Research And Implementation Of Traffic Flow Forecasting Based On Remote Sensing Image And Graph Network

With the maturity of deep learning technology and the vigorous development of intelligent transportation systems,research on traffic flow prediction based on remote sensing images has attracted more and more attention.Remote sensing image can describe traffic information intuitively.Traffic flow prediction is closely related to urban traffic distribution,path planning,road management,map navigation,etc.In practical application scenarios,there are various challenges in traffic flow prediction.On the one hand,conventional traffic information data is collected based on physical sensor equipment,which has problems such as high equipment costs,high data redundancy,large calculation costs,and poor real-time performance;On the other hand,road topological relationships in remote sensing images are difficult to extract,vehicle targets are small,and background complexity is high.It is difficult to directly use remote sensing images to construct traffic information datasets.Therefore,how to effectively use the traffic information contained in the remote sensing image to make traffic flow prediction has great challenges.Aiming at the above problems,this paper proposes a road extraction and road vehicle detection algorithm based on remote sensing images,and a method of constructing traffic information datasets.A traffic flow prediction algorithm based on graph networks is designed,and based on this,a design prototype system for traffic flow prediction based on remote sensing images and graph networks.First,the related algorithms are implemented and the traffic information extraction of remote sensing images is completed.This article improves the road extraction accuracy by improving the ASPP structure,and improves the accuracy of road vehicle detection through high and low feature fusion and detection suppression algorithms;Secondly,using the previously extracted road and vehicle information,a traffic information relationship data set was constructed;Then,the traffic flow prediction is completed,and the road network flow prediction is realized by using the graph network structure that can integrate the spatiotemporal characteristics.Finally,the prototype system implemented in this paper is verified.In actual application scenarios,a comparative experiment is performed on the traffic information extraction network and traffic flow prediction method proposed in this paper.The results show that the traffic flow prediction algorithm implemented in this paper has better accuracy and real-time performance,and has usability to achieve the expected research goals.In this paper,the method of combining remote sensing image and graph network to predict traffic flow has a broad application prospect in the field of satellite and intelligent transportation.