Research On SF₆ Leakage Detection Based On Deep Learning

Sulfur hexafluoride(SF6)gas has very high stability,but also has excellent negative electricity and high thermal conductivity,so it is widely used in the power and chemical industry.As an industrial gas,its leakage problem cannot be ignored.Due to the colorless,odorless and dense nature of SF6 gas,its leakage is not easy to detect on the one hand;on the other hand,SF6 leakage will quickly gather in low-lying areas and cause harm to the human body.In severe cases,suffocation symptoms may occur.Traditional detection methods have many problems in real-time,detection efficiency,detection range,detection accuracy and so on.The application of infrared imaging technology visualizes SF6 leak detection.Therefore,it is widely used in SF6 detection in the electric power and chemical industries.However,there are still problems such as time-consuming and labor-intensive,remote real-time detection.In response to these problems,this paper uses machine learning target detection algorithms to achieve more efficient SF6 leak detection based on infrared imaging technology.The main research work is as follows:1.In view of the problem that there is no data set for SF6 gas infrared images,more than 30,000 infrared SF6 leakage image data sets have been established.Through normalization,data enhancement,etc.,the data set is processed,and finally two types of data sets of leaked images and nonleaked images are obtained.2.The SF6 gas under infrared conditions has rich characteristics.In order to further improve the detection rate and reduce the suspected leakage area,the detection effect of a variety of machine learning algorithms for moving target detection on infrared SF6 leakage videos is studied.At the same time,we also try to extract various features of SF6 under infrared conditions,and effectively extract the suspected leakage area of SF6 gas.3.Aiming at the problems of traditional convolutional neural network in infrared SF6 leakage detection:(1)SF6 gas under infrared conditions has rich characteristics,and the leakage area is small,so the direct use of the network detection effect is not good.(2)The spatial information of the pooling layer in the convolutional neural network is lost,which affects the classification effect.In this paper,the convolutional neural network is optimized,the expansion convolution mechanism is added,and the machine learning combination algorithm is used to detect SF6 gas under infrared conditions.The expanded convolution mechanism can effectively reduce the loss of spatial information caused by the pooling layer,thereby obtaining a better classification effect.This paper verifies through experiments that this method can more accurately detect SF6 leakage under infrared conditions,and can accurately mark the SF6 leakage area by detecting SF6 leakage videos of different equipment and different parts.The detection accuracy rate can reach82.71%,and the environment changes.The noise interference brought by has a certain degree of adaptability.