Abstract：In response to the challenges posed by the large sample and the scarcity of relevant drawing and documents for the study of Singapore’s shophouse types, this paper proposes a method for detecting and identifying building types based on satellite images, integrating a general large model and image deep learning algorithms. Utilizing the SAM general large model, rapid extraction of semantic information from urban satellite image elements is achieved. Employing the Mask R-CNN framework, high-precision segmentation and identification of building types are accomplished. The paper focuses on Singapore’s Chinese Street shophouses as the subject, validating the improved Mask R-CNN building type recognition framework. The results indicate that this method achieves an accuracy of 90.9% in recognizing shophouse building types, demonstrating high robustness and recall rate. The paper provides technical examples and support for automatically and efficiently detecting and identifying residential building types, and discusses the application prospects and current limitations of this method in similar research.
李泽辉1，2，朱宁宁3，赵冲4*，聂复丹1，赵逵5. 基于卫星图像深度学习的建筑类型识别方法研究
——以新加坡“店屋”建筑为例[J]. 新建筑, 2025, 43(2): 93-98.
LI Zehui, ZHU Ningning, ZHAO Chong, NIE Fudan, ZHAO Kui. Research on Building Type Recognition Methods Based on Satellite Image Deep Learning: The Case of Singapore's Shophouse Architecture. New Architecture, 2025, 43(2): 93-98.
版权所有 © 2017 新建筑杂志社