Power Meter Visual Proofreading System Based On Deep Learning

The accuracy of electric measuring instruments is very important for the measurement,and it needs to be checked regularly to ensure its accuracy.However,not all traditional digital display instruments have a communication interface,and they cannot directly obtain the values of the standard measuring instrument and the measured instrument.In this paper,the images of their rapid boost and buck process are collected by dual cameras,and the instrument readings are automatically recognized.Finally,compare the results of the standard table and the measured table to determine the accuracy of the measured table.Based on the above problems,a visual proof system for electric instrumentation based on deep learning is designed and implemented.The system mainly includes two parts: terminal and remote center.Among them,the terminal mainly takes the Raspberry Pi controller as the core,and is responsible for the dual-meter display data acquisition and image compression processing;the function of the remote center is mainly responsible for the image recognition of the instrument,and the recognition results are stored in the database to achieve data storage and supervision.The main work of this article includes the following aspects:The core of the terminal adopts the Raspberry Pi 3b as the controller,which is more stable and powerful than the self-built embedded hardware equipment.The digital camera is acquired by the dedicated camera module acquisition instrument that comes with the Raspberry Pi,and the onboard wifi module is passed through the Raspberry Pi communicate with the remote center.In the process of image transmission,the network bandwidth and data size will affect the realtime performance of the meter calibration.Therefore,in the terminal program design,the H.264 video encoding algorithm is used to compress and encode the transmitted image,and the performance of the H.264 video encoder is verified through experimental comparison.By comparing and analyzing the network structure and network performance of commonly used convolutional neural networks such as ZFNet,AlexNet,VGG-16,and combining the experimental results on the Image Net data set,the feature extraction layer is designed using VGG-16,target detection algorithm choose Faster-Rcnn's plan.Increase the sample size by scaling the data set,use the Labelimg labeling tool to label,train the target detection model under the built system platform,and optimize the model by means of fine-tuning model and parameter adjustment.The experimental results prove the effectiveness of the model,which can accurately locate the position of the frame and identify the correct result.The system designed in this paper shows through experiments and tests that the system can complete the functions required by the project and operate stably.The image in the transmission module can be sent to the remote center stably and correctly;in the remote center image recognition module,the program can correctly identify the instrument characters,and the recognition time is only 0.3s,which meets the real-time recognition requirements of the project.