Machine Learning Based Remote Metering System

The process of metrology involves the calibration and traceability of measuring instruments.High-precision and efficient metrology ensures the validity of data in industrial and scientific research,thereby accelerating industrial production efficiency.However,the current centralized metrology method for inspected measuring instruments in laboratories located in different places is inefficient and cannot meet the requirements for metrology efficiency in the process of modernization.Therefore,there is an urgent need for a visualized and automated high-efficiency metrology method.This article focuses on the remote metrology platform for the retroreflective measurement equipment of traffic signs,and proposes an intelligent integrated remote metrology system that can automatically identify instrument readings and upload them to the metrology cloud platform,which has significant practical significance for improving the metrology method and efficiency.The specific research contents of this article are as follows:(1)A remote retroreflective metrology framework based on the Internet was constructed,and a metrology visual image acquisition and transmission module was designed.Starting from the retroreflective metrology experimental requirements,the functional and non-functional requirements of the remote metrology system were analyzed,and a networked remote metrology system framework was established,mainly including: a hardware physical layer based on image acquisition module,a network transmission layer based on HTTP protocol,a cloud-based database layer,and a system function layer for data visualization;the hardware and software parts of the visual image acquisition unit were designed to enable the reading system to reliably obtain the video image of the remote metrology dial.(2)A DeepLabV3+ based algorithm for pointer gauge reading recognition is proposed.The algorithm mainly completes the gauge segmentation and pointer segmentation tasks.In order to improve the recognition accuracy of pointer gauges,YOLOv8 is used to detect dials and DeepLabV3+ segmentation network to extract dials and hands and other related information,and data sets are built to train and test them.The experimental results show that the relative error of recognition results of the algorithm in this paper is less than 0.35%.(3)A traffic sign retro-reflective digital meter reading recognition algorithm based on deconvolution network is proposed.The algorithm uses DB-Net and ABI-Net models to locate and recognize digital meter characters in view of the limitations of traditional meter readings.It solves the problem that the traditional digital meter reading recognition process is easily affected by the character area positioning and character segmentation effect,and effectively improves the accuracy and reliability of digital meter recognition.(4)A cloud-based networked remote metering system for traffic retroreflectivity coefficient measurement is constructed based on the above.Software programs for the streaming acquisition software unit,the reading recognition module and the database management module were designed,while the system interface was built,and the optimisation of the data tables in the Mysql database was completed and deployed to the cloud server.Remote metrology experiments were carried out in an in-situ remote laboratory,and the results showed that the system was able to achieve remote metrology of traffic retroreflectometers,and the recognition accuracy of the reading system was as high as 99%,meeting the demand for remote in-situ efficient metrology of retroreflectometers in the traffic field.