Research On Insulator Missing Detection In Aerial Photography Based On YOLOv3

In the actual production process,due to the influence of natural environment and objective factors,the phenomenon of insulator self explosion often occurs,so it needs planned inspection to ensure the safety of power system.With the deepening of power grid intelligence,UAV is widely used in power system,especially in patrol inspection,which means a lot of manpower and time investment,and the working experience of screening personnel is likely to affect the final results.Therefore,a fast and accurate image recognition method is needed to achieve the intelligent identification and defect detection of inspection images,which is of great significance to ensure the reliability of power grid.In this study,template matching method based on image feature recognition and learning model recognition method based on yolov3 are used to detect insulator missing.The following is the main work and experimental results of this study:(1)According to the actual inspection line of Liaoyang power supply company,UAV inspection is carried out.A total of 1051 insulator images in natural environment were obtained as sample set,and the insulator target position was manually marked on the original size image by using image marking software labelimg.Through the random combination of mirror,scale,rotation,clipping and translation,the samples were increased to 10962.(2)This paper analyzes the principle and characteristics of artificial neural network,introduces the working principle of convolution neural network and YOLOv3 network,sets the boundary frame confidence of YOLOv3 network according to the characteristic information of insulator image,introduces the three-layer network structure of YOLOv3 in detail,and sets the parameters suitable for insulator image training in this study.(3)By using the learning model recognition method based on YOLOv3,the problems of weak robustness and generalization ability of traditional detection algorithm are solved.Firstly,the structure of YOLOv3 network is analyzed in detail,and the K-means clustering algorithm is used to solve the problem that YOLO network is not sensitive to small targets.Secondly,in view of the sample imbalance problem in YOLOv3,the original loss function is improved by focal loss function.Finally,the Mish activation function is used to replace the Leaky Re LU activation function to further improve the accuracy and stability of the network.(4)By comparing the test results of the improved YOLOv3 algorithm,it can be seen that the accuracy rate of the improved algorithm is increased by 3.4%,the average precision average(map)is increased by 3.7%,and the detection speed is reduced by 1 frame / s.It isproved that the improved algorithm can achieve the purpose of intelligent and rapid identification of insulator loss.