Design And Implementation Of Remote Automatic Gas Meter Reading System Based On Image Recognition And LoRa

With the accelerated economic growth and the implementation of "One House One Meter" policy,the number of users of civil gas continues to increase.Due to historic reasons,half of them still use mechanical meter which needs manual meter reading and charging.This kind of method not only disturbs the users,but also wastes manpower and material resources.Although the remote automatic meter has been used with a certain area,it is nearly impossible to replace all the mechanical meter into them at one stage because of the cost and the complication of the rebuild.Therefore,it is urgently needed that a set of solutions which can change the traditional meter reading and payment method in a low-cost,quick and convenient manner,to allow the meter users to enjoy the functions of the remote automatic meter.A lot of researchers propose a remote automatic meter reading system based on image recognition.Not only most of their communication are based on the AD Hoc network or cellular network,but also the research of them are focused on the image recognition algorithm.It's worthless if the system structure,the mount of nodes,and the actual work environment or anything like that are not taken into consideration.Under the premise of non-changing the structure of the original gas meter,the main work of this thesis uses camera and IoT to design and implement a remote automatic gas meter reading system based on image recognition and LoRa.The gas dial image is obtained by a camera of the node and transformed into a string after local identification,which contains the result of the meter readings.After that,the string is packed and sent to a gateway through LoRa and forwarded to the server to store and managed centrally by the gas company.The main work of this thesis includes the following aspects:(1)This thesis designs a gas meter retrofit node structure which can work on majority kinds of gas meter,and possess easy-mounted and universality.(2)This thesis built a hardware platform and the embedded software of image recognition and LoRa communication with an image processing module and a LoRa wireless module whose cores are Raspberry Pi 0 and SX1276.(3)The processing is composed of four steps,which are preprocessing,number area extraction and localization,number segmentation and number recognition.The localization algorithm of the display of color features shows that the color binarization of the traditional look-up table cannot be used for color shifting images.Also,the convolutional neural network based on the Inception-VI architecture is used for character recognition to deal with the problem that the machine vision algorithm rely too much on the quality of the image.(4)Finally,the system is tested and verified.The result shows that the system is stable,the mount of the node is easy,and algorithm is robust.