Abstract:Due to the acceleration of urbanization process in modern society, there are many metropolises throughout the world, such as New York, Tokyo, and Shanghai. Such large cities typically have a carefully designed space planning. For example, the residential area should be distant from the industrial area due to the potential environmental pollution issue. In order to cope with public health emergencies, the infrastructure needed for modern urban governance needs to be improved. Notably, establishing a quality visual model to exploit such sophisticated spatial configurations is an important but challenging task. Such a model can facilitate many applications such as urban planning, environmental evaluation, smart transportation, and urban governance. However, the flexible spatial interactions among multiple regions make it difficult to apply a traditional visual model to encode them. In this work, a quality-guided feature selection framework is proposed to obtain a set of high-quality topologies to model the discriminative structures from different land spatial city regions. Given a city region from a metropolitan area, the well-known super pixel algorithm SLIC is used to decompose each land spatial city region image into multiple atomic regions. Based on this, a binary graph is used to model the spatial interactions among these regions. Each binary graph is then decomposed into multiple subgraphs and a topology selection algorithm is proposed to discover subgraphs with highly discriminative topologies. By leveraging these high-quality subgraphs, the image kernel machine is used to convert the high quality subgraphs from each city region’s image into a feature vector. Afterward, a multi-category support vector machine (SVM) is learned to classify each city region’s image into one particular category. Comprehensive experimental results by comparing with many state-of-the-art have shown the competitiveness of this method. Furthermore, the selected high-quality topologies have demonstrated that highly representative spatial interactions are nicely encoded. The results are of great significance to the establishment of the land spatial database management system.

Determining the optimal generalization operators for building footprints using an improved graph neural network model

NGAT: Attention in Breadth and Depth Exploration for Semi-Supervised Graph Representation Learning

Semantic-Enhanced Graph Convolutional Neural Networks for Multi-Scale Urban Functional-Feature Identification Based on Human Mobility

A graph-based neural network approach to integrate multi-source data for urban building function classification

High-quality Topological Structures Selection for Smart City Land Spatial Understanding and Governance.

Multiscale Urban Functional Zone Recognition Based on Landmark Semantic Constraints

Graph isomorphism network with weighted multi‐aggregators for building shape classification

Inferring building function: A novel geo-aware neural network supporting building-level function classification

Building pattern recognition by using an edge-attention multi-head graph convolutional network

Evaluating the generalization ability of convolutional neural networks for built-up area extraction in different cities of China

Building Simplification Using Backpropagation Neural Networks: a Combination of Cartographers' Expertise and Raster-Based Local Perception

Multi-scale attention integrated hierarchical networks for high-resolution building footprint extraction

Automated Building Generalization Based On Urban Morphology And Gestalt Theory

A road generalization method using graph convolutional network based on mesh-line structure unit

Building Footprint Generation Using Improved Generative Adversarial Networks

Ada-GNN: Adapting to Local Patterns for Improving Graph Neural Networks

Partitioning building groups at multiple scales based on image segmentation

Representation and assessment of spatial design using a hierarchical graph neural network: Classification of shopping center types

Integrating OpenStreetMap Tags for Efficient LiDAR Point Cloud Classification Using Graph Neural Networks

Optimizing the Spatial Scale for Neighborhood Environment Characteristics Using Fine-Grained Data.

Semantic Segmentation-Based Building Footprint Extraction Using Very High-Resolution Satellite Images and Multi-Source GIS Data