by 1.1%to 9%,providing an effective technical support for comprehensive and accurate diagnosis of substation equipment faults.