The continuous generalization of geographic information poses a huge challenge to the classic geographic information analysis modes. Networked knowledge services will gradually become a new mode for geographic information applications, facilitating to transform the form of geographic computing into social computing. Geographic knowledge services need to connect people, institutions, natural environments, geographical entities, geographical units and social events, so as to promote knowledge assisted data intelligence and computational intelligence. Facing the urgent need for spatiotemporal knowledge acquisition, formal expression and analysis, this paper firstly introduces the concepts and characteristics of spatiotemporal knowledge graph. The spatiotemporal knowledge graph is a directed graph composed of geographic spatiotemporal distribution or geo-locational metaphors of knowledge that is a knowledge graph centered on spatiotemporal distribution characteristics. Secondly we proposes a research framework for spatiotemporal knowledge graph. The framework includes various levels from multimodal spatiotemporal big data to spatiotemporal knowledge services that contain ubiquitous spatiotemporal big data layer, spatiotemporal knowledge acquisition technique layer, spatiotemporal knowledge management layer, spatiotemporal knowledge graph layer, software/tools layer, and industrial application layer. Thirdly this paper introduces relevant research progress from text implied geographic information retrieval, heterogeneous geographic semantic web alignment, spatiotemporal knowledge formalization and representation learning. Combined with application practice, we then enumerate the construction and application approaches of domain oriented spatiotemporal knowledge graph. Finally, it discusses the key scientific issues and technical bottlenecks currently faced in the research of spatiotemporal knowledge graph. It is argued that in the era of large models, constructing explicit spatiotemporal knowledge graph and conducting knowledge reasoning to meet domain needs is still the only way for spatiotemporal knowledge services.

Geoscience Knowledge Graph (GKG) has strong capabilities of knowledge representation and semantic reasoning, thereby becoming a required infrastructure for the development of geoscience big data and geoscience artificial intelligence. However, existing studies on GKG were mainly conducted under the experimental scenarios. Because of a lack of research on the general framework of construction methods, sharing, and application of large-scale GKG for practical applications, it has not been used in practical applications in the geoscience field. For this reason, towards the needs of research and applications of geoscience big data and artificial intelligence for GKG, this paper first studied the construction techniques of large-scale GKG. Then, a general framework for covering the lifecycle of GKG including its construction, sharing, and application was proposed. Taking the big science program “Deep-Time Digital Earth (DDE)” as an example, the practice of developing GKG platform towards the practical application of DDE was carried out. Using this platform, this paper realized the construction of DDE large-scale GKG, the open sharing and application of built GKG, proving that the proposed framework can effectively support the construction, sharing, and application of large-scale GKG. This paper plays an important role in promoting the realization of the practical application value of GKG.

The new combat style places new requirements for battlefield environment service support. The intelligent service of battlefield environment urgently needs to improve knowledge based on the global multidimensional battlefield environment data. In view of the knowledge modeling problem of intelligent cognition of battlefield environment, this paper puts forward the classification method of battlefield environment knowledge and considers the battlefield environment knowledge graph as a new form of battlefield environment knowledge representation under the context of big data and artificial intelligence. To solve the fragmentation problem of triplet knowledge representation, a temporal hypergraph representation model of battlefield environment is constructed, a multi-level unified graph model combining entity knowledge, event knowledge, influence process knowledge, and service decision-making knowledge is realized, and all kinds of knowledge are represented as a unified knowledge hypergraph network with spatiotemporal and scene characteristics. Finally, the experimental verification is carried out based on the data of map, event, impact process, and combat impact effectiveness. The hypergraph network realizes the correlation of various battlefield environment knowledge from the semantic level, which can provide support for the further realization of intelligent reasoning and service decision-making based on hypergraph.

The assurance of a consistent supply of daily necessities in megacities is pivotal in fortifying community supply resilience. It is axiomatic that a community system is not an insular entity; rather, it intricately intertwines with various elements of urban systems. As a foundational unit of urban governance, the urban community is instrumental in facilitating a congruent nexus between supply and demand, thereby augmenting urban resilience. This study proposes an exploratory evaluation method for the urban community supply support and resilience based on complex network theory, attempting to achieve a breakthrough in the underlying theoretical framework of resilience assessment from "single system assessment" to "multi-system correlation assessment". Taking the six districts in the central city of Guangzhou as an example, we build a supply-demand network based on citizens' spatio-temporal behaviors using multi-source data such as mobile phone signaling data and other data. The attacking strategies of network are based on five community resilience indicators. Besides, the cascade failure mechanism is introduced to evaluate the network resilience, and the entropy-weighted method is employed to obtain resilience evaluation results. The influence mechanism of community resilience on the supply system is further analyzed by studying the factors affecting community node failure at different stages of supply network. The findings are as follows: (1) The proposed evaluation model of the community supply support and resilience can effectively simulate urban community supply-demand networks and evaluate the resilience of communities. Low-resilience communities are mainly categorized into three spatial types: old blocks, urban villages, and suburban blocks; (2) Through the analysis of network resilience under five different attack strategies, it is found that the dominant influencing factors are different, with the population density being the primary factor; (3) There exists a complex bidirectional relationship between community resilience and supply security, including the obvious vulnerability of low-resilience communities. And the community self-organization ability, the supply facility layout, and the linkage scheduling between supply points all affect the overall community resilience.

With the deepening of global economic integration, maritime traffic congestion and ship accidents occur frequently. In order to supervise and analyze the marine ship activities, the traditional methods mainly use the ship positioning data for data mining without combining other marine multi-source data for the analysis of ship spatiotemporal activity process and behavior pattern, and thus lack deep knowledge mining. Therefore, this paper makes comprehensive use of multi-source data and constructs the ship activity knowledge map based on extracting the semantic information of trajectory, which provides an effective way for the transformation of trajectory spatiotemporal point sequence with low knowledge density to high-order semantic knowledge. Specifically, firstly, by analyzing the characteristics and constituent elements of ship activities, the ontology layer of ship activity knowledge map is designed based on the core idea of "process-event-behavior"; Then, the track semantic information is extracted by Stop/Move model, and the ship emergencies are extracted by DMCNN model to complete the filling of instance layer; Finally, the above model and method are verified by constructing a prototype system. The results show that the ship activity knowledge map constructed in this paper can support the knowledge representation of ship routine activities and emergencies, and realize spatiotemporal activity query and backtracking, so as to achieve the effect of semantic enhancement, which has a certain application value.

Convolutional neural networks have been widely used in the task of Remote Sensing Image Scene Classification (RSISC) and have achieved extraordinary performance. However, these excellent models have large volume and high computational cost, which cannot be deployed to resource-constrained edge devices. Moreover, in the RSISC task, the existing knowledge distillation method is directly applied to the compression model, ignoring the intra-class diversity and inter-class similarity of scene data. To this end, we propose a novel class-centric knowledge distillation method, which aims to obtain a compact, efficient, and accurate network model for RSISC. The proposed class-centric knowledge distillation framework for remote sensing image scene classification consists of two streams, teacher network flow and student network flow. Firstly, the remote sensing image scene classification dataset is sent into the teacher network pre-trained on a large-scale dataset to fine-tune the parameters. Then, the class-centric knowledge of the hidden layer is extracted from the adjusted teacher network and transferred to the student network based on the designed class center distillation loss, which is realized by constraining the distance of the distribution center of similar features extracted by the teacher and student network, so that the student network can learn the powerful feature extraction ability of the teacher network. The distillation process is combined with the truth tag supervision. Finally, the trained student network is used for scene prediction from remote sensing images alone. To evaluate the proposed method, we design a comparison experiment with eight advanced distillation methods on classical remote sensing image scene classification with different training ratios and different teacher-student architectures. Our results show that: compared to the best performance of other distillation methods, in the case of the teacher-student network belonging to the same series, the overall classification accuracy of our proposed method is increased by 1.429% and 2.74%, respectively, with a given training ratio of 80% and 60%; and in the case of teacher-student networks belonging to different series, the classification accuracy is increased by 0.238% and 0.476%, respectively, with the two given ratios. Additionally, supplementary experiments are also carried out on a small data set of RSC11 with few classes and few samples, a multi-scale data set of RSSCN7 with few classes and multiple books, and a large complex data set of AID with many classes of heterogeneous samples. The results show that the proposed method has good generalization ability. Trough the comparison experiments with similar techniques, it is found that the proposed method can maintain excellent performance in challenging categories through confusion matrix, and the proposed distillation loss function can better deal with noise through testing error curve. And visualization analysis also shows that the proposed method can effectively deal with the problems of intra-class diversity and inter-class similarity in remote sensing image scenes.

With the development of science and technology, cyberspace has been deeply integrated with people's daily lives and represents a new spatial form of human activities. The cyberspace correlates to the real world, but on the other hand it also differs from it. Cyberspace has distinct geographical characteristics, and the spatial-temporal relationship in geograph remains an indispensable element in cyberspace. Therefore, it is of great significance to apply geographical thinking to the cognition of cyberspace in order to describe the situation of cyberspace and maintain cybersecurity. In this paper, we review the emergence and development of cyberspace, analyze the basic structure and characteristics of cyberspace, and examine the geographical properties of cyberspace based on different views of cyberspace. From the perspective of the three laws of geography, this paper discusses how to use geographical thinking and Geographic Information Science (GIS) methods to describe cyberspace, and takes the visualization of cyberspace, the construction of geographic knowledge graph of cyberspace, and the intelligent analysis of cyberspace behavior as examples to illustrate how to apply geographical thinking to the analysis and research of cyberspace. Exploring the geographical properties of cyberspace and applying geographical techniques to cyberspace protection can provide new insights into the comprehensive governance of cybersecurity, thus improving the cognitive level and governance capabilities of cyberspace in the new era.

Digital city is one of the main requirements of three-dimensional (3D) real scene and leads the direction of future smart city construction. Digital city usually uses the 3D model of the real scene as the spatial data volume and integrates the object-linked data of various sensors to achieve virtual-real fusion. The integration of spatial data volume and object-linked perception data is the key to digital city applications. Visual sensor is an important sensor type which is widely used in urban life, such as surveillance cameras, vehicles, and other devices. The key to digital city application is registering the visual sensors with virtual 3D model accurately. The purpose of spatial registration for visual sensor is to estimate or optimize the position and orientation of the visual sensor and to get the accurate spatial position of any object in the image. It is one of the key technologies for applications such as Augmented Reality and Video GIS. Currently, the spatial registration methods for visual sensors can be divided into hardware-based and vision-based methods. Due to the popularity of vision sensors, vision-based registration methods have been widely used. However, in digital city applications, seasons and weather always change, there are often large differences in appearance between the real image taken by visual sensor and the image of virtual scene. Therefore, the accuracy of outdoor 6 Degree of Freedom (DOF) position obtained by existing methods is usually insufficient, resulting in low registration accuracy of the visual sensor. In order to improve the accuracy of visual sensor spatial registration in digital city scene, this paper presents a method of virtual-real registration for digital city scene with structural semantic information in urban area. Firstly, the virtual perspective image of digital city scene is obtained, the plumb line which contains structural semantic information is extracted from the target image, and the properties of global constraints of the plumb line is used to restore the camera's position accurately and achieve the registration of monocular image in the virtual digital scene. Experiments show that this method achieves accurate registration of virtual and real images with large differences in appearance. Compared with the existing methods, the position and orientation errors are reduced by 35.9% and 39.3%. This method can effectively optimize the initial pose and improve the registration accuracy of visual sensors in digital city scene. A lightweight cloud-edge registration framework is designed and can be used in image geolocation tasks based on portable devices.

Natural language is an effective tool for humans to describe things, with diversity and ease of dissemination, and can contain human spatial cognitive results. How to use natural language to describe geographic spatial scenes has always been an important research direction in spatial cognition and geographic information science, providing important application values in personalized unmanned tour guides, blind navigation, virtual space scene interpretation, and so on. The essence of natural language description of geographic spatial scenes is the process of transforming the two-dimensional vector of geographic space into a one-dimensional vector in word space. Traditional models perform well in handling spatial relationships, but are somewhat inadequate in natural language description: (1) spatial relationship description models are one-way descriptions of the environment by humans, without considering the impact of the environment on the description; (2) spatial scenes emphasize traversal-based descriptions of spatial relationships, where each set of spatial relationships is equally weighted, which is inconsistent with the varying attention paid by humans to geographic entities and spatial relationships in the environment; (3) the spatial relationship calculation of traditional models is a static description of a single scene, which is difficult to meet the requirement of dynamic description of continuous scenes in practical applications; (4) the natural language style of traditional models is mechanical, lacking necessary knowledge support. This article proposes a spatial scene natural language generation framework Map2Text (M2T) that integrates multiple knowledge graphs. The framework establishes knowledge graphs for spatial relationships, language generation style, and spatial attention, respectively, and realizes the fusion of multiple knowledge graphs and the generation of natural language descriptions of spatial scenes within a unified framework. The spatial scene description knowledge graph solves the pruning problem of traversing spatial relationships, and establishes the relationship between spatial scenes by building a spatial relationship graph, supporting continuous expression of spatial scenes; the natural language style knowledge graph establishes the relationship between spatial expression and language style, achieving diversified language styles that are appropriate for spatial natural language expression; the spatial attention knowledge graph captures the nuances of natural language spatial expression by establishing an attention matrix based on the interaction state between the subject and object of the spatial scene. An experimental prototype system designed based on the Beijing Forbidden City demonstrates that the system-generated results are close to human travel notes, with more complete content coverage and more diverse styles, verifying the effectiveness of the M2T framework and demonstrating the potential value of natural language description of spatial scenes.

With the development of Unmanned Aerial Vehicle (UAV) technology, it has been applied to various tasks in different fields. The prerequisite for a UAV to perform successful aerial tasks is accurate localization of its own position. Generally, traditional UAV navigation relies on the Global Navigation Satellite System (GNSS) for localization. However, this system has disadvantages such as instability and susceptibility to interference, leading to situations where UAV cannot use GNSS for positioning, known as GNSS-denied environments. This study focuses on the navigation and positioning of UAV in GNSS-denied environments and proposes a UAV visual retrieval and positioning method that comprehensively utilizes local and global deep learning features of known satellite orthophotos. Specifically, ConvNeXt is used as the backbone network, combined with generalized mean pooling, to form a retrieval feature extraction algorithm for extracting global features of satellite and UAV images. A triplet loss function considering the overlapping area between images is designed for the retrieval and positioning tasks, and a corresponding training data set is established to train the feature extraction algorithm. Then, the satellite images within a certain range are retrieved according to the extracted global features, and the preliminary retrieval results are obtained. In order to further improve the accuracy of the retrieved target images, the LoFTR algorithm based on deep learning local features is used for matching and reordering. Since the LoFTR algorithm has many mismatches, RANSAC is used to screen the matching results. Experiments using the test datasets we established demonstrate that the proposed method obtains an average accuracy of 90.9% and an average time cost of 2.22 seconds for retrieving satellite images in different seasons from fully overlapped UAV simulated images. The accuracy of the UAV real image test is 87.5%, which can meet the UAV positioning requirements.

China is one of the countries that are most severely affected by typhoons. The direct economic losses caused by typhoons amount to more than 10 billion yuan, and the affected population is more than one million each year in China. Typhoon disasters seriously threaten the social economy and natural environment of coastal areas. As a result, there is an immediate necessity to improve disaster emergency management and comprehensive disaster prevention and mitigation. With the concept of knowledge graph proposed by Google in 2012, it has gradually become a research hotspot in the field of artificial intelligence and played a role in applications such as information retrieval, question answering, and decision analysis. The information integration and representation capability of the knowledge graphs can provide effective support for dynamic monitoring and management decisions of typhoon disaster events. There are problems with the current typhoon disaster models in the representation of spatiotemporal processes. Most disaster knowledge graphs are analyzed for single elements of disaster events, and the research on ontological representation and analysis of disaster development process is still lacking. Firstly, we propose a typhoon disaster events knowledge representation model established from five levels: concept, object, state, characteristic, and relationship, by analyzing the components and dynamic characteristics from the typhoon disaster mechanism. Second, this model considers the multi-granularity of typhoon event information, and unifies the different feature information into each object. Besides, in order to highlight the evolutionary characteristics of typhoon events, the state is taken as the cross-section of the process in a specific spatiotemporal feature. The state is an action or record of an object occurring in time and space. The process is the procedure that properties, forms, and patterns undergo as it gradually changes over time. Finally, we construct a knowledge graph of typhoon disaster events using the 2021 severe typhoon In-Fa as a case study. The results show that the model not only highlights the characteristics of different objects portrayed in the event, but also expresses the spatiotemporal processes of typhoon disaster events through the state sequences of multiple objects. The knowledge graph can be applied as a primary knowledge source in the emergency decision management of typhoon disaster events, which can undoubtedly enable relevant decision-makers to better perceive the spatial and temporal development of typhoon disaster situations. In disaster assessment, based on the rich contents covered in the knowledge graph, it can cope with assessing various aspects such as human casualties, economic losses, and secondary disasters.

Geological and mineral resource survey and scientific research in "geology, geophysics, geochemistry, and remote sensing " have established a large amount of geological and mineral survey data, which contain rich knowledge related to mineralization and distribution of gold mine, such as the metallogenic and tectonic setting, geological environment of occurrence, geological characteristics of mineral mine, genesis and metallogenic model of mine, and so on. The transformation from massive mineral related data to effective metallogenic knowledge has become one of the most important breakthroughs to improve the accuracy of geological prospecting. To solve this problem, through the in-depth analysis of knowledge representation, information extraction, and knowledge fusion in knowledge engineering, this paper explores the knowledge graph construction method of gold mine based on ontology. Firstly, referring to industry norms, gold mine knowledge base, and reference material of geological and mineral resource exploration, the metallogenic model of gold mine is sorted out, and the gold mine concept, gold mine entity, gold mine relationship, gold mine geological attribute, and gold mine metallogenic attribute are determined. In addition, the schema layer of gold mine knowledge graph is constructed by using the top-down ontology knowledge representation method, which represents the conceptual model and logical basis of gold mine knowledge graph. Secondly, based on structured, semi-structured, and unstructured multi-source heterogeneous geological data, the deep learning model is used to realize gold mine knowledge extraction, semantic analysis, and knowledge fusion, which enriches the data layer of gold mine knowledge graph and provides data support for gold mine knowledge graph. The gold mine knowledge graph is constructed in a bottom-up way, and the gold mine knowledge triplet is stored by Neo4j graph database, in which nodes represent gold mine concept, gold mine entity, and gold mine attribute value, while edges represent relation and attribute. Finally, the gold mine knowledge management system is developed based on the graph database. It can be applied to the management of gold mine data, acquisition of knowledge, visualization representation of gold mine knowledge graph, inquiry of knowledge, management and presentation of knowledge base, and other functions well, so as to lay a foundation for the intelligent analysis and mining of geological big data. This study develops a geological prospecting method driven by data and knowledge, and provides a reference for identifying, controlling, and managing mineral resources, which can improve the prospecting accuracy in geological exploration.

Wetland is of great significance to biodiversity and climate change, and it is also one of the basic living environments of human beings. In order to better understand and express wetland knowledge and the relationship between classifications, this paper proposes an ontology-based wetland knowledge graph construction method. Based on the land cover classification system of GlobeLand 30, this paper establishes the conceptual structure of wetland data and the rich semantic relationship between the elements around wetland type definition, spatial pattern, case distribution, and trend change. Firstly, based on the prior knowledge of wetlands, taking the wetland types in the GlobeLand 30 classification system as an example, we analyze the wetland domain knowledge around the wetland types, feature distribution, and other elements, extract the semantic relationship between knowledge, and construct the ontology database of wetland knowledge by combining top-down and bottom-up methods. The conceptual framework of wetland knowledge graph is formed through ontology modeling. Secondly, based on the wetland knowledge automatically extracted from the technical specification text and encyclopedia website, the extracted conceptual knowledge is stored in the model layer, and the data layer is constructed from bottom to top. The main contents include knowledge acquisition and knowledge fusion. According to the concepts contained in wetland knowledge, the relationship extraction of wetland knowledge is carried out, mainly including attribute relationship, spatial relationship, and temporal relationship. Using the wetland directory crawled from the wetland China website, the wetland entity name and knowledge are directly extracted from Baidu Encyclopedia by means of web crawler to form a triple. Finally, Through the above construction processes of wetland knowledge graph, the wetland related data with different structures are transformed into structured knowledge triple data, and the graph database Neo4j is used for semantic relationship storage with the "node relationship" storage model. Knowledge graph provides a new idea for the study of rich knowledge representation and storage in the field of land cover. It is a bridge between the basic geographic data of surface coverage and spatial knowledge service. It is of great significance to promote the sharing and reasoning analysis of surface coverage data. Taking the wetland land cover type as the research example, the knowledge graph constructed in this paper expands the conceptual description information of wetland entities, explores the wetland knowledge representation method by considering the temporal and spatial characteristics, and provides a new perspective and application demonstration for the expression of land cover knowledge.

Geometric and semantic integration of 3D building models are important infrastructure data for smart city, they are conducive for promoting the refined management and intelligent application of building facilities. However, most of the existing point cloud-based modeling methods focus on the reconstruction of geometric models with simple roof structure, and semantic and topological relations are ignored. Moreover, these methods are sensitive to noise, which are difficult to assure topological consistency and geometric accuracy. To solve these problems, this paper proposes a 3D primitive fitting algorithm for automatically reconstructing building models with complex roof structure from point clouds. Firstly, a 3D building primitive library is designed, including various 3D building primitives with simple and complex roof types. Secondly, an individual building point cloud input is segmented into multiple planes using RANSAC algorithm. The Roof Topology Graph (RTG) is then generated according to the relationship of roof planes, and the roof type of point cloud is subsequently recognized by comparison of RTG between point cloud and building primitives. Thirdly, the reconstruction is formulated as an optimization problem that minimizes the Point-to-Mesh Distance (PMD) between the point cloud and the candidate meshed building primitive. The sequential quadratic programming optimization algorithm with necessary constraints is adopted to perform holistically primitive fitting, so as to estimate the shape and position parameters of a 3D primitive. Finally, the parameterized model is automatically converted into City Geography Markup Language (CityGML) building model based on the prior 3D building primitive. The generated CityGML LoD2 (second level of detail) models are different from mesh models created by conventional building modeling methods, which are represented with geometric, semantic, and topological information. To evaluate the quality and performance of the proposed approach, airborne lidar and photogrammetric building point clouds with different roof structures are collected from public datasets for test. Several building models with complex roof types are successfully reconstructed by using this approach, and the average PMD of five models is 0.17 m. The proposed algorithm is also compared with three other methods. Experimental results indicate that the proposed method achieves the best geometric accuracy, because the average PMD of each model is less than that of other methods. Moreover, this automatic primitive fitting method is efficient, and it is also robust to noise and local data missing. This study demonstrates that the resulting building models can well fit the input point cloud with topologic integrity and rich semantic. This method provides great potential for accurate and rapid reconstruction of geometric-semantic coherent building models with complex roof condition.

The graph structure-based knowledge graph plays important roles not only in the description and reasoning of semantic network, but also in the structured abstraction and spatial reasoning of spatial entities. The relational information of spatial entities is recorded in edges in the knowledge graph. Through the edge-based knowledge graph computational reasoning such as path detection, sub graph alignment, pattern discovery, etc., it can play an important role in spatial scene cognition. Geographic knowledge graph is a knowledge system that formally describes geographic concepts, entities, and their interrelationships. It has both the connotation and characteristics of general knowledge and the specific spatiotemporal characteristics of geographic knowledge. It can connect semantic models with spatiotemporal models to describe semantic relations, spatial relations, and temporal relations, and has great application potential in the expression, understanding, acquisition, and reasoning of geographic knowledge. The existing research work of geographic knowledge graph is mostly focused on semantics, and the extraction and expression of semantic relations are very rich and comprehensive, which can support further functions such as semantic search and association analysis of geographic knowledge. However, the knowledge expression of geographic knowledge graph in spatiotemporal model is relatively lacking, and the existing spatial relationship is limited between elements, rarely involving the further distribution situation and spatial pattern in spatial cognition. Thus, the geographic knowledge graph needs to be strengthened in terms of spatial semantic knowledge. Based on the principle of knowledge graph construction, this paper takes the construction of geographic knowledge graph of buildings as an example to realize the grid-pattern recognition of buildings. Firstly, the buildings are abstracted into entities and expressed as nodes of the graph, and the spatial neighborhood relations between buildings is extracted based on geometric proximity analysis, so as to build the geographic knowledge graph of the building group. On this basis, combined with the domain knowledge of building pattern recognition, it further infers and constructs other spatial semantic relations, and improves the geographic knowledge graph. Then the grid-pattern of the buildings complex scene is expressed as the rules of the knowledge graph, which is based on NoSQL language for reasoning. The results show that this method can effectively extract the linear pattern of buildings and further deduce the grid-pattern, which demonstrates the important role of geographic knowledge graph in spatial reasoning and its good adaptability in domain problem research, and provides ideas for the application of geographic knowledge graph in the field of spatial cognition.

With the development of big data and artificial intelligence, the scope of digital earth modeling has extended to full-time holography beyond the earth surface. However, the current data model of digital earth still remains in the data modelling of earth tile or grid subdivision. This model severely limits the application of scenario-based and intelligent digital earth development. This paper proposes the concept of digital holographic earth and a corresponding data organization model of earth data cube. By using global multi-level grid reference system to describe and express multi-scale space and using two or three-dimensional grid cells to describe spatial positions, the traditional spatiotemporal description of "longitude, latitude, elevation, and time" is transformed to a new spatiotemporal description system of "time granularity, time coverage, grid position, and grid scale". The proposed model is characterized by the dimensions of "time-space-scale-attribute" based on spatiotemporal big data in the digital earth. The model encapsulates vectors, rasters, grids, time series arrays, and 3D models into an unified system. This unified system ensures that any data value of a specific earth data cube is aligned perfectly in time, space, and scale, which solves the problem of multi-dimensional or spatiotemporal dynamic fusion of big earth data. Finally, this paper develops a deep-time and spatiotemporal dynamic visualization simulation system to verify the data model based on the requirements of the Deep-time Digital Earth International Science Program.

Knowledge graphs are an important data infrastructure in AI technologies and applications, and have become a hot research topic in geosciences. The size and topological features in geographic knowledge graphs are usually different from universal knowledge graphs, which are not typical small-world networks. However, existing studies often use the default network search depth when learning geographic knowledge graph representations, and its rationality needs further demonstration. For this purpose, this paper constructs a metro travel knowledge graph based on the topological structure features of metro line network, combined with passenger flow data, POI (Point of Interest) data and built environment data, etc.; then GraphSAGE model is used to learn node multidimensional feature embedding and combine POI data for semantic recognition of station classification results to verify the suitable network search depth for metro travel knowledge graph. The results showed that, compared to the default 2 layers search depth, the node embedding features of this metro travel knowledge graph work optimally when the search depth is 3 layers. This study shows that the hyperparameter selection of the geographic knowledge graph representation is supposed to take into account the geographic features, and it is important to avoid the use of results from fields such as computer science that have not been distinguished. When the search depth is 3 layers, the metro station classification results are also more reasonable and explanatory, which can provide a basis for station planning and passenger flow prediction using knowledge graph and AI methods.

Named Entity Recognition (NER) is the basis of many researches in natural language processing. NER can be defined as a classification task. The aim of NER is to locate named entities from unstructured texts and classify them into different predefined categories. Compared with English, Chinese have the features of flexible formation and no exact boundaries. Because of the features of Chinese and the lack of high-quality Chinese named entity datasets, the recognition of Chinese named entities is more difficult than English named entities. Fine-grained entities are subdivisions of coarse-grained entities. The recognition of Chinese fine-grained named entities especially Chinese fine-grained geographic entities is even more difficult than that of Chinese named entities. It is a great hardship for Chinese geographic entity recognition to take both accuracy and recall rate into account. Therefore, improving the performance of Chinese fine-grained geographic entities recognition is quite necessary for us. In this paper we proposed two Chinese fine-grained geographic entity recognition models. These two models are based on joint lexical enhancement. Firstly, we injected the vocabulary into the experimental models. The vocabulary was considered as the 'knowledge' in the models. Then we explored the appropriate fine-grained named entity recognition method based on vocabulary enhancement. And we found two models, BERT-FLAT and LEBERT, that were suitable for fine-grained named entity recognition. Secondly, to further improve the performance of these two models in fine-grained geographical named entities recognition, we improved the above two models with lexical enhancement function in three aspects: pre-training model, adversarial training, and stochastic weight averaging. with these improvements, we developed two joint lexical enhancement models: RoBERTa-wwm-FLAT and LE-RoBERTta-wwm. Finally, we conducted an ablation experiment using these two joint lexical enhancement models. We explored the impacts of different improvement strategies on geographic entity recognition. The experiments based on the CLUENER dataset and one microblog dataset show that, firstly, compared with the models without lexical enhancement function, the models with lexical enhancement function have better performance on fine-grained named entities recognition, and the F1-score was improved by about 10%; Secondly, with the improvements of pre-training model, adversarial training, and stochastic weight averaging, the F1-score of the fine-grained geographic entity recognition task was improved by 0.36%~2.35%; Thirdly, compared with adversarial training and stochastic weight averaging, the pre-trained model had the greatest impact on the recognition accuracy of geographic entities.

Most of the precipitation datasets in Central Asia have problems such as data missing, geographical bias and outliers, low resolution, and so on. The normal prediction results obtained by most machine learning methods are usually hard to interpret, not only due to the uncertainties from input information but also due to the complicated global geographical environments as well as the underlying local geographical conditions. In this paper, to overcome this problem, we proposed a novel downscaling precipitation model to adjust and optimize the precipitation computation results from Conditional Generative Adversarial Networks (CGAN) using an inverse distance weighting method based on the prior information of geographical differences of local digital terrain model and multiple weather stations. In this study, the Amu Darya River Basin was selected as the research area due to its various geographical environment and complicated topographic and geographical conditions. First, the input Climate Research Units (CRU) precipitation data with 55 km resolution were spatially corrected based on the topographic map using the spatial deformation model. The spatial deformation model was extended from spatial transformation network methods. Second, we input the corrected CRU precipitation data, temperature, wind speed, humidity equivalent data, and remote sensing data to the CGAN computing framework for high-resolution precipitation reconstruction. The experiment adopted the cross-validation method, taking 80% of the data as the training set, and the remaining 20% as the verification set. The test set contained 20 raster maps of annual precipitation from 2000 to 2019. The model was built based on pytorch 1.10.0, the batch size was 16, and the learning rate was 0.000 3. The epoch was 8 000 iterations in the Adam optimizer for gradient descent. Finally, the precipitation data of meteorological stations were used as the true values for analyzing the geographical differences of inverse distance weights and the accuracy of the corrected precipitation grid data. The results show that the proposed method can improve the resolution and accuracy of precipitation data，especially for the complex terrain and mountainous area. And Experiments on the Amu Darya in Central Asia show that the Root Mean Square Error (RMSE) of the downscaling result within the watershed was 15.96 mm, the Mean Absolute Error (MAE) was 11.82 mm, the R² value was 0.83, and the deviation was 0.08. This study provides a robust, accurate method for improving the spatial resolution of precipitation data in complex geographical areas.

The development of financial network has profoundly changed the form of knowledge spillover between cities and further affected the level of urban economic growth. The research on the externality of financial network has gradually become a research hotspot in financial geography. With the assistance of the data of headquarters and branches of enterprises in China's financial industry from 2005 to 2020, this paper combines the methods of social network analysis and recursion thought to calculate the centrality of financial network to analyze its spatial and temporal distribution pattern. Meanwhile, we use the OLS model to analyze the influence mechanism of financial network centrality on urban economic growth in China. The study finds that: (1) From 2005 to 2020, the centrality of financial network showed a "core-edge" distribution pattern, but the spatial agglomeration degree of financial network was decreasing. In this paper, with the help of the Core/Periphery algorithm in UCINET software, cities are divided into core and peripheral cities according to the 2020 financial network，and finally 52 cities with core status, such as Beijing, Guangzhou, and Hangzhou, and 239 cities with peripheral status, such as Anshan, Binzhou, and Foshan, are identified. Cities in the eastern region play an important role in the allocation of financial network resources, forming a "core-periphery" pattern with Beijing, Shanghai, and Shenzhen as the core cities and the other cities as the periphery cities； (2) The development of financial network can not only alleviate financing constraints, but also affect urban economic growth by promoting knowledge spillovers under the influence of network externalities. Cities with larger network links and better accessibility have higher levels of economic growth, and the knowledge spillover effect is more dependent on network proximity than location advantage in "place space"; (3) The degree of knowledge spillover promoted by financial network centrality shows spatial heterogeneity, and the core city which plays the role of "knowledge gatekeeper" can obtain greater benefits from network links. Due to the lack of "knowledge gatekeepers", peripheral cities are unable to make full use of external resources, which exacerbates the risk of being at the low end of the value chain and finally leads to the economic growth level of core cities much higher than that of peripheral cities. In the future, we should attach great importance to the construction of financial network and give full play to the driving role of financial network in urban economic growth in China.

There are various reasons for the inconsistency among land cover products, such as the difficulty in determining pixel into categories at different image resolutions. The inconsistency caused by image resolution is determined by the attributes of the image itself and cannot be changed by increasing the number of samples, improving classification methods, and other measures. Especially in mountainous and hilly areas with high ground fragmentation, the resolution has a greater impact on the quality of land cover products. Therefore, taking the cultivated land of GlobeLand 30 and WorldCover products as an example, this paper introduces information entropy to analyze the consistency of multi-source land cover products. Firstly, the information entropy of different bands of the original images in the cultivated land is calculated to reflect the overall uncertainty of the cultivated land category. Secondly, the local information entropy is constructed to describe the local uncertainty of the images. Finally, the uncertainty of the two products is overlapped and the consistency is analyzed. The results show that: (1) The information entropy can reflect the uncertainty changes with more details in the cultivated land based on the spatial distribution of products, and the regional distribution of categories that are difficult to determine can also be easily detected; (2) The overall uncertainty of GlobeLand30 cultivated land with 30m resolution is greater than that of WorldCover cultivated land with 10 m resolution. The large uncertainty of the two products mainly exists in the transitional areas between cultivated land and other ground features, where there are more mixed pixels and fuzzy ground feature categories; (3) Based on the overlapped uncertainty of the two products, the uncertainty of cultivated land is relatively low in products with higher resolution and relatively high in products with lower resolution. The difference of cultivated land uncertainty caused by resolution difference are that 34.54% in blue band, 51.13% in green band, 46.03% in red band, and 61.48% in near-infrared band; (4) For areas where the spatial distribution of the two products is inconsistent, the uncertainty of GlobeLand30 product is greater, which means it is greatly affected by the resolution, while the uncertainty of WorldCover product is smaller, which means it is less affected by resolution.

Image geo-localization is a technique that obtains the geographic location information of an image through a series of methods, so as to establish a mapping relationship with the real geographic space. This technique is important for further image information mining and has potential application value in cyberspace surveying and mapping, intelligence acquisition, user outdoor positioning, and augmented reality. Despite the tremendous progress in the field of computer vision, high-precision automatic geo-localization of images still needs to be addressed due to the involvement of multiple fields such as image feature extraction, large-scale data retrieval, large-scale point cloud processing, deep learning, geographic information feature extraction, geometric modeling and reasoning, semantic scene understanding, context-based reasoning, and multiple data fusion. This paper reviews the progress of image geo-localization research, mainly including image geo-localization methods, image geo-localization datasets, image geo-localization evaluation methods, and summary and prospect of image geo-localization. Firstly, image geolocation methods are classified into three categories, i.e., image retrieval, 2D-3D matching, and cross-modal retrieval, according to the relevance of the research content. Secondly, the datasets and evaluation methods used for image geo-localization research are categorized and summarized. The geo-localization datasets include image datasets, cross-view datasets, Structure from Motion (SfM) datasets, and multimodal datasets, etc. The image geo-localization evaluation metrics include Top-k candidates, localization error, position and orientation error per video frame, and accuracy/recall. Finally, the current status of image geo-localization research is analyzed, and the future research directions of image geo-localization are outlined in terms of global geo-localization, natural area geo-localization, multi-method fusion for geo-localization, Point of Interest (POI) data-based geo-localization, and pre-selected location refinement.

Location description is the natural language expression of human spatial cognition. Since natural language is the primary and basic means of information transmission in human society, location description is an important medium for transmitting spatial location information in human communication. Spatial positioning based on spatial location description is the key to intelligent transformation of location-based services in the era of big data. To solve the problem that the vagueness of location description in different contexts is significantly different and results in difficulty in positioning, this paper proposes a representation method and reasoning mechanism for vague location description. Firstly, by combing the law of human spatial cognition, the types of elements concerned in the description of natural language are clarified. Based on the analysis of the sources of vagueness, a formal representation of vague location description is constructed. Different from the traditional spatial information modeling which focuses on spatial relationship, the formal representation proposed in this paper establishes the vagueness relation and influence among different information factors by the strategy of multi-factors representation. The formal representation also enhances the semantic analysis ability for the vagueness of location description. Secondly, based on supervaluation theory, the reasoning mechanism of vague location description is proposed from three aspects: spatial object, distance relation, and direction relation. Considering the context semantics of spatial location description, the threshold of observation value is used to carry out spatial reasoning. By being super-valued to different contexts, the reasoning results in different situations are obtained. The aim of the reasoning mechanism is to establish the mapping relationship between vague location description and real spatial location. Thirdly, a Question-Answering (Q&A) system is designed to collect contexts of location description, and a case study on the method is conducted. In the case study, a group of users' viewpoints from Q&A on spatial cognition are transformed into the spatial scope in the real world. These spatial scopes can establish the relationship between qualitative spatial concepts and quantitative spatial data, so as to realize the representation of vague location description in GIS. The results show that the proposed method in this paper can adjust the granularity of formal representation of location description in time according to actual application scenarios, and the spatial reasoning results fit intuitive cognition. In the future, knowledge graphs will be introduced to further improve the semantic reasoning ability and positioning accuracy for vague location description.

The term "equality " refers to the fairness of rights or resource distribution among social group members, which is crucial for traffic planning. The accessibility of public transport reflects whether residents of corresponding geographical analysis unit have reasonable access to public transports given the size of their population. Public taxi service (including both ride-hailing taxi and traditional taxi) is the primary means to meet urban residents’ personalized daily demand for public transport. Exploring the equality of public taxi services can provide support for optimizing personalized travel service in urban public travel. In this study, based on the taxi trajectory and dynamic population data of Xiamen, we used Gini coefficient and Theil index to calculate the equality of taxi service for three types of taxi service, i.e., cruise only, ride-hailing only, and mixed receiving taxi service. Then we analyzed the contribution of urban functional areas to equality of tax service and extracted areas with potential inequity. The results show that the proposed method can quantify the inequality of different taxi service and its changes under the context of COVID-19, based on the reclassification of taxis service. Specifically, (1) after the COVID-19 outbreak, the number of vehicles providing ride-hailing service decreased by 18.8% in Xiamen, with a 10% increase in service inequality relative to the dynamic population, while the number of taxis using cruising only to pick up orders remained relatively stable, and the inequality was also at a high level constantly; (2) In Xiamen city the overall inequality of ride-hailing only service was 66% that of cruise only service. The use of online car hailing platform improved the fairness of personalized public travel services for urban residents; (3) the equality of taxi service differed significantly among different types of urban functional areas. For cruising only service, 57.49% of the overall inequality was contributed by commercial land use in Xiamen. For ride-hailing only service, industrial and tourism land contributed 35.07% and 19.32% of the inequality, respectively. After the COVID-19 outbreak, the contribution of tourism land to the inequality of online shopping service increased significantly, by 81.67% compared to the pre-epidemic period; (4) Xiamen City contains two parts: inside-island and outside-island. The number of potentially inequitable service areas was much larger outside-island than inside-island. Areas with a low level of passenger service but a high level dynamic population are typically located in the island's natural parks, while areas with a high level of passenger service but a low level dynamic population are located beyond the island's natural parks; (5) Compared with the results based on the dynamic population, the inequality based on the static population assessment was overestimated by up to 89.25%. The research results of this paper demonstrate that the theoretical framework and analysis method can systematically reveal the inequality characteristics of public taxi service, providing guiding principle in urban traffic optimization.

The converting evolution of cascading disaster scenario refers to that in the process of disaster scenario evolution, the disaster bearing bodies transform into new disaster hazards, forming a disaster chain. Rainstorm can easily cause serious secondary disasters such as waterlogging, debris flow and flood, and the combination of these secondary disasters will make the city more vulnerable. However, existing research on rainstorm cascading scenario deduction lacks the analysis of specific scenario evolution situations such as multi disaster combination, scenario element converting, and human-induced emergencies. Meanwhile, traditional research often relies on the probability inference based on existing scenario evolution networks, without providing a construction method for scenario evolution networks, making it difficult to adapt to the knowledge requirements of actual scenario situation converting deduction. To address the scenario converting evolution problems of urban rainstorm cascading disasters, this paper proposes a scenario converting deduction method for rainstorm cascading disaster response based on multi-source spatial data and probability analysis tools. First, based on local and non-local historical emergency cases, the scenario elements involved in the rainstorm cascading disaster scenarios and their potential converting paths are identified. Next, with the support of Baidu Encyclopedia and Wikipedia network knowledge resources, relevant scenario element features and their associations are extracted, and a Group Lasso machine learning method is adopted to achieve feature selection of involved scenario elements. Then, considering the multi-stage and complex scenario correlation in the process of cascading scenario evolution, a dynamic Bayesian network model for scenario converting deduction is constructed. Finally, a Markov chain Monte Carlo method is used to solve the Bayesian network and generate the converting probabilities. The proposed method is applied to the rainstorm response practice of Wuhan High-tech Zone. The use case results show that the proposed method can combine historical cases and network data to achieve rapid and effective generation of key scenario elements and their features, helping to improve the reliability of scenario converting deduction. At the same time, the proposed method supports the scenario converting deduction of small-scale disaster-bearing bodies such as geographic grids, which helps to provide more accurate rainstorm emergency decision-making support and provide good performance in visual analysis. The uncertainty analysis of the proposed method shows that the precision of original probabilities of key scenario element features and the size of generated geographic grids significantly affect the scenario converting deduction results. These findings provide important information for the local area and are expected to help the rainstorm disaster management of other jurisdictions.

The multispectral LiDAR system can simultaneously provide the 3D space and spectral information of the target ground object, which is convenient for ground object recognition, land cover/use classification, and scene understanding. However, most multispectral LiDAR point cloud classification methods cannot fully mine the geometric information of point clouds and achieve poor performance in fine-scale classification. To overcome this limitation, this paper presents a continuous kernel point convolutional network which uses local point cloud geometric information to enhance features. Firstly, the network combines a random sampling with a farthest point sampling to quickly process large-scale multispectral LiDAR point clouds. Then, an enhanced convolution module based on continuous variable convolution is designed to improve the semantic information expression of multispectral LiDAR point cloud data. In order to address the problem that kernel point convolution simply using the distance relationship between the geometric space and feature space of neighboring points and centroids is insufficient to express the local information as a complementary feature of the kernel point convolution network, the local features given to the kernel points are enhanced by using the position relationship between neighboring points and centroids while aggregating the local features to provide richer semantic information for the multispectral LiDAR point cloud classification network. Finally, the weighted label smoothing loss and the Lovasz-Softmax loss are combined to further improve the classification performance. The results on the Titan multispectral LiDAR dataset show that the proposed network achieves an overall accuracy of 96.80%, a macro-F₁ index of 88.51%, and a mIoU value of 83.42%, which is superior to the state-of-the-art (SOTA) multispectral LiDAR data networks. The proposed model uses the combination of grid sampling and KD-Tree to better preserve the geometric features of the original point cloud. In the case of a single batch of 65,536 points, the point cloud sampling time is reduced by 28 261.79 ms compared with similar multispectral LiDAR point cloud classification networks. This Study demonstrates the potential of enhanced feature kernel points convolutional network for multispectral LiDAR point cloud classification tasks.

The spatial distribution of world population varies considerably from region to region due to differences in geography and economic development, which requires a specific scale of population grid for population spatialization in different regions. A uniform grid scale can only represent the spatial distribution of population at one grid scale, which leads to poor representation for more densely or sparsely populated areas. In order to express the spatial distribution of the population more accurately and determine the appropriate scale of the population grid for a given area, this paper proposes a method to determine the scale of suitable grids based on a multi-indicator analysis. We construct an innovative index system based on the shape-centered difference distance, by combining spatial suitability, numerical suitability, and spatial relationship suitability, and use spatial autocorrelation, coefficient of variation, and geographic detector model to get rid of the scale effect of landscape index from the spatial and numerical perspectives. In this paper, the process of population spatialization employs geographic country data combined with occupancy spatial attributes (i.e., number of building floors and floor area). The suitable scale threshold of the grid under each index is determined based on the analysis of the change of the characteristic points of each index, and then the common scale threshold is identified as the overall suitable grid scale. The geometric mean of the absolute relative error against the real population value is calculated to verify the index analysis results and the suitable scale of population grid. This paper takes the Guye District of Tangshan city in the Capital Economic Circle as the empirical analysis area. The results of the index system analysis show that the differences between clusters of population patches gradually become smaller and their distribution patterns become less expressive as the scale of the girds increases, and the overall expressiveness of the population grid gradually decreases as the scale increases. From perspectives of spatial suitability, distribution morphology suitability, numerical suitability, and spatial relationship suitability, 90 m and 100 m are identified as the suitable grid scales, under which the population spatialization could obtain better quality. The validation analysis demonstrates that the method of determining suitable grid scale for population spatialization in this study has a certain degree of validity and reliability.

Terrain fragmentation is an important factor that can result in poor spatial connectivity and accessibility in mountainous areas, which seriously restricts regional transportation accessibility and urban-rural integration development. This study takes the Loess Plateau with dense valleys and highly fragmented terrain as an example and constructs a terrain fragmentation index system based on DEM data for overall fragmentation, positive and negative terrain, and transition terrain. We use the spatial clustering method (Automated Zoning Procedure-Simulated Annealing) and the objective weighting method (Criteria Importance Though Intercrieria Correlation) to generate spatial zoning and evaluation grading of topographic fragmentation at the county level and explore the spatial heterogeneity of topographic fragmentation on the Loess Plateau. The results show that: (1) The general spatial distribution of terrain fragmentation indicators are characterized by contiguous clusters, among which high values of Elevation Standard Deviation and Terrain Relief are mainly located in the Longzhong Plateau and the area near the Qinling Mountains, and the low and medium values are mainly distributed in the north of Liupanshan-Weihe River; (2) The Loess Plateau can be divided into eight spatially continuous topographic fragmentation zones, and the four largest zones are located in the northern Shaanxi Plateau within the central part of the Loess Plateau, the Lvliang Mountains, and the Ordos Plateau in the north. They account for 66.37% of the total area and are distributed in a shape of Chinese character "田", while the rest of the subregions located along the western and southeastern edges of the plateau are small areas with long and narrow shapes. The spatial differentiation of the topographic fragmentation of the Loess Plateau has a simple characteristic in central area and are complex at the edge area; (3) The terrain fragmentation degree of the Loess Plateau can be divided into five levels. The highest fragmentation area accounts for 13% of the total area of the plateau, which is mainly distributed in the Longzhong Plateau and the northern part of the Qinling Mountains. And 55% of the total area of the plateau presents an spatial pattern of high in the west and low in the north, which is mainly distributed in the northern Shaanxi plateau and Shanxi plateau located in the central and eastern part of the plateau. This study provides references for the formulation of urban-rural integration development policy and transportation infrastructure planning in the Loess Plateau.

The vegetation and soil fraction in the sensor's field of viewing will be varied with different light and observation geometry, and such variation can be used by the remote sensing geometric-optical model to simulate canopy multi-angle reflectance. As a result, the four components, i.e., lit and shaded vegetation, as well as lit and shaded soil are important input parameters for the geometric-optical model. In this paper, an algorithm for extracting the four geometric-optical components with the combination of solar illumination information and multi-scale clusters derived from a k-means process was proposed. Firstly, the clustering space was formed by synthesizing a new color index, then the multi-scale image hierarchical model was constructed by using the lit and shaded component in the subgraphs of the images respectively, and then the k-means clustering was performed in the multi-scale image hierarchical model to obtain the vegetation component and soil component results. Finally, the obtained results in the above subgraphs were combined as the output to achieve the extraction of four geometric-optical components. Validation on the proposed method was conducted on fifty-two vegetation canopy images which were acquired under natural lighting conditions. We compared our results with those of OTSU threshold on ultra-green index, Fisher linear algorithm, and SHAR-LABFVC algorithm. The results showed that the proposed algorithm performed well in mapping accuracy and user accuracy in the classification of shaded components, and the highest Kappa coefficient (0.82) was achieved. Good and stable classification results were observed under the conditions of continuous changing canopy cover and solar altitude angle, and this promising result suggests that the proposed method has the potential in long-term vegetation monitoring as well as measuring vegetation four-component changes even in a single day. The advantages of this algorithm are to improve the classification accuracy of the shadow component and to solve the extraction problem of the four components under high vegetation coverage. However, reducing computational cost and thus to improve the applicability of this algorithm in complex scenes will need further efforts in the future work.

The road boundary extraction from the vehicle-borne LiDAR point cloud can be easily affected by the occlusion of vehicles and pedestrians in urban environment. These occlusion phenomena will cause two problems: one is the generation of pseudo-boundary points, and the other is the discontinuity of boundary lines. We find that the elevation standard deviation constraint can effectively deal with the problems caused by occlusion. First, data pre-processing is carried out, including point cloud subdivision, cloth simulation filtering, and scanline storage. Second, a continuous double window is established based on scanlines. The bidirectional moving window method is adopted to construct elevation difference constraint, angle value constraint, and elevation standard deviation constraint to obtain candidate road points. Then, based on the continuity of road boundary, the DBSCAN algorithm in density clustering is used to generate relatively continuous and accurate road boundary points. Finally, the cumulative curvature value and distance of the boundary breakpoint region are calculated to determine whether a location is a junction. The boundary breakpoint is not considered connected if it is a junction; otherwise, it is regarded as a breakpoint caused by vehicle or pedestrian occlusion. The quadratic polynomial curve is used to smooth the boundary points, and the mathematical parameter model of the boundary is obtained. The experimental results show that the accuracy of road boundary extraction can exceed 80% in an urban environment with more occlusion.