Insulator Identification And Discharge Severity Evaluation Based On YOLO Algorithm

Insulators have a large amount of use and a large geographical span in the power grid.Conventional inspection methods mainly rely on manual experience,and it is difficult to find weak discharges on the surface of equipment.Solar blind ultraviolet imaging detection has the advantages of high sensitivity,non-contact and intuitive characterization.Based on the characteristics of dual-channel imaging of ultraviolet imagers,this paper proposes a method for identifying insulators of visible light channel images based on artificial intelligence.Based on this,discharge severity assessment of the ultraviolet channel insulator image is realized.A deep learning hardware and software research platform based on the TensorFlow and Darknet frameworks under the Linux environment was established.Based on the pictures collected in the laboratory and the field,a sample database of insulators was established and the sample pictures were marked and the YOLO network training was completed,the effects of training sample number and network depth on recognition accuracy are studied,and a matching scheme is given.After optimization,the recognition accuracy reached 95%.Based on the artificial climate chamber,the research on the pollution discharge experiments of insulators was carried out.Ultraviolet videos at different discharge intensities were obtained.The relative photon number and spot area parameters were defined.The relationship between the above parameters and leakage current was analyzed.The ultraviolet image database was established,the classification of ultraviolet images and the labeling of the four discharge severities were completed.The effects of learning rate and activation function on recognition accuracy and training errors were analyzed,and the optimization methods were proposed.The degree of realization is evaluated.The recognition accuracy on the test set reached 94.5%.Based on the QT language,the software for extracting UV image quantization parameters was developed to realize image segmentation and parameter calculation.Based on the PHP language and my SQL database,a comprehensive information management system based on the browser/server model was developed,which realized effective management of inspection UV imaging videos and basic information of equipment.The preliminary field applications were carried out with good results.