Detection Of Power Equipment And Its Temperature Defect Based On Infrared Image

In recent years,with the development of society,the scale of the power grid is increasing rapidly,and the detection of substation’s power equipment is becoming more and more complicated.The traditional defect detection method of power equipment based on infrared temperature imaging adopts template matching method,which requires fixed distance,focal length and Angle.Therefore,when the substation scene changes,these conditions need to be fixed again.It is difficult to detect power equipment accurately only by relying on this algorithm.So it is urgent to study a new and more reliable power equipment detection method.A method of power equipment detection and infrared temperature defect detection based on infrared image combining deep learning and component location is givend.Firstly,an improved deep learning target detection algorithm model DEDID is given.The accuracy of target detection algorithm for power equipment identification is improved by improving the determination method of Gaussian kernel radius of model.The circular Gaussian convolution kernel is changed into an elliptical Gaussian convolution kernel to distinguish width and height,and the shape of the thermal map can change with the different shape of the infrared image of the power equipment to be detected.Aiming at the problem that the model is not ideal in the detection speed,a method of using Gaussian kernel to generate high quality training samples is proposed.The effect of Batch＿size is increased by generating more monitoring signals for model regression.At the same time,the loss function is improved to improve the detection speed of the model.Then,based on the target detection results,a structured positioning algorithm and a template matching legal bit algorithm are designed and analyzed to realize the positioning and detection of different structural parts of power equipment from complex infrared images.The surface temperature of power equipment is extracted and calculated by threshold segmentation method.To design and develop the judgment criterion for temperature defects of power equipment,according to which the relevant defects of power equipment can be judged.In this paper,10 types of infrared image data sets of power equipment are used,with a total of 3140 images.Experiments were carried out on DLA-34,Res Net-101 and Res Net-18 basic networks.The experimental results show that the detection accuracy of current transformers,circuit breakers,voltage transformers and lightning arrester reaches more than 90.3%,and the average accuracy of all the 10 types of power equipment reached more than 88.8%.The results show that the research results can meet the high requirements of infrared detection of power equipment in the actual substation scene.It has a wide application prospect.