Recognition And Research On Instrument Data Of Gas Gathering Station Based On Machine Vision

With the rapid development of the natural gas industry,as a key transfer station in natural gas extraction and transportation operations,the normal operation of the gas gathering station is particularly important.A large number of meters in the gas gathering station constantly monitor and display the operation of the gas gathering station.Manually counting the instrument data is particularly cumbersome,time-consuming,labor-intensive,and inefficient.At the same time,the location of the gas gathering station is remote and the environment is harsh.Natural gas itself is a dangerous gas,and manual inspection has safety risks.Therefore,the research of inspection robots for the gas gathering station is imperative.This paper takes the inspection robot of gas gathering station as the carrier,and the pointer instruments and digital meters in the gas gathering station as the research objects,to study the meter reading algorithm based on machine vision that can be used in the gas gathering station environment.The main research contents of this paper are as follows:(1)Research on the image enhancement algorithm of dark light instrument that can be used in the inspection robot of gas gathering station.The gas-gathering station is located in an open and open environment,and the imaging light of the robot is insufficient during morning and evening inspections,and it is easily affected by the weather conditions and the low-light photos taken cannot be recognized.To solve this problem,this paper studies the dark-light image enhancement algorithm based on Retinex theory.Two algorithms of MSRCR and MSRCP are implemented respectively,and the instrument image enhancement effect and algorithm efficiency of the two algorithms under different dimming degrees are compared.Finally,the Retinex Net convolutional neural network based on Retinex theory is introduced to collect and produce a gas gathering station dark-light instrument image comparison data set.After training the network,compare it with MSRCR and MSRCP.The results show that Retinex Net has high color reproduction and good detail retention.Which greatly increases the speed of enhanced image,is more suitable for the enhancement of dark-light instrument images in the actual gas gathering station environment,and lays the foundation for the next step of instrument detection and reading.(2)Aiming at the problem that the traditional SIFT feature point matching algorithm cannot accurately locate the dial under the complex background environment of the gas gathering station,this paper studies the dashboard detection algorithm based on the improved YOLOv4 neural network.The gas gathering station dashboard data set was produced and trained,and the prior frame was adjusted by introducing the K-means++ algorithm to improve the recognition accuracy and speed.The experimental results show that the target detection algorithm based on the improved YOLOv4 tilts the shooting angle when affected by the ambient light It can quickly detect the location of the dashboard.At the same time,some meters in the gas gathering station have special positions,and the photos taken by the inspection robot have the angle of the dial.To solve this problem,a set of meter image tilt correction algorithm flow is given.Firstly,the instrument feature points are extracted through the ORB algorithm,and the violent matching algorithm is used to match the feature points on the template map and the tilt map.Finally,the RANSAC algorithm is used to filter the feature points and then the perspective transformation algorithm is used to correct the tilt.(3)A set of reading algorithm flow of pointer meter suitable for gas gathering station environment is given.By comparing the denoising effects of different filter algorithms on the dial image,it is verified that the bilateral filter algorithm can remove noise while retaining the clarity of the dial scale area;after the dial image is binarized,the Hough transform is used to initially detect the dial outline and remove the circle External noise interference;then conduct connected domain detection,identify the position according to the characteristics of the scale area and the pointer,refine the pointer area and then detect the pointer direction;finally calculate the display value of the meter through the angle method combined with the known range.The test was carried out using the actual meter photos taken in the gas gathering station.The experimental results showed that the algorithm process can quickly and accurately identify the meter indication.(4)Design a set of reading algorithm flow based on Le Net-5 neural network that can be used for digital meters in gas gathering stations.First,convert the digital instrument image to HSV color space,segment the digital area according to the color features;then use the edge detection algorithm based on the Canny operator to identify the digital contour;finally compare the effect of the SVM algorithm and the neural network algorithm based on Le Net-5 on the number classification,Experiments with the digital instrument picture of the gas gathering station,the results show that the algorithm based on Le Net-5 has higher recognition accuracy and faster recognition speed,which can be applied to the real-time digital instrument recognition of the gas gathering station.