Research On Street Tree Recognition Method Based On Deep Learning

Street trees are an important part of urban component management.The identification and extraction of street trees is a cumbersome task when conducting street tree greening surveys.When the road tree is identified and extracted,it is often done by means of visual interpretation.This mode of operation is time-consuming and laborious,which greatly increases the cost of the greening of the street tree.Therefore,exploring the use of new techniques of deep learning for the automatic identification and extraction of street trees has positive and far-reaching significance for the timeliness and cost control of greening investigations.Based on the deep learning technology,this paper uses the panoramic image data collected by the in-vehicle mobile measurement system to carry out related research work on the needs of automatic recognition and extraction of street trees.The main work content and research results of this paper include the following aspects:(1)Firstly,the significance and importance of the identification and extraction of street trees in road greening are briefly introduced.The theoretical knowledge related to deep learning is systematically introduced.The principle and network structure of convolutional neural networks are introduced.Four depths are analyzed.Learn the target detection algorithm and perform performance comparison analysis.(2)Constructing a street tree data set using panoramic image data collected by the in-vehicle mobile measurement system.The deepened convolutional neural network model is trained using the established street tree data set until the deep network model converges,and then the test data set is used to test the generalization ability of the constructed deep convolutional network model.The experimental results in this paper show that the recall index(Recall),mean mean accuracy(MAP),and average crossover ratio(Region Avg IOU)of the deep convolutional network model are 82.2%,83.6%,and 81.7%,respectively.Deep learning technology is feasible for automatic recognition and extraction of street tree images.(3)Using ArcObjects component and C# language to develop the automatic recognition and extraction plug-in based on deep learning,realize the following functions:(1)Implement the hexahedral cutting function,process the spherical panoramic image data collected by the in-vehicle mobile measurement system,and obtain the left and right views as the street tree.Identify and extract the data source;(2)Using the DNN module in OpenCV,call the trained deep network model to realize the automatic recognition and extraction of the street tree;(4)In ArcMap,based on the data collected by the vehicle mobile measurement system,the automatic recognition and extraction plug-in based on deep learning based on deep learning is developed to realize the automatic recognition and extraction of the street tree based on image data.Compared with the pure manual extraction method,its in-house processing efficiency can be increased by more than 40%.