Research And Application Of Data Recognition System For Digital Display Screen Of Metering Equipment Based On OCR Technology

With the rapid development of science and technology,metering equipment has been widely used in industry,scientific research,medical care,environmental protection and other fields,and each field has put forward higher requirements for measurement quality and efficiency.However,most of the traditional measurement devices lack data transmission interfaces or data transmission protocols are not unified,resulting in the measurement data can only rely on manual reading,which has low efficiency,large errors,high costs and other problems to meet the requirements of the rapid development of the measurement industry.Based on the above reasons,this paper applies Optical Character Recognition(OCR)technology to the field of metrology,and designs a data recognition system based on OCR technology to automatically detect and recognize data on digital display screens,taking the metrology equipment that reads data through digital display screens as the research object.The main research contents are as follows:(1)Design the overall architecture of the digital display screen data recognition system according to the actual measurement standards of the measurement equipment,analyze and select the image acquisition module of the system according to the requirements of the measurement scenario,and design the research plan of the algorithm module and the software module.(2)Research on two-stage OCR algorithm based on deep learning:In the detection phase,the DBNet algorithm is applied to the self-made metering equipment digital display screen detection dataset for training and testing.The results show that the algorithm has poor detection accuracy for unevenly distributed data.To address this problem,an improved DBNet algorithm is proposed to incorporate PAN structure to improve the localization effect of character edge pixels using accurate position information;to replace FPN with AF-FPN to reduce the information loss caused by channel changes and enhance the feature expression of the network;to introduce a focal loss function in the calculation of probability map loss to balance the category weights and difficulty weights of different samples to improve the detection effect of difficult samples.The experimental results show that the improved DBNet,the detection accuracy can reach 95.2%,which can meet the detection accuracy requirements.In the recognition phase,the CRNN algorithm is applied to the recognition dataset of homemade metering equipment digital display screen for training and testing.The results show that the algorithm is less effective in recognizing data in complex environments,with an accuracy of only 88.1%,which cannot meet the recognition requirements.To address this problem,an A-CTC scheme is proposed in the transcription layer of CRNN,which uses a mixture of Attention and CTC to decode feature sequences,learn better feature alignment and representation,and improve the generalization ability of the model.Experimental results show that the accuracy of the improved algorithm can reach 96.4%,which meets the accuracy requirements of practical application scenarios.(3)The trained models are deployed in tandem to the server to generate the calling interface.A multi-threaded approach is used to design the human-computer interaction interface,which can display the recognition results and view the measurement dynamics in real time.Through testing,the practicality and effectiveness of the system designed in this paper are verified.