Research On Insulator Fault Detection Algorithm Based On Mask R-CNN

The insulator plays the role of electrical insulation at the same time as connecting the conductor.It is a necessary device for power transmission lines and is widely used in power systems.The operating state of the insulator affects the reliability and safety of the power system,and even affects the operating life of the entire power system.Regular inspections and timely detection of faulty insulators are of great significance for maintaining the normal operation of the power grid.With the development of automation technology,drones have been used to inspect the state of insulators,but the complexity and variability of application scenarios have brought great challenges to the automatic identification of insulator fault images in a big data environment.Based on deep learning,this paper studies the automatic detection of insulator fault images.Firstly,the implementation model of Convolutional Neural Network(CNN)is compared and analyzed.Combined with the characteristics of aerial images of insulators,a mask region-based convolutional neural network(Mask R-CNN)is selected to identify insulator fault images.The feature extraction network of Mask R-CNN is studied,the influence of the training parameters of Mask R-CNN network on the learning insulator sample is studied,and the network structure and parameter configuration suitable for learning the insulator fault image are designed.An insulator fault image detection algorithm based on Mask R-CNN is proposed,and a Mask R-CNN model suitable for insulator fault image detection is established.A sample library of insulator fault images is built using Labelme to realize the learning and detection of insulator fault images.In the Python environment,the proposed algorithm is implemented by software using Tensor Flow and Keras development tools.The proposed algorithm was tested using measured aerial insulator self-explosion fault images,and the results were as follows: The average test average accuracy(m AP)of 710 images was 0.986,of which the m AP of a single target image can reach 1.000,and the m AP of a multi-target image can be Up to 0.948,it can be seen that the algorithm can well detect the insulator self-explosion fault image.By analyzing the generalization ability and environment adaptability of the proposed algorithm,it is found that the algorithm has certain generalization ability and environment adaptability,and can improve the detection ability of fault image by supplementing and updating training samples,which provides certain technical support for insulator fault automatic identification.