Using semi-supervised machine learning to assist classification and recognition of Chinese vernacular architecture

In order to construct a more refined, comprehensive and systematized classification system for Chinese vernacular architecture, this paper proposed a classifying method based on semisupervised machine learning using architecture image data. By adding type labels based on experts' prior knowledge as a guide to part of the training samples, it allows generalized machine clustering to effectively combine image feature mining with characteristics of special objects. Taking the scheme of machine clustering as a reference, a classification system for Chinese vernacular architecture with 9 major categories and 23 subcategories was established, which boasts extensive coverage of vernacular architectural types and has strong distinguishability. This method integrates the advantages of expert knowledge and computer algorithm, and can be utilized as a reference for studies where unsupervised or fully supervised machine learning is challenging to apply. In addition, the YOLOv8 object detection algorithm of deep learning is also applied to construct a corresponding recognition model for these architectural categories, which performs well in terms of recognition accuracy and robustness. This tool holds significant potential for application in heritage protection and urban and rural planning.