Research On Recognition Method Of Transformer Infrared Image

As the overall scale of the power grid continues to expand,the number of transformers is also increasing year by year.As an important part of the power grid,the operating status of the transformer greatly affects the reliability of the power supply and the stability of the system,so it is necessary to monitor the health of the transformer.Because the infrared thermal imager,as a temperature measuring device,can effectively present the temperature status of the device,and then reflect its health status,it has been widely used in power equipment inspections.The current inspection method mainly relies on manual analysis of the collected infrared images,but when faced with massive amounts of data,it is difficult to complete it by manual alone.So this article puts forward the image recognition method based on transformer infrared image.In order to achieve good image recognition effect,this paper deals with the infrared image of the transformer from two aspects: image denoising and enhancement,aiming at the characteristics of low signal-to-noise ratio,low resolution and poor contrast of infrared images.1)In order to take into account both image denoising and texture details,a denoising algorithm based on Nonsubsampled Shearlet Transform(NSST)is proposed.First,NSST is used to decompose the image into different frequency bands,then the noise in the high frequency subbands is filtered by local entropy threshold method,and finally all frequency bands are inversely transformed to obtain a new image.2)An image enhancement algorithm based on Contrast Limited Adaptive Histogram Equalization(CLAHE)and guided filtering is proposed.The basic layer of the image is separated through guided filtering,and then the gray intensity is enhanced by improved CLAHE algorithm.Finally,experiments are used to prove the effectiveness of the denoising and enhancement algorithms in this chapter.Power equipment often presents the characteristics of visual saliency in infrared images,so the saliency detection via graph-based manifold ranking is used to detect the target of transformer infrared images,and the algorithm is improved for the characteristics of low segmentation accuracy and great influence by the image boundary background.First,the superpixel segmentation method of simple linear iterative clustering is improved by gray scale variance analysis and the Otsu to achieve fine segmentation.Then,the boundary contrast method is proposed to divide the boundary background into "real background" and "false background",and t he saliency mapping matrix is improved.Through comparative experiments,it is demonstrated that the improved algorithm can finely segment power equipment and achieve target detection under the condition of accurately distinguishing the background.The deep convolutional neural network has the characteristics of good robustness,strong anti-interference ability and high recognition accuracy in target recognition,so it is very suitable for the recognition of infrared images of power equipment.This article chooses the lightweight algorithm Tiny-YOLOv3 to improve on the basis of comprehensive consideration of model size,recognition speed and recognition accuracy,mainly in three aspects: 1)Add convolutional block attention modules to the network feature extraction layer;2)Add 52×52 network detection layer to reduce the missed detection rate of small targets;3)Replace the original bounding box loss with the complete intersection over union.Finally,through comparative experiments,the model in this paper has high recognition speed and accuracy while maintaining lightweight characteristics,and can basically meet the requirements of transformer inspection based on infrared images.This article provides a complete solution for the recognition of transformer infrared images,and lays a foundation for further realization of unmanned monitoring and intelligent inspection of power equipment.