Research On The Algorithm Of Inspection For Smart Meter Display Defects Based On Deep Learning

With the rapid development of the electric power industry,the State Grid has been greatly promoted.As an important part of the “smart grid”,Smart meters have been promoted countrywide for 11 years.China is accelerating the replacement of expired electric meters.The expired meters must be inspected with the stand ard of the State Grid Metrology Centre.According to the display defects existing in the expired smart meter,the design scheme of the algorithm of inspection for expired smart meter display defects is developed by means of deep learning technology to ensu re the quality of the smart meter,improve the detection efficiency.The main research contents of this paper include:(1)Analysis of inspection requirements,defect characteristics and related background of smart meter;understanding the current research and application of deep learning technology in text detection and recognition and understanding the current research status of inspection for smart meter display defects by consulting relevant literature.(2)Research and comparative analysis of defect lo cation algorithm for electric meter display.Firstly,the traditional methods of locating defects of smart meter display based on template matching and projection are introduced,due to the traditional algorithms have deficiencies of poor computional efficiency and are susceptible to interference,a method of locating for smart meter display defects based on R-FCN(region-based Fully Convolutional Networks)is proposed,and design the corresponding comparative experiment to verify the feasibility of the met hod.The method proposed can achieve99.38% accuracy and 97.25% recall rate without relying on the template image.(3)Research and comparative analysis of defect recognition of smart meter display.Firstly,introducing the traditional method of defects recognition of smart meter display based square deviation matching and Knn(K-Nearest Neighbor),to solve the problem that the tradi tional methods are susceptible to image noise and time-consuming,a method of recognition for smart meter display defects based on CNN is proposed,and a comparison experiment is carried out on a test set composed of character images of different types of smart meters.The method proposed can obtain the 92.53%recognition rate.(4)Design the experimental apparatus of inspection f or smart meter display defects and carry out the comparison experiment of the algorithm.According to the test standard,built the experiment platform,selected industrial camera and lens;the accuracy of the algorithm is validated by multi-type sampling tests,the algorithm can achieve 92.85% accuracy and the average processing time of single image is 903 ms.At the same time,the causes of misjudgment are analyzed to improve the accuracy of the algorithm.