Implementation Of Crowdsourcing-based Image Processing System For Oilfield Gas Gathering Station Monitoring

Gas gathering station is an important infrastructure in the process of natural gas extraction,and with the development of digital information technology,China is gradually promoting unattended gas gathering stations in oil and gas fields.Valve is one of the important components in the gas gathering station,if the valve has serious corrosion there will be natural gas leakage problems,which will cause serious safety accidents.Therefore,for the monitoring and management of unattended gas gathering stations,an important element is to monitor the corrosion of valves.At present,unattended gas gathering stations are equipped with centralized transmission station automation monitoring,distribution automation and other systems,but also equipped with video monitoring system,but the video monitoring system is only used for remote manual monitoring,can not identify the valve corrosion in a timely manner,therefore,the introduction of artificial intelligence technology in the existing video monitoring system to achieve intelligent and rapid identification of valve corrosion,has important practical significance.Unattended gas gathering station monitoring image(valve image as an example)recognition technology is an important technology in the intelligent monitoring system.Based on the current research results in the field of computer vision,this thesis proposes a valve monitoring image recognition algorithm model based on the fusion of deep learning and crowdsourcing.The model firstly uses crowdsourcing technology to obtain labels for a large dataset of valves from gas collection stations,and secondly uses deep learning technology to detect and identify and classify valve images.The main research elements are as follows:(1)This thesis proposes an improved valve detection model based on YOLOv5s(You Only Look Once,single-stage target detection algorithm),and the algorithm uses YOLOv5 s +Dense Net model.The target valve in the original image is detected by YOLOv5 s to achieve the purpose of distinguishing the background from the target.Secondly,drawing on the advantages of the Dense Net network structure,the dense connection approach is applied to the residual module of the YOLOv5 s backbone network,and the cross-connection approach is applied to the Neck network,thus enhancing the capability of valve detection and recognition.Finally,the network model is optimized through fine screening and classification of the valve dataset to achieve the purpose of improving the detection and recognition accuracy.The experimental results show that the improved fused CSPD＿x+CSPS＿x+dataset optimization+YOLOv5s model can obtain deeper valve feature information and higher accuracy of corrosion valve detection.(2)In order to reduce the cost of dataset annotation and increase the number of valve calibration datasets,this paper uses crowdsourcing techniques to annotate the datasets.Each user annotates the dataset to generate an independent annotation record file,and passes it to the same model for iterative updates,and then trains the annotated dataset.At this point,the generalization ability of the model is improved,and it can detect and identify the valves of various scenes.(3)Based on the valve detection algorithm model based on the fusion of deep learning and crowdsourcing technology,a valve monitoring image processing system for unattended gas gathering stations is designed using MATLAB.The system mainly includes user management module,image annotation module,dataset training module,target detection module,etc.to realize corrosion detection of valves in unattended gas collection stations.