Pattern Recognition Of River Network Based On Graph Convolutional Neural Network

With the rapid development of information and communication technology,highprecision geospatial data in the global scale presents a blowout growth,how to mine the inherent laws from the massive geospatial data has become a research hotspot.The successful practice of artificial intelligence technology brings new possibilities to the analysis and application of geographical big data.However,due to the particularity of geographical spatial vector data structure,it hinders the application of deep learning method represented by convolution neural network.Graph convolution neural network,which has been developed recently,is particularly suitable for applications in irregularly arranged geospatial vector data.River network data is an important part of geospatial big data.River network spatial pattern recognition can be applied in the fields of geological phenomenon analysis and cartography.How to formally express the semantic,geometric,topological,local and global characteristics of river network?How to design a pattern recognition model of river network based on graph convolution network?In response to these issues,this paper has carried out the following research work:(1)This paper discusses the necessity and feasibility of applying deep learning method in river network pattern recognition under the background of geographic big data.Human cognitive behavior is extremely subjective,and it is often difficult to express accurately with data.Deep learning method can continuously extract high-level feature information from data by optimizing model parameters,which reduces the incompleteness of feature design.Therefore,the application of deep learning method may realize the automation of river network pattern recognition research.(2)Study the method of structured representation of river network data.On the basis of summarizing the existing research,combined with the characteristics of river network data,the binary tree is used to construct the tree structure based on river segment.The hierarchical structure of the binary tree is determined by identifying the upstream and downstream relationship of the river network.On this basis,the angle between upstream and downstream is calculated,the left and right subtrees are distinguished,and the river network is coded to realize the river network data structure.In addition,the length,tortuosity and other features are embedded into the river network structure to comprehensively describe the river network.(3)A pattern recognition model of river network based on graph convolution neural network is constructed.The input of the model is the structured river network sample,and the output of the model is the result of sample label,which realizes the end-to-end data processing.The experimental results show that the river network structured method used in this paper can meet the needs of river network pattern recognition,the selected river network characteristics can effectively express the geometric,spatial and other attributes of river network.The designed model has good classification performance,and the recognition results are consistent with human cognition and have practical value.