Research On Typical Spatial Distribution Pattern Recognition In Drainage System

One of the core tasks of automatic map generalization is to identify,describe,generalize and represent the important characteristics of geographic elements.As the spatial knowledge implicit in geographic elements,the spatial pattern is the product of a high degree of generalization of typical distribution characteristics,and is the important characteristics that need to be paid attention to in map generalization.Identifying spatial distribution patterns is an important method for spatial reasoning and data mining,which can promote the development of map generalization,multi-scale representation,geospatial analysis and modeling based on spatial data.Considering the skeleton role of drainage system in the map,this paper focuses on the spatial patterns of them and conducts in-depth research on the theory and method of identifying the typical spatial patterns of drainage system.The main work and innovations of the paper are detailed as follows:(i)The research background and significance of the spatial distribution pattern of drainage system is clarified first.The main contents of spatial pattern recognition in buildings,road networks,drainage system and islands are summarized with achievements and disadvantages of pattern recognition in the recent year.Then this paper focuses on the spatial patterns of drainage system.According to difference of drainage system types,the detailed research content of this paper is divided and a framework for pattern recognition of drainage system combined with various data and technologies is built based on geomorphological principles and human visual perception theory.(ii)The classification and definition of spatial pattern of drainage system based on spatial cognition theory and geomorphological principles.The difference and relationship between geographic environment,map space and spatial distribution pattern are analyzed.On this basis,the connotation and extension of spatial pattern in maps are clarified and the scope of this research is proposed.In the aspect of human visual perception theory,the main cognitive principles of spatial pattern recognition and their corresponding calculation models are summarized to guide the cognition and modeling of spatial patterns of drainage system.In terms of geomorphological principles,the main factors which influences the formation and development of drainage system are sorted out,so as to support the classification of spatial patterns of drainage system.Finally,different types of spatial patterns in drainage system are defined clearly to meet different requirements of map generalization,so the tasks of pattern recognition are divided scientifically and reasonably.(iii)The recognition method of linear pattern in manual ponds is proposed.Aiming at complex distribution characteristics of manual ponds,which is aggregated and regular globally and broken locally,the method takes into account global homogeneity and local heterogeneity to detect the complex linear pattern.Under the guidance of the theory of visual integrity and global-first principle,a multilevel cognitive structure composed of major-and-minor relationship,parallel relationship,simple linear pattern and complex linear pattern is constructed.The major-and-minor relationship,parallel relationship and simple linear patterns are identified sequentially by the different constraints of geometric,topological,and arranging characteristics of ponds.Aiming at the problem that the obvious heterogeneity of local pond groups leads to the fracture and omission of simple linear patterns,the strategies of “local optimization” and“aggregation” are designed to identify these heterogeneous ponds and consider them as a whole.Finally,the complex linear pattern of ponds can be detected successfully.(iv)The recognition method of hierarchical spatial pattern for ditch networks is designed.Aiming at the contradiction of good continuity of the whole ditch network and multiple local breakpoints,the research of hierarchical structure identification on global and local perspectives is carried out.At the global level,the “Stroke” structure of ditch network is identified using the semantic,topological and geometric characteristics of the ditches,then the importance of the ditches is evaluated from the functional point of view based on the network centrality.The global hierarchical structure of ditch network,consisting of the important ditches,the main ditches and the minor ditches,are designed.At the local level,guided by visual integrity and functional continuity,a local hierarchy model with collinear relationships,parallel relationships and main-tributary relationships is constructed.Based on the geometric and distribution characteristics,the groups of collinear ditch segments are preferentially identified and considered as a whole to eliminate the influence of heterogeneity.On this basis,the parallel ditches are identified by the direction and length constraints.Additionally,the main-tributary relationships between parallel ditches and the main ditches are detected by proximity distance and arrangement angle.Finally,the aim of detecting the local hierarchical structure of ditch network is realized.(v)The method of lakes clustering is proposed.Due to the complex distribution characteristics of lakes that are globally loose,locally aggregated and with different degrees of aggregation,a novel parameter to describe the distribution density of polygons is proposed,and based on this,a clustering method of lakes that meets the requirements of different scales is designed.The Voronoi diagram is used to divide the area and the influence of the adjacent polygons to the central polygon is considered,then the density parameter,aggregation degree,is firstly proposed and its effectiveness is validated by different simulation data.According to the aggregation degree and the nearest adjacent distance,the aggregation degree vector is constructed.Moreover,the clustering centers and initial clustering results are extracted.In order to meet the requirements of different levels of details in clustering results for different scales,a method of edge detection and group merging is designed with average aggregation degree and average nearest neighbor distance.Furthermore,the clustering method also shows good performance in both lakes and islands,which has good applicability.(vi)The recognition method of various spatial patterns of drainage networks is proposed.With the complex tectonic conditions and diverse shapes of drainage network,the identification of the spatial pattern of single drainage network and the combined spatial pattern of multiple drainage networks are carried out.According to the differences in the connectivity of different drainage networks,it is clear that the spatial patterns of a single drainage system and the radial drainage network,which is composed by multiple drainage networks,are used as identification targets.The graph structure data is converted from vector drainage data based on the connectivity of drainage segments and the characteristics of drainage segments,basin and the whole network.Meanwhile,a deep learning model with spatial graph convolution,self-attention pooling and fully connected classification as the main modules is designed.Then the model is trained by these graph structure data to mine the deep-level characteristics of different patterns and realize the intelligent identification of multiple spatial patterns of single drainage networks.Throughout making full use of the dome topographic characteristics of the radial drainage network,the mountain points are detected base on DEM data.Furthermore,a novel method for identification of the radial drainage network is proposed constrained by the characteristics of drainage segments and basin.According to the different relationships between radial drainage networks,the attribution decision model of the boundary drainage segments is constructed.Finally,Automatic identification of radial river system based on multi-source data fusion analysis is realized for the first time.