Machine Vision-based Intelligent Algorithm For Meter Reading And Its Applications

As one of the crucial links in the power network,substation plays a vital role in the reliable operation of power system.Workers can keep track of real-time monitoring information about the meter reading in substations so as to find anything unusual timely.Owing to lack of communication interface for most traditional meters in substations,the meter reading needs to be acquired manually.However,manual reading is not only costly and inefficient,but also has a high risk of misjudging.Therefore,it is necessary to bring forward a set of intelligent recognition algorithms to meet the urgent reliability requirement of substations.This thesis systematically studies the methods of meter image preprocessing,meter dial area extraction,dial tilt correction,and meter reading recognition for two common meters including pointer type and digital-display type.A machine vision-based intelligent algorithm for meter reading in substations is put forward in view of the current insufficiency of most method,such as the impossibilities of recognizing the meter image at night and extracting the dial position due to the surface reflection.The major contributions of thesis are as follows:(1)Studying an image preprocessing method for low light images.On the basis of a bilateral filtering based denoising algorithm for meter image and the Retinex algorithm for enhancing the measurable parts of image,the dark-light image preprocessing is implemented for the meter images taken at night by means of full convolutional network,which can effectively eliminate the interference of the low light environment on the meter recognition.(2)Studying a meter dial area extracting method from complex background.The traditional object detection method and the deep learning-based object detection method are investigated and analyzed.The conclusion is that the traditional target detection algorithm cannot solve the problem of reflective edge of the instrument dial.Therefore,this thesis proposes a method of applying the YOLO algorithm to the detection of instrument dial.Not only can the dial panel be extracted accurately,but the detection result is not affected by the reflective factor.Finally,the obtained dial images processed by slant correction using the perspective transformation algorithm to eliminate dial deformation.(3)Studying an intelligent recognition method for pointer meter.Firstly,the connected component is labeled and filtered in binary images of meter so as to extract the pointer and scale area of the meter.The least squares method is used to accurately fit the line where the pointer is located.Finally,the meter reading is estimated based on the angle method.The accuracy of the reading algorithm is verified by the recognition experiment of pointer meter.(4)Proposing an intelligent recognition method for digital meter.Combining the analysis of the existing digital character recognition methods with the characteristics of substation digital-display meters,this thesis proposes an improved BP neural network algorithm.On the premise of guaranteeing the accuracy,the algorithm can speed up the recognition of digital characters and reduce the total runtime.Finally,the effectiveness of the algorithm is verified by the recognition experiment of digital-display meter.