A Spatial Adaptive Algorithm Framework for Building Pattern Recognition Using Graph Convolutional Networks

Area of Science:

• Computer Science
• Artificial Intelligence
• Geographic Information Systems

Background:

• Graph learning, particularly graph convolutional networks, excels at processing topological data.
• Real-world relationships extend beyond adjacency, incorporating spatial correlations like those in building vector data.
• Existing methods may struggle with diverse spatial distributions and require ancillary data.

Purpose of the Study:

• To develop a spatial adaptive algorithm framework for building group division and pattern recognition.
• To create a data-driven, multi-stage approach insensitive to geographical variations in building distribution.
• To leverage deep learning for analyzing building vector data without external information.

Main Methods:

• A spatial adaptive algorithm framework utilizing a data-driven, multi-stage design.
• Integration of graph convolution methods and deep neural networks (DNNs).
• Supervised training on descriptive building vector data for multitask learning.

Main Results:

• The proposed framework demonstrates satisfactory performance in building group division and pattern recognition.
• The method for expressing buildings effectively captures relevant spatial relationships.
• The algorithm framework is robust across different geographical regions and building distributions.

Conclusions:

• Deep learning methods are effective for building group division and pattern recognition tasks.
• The developed spatial adaptive algorithm framework shows significant potential for further research and application.
• The approach offers a powerful tool for analyzing urban spatial data.