Temperature Anomaly Detection System Of Power Equipment Based On Deep Learning

With the development of computer vision and the proposed goal of smart grid,the integration of deep learning technology and traditional industrial scenes has become an important direction.The temperature monitoring during the operation of power equipment is the top priority to ensure the safety of equipment.The traditional manual monitoring method is inefficient and has potential security risks.This paper develops a set of electric equipment temperature anomaly detection system based on deep learning through infrared thermal imaging,Object detection,image classification,image segmentation and other technologies.It realizes the temperature monitoring of electric equipment and alarm for abnormal high temperature,which improves the efficiency of temperature monitoring.The main contents of this article are as follows:(1)Analyze the infrared thermal image information in the electric power scene,and successfully obtain the value,symbol,Chinese character and other information in the image by combining Object detection and image classification,so as to obtain the highest and lowest temperature values in the image.Use the Yolov5 Object detection model to detect all numerical regions,symbol regions,palette regions,and other text information regions in the current scene image;The Resnet50 classification model is used to obtain the corresponding number of each numerical region.Finally,the symbol information and digital information are spliced in order from small to large according to the abscissa of the center point of the detection region to obtain the highest and lowest temperature values in the current monitoring region.The experimental results show that the accuracy of the temperature detection in the image is significantly improved by using the method in this paper compared with the detection method alone.(2)The static mask mechanism and the improved mask R-CNN instance segmentation method are used to complete the acquisition of single device information in the infrared thermal imaging image,and the comparison is made.Finally,the improved mask R-CNN instance segmentation model is used.In the data preprocessing stage,the model adds the directional gradient histogram feature(Hog feature)and edge feature of the image.Adding Hog feature can effectively reduce the impact of illumination on the image,including uneven illumination and local shadow.Adding edge feature can strengthen the edge information of the power equipment in the infrared thermal imaging image;Non-local attention mechanism is added in the feature extraction process to obtain global information in the feature extraction process,so as to more accurately segment power equipment and achieve better instance segmentation effect.(3)Complete the temperature identification of each electric equipment in the infrared thermal imaging map,and accurately calculate the actual temperature of the electric equipment.According to the palette information and the highest and lowest temperature information in the infrared thermal imaging image,the temperature of each pixel value is calculated using the European distance matching method,and the temperature fitting of electric equipment is realized.(4)Completed the development of the abnormal temperature detection system for power equipment,including the front-end monitoring interface and the back-end computing platform.The Front-end monitoring interface includes:equipment temperature monitoring curve,high temperature abnormal record,high temperature pop-up alarm;Back-end computing platform,responsible for controlling the operation of camera,temperature acquisition,power equipment segmentation,and power equipment temperature calculation.