Meter Reading System Based On Computer Vision And Deep Learning

The instrumentation has a very broad application scenario,such as in the field of production,transportation,industrial production,medical equipment,agricultural development,etc.,the influence is deep.As for the data collection,data monitoring records,there is a very important application tool in various scenarios,which is very important for conventional,convenient,safe production and life.Automatic reading can not only reduce human cost,reduce the difficulty of reading tables,dangerous,and improve reading,and increase reading accuracy,and has great practical value.Therefore,it is very practical that the instrument reading system based on computer vision and depth learning.This article aims to design a high accuracy and robust instrument automatic reading system,the main work content includes:1.Based on the current research results YOLOV5 as the target detection model,the self-made electric meter data set is used for training detection.Small models,fast speed,high accuracy recall rate,suitable for edge devices and achieve optimal speed and accuracy combination.2,Combined with the advantages of the traditional computer vision algorithm and the depth convolutional neural network,a series of pretreatment is performed,and then input into the neural network,and the degree of detection accuracy is improved.3.Build high quality meter data sets and printers digital data sets.In response to the current lack of identification,the annotation meter data set and the printed digital data set,the author of this article constructs a robust meter and Printing digital data set for target detection recognition.4.The way the scale and the digital combination is proposed,and the phenomenon that can only read the digital or only read a specific instrument scale,which greatly improves the accuracy and robustness of the instrument reading,and has a very high actual application value.