Research On The Algorithm Of Substation Power Meter Detection And Reading Recognition

Gauge detection and gauge automatic reading are the main tasks of substation robots when conducting inspections in substations.However,due to the complex outdoor environment,the outdoor Gauge detection and gauge automatic reading algorithm are faced with the problems of high false detection rate and the algorithm can’t meet actual needs.The thesis conducts related research on the outdoor condition table measurement and gauge reading recognition algorithm and propose solution.Solutions are proposed for the gauge detection task:(1)A gauge detection algorithm combining the positioning information of the inspection robot is proposed.The algorithm uses cascaded Adaboost classifier,and uses Merlin Fourier transform,phase correlation and other algorithms to make full use of the robot positioning error of about 5cm to screen the results of the classifier.The algorithm has a correct rate of 99.3% with a recall rate of 97.7%,and the algorithm execution speed reaches 5frames/second.The detection accuracy and real-time performance meet the requirements of the substation inspection robot.(2)Using the object detection algorithm Faster Rcnn,the gauge detection of the deep learning method is completed.Firstly,this thesis uses migration learning technology to solve the problem that the collected data does not meet the amount of data needed for deep learning.Secondly,it is trained on different feature networks and compares the results.After the training is completed,the model can simultaneously detect a variety of meters and identify their types.The final gauge detection accuracy can reach more than 95%,and the detection time is controlled below 0.2s.Solutions for gauge identification tasks:(1)For various types of substation gauges such as pointer gauge s,digital gauges,oil level gauge s,etc.,traditional image processing algorithms are used to identify them.(2)A key point detection of pointer gauge algorithm based on convolutional neural network is proposed,and the algorithm is used to complete the automatic reading task of the pointer gauge.This thesis defines seven key points of the pointer gauge.Using the deep learning regression algorithm to transform the pointer detection problem into the gauge key point detection problem.After detecting the key points of the gauge,the direction vector of the pointer can be calculated and the gauge’s number can be recognized.In order to solve the problem that the collected data does not meet the amount of data required for deep learning,this thesis uses the migration learning technology to pre-train the model,and then continues to train the model with the produced gauge key point data set.The final test results show that the correct rate of the key point detection model of the pointer gauge is 95%;the correct reading rate of the gauge is 92%;the detection speed is about 12frames/s,which meets the real-time and accuracy requirements of the inspection robot.