TopicIssue
Top ViewTop Cited
- Probability Estimation of Future Earthquakes in China Based on Improved PSHA Model
- Self-adaptive Window Extension Strategies of Discontinuity Features for MVLL Algorithm
- Big Data GIS
- Experimental Geography Based on Virtual GeographicEnvironments(VGEs)
- Big Data in Smart City
- Study on GPS/GLONASS Precise Point Positioning Using UofC Ionosphere-free Combination
- An Improved Cellular Automata Model for Simulating Pedestrian Evacuation
- Introduction to Urban Computing
- Detecting “Hot Spots” of Facility POIs Based on Kernel Density Estimation and Spatial Autocorrelation Technique
- On Construction of China’s Space Information Network
- Reflections on Issues in National Geographical Conditions Monitoring
- LUR-based Simulation of the Spatial Distribution of PM2.5of Wuhan
- Oblique Image Based Automatic Aerotriangulation and Its Application in 3D City Model Reconstruction
Download CenterMore >
LinkMore >
Display Method:
2020, 45(12): 1829-1835.
doi: 10.13203/j.whugis20200536
Abstract:
Physical space and social space constitute the dual space of human society. With the rapid development and popularization of information and communication technology (ICT), information space has grown into the third space, which constitutes the ternary space of human society together with physical space and social space. Smart cities are the intelligent transformation of cities under the condition of ternary space. Its basic engineering logic is to establish the mapping from physical space and social space to information space, and then feedback physical space and social space through information space to optimize the urban system and solve the urban problems. Geographic information system (GIS) is a technical science that depicts and expresses physical space and social space from the perspective of space. Establishing the intrinsic relationship of ternary space through GIS will be the logical basis for the construction of smart cities. Firstly, the GIS technology requirements of smart cities is analyzed from four aspects: Digitization, intelligence, visualization and openness. Then, a GIS framework for smart cities is proposed, the core of which is shielding complex heterogeneous data downwards and enabling diverse applications upwards. Finally, the key research contents that need to be focused on is discussed, including unified urban data platform construction, digital twins and visualization, two-way synchronized spatial intelligence, and open platform services.
Physical space and social space constitute the dual space of human society. With the rapid development and popularization of information and communication technology (ICT), information space has grown into the third space, which constitutes the ternary space of human society together with physical space and social space. Smart cities are the intelligent transformation of cities under the condition of ternary space. Its basic engineering logic is to establish the mapping from physical space and social space to information space, and then feedback physical space and social space through information space to optimize the urban system and solve the urban problems. Geographic information system (GIS) is a technical science that depicts and expresses physical space and social space from the perspective of space. Establishing the intrinsic relationship of ternary space through GIS will be the logical basis for the construction of smart cities. Firstly, the GIS technology requirements of smart cities is analyzed from four aspects: Digitization, intelligence, visualization and openness. Then, a GIS framework for smart cities is proposed, the core of which is shielding complex heterogeneous data downwards and enabling diverse applications upwards. Finally, the key research contents that need to be focused on is discussed, including unified urban data platform construction, digital twins and visualization, two-way synchronized spatial intelligence, and open platform services.
2020, 45(12): 1836-1846.
doi: 10.13203/j.whugis20200198
Abstract:
With the increasing volume of remote sensing data, the efficient online data management and service system has become an essential infrastructure for research on massive remote sensing data application. At present, different remote sensing data storage and service application schemes have brought difficulties to the system implementation and application due to the lack of independent intermediate framework design. There is an urgent need for a remote sensing data online distributed management and service framework-independent of data storage and service application, and to reconstruct the implementation mode of data management and service, reduce the difficulty of massive remote sensing image online management and service. Against this background, we design and propose a platform-independent online distributed management and service framework for remote sensing(RS-ODMS) by abstracting the remote sensing data management and computing model. This framework constructs a remote sensing data distributed hybrid storage model (DHS) and a remote sensing image data distributed parallel computing model(DPC) to improve the remote sensing data management ability and service efficiency. Experimental results show that the prototype system based on RS-ODMS can support an expansion capability of multiple distributed storage backends on remote sensing image management. And it is endowed with high efficiency, stable service ability and extensive applicability, which can provide reference for the online management and service application research of massive remote sensing data.
With the increasing volume of remote sensing data, the efficient online data management and service system has become an essential infrastructure for research on massive remote sensing data application. At present, different remote sensing data storage and service application schemes have brought difficulties to the system implementation and application due to the lack of independent intermediate framework design. There is an urgent need for a remote sensing data online distributed management and service framework-independent of data storage and service application, and to reconstruct the implementation mode of data management and service, reduce the difficulty of massive remote sensing image online management and service. Against this background, we design and propose a platform-independent online distributed management and service framework for remote sensing(RS-ODMS) by abstracting the remote sensing data management and computing model. This framework constructs a remote sensing data distributed hybrid storage model (DHS) and a remote sensing image data distributed parallel computing model(DPC) to improve the remote sensing data management ability and service efficiency. Experimental results show that the prototype system based on RS-ODMS can support an expansion capability of multiple distributed storage backends on remote sensing image management. And it is endowed with high efficiency, stable service ability and extensive applicability, which can provide reference for the online management and service application research of massive remote sensing data.
2020, 45(12): 1847-1856.
doi: 10.13203/j.whugis20200422
Abstract:
Public health and public security events present the importance of crowd dynamics observation. The crowd dynamics observation should be a key component of human-orient observation, and the necessary elements of global observation system framework consisting observations of space, sky, land, ocean and people. This paper firstly introduces a concept of crowd dynamics observation, defines its scientific connotations and boundaries, and highlights the role and significance of this theory. Then, from the perspective of public health and safety emergency, this paper analyzes the characteristics of accuracy, scientific city and on-demand in crow dynamics observation requirements, and introduces the state of art of crowd dynamics observation including data perception of crowd dynamics, sampling and expansion of population scale, and quantitative modeling of crowd dynamics, crowd dynamics research field, etc. Next, this paper thinks about the observation tasks of crowd dynamics from four important aspects, such as, rigorous distinction between individuals and groups, abnormal observations centered on spatiotemporal processes, breakthroughs in observational thinking under missing data, and scientific predictions of crowd dynamics under specific scenarios, etc. Finally, this paper summarizes three important research challenges combined with public health and safety emergency tasks, namely, the challenge of improving observation capabilities, the challenge of fusion observation of human and earth observation, and the research ethical challenge of human observation. In short, the concepts, thinking dimensions and challenges proposed in this paper provide basic theories and methods for major public health and safety emergencies, clarifying the laws of urban multi-type spatial development, mastering the differentiation mechanism of spatial evolution, and controlling urban spatial development paths. All of them show assignificance of the frontier exploration.
Public health and public security events present the importance of crowd dynamics observation. The crowd dynamics observation should be a key component of human-orient observation, and the necessary elements of global observation system framework consisting observations of space, sky, land, ocean and people. This paper firstly introduces a concept of crowd dynamics observation, defines its scientific connotations and boundaries, and highlights the role and significance of this theory. Then, from the perspective of public health and safety emergency, this paper analyzes the characteristics of accuracy, scientific city and on-demand in crow dynamics observation requirements, and introduces the state of art of crowd dynamics observation including data perception of crowd dynamics, sampling and expansion of population scale, and quantitative modeling of crowd dynamics, crowd dynamics research field, etc. Next, this paper thinks about the observation tasks of crowd dynamics from four important aspects, such as, rigorous distinction between individuals and groups, abnormal observations centered on spatiotemporal processes, breakthroughs in observational thinking under missing data, and scientific predictions of crowd dynamics under specific scenarios, etc. Finally, this paper summarizes three important research challenges combined with public health and safety emergency tasks, namely, the challenge of improving observation capabilities, the challenge of fusion observation of human and earth observation, and the research ethical challenge of human observation. In short, the concepts, thinking dimensions and challenges proposed in this paper provide basic theories and methods for major public health and safety emergencies, clarifying the laws of urban multi-type spatial development, mastering the differentiation mechanism of spatial evolution, and controlling urban spatial development paths. All of them show assignificance of the frontier exploration.
2020, 45(12): 1857-1864.
doi: 10.13203/j.whugis20200650
Abstract:
The rapid development of deep learning provides an important technical means for intelligent analysis of remote sensing big data. Firstly, this paper mainly introduces the deep learning modes in remote sensing data recognition and application, and proposes a deep reinforcement learning, multi-task learning and sub-pixel-pixel-super-pixel feature learning models for object features recognition from LiDAR point clouds, optical remote sensing images and hyperspectral images. The model parameters are basically obtained by learning, and thus the workload of the parameter adjustments is small. The spatial and contextual information, texture and spectral characteristics between ground objects are fully taken into account, so the presented models have good generalization abilities. Then, it describes the progress in terms of the joint deep learning and multi-source remote sensing data in accurate poverty alleviation assessment, wetland change and spatial analysis in Qinghai-Tibet Plateau in the past 20 years, and corn yield estimation. In order to better promote the transformation from remote sensing data to knowledge, it is necessary give full play to the advantages of deep learning in remote sensing big data processing, and develop new data processing algorithms and technologies.
The rapid development of deep learning provides an important technical means for intelligent analysis of remote sensing big data. Firstly, this paper mainly introduces the deep learning modes in remote sensing data recognition and application, and proposes a deep reinforcement learning, multi-task learning and sub-pixel-pixel-super-pixel feature learning models for object features recognition from LiDAR point clouds, optical remote sensing images and hyperspectral images. The model parameters are basically obtained by learning, and thus the workload of the parameter adjustments is small. The spatial and contextual information, texture and spectral characteristics between ground objects are fully taken into account, so the presented models have good generalization abilities. Then, it describes the progress in terms of the joint deep learning and multi-source remote sensing data in accurate poverty alleviation assessment, wetland change and spatial analysis in Qinghai-Tibet Plateau in the past 20 years, and corn yield estimation. In order to better promote the transformation from remote sensing data to knowledge, it is necessary give full play to the advantages of deep learning in remote sensing big data processing, and develop new data processing algorithms and technologies.
2020, 45(12): 1865-1874.
doi: 10.13203/j.whugis20200597
Abstract:
The technological progress in the field of artificial intelligence (AI) has brought new opportunities and challenges to the intelligent development and innovative research in geospatial related fields. Geospatial artificial intelligence (GeoAI) refers to the interdisciplinary research direction that combines geography, earth science and artificial intelligence, and seeks to solve major scientific and engineering problems in human-environmental interaction systems through the research and development of spatial intelligence in machines to improve the dynamic perception, intelligent reasoning and knowledge discovery of geographic phenomena and earth science processes. This paper briefly summarizes the historical origins of GeoAI development, introduces spatially explicit and implicit AI models, reviews recent GeoAI research and applications(including spatial representation learning, spatiotemporal prediction and spatial interpolation, monitoring of geographic resources and environment, cartography, and geo-text data semantic analysis), and identifies several potential research challenges and directions for the future development of GeoAI.
The technological progress in the field of artificial intelligence (AI) has brought new opportunities and challenges to the intelligent development and innovative research in geospatial related fields. Geospatial artificial intelligence (GeoAI) refers to the interdisciplinary research direction that combines geography, earth science and artificial intelligence, and seeks to solve major scientific and engineering problems in human-environmental interaction systems through the research and development of spatial intelligence in machines to improve the dynamic perception, intelligent reasoning and knowledge discovery of geographic phenomena and earth science processes. This paper briefly summarizes the historical origins of GeoAI development, introduces spatially explicit and implicit AI models, reviews recent GeoAI research and applications(including spatial representation learning, spatiotemporal prediction and spatial interpolation, monitoring of geographic resources and environment, cartography, and geo-text data semantic analysis), and identifies several potential research challenges and directions for the future development of GeoAI.
2020, 45(12): 1875-1883.
doi: 10.13203/j.whugis20200535
Abstract:
City is the place aggregated massive human activities. City is the exchange hub of population flow, goods flow, information flow and currency flow, which is highly dynamic and complex. Smart city provides various tools to acquire spatiotemporal big data, such as satellite and drone remote sensing, mobile sensing, social sensing, crowdsourcing sensing, etc., which enable us to sense urban dynamics. This paper introduces the framework of urban dynamic sensing, describes the typical applications of spatial dynamics, human behavior dynamics and space-behavior interaction dynamics, and discusses the problems, such as the uncertainty in spatiotemporal big data, the multi-view ensemble learning in urban sensing, the verification of the urban dynamic results and the cascading influence of multi-urban factors. Outlooking the future, the study of urban dynamics should combine with real-time Internet of things data to sense multi-dimensional, multi-spatiotemporal resolution urban dynamic to enable refined urban governance and to effectively solve urban problems.
City is the place aggregated massive human activities. City is the exchange hub of population flow, goods flow, information flow and currency flow, which is highly dynamic and complex. Smart city provides various tools to acquire spatiotemporal big data, such as satellite and drone remote sensing, mobile sensing, social sensing, crowdsourcing sensing, etc., which enable us to sense urban dynamics. This paper introduces the framework of urban dynamic sensing, describes the typical applications of spatial dynamics, human behavior dynamics and space-behavior interaction dynamics, and discusses the problems, such as the uncertainty in spatiotemporal big data, the multi-view ensemble learning in urban sensing, the verification of the urban dynamic results and the cascading influence of multi-urban factors. Outlooking the future, the study of urban dynamics should combine with real-time Internet of things data to sense multi-dimensional, multi-spatiotemporal resolution urban dynamic to enable refined urban governance and to effectively solve urban problems.
2020, 45(12): 1884-1889.
doi: 10.13203/j.whugis20200423
Abstract:
Urban simulation research originated between the 1980s and 1990s. Today urban simulation has become a new paradigm of urban research, which is an important outcome of computational thinking in urban research. Urban simulation methods are usually based on cellular automata (CA) and machine learning. A series of urban CA models have been developed to simulate complex urban evolution processes and associated multi-scenario analysis. This paper reviews the origin and progress of urban simulation research. With the discussion of urban CA's general structure, we explains the necessity and feasibility of machine learning methods to support urban simulation. Furthermore, we reviews the integration of machine learning and CA in urban research, and also discusses its new trends and emerging challenges.
Urban simulation research originated between the 1980s and 1990s. Today urban simulation has become a new paradigm of urban research, which is an important outcome of computational thinking in urban research. Urban simulation methods are usually based on cellular automata (CA) and machine learning. A series of urban CA models have been developed to simulate complex urban evolution processes and associated multi-scenario analysis. This paper reviews the origin and progress of urban simulation research. With the discussion of urban CA's general structure, we explains the necessity and feasibility of machine learning methods to support urban simulation. Furthermore, we reviews the integration of machine learning and CA in urban research, and also discusses its new trends and emerging challenges.
2020, 45(12): 1890-1902.
doi: 10.13203/j.whugis20200334
Abstract:
Point-of-interest (POI) recommendation has emerged as a focal point in the research of location-based social network (LBSN) in recent years. It can help users find their favorite venue and bring considerable benefits to businesses. Nowadays, deep learning is gradually applied to the task of recommendation system because it can capture the nonlinear relationship between users and items more effectively. This paper thus focuses on recent research on POI recommendation combined with deep learning. Firstly, we introduce the difference between POI recommendation and other traditional recommendation tasks and illustrate various influencing factors that can improve the performance of the model. Then, the methods of applying deep learning to POI recommendation are divided into four categories, including POI embedding, deep collaborative filtering, feature extraction from side information, and sequence recommendation using recurrent neural network (RNN). We also investigate the development of user models performance and advantages combined with deep learning in these different aspects of applications. Finally, we summarize and look forward to the development of POI recommendation research combined with deep learning.
Point-of-interest (POI) recommendation has emerged as a focal point in the research of location-based social network (LBSN) in recent years. It can help users find their favorite venue and bring considerable benefits to businesses. Nowadays, deep learning is gradually applied to the task of recommendation system because it can capture the nonlinear relationship between users and items more effectively. This paper thus focuses on recent research on POI recommendation combined with deep learning. Firstly, we introduce the difference between POI recommendation and other traditional recommendation tasks and illustrate various influencing factors that can improve the performance of the model. Then, the methods of applying deep learning to POI recommendation are divided into four categories, including POI embedding, deep collaborative filtering, feature extraction from side information, and sequence recommendation using recurrent neural network (RNN). We also investigate the development of user models performance and advantages combined with deep learning in these different aspects of applications. Finally, we summarize and look forward to the development of POI recommendation research combined with deep learning.
2020, 45(12): 1903-1909.
doi: 10.13203/j.whugis20200209
Abstract:
By combining geo-computing tasks and spatial-temporal data resources with specific business logic, users can deal with complex geographic problems in specific domains. Traditional geo-computing processes require manual participation if complex iterations are built. Scientific workflows support automated process to assist researchers in geo-computing and knowledge discovery. A method for complicated geospatial flow processing based on scientific workflow is proposed. Firstly, exist open source scientific workflow management systems are compared. Subsequently, the core problems of geo-computing processes are analyzed: Heterogeneous multi-source data access, the description of complex computing, the scalability of computing, and the interaction and sharing of data analysis processes. The complex geo-computing flow processing method supported by scientific workflow is put forward, which can combine the data dependencies between complicated applications. Based on Swift's excellent capabilities in process structure, dispatchable resource types, and fault tolerance levels, geospatial flow can control the sequence of different parts to be completed under the constraints of time, storage and resources. Therefore, spatial-temporal data can be searched, moved, analyzed, processed and visualized. Finally, a housing site selection case based on this method is presented to show the feasibility.
By combining geo-computing tasks and spatial-temporal data resources with specific business logic, users can deal with complex geographic problems in specific domains. Traditional geo-computing processes require manual participation if complex iterations are built. Scientific workflows support automated process to assist researchers in geo-computing and knowledge discovery. A method for complicated geospatial flow processing based on scientific workflow is proposed. Firstly, exist open source scientific workflow management systems are compared. Subsequently, the core problems of geo-computing processes are analyzed: Heterogeneous multi-source data access, the description of complex computing, the scalability of computing, and the interaction and sharing of data analysis processes. The complex geo-computing flow processing method supported by scientific workflow is put forward, which can combine the data dependencies between complicated applications. Based on Swift's excellent capabilities in process structure, dispatchable resource types, and fault tolerance levels, geospatial flow can control the sequence of different parts to be completed under the constraints of time, storage and resources. Therefore, spatial-temporal data can be searched, moved, analyzed, processed and visualized. Finally, a housing site selection case based on this method is presented to show the feasibility.
2020, 45(12): 1910-1918.
doi: 10.13203/j.whugis20200185
Abstract:
Crowdsource geospatial data is one kind of open geographic data collected by the public. A wealth of spatial information and knowledge are hidden in crowdsource geospatial data. Check-in data is one of the representative crowdsource geospatial data. Most existing work on spatial-textual objects, such as evaluating the similarity of two spatial-textual objects in spatial keyword query, considers that spatial similarity and text similarity are independent of each other. According to the first law of geography: Everything is connected; the closer two objects are, the stronger their connection is. We explore the correlation between spatial information and textual information in the real check-in data scrawled from the location based social networks. After data preprocess and geographic mapping, we computed the textual attribute values in each region. Then, we use the exploratory spatial analysis to analyze the global spatial autocorrelation and local spatial autocorrelation in different the spatial scales, that is different states in United States and the two cities such as New York and Los Angeles, respectively. The results show that different textual attributes in different regions have different global spatial autocorrelation; the results obtained from the local autocorrelation analysis show the phenomenon that the textual attributes get together. Both the above results provide the basis for research on the assumption that "the texts are similar in the near space". Furthermore, the conclusion can help departments or enterprises to make reasonable decisions.
Crowdsource geospatial data is one kind of open geographic data collected by the public. A wealth of spatial information and knowledge are hidden in crowdsource geospatial data. Check-in data is one of the representative crowdsource geospatial data. Most existing work on spatial-textual objects, such as evaluating the similarity of two spatial-textual objects in spatial keyword query, considers that spatial similarity and text similarity are independent of each other. According to the first law of geography: Everything is connected; the closer two objects are, the stronger their connection is. We explore the correlation between spatial information and textual information in the real check-in data scrawled from the location based social networks. After data preprocess and geographic mapping, we computed the textual attribute values in each region. Then, we use the exploratory spatial analysis to analyze the global spatial autocorrelation and local spatial autocorrelation in different the spatial scales, that is different states in United States and the two cities such as New York and Los Angeles, respectively. The results show that different textual attributes in different regions have different global spatial autocorrelation; the results obtained from the local autocorrelation analysis show the phenomenon that the textual attributes get together. Both the above results provide the basis for research on the assumption that "the texts are similar in the near space". Furthermore, the conclusion can help departments or enterprises to make reasonable decisions.
2020, 45(12): 1919-1929.
doi: 10.13203/j.whugis20200488
Abstract:
Spatiotemporal data mining is the core research topic of geographic information science. In the era of big data, the explosive growth of geographic spatiotemporal data puts forward an urgent demand for spatiotemporal knowledge discovery, which promotes the continuous development of spatiotemporal data mining technology. However, the universal heterogeneity and sparse distribution characteristics of spatiotemporal big data restrict the realization of spatiotemporal data mining algorithms, and significantly affect the description and analysis capabilities of natural and social complex systems. Thus, this paper focuses on the series of bottlenecks faced in the expression and application of heterogeneous and sparsely distributed spatiotemporal data. We systematically summarized the research status and existing problems of several key spatiotemporal mining tasks including missing spatiotemporal data interpolation, sparse spatiotemporal data reconstruction, and spatiotemporal state prediction, condensed four key scientific problems, and gave four corresponding solutions. The proposed methods are expected to enrich the method system in the field of spatiotemporal data mining and improve the quality and application value of spatiotemporal data modeling.
Spatiotemporal data mining is the core research topic of geographic information science. In the era of big data, the explosive growth of geographic spatiotemporal data puts forward an urgent demand for spatiotemporal knowledge discovery, which promotes the continuous development of spatiotemporal data mining technology. However, the universal heterogeneity and sparse distribution characteristics of spatiotemporal big data restrict the realization of spatiotemporal data mining algorithms, and significantly affect the description and analysis capabilities of natural and social complex systems. Thus, this paper focuses on the series of bottlenecks faced in the expression and application of heterogeneous and sparsely distributed spatiotemporal data. We systematically summarized the research status and existing problems of several key spatiotemporal mining tasks including missing spatiotemporal data interpolation, sparse spatiotemporal data reconstruction, and spatiotemporal state prediction, condensed four key scientific problems, and gave four corresponding solutions. The proposed methods are expected to enrich the method system in the field of spatiotemporal data mining and improve the quality and application value of spatiotemporal data modeling.
2020, 45(12): 1930-1941.
doi: 10.13203/j.whugis20200491
Abstract:
The concept of place plays an important role for coupling between human behaviors and geographic environments, and thus provides a human-centric research perspective for geographical analysis. In early place studies, it was difficult to capture and represent the comprehensive semantics of a place in computer systems due to the inherent vagueness, uncertainty, and ambiguity of places, as well as the high cost of obtaining place-based behavioral data at the individual level. Recently, multi-sourced big geo-data and artificial intelligent methods have brought new opportunities for place modeling and sensing. Moreover, place modeling is an important component of knowledge representation in GeoAI. This article reviews the progress of two directions of place modeling and place sensing.Place modeling aims to construct a platial knowledge model that is more in line with human spatial cognition and communication, so as to improve the intelligence and human-computer interaction of geographical information retrieval and spatial query answering. Place sensing aims to capture human perception and cognition about a place to better understand the intrinsic connections between human activities and the geographic environments, and to represent the rich geographical context of a place. The ultimate objective is to construct "white-box" place-based knowledge representation and reasoning for promoting the development of GeoAI.
The concept of place plays an important role for coupling between human behaviors and geographic environments, and thus provides a human-centric research perspective for geographical analysis. In early place studies, it was difficult to capture and represent the comprehensive semantics of a place in computer systems due to the inherent vagueness, uncertainty, and ambiguity of places, as well as the high cost of obtaining place-based behavioral data at the individual level. Recently, multi-sourced big geo-data and artificial intelligent methods have brought new opportunities for place modeling and sensing. Moreover, place modeling is an important component of knowledge representation in GeoAI. This article reviews the progress of two directions of place modeling and place sensing.Place modeling aims to construct a platial knowledge model that is more in line with human spatial cognition and communication, so as to improve the intelligence and human-computer interaction of geographical information retrieval and spatial query answering. Place sensing aims to capture human perception and cognition about a place to better understand the intrinsic connections between human activities and the geographic environments, and to represent the rich geographical context of a place. The ultimate objective is to construct "white-box" place-based knowledge representation and reasoning for promoting the development of GeoAI.
2020, 45(12): 1942-1948.
doi: 10.13203/j.whugis20200081
Abstract:
In large-scale road environment, point-based methods require dynamic calculations, and voxel-based methods often lose a lot of information when balancing resolution and performance. To overcome the drawbacks of the above two classical methods, this paper proposes a general network architecture that combines bi-level convolution and dynamic graph edge convolution optimization for multi-object recognition of large-scale road scenes. The framework integrates the convolution operations of two different domains of points and supervoxels to avoid redundant calculations and storage of spatial information in the network. Coupled with the dynamic graph edge convolution optimization, our model enables it to process large-scale point clouds end-to-end at once. Our method was tested and evaluated on different datasets. The experimental results show that our method can achieve higher accuracy in complex road scenes, which is superior to the existing advanced methods.
In large-scale road environment, point-based methods require dynamic calculations, and voxel-based methods often lose a lot of information when balancing resolution and performance. To overcome the drawbacks of the above two classical methods, this paper proposes a general network architecture that combines bi-level convolution and dynamic graph edge convolution optimization for multi-object recognition of large-scale road scenes. The framework integrates the convolution operations of two different domains of points and supervoxels to avoid redundant calculations and storage of spatial information in the network. Coupled with the dynamic graph edge convolution optimization, our model enables it to process large-scale point clouds end-to-end at once. Our method was tested and evaluated on different datasets. The experimental results show that our method can achieve higher accuracy in complex road scenes, which is superior to the existing advanced methods.
2020, 45(12): 1949-1959.
doi: 10.13203/j.whugis20200189
Abstract:
Cox regression model is an important method for event data survival analysis, which can effectively identify the correlations between event occurrence time and its influencing factors. The basic Cox regression model can only study on a single unique event, and the model can be adjusted expanded through event grouping and event start time setting, which can be used to deal with repeated events. Although the existing Cox model can also be easily modified to deal with time-varying explanatory variables, it is still lack of considering the comprehensive influence from temporal and spatial factors. In light of the repeated national survival status, this paper takes the occurrence probability of national terrorist attacks high risk status as the research object, and establishes the repeated event Cox model considering multiple spatial and temporal factors based on global terrorism database. Additionally, this paper has proposed a new terrorist attack risk level as well as spatial lag calculating method based on composited weight factor and similarity to ideal solution, which provide research foundation for the following expanded Cox model establishment. The results show that political and military factors have a more important influence on the occurrence probability of high-risk state from 1995 to 2016, compared with economic social and geographical factors. The regression effect of the repeated event Cox model with introducing time interaction and spatial lag explanatory variables have been improved, and time interaction as well as spatial lag explanatory variables have significant effects on the occurrence probability of high-risk state.
Cox regression model is an important method for event data survival analysis, which can effectively identify the correlations between event occurrence time and its influencing factors. The basic Cox regression model can only study on a single unique event, and the model can be adjusted expanded through event grouping and event start time setting, which can be used to deal with repeated events. Although the existing Cox model can also be easily modified to deal with time-varying explanatory variables, it is still lack of considering the comprehensive influence from temporal and spatial factors. In light of the repeated national survival status, this paper takes the occurrence probability of national terrorist attacks high risk status as the research object, and establishes the repeated event Cox model considering multiple spatial and temporal factors based on global terrorism database. Additionally, this paper has proposed a new terrorist attack risk level as well as spatial lag calculating method based on composited weight factor and similarity to ideal solution, which provide research foundation for the following expanded Cox model establishment. The results show that political and military factors have a more important influence on the occurrence probability of high-risk state from 1995 to 2016, compared with economic social and geographical factors. The regression effect of the repeated event Cox model with introducing time interaction and spatial lag explanatory variables have been improved, and time interaction as well as spatial lag explanatory variables have significant effects on the occurrence probability of high-risk state.
2020, 45(12): 1960-1969.
doi: 10.13203/j.whugis20200022
Abstract:
Road networks are the skeleton of a city, its pattern recognition plays an important role in urban landscape analysis, municipal planning, and traffic flow analysis. Road networks pattern recognition is an intelligent recognition problem that combines road geometric features, semantic features, and contextual relationships. The current road pattern recognition methods are mainly using geometric measures and statistical indicators to detect pattern recognition rules, and based on the edge-node graph structure and the polygon group structure of road network.As the expression of road patterns is influenced by human spatial cognition and visual psychology, rule-based reasoning and statistical analysis are difficult to obtain results consistent with human judgment. Driven by artificial intelligence technology, this paper introduces a novel deep learning model for graph structure, namely graph convolution network (GCN), to identify the grid-pattern in a road network. Firstly, to construct the graph structure, the road intersections are taken as nodes and the roads are taken as edges, all nodes are classified into grid and non-grid to act as the label and prediction target of the model. Road segments are then broken into consecutive linear units to extract the characteristics of the nodes which are also the input of the model. Then, the grid pattern is recognized by extracting the roads whose endpoints are all predicted to be a grid. The experimental results show that the accuracy of the method can reach 89.2% for road network's node classification and 86.1% for road segment classification, so it is effective for identifying grid pattern in a road network.
Road networks are the skeleton of a city, its pattern recognition plays an important role in urban landscape analysis, municipal planning, and traffic flow analysis. Road networks pattern recognition is an intelligent recognition problem that combines road geometric features, semantic features, and contextual relationships. The current road pattern recognition methods are mainly using geometric measures and statistical indicators to detect pattern recognition rules, and based on the edge-node graph structure and the polygon group structure of road network.As the expression of road patterns is influenced by human spatial cognition and visual psychology, rule-based reasoning and statistical analysis are difficult to obtain results consistent with human judgment. Driven by artificial intelligence technology, this paper introduces a novel deep learning model for graph structure, namely graph convolution network (GCN), to identify the grid-pattern in a road network. Firstly, to construct the graph structure, the road intersections are taken as nodes and the roads are taken as edges, all nodes are classified into grid and non-grid to act as the label and prediction target of the model. Road segments are then broken into consecutive linear units to extract the characteristics of the nodes which are also the input of the model. Then, the grid pattern is recognized by extracting the roads whose endpoints are all predicted to be a grid. The experimental results show that the accuracy of the method can reach 89.2% for road network's node classification and 86.1% for road segment classification, so it is effective for identifying grid pattern in a road network.
2020, 45(12): 1970-1980.
doi: 10.13203/j.whugis20200068
Abstract:
Predicting the future activity location and trajectory of residents can provide essential information for smart urban management such as epidemic control, traffic facilitation, public security, etc. However, the current personal location prediction methods generally focus on the mining of individual's historical travel patterns, and seldom consider the feature of different travel stay points. This paper aims to propose a location prediction model utilizing stay point feature extraction. The model firstly constructs the historical trajectory links based on trajectory data, performs the location discovery rules to transform historical trajectory links into stay point links and clusters the stay points to form clustering links. Secondly, the model extracts the feature information (entry time, departure time, weather and land use) from different stay points and extracts the space feature from clustering links. Finally, the cluster links with feature information is introduced into long short-term memory (LSTM) network for customization to implement the personal location prediction capability. By using a 23-day trajectory location data of more than 3 million volunteer users in Shenzhen city, China. The results show that the location prediction F-score of our proposed model is better than variable order Markov model (about 5.5% performance gains) and the traditional N-order Markov model (about 7% performance gains). The model also introduces approximately 6.6% performance gains by utilizing the temporal and weather features of stay points. The model shows the capability of utilizing travel stay point feature for personal next location prediction.
Predicting the future activity location and trajectory of residents can provide essential information for smart urban management such as epidemic control, traffic facilitation, public security, etc. However, the current personal location prediction methods generally focus on the mining of individual's historical travel patterns, and seldom consider the feature of different travel stay points. This paper aims to propose a location prediction model utilizing stay point feature extraction. The model firstly constructs the historical trajectory links based on trajectory data, performs the location discovery rules to transform historical trajectory links into stay point links and clusters the stay points to form clustering links. Secondly, the model extracts the feature information (entry time, departure time, weather and land use) from different stay points and extracts the space feature from clustering links. Finally, the cluster links with feature information is introduced into long short-term memory (LSTM) network for customization to implement the personal location prediction capability. By using a 23-day trajectory location data of more than 3 million volunteer users in Shenzhen city, China. The results show that the location prediction F-score of our proposed model is better than variable order Markov model (about 5.5% performance gains) and the traditional N-order Markov model (about 7% performance gains). The model also introduces approximately 6.6% performance gains by utilizing the temporal and weather features of stay points. The model shows the capability of utilizing travel stay point feature for personal next location prediction.
2020, 45(12): 1981-1988.
doi: 10.13203/j.whugis20200035
Abstract:
Urban traffic time series pattern recognition is mainly based on two methods: Distance calculation and feature extraction. The recognition results based on distance calculation are affected by the non-linear features of the time series, and it is also difficult to extract multi-time period Shapelet sub-sequences based on the feature extraction algorithm. Therefore, in order to accurately identify the data pattern of urban traffic index, a method for pattern recognition of traffic time series data based on convolutional neural network(CNN) was proposed. Firstly, the time series data are pre-processed and transformed into N-dimensional matrix to determine the input data of the neural network model. Then, the characteristic information of the input layer is extracted through repeated training of the characteristic network, and the pattern type of the traffic index is determined. Finally, Softmax classifier is used to classify the modes and realize the classification of traffic index. The time series pattern recognition of Beijing's traffic index from 2016 to 2018 was taken as an example. Compared with the traditional machine learning algorithm based on distance calculation, it was found that the CNN can more accurately divide the time series data of traffic index into five types of patterns that are consistent with the real situation. The experimental results show that the CNN is more accurate and effective for pattern recognition of time series data, and is suitable for correlation research and pattern discovery of time-phase change data.
Urban traffic time series pattern recognition is mainly based on two methods: Distance calculation and feature extraction. The recognition results based on distance calculation are affected by the non-linear features of the time series, and it is also difficult to extract multi-time period Shapelet sub-sequences based on the feature extraction algorithm. Therefore, in order to accurately identify the data pattern of urban traffic index, a method for pattern recognition of traffic time series data based on convolutional neural network(CNN) was proposed. Firstly, the time series data are pre-processed and transformed into N-dimensional matrix to determine the input data of the neural network model. Then, the characteristic information of the input layer is extracted through repeated training of the characteristic network, and the pattern type of the traffic index is determined. Finally, Softmax classifier is used to classify the modes and realize the classification of traffic index. The time series pattern recognition of Beijing's traffic index from 2016 to 2018 was taken as an example. Compared with the traditional machine learning algorithm based on distance calculation, it was found that the CNN can more accurately divide the time series data of traffic index into five types of patterns that are consistent with the real situation. The experimental results show that the CNN is more accurate and effective for pattern recognition of time series data, and is suitable for correlation research and pattern discovery of time-phase change data.
2020, 45(12): 1989-1996.
doi: 10.13203/j.whugis20200159
Abstract:
Understanding and accurately predicting the trajectory of pedestrians is of great significance to improve the level of auto-driving technology and reduce the occurrence of traffic accidents. Aiming at the problems of low prediction accuracy and insufficient utilization of pedestrian interaction information, a pedestrian trajectory prediction model based on self-attention mechanism and group behavior characteristics is presented. The model considers the movement information of each pedestrian and its interaction with the surrounding pedestrians. The recurrent neural network and graph convolutional network are used to model the pedestrian's walking state and the pedestrian's interaction separately. In the graph convolutional network, nodes of the graph represent the movement information of pedestrians, and edges of the graph represent the interaction between pedestrians. Self-attention mechanism is used to calculate pedestrian interaction. In addition, in order to increase the ability of the network to capture the group walk behavior and decrease the prediction error, a peer loss function is proposed. Experiments on public datasets show that the proposed model in this paper improves prediction accuracy significantly.
Understanding and accurately predicting the trajectory of pedestrians is of great significance to improve the level of auto-driving technology and reduce the occurrence of traffic accidents. Aiming at the problems of low prediction accuracy and insufficient utilization of pedestrian interaction information, a pedestrian trajectory prediction model based on self-attention mechanism and group behavior characteristics is presented. The model considers the movement information of each pedestrian and its interaction with the surrounding pedestrians. The recurrent neural network and graph convolutional network are used to model the pedestrian's walking state and the pedestrian's interaction separately. In the graph convolutional network, nodes of the graph represent the movement information of pedestrians, and edges of the graph represent the interaction between pedestrians. Self-attention mechanism is used to calculate pedestrian interaction. In addition, in order to increase the ability of the network to capture the group walk behavior and decrease the prediction error, a peer loss function is proposed. Experiments on public datasets show that the proposed model in this paper improves prediction accuracy significantly.
2020, 45(12): 1997-2004.
doi: 10.13203/j.whugis20200184
Abstract:
With the continuous improvement of user needs and the continuous enrichment of spatial data, the same geospatial scope includes different types of spatial data such as terrain, remote sensing images, and 3D models of oblique photogrammetry. 3D WebGIS cannot effectively load different types of spatial data in the same scene. This paper proposes a spatial data loading optimization method for 3D WebGIS to optimize the efficient loading of multiple types of spatial data. First, using the structural features of 3D Tiles with regular enclosing volume shapes and overlapping regions, a fast growth algorithm based on binary tree is proposed to locate and extract the outer boundary of 3D model. Then, there were jagged protrusions on the outer boundary of the roughly extracted polygon. In order to solve the problem, a polygon simplification method based on two neighboring points was proposed to optimize the outer boundary of the polygon to achieve accurate extraction of the outer boundary of the 3D model. Finally, by judging the inclusion of map tile and terrain tile data relationship to achieve selective loading of tile data. Experiments show that when the viewpoint height is below 300 meters, the total load of map tile data can be reduced by about 40%, indicating that the optimization method in this paper can significantly improve the data loading efficiency of the system without affecting the visualization of 3D WebGIS.
Articles online first have been peer-reviewed and accepted, which are not yet assigned to volumes /issues, but are citable by Digital Object Identifier (DOI).
Display Method:
, Available online ,
doi: 10.13203/j.whugis20200221
Abstract:
The development of ICT (Information and Communications Technology) provides new challeng⁃ es and opportunities for modern cartography. Totally different from the traditional cartography in the fields of the research methods and themes, modern cartography trends to digital cyberspace from paper map, and cartographic theories are urged to be improved. However, the understanding of the theories about carto⁃ graphic virtual space have great differences within the professional domain, and the internal contradictions and transboundary invasion make map gradually become an appurtenance of Internet products. To learning from certain domains to analyse and handle them, challenges have prompted cartographic scholars to ex⁃ plore new fields. The development of "digits and culture" has pushed the game to the forefront of hightech. Games are often geographically significant, and maps of game are a combination of ICT and maps. Based on the cyberspace in game map with instances as the research subject, the paper explores the charac⁃ teristics and essence of game map and its implications to the real physical world by the dimensionalities, in⁃ teractivity, narrativity, exploration and cultural dissemination of game maps to discover the extension of cartography, which would replenish and perfect modern cartographic theory and promote its development.
The development of ICT (Information and Communications Technology) provides new challeng⁃ es and opportunities for modern cartography. Totally different from the traditional cartography in the fields of the research methods and themes, modern cartography trends to digital cyberspace from paper map, and cartographic theories are urged to be improved. However, the understanding of the theories about carto⁃ graphic virtual space have great differences within the professional domain, and the internal contradictions and transboundary invasion make map gradually become an appurtenance of Internet products. To learning from certain domains to analyse and handle them, challenges have prompted cartographic scholars to ex⁃ plore new fields. The development of "digits and culture" has pushed the game to the forefront of hightech. Games are often geographically significant, and maps of game are a combination of ICT and maps. Based on the cyberspace in game map with instances as the research subject, the paper explores the charac⁃ teristics and essence of game map and its implications to the real physical world by the dimensionalities, in⁃ teractivity, narrativity, exploration and cultural dissemination of game maps to discover the extension of cartography, which would replenish and perfect modern cartographic theory and promote its development.
, Available online ,
doi: 10.13203/j.whugis20200172
Abstract:
Aiming at the mixed observation data of multiple distribution forms, a P-model distributed mixed model is established. Considering that the mixed number in the model belongs to incomplete data, the EM algorithm is introduced to estimate the parameters of the mixed model and the P-model mixed model parameters are derived in detail The estimated iteration formula and the corresponding iteration steps are given. Finally, the mixed Gaussian distribution data, Laplacian distribution and Gaussian distribution mixed data, and the residual data of measured GPS observations are used to verify the correctness and adaptability of the formula in this paper. The results of the calculation examples show that, compared with the single probability distribution, the P-parameter hybrid model can accurately reflect the actual situation of the data distribution, and the model parameters estimated by the EM algorithm have higher accuracy.
Aiming at the mixed observation data of multiple distribution forms, a P-model distributed mixed model is established. Considering that the mixed number in the model belongs to incomplete data, the EM algorithm is introduced to estimate the parameters of the mixed model and the P-model mixed model parameters are derived in detail The estimated iteration formula and the corresponding iteration steps are given. Finally, the mixed Gaussian distribution data, Laplacian distribution and Gaussian distribution mixed data, and the residual data of measured GPS observations are used to verify the correctness and adaptability of the formula in this paper. The results of the calculation examples show that, compared with the single probability distribution, the P-parameter hybrid model can accurately reflect the actual situation of the data distribution, and the model parameters estimated by the EM algorithm have higher accuracy.
Articles just accepted have been peer-reviewed and accepted, which are not yet assigned to volumes /issues, but are citable by Digital Object Identifier (DOI).
Display Method:
, Available online ,
doi: 10.13203/j.whugis20200098
Abstract:
Unmanned Aerial Vehicle; Antarctic Expedition; Climate Change; Photogrammetry; Glaciology; Ecology; Geomorphology
Unmanned Aerial Vehicle; Antarctic Expedition; Climate Change; Photogrammetry; Glaciology; Ecology; Geomorphology
Author Login
Review Login
Editor Login
Reader Login
Editorial Login
Editor-in-Chief Login
Start in 1957 Monthly
ISSN 1671-8860
CN 42-1676/TN
中国中文核心期刊
中国科技核心期刊
Ei核心期刊
Special Recommend
- All visit:
- Today's visit:
