The Research Of Transformer Image Processing Method Based On Infrared Image

Transformer is the important equipment in power distribution facilities.Prevention,detection,and promptly elimination of equipment failure are the necessary conditions to the maintain of power grid's security operations.Correct diagnosis and the repair of equipment failure have their practical significance.The traditional transformer fault detection method has some shortcomings,such as the problem of not being able to run the charge and the low safety,so it is difficult to meet the rapid growth of detection needs.In recent years,infrared imaging is maturing,equipment manufacturing costs are reduced,so the technology gradually is applied to the equipment fault detection.Transformer detection method based on infrared image has achieved a certain effect in the practical application,but there are still some shortcomings,such as the low resolution pictures,the difficulties of separating the equipment from the image effectively,the high professional requirements and the low degree of automation intelligence.In this paper,based on the current test method we further study the test methods and put forward some preprocessing methods of image before the diagnosis of the equipment.Hoping to get better diagnosis effect and to improve the degree of automation intelligent diagnosis.What's more,we design verification experiments to verify the correctness of the approach.In this paper,the processing method is mainly divided into three parts.The first part is about stitching the local image to obtain image of the complete equipment with high resolution.In the second part,we study the automatic identification of the device type on the infrared picture being captured.And the last part is about automatically separate the Images which have been identified for the transformer in the equipment.In the infrared image mosaic of transformer,this paper studies the commonly used image mosaic algorithm,and the detection results of common feature detection methods in transformer infrared images are compared,This paper presents a new algorithm to remove false matching based on SIFT feature matching and spatial constraint,The algorithm is used to splice the transformer infrared image and get better stitching effect,so as to get a larger view of the equipment infrared image.In the automatic identification of equipment type,we introduce the traditional classification method,which is to segment the image firstly and then extract the feature,try to use depth learning method to automatically extract feature recognition equipment,I analyze the principle of traditional neural network,convolution neural network and the advantage of the latter in dealing with large data,The method of convolution neural network is used to classify transformer and insulator,and experimental results show,the method of convolution neural network has a good classification accuracy in equipment type recognition.In the aspect of automatic segmentation of transformer image,this paper studies the commonly used methods of image segmentation and the segmentation results,which show that it is difficult to separate transformer equipment from complex background by a single method.Then,an integrated method is proposed to realize the accurate segmentation of the transformer in the infrared image,the method combines the MSER region,the gray level distribution characteristics,Hough detection and mathematical morphology processing method,and the experimental results show that the method has a better segmentation accuracy.