Insulator Defect Recognition System Based On Convolutional Neural Network

Overhead lines are an important part of the national power system responsible for the transmission of electrical energy,and have experienced long-term damage from external environments such as wind,sun,lightning,rain and snow,pollution and subsidence.As one of the core components of overhead lines,insulators are related to the reliable operation of transmission lines.Therefore,the detection of self-explosion defects of insulators is particularly important.Aiming at the problems of long operation cycle,low efficiency and low security of existing insulator defect detection methods,this paper combines new artificial intelligence technology to design an insulator defect recognition system based on convolutional neural network,and realizes fast and accurate insulator defect detection function,has certain application value.The main work of the paper can be divided into the following three points:(1)Construction of insulator defect dataset and determination of benchmark network.To address the problems of complex background and tiny targets in the constructed aerial insulator defect detection dataset,the dataset is first manually labeled,then the dataset is analyzed with the labeling results,followed by offline image enhancement to process the images in the dataset and their corresponding labels to obtain a larger number and higher quality insulator dataset,which provides the data basis for insulator defect detection in this paper Finally,experiments are conducted on the constructed dataset,and the benchmark network YOLOv5 is obtained by comparing and analyzing the experimental results.(2)An improved YOLOv5 insulator defect detection algorithm is proposed for the insulator defect detection task.First,the bipartite K-means clustering algorithm is used to obtain an anchor box set that is more suitable for the constructed dataset,and then the sandglass block and the Convolutional Block Attention Module(CBAM)are introduced to optimize the YOLOv5 network.The experimental results show that the improved algorithm model can accurately and quickly detect insulator defects.(3)Realize the reasoning and deployment of the algorithm model on edge devices and build an insulator defect identification system based on QT(application development framework).In terms of model conversion,the weight file is first converted into an ONNX package,and then the ONNX package is converted into a Tensor RT engine with inference acceleration function suitable for the chip framework;in the construction of the insulator defect recognition system,the user interface is built and the system is integrated.The function of the insulator is formulated and improved,so that the preliminary work is integrated into the insulator defect identification system.The test results show that the transformed model can achieve fast and accurate insulator defect detection function with96.33% recognition accuracy when used in conjunction with the built system.Figure [42] Table [13] Reference [45]...