Infrared Image Recognition And Thermal Fault Diagnosis Of Electrical Equipment Based On Deep Learning

Most of the faults of electrical equipment in the early stage will be accompanied by abnormal heating phenomenon,and infrared temperature measurement technology as a non-power outage equipment temperature detection method,has been widely used in the inspection of electrical equipment.At present,the analysis of infrared images of electrical equipment mainly relies on manual work,which is inefficient and difficult to deal with the massive infrared images generated in the operation of substation.Therefore,this paper proposes an automatic analysis meth od for infrared images of electrical equipment to realize intelligent identification and fault diagnosis of electrical equipment in infrared images.The infrared image of electrical equipment is often of low quality.This paper preprocesses the infrared image from two aspects of denoising and enhancement.1)to reduce the noise of infrared image,this paper puts forward a kind of denoising algorithm based on improved Dn CNN neural network,introduced in the original Dn CNN network capability of noise points fitting for chaos of deformable convolution module,training network learning noise of infrared image information,minus the noise component with noise of infrared image is used to achieve the goal of image denoising.2)In view of the problems of poor contrast and low imaging quality of infrared images,this paper proposes an image enhancement method based on the combination of weighted guided filtering layering and improved Single-Scale Retinex(SSR).The weighted guided filtering is used to separate the base layer of the image,and the gray enhancement is carried out through the improved SSR algorithm.Finally,the effectiveness of denoising and enhancement algorithm is proved by experiments.Deep neural network has the advantages of high recognition a ccuracy and strong robustness in image target detection,which is suitable for recognition of electrical equipment in complex background.In this paper,aiming at the characteristics of the electrical equipment in infrared image,is proposed based on impr oved Retina Net electrical equipment of the infrared target fine detection model,the main improvement has three aspects: 1)the original Retina Net network of horizontal rectangle instead of rectangle with rotation Angle,arrangement of dense,fine position ing identification with Angle of electrical equipment.2)CBAM attention mechanism module is introduced into the original network to reduce the interference of irrelevant background in the infrared image.3)The PAN network module is introduced into the original network to further carry out feature fusion for feature maps of different scales to improve detection accuracy.Finally,comparative experiments are designed to prove the advantages of this algorithm in recognition accuracy and detection refinement.On the basis of realizing the identification of electrical equipment,this paper analyzes and verifies the thermal fault diagnosis of transformer bushing,isolating switch and voltage transformer with the combination of similar equipment comparison method and Deeplab V3+ semantic segmentation model.First of all,we use modified Retina Net to recognize the same type of equipment,then we use Deeplab V3+ semantic segmentation model to separate three types of key weak structures which are prone to thermal failure,and use temperature difference information of the same type of equipment to realize thermal fault diagnosis of key parts of electrical equipment.