Classification and Automated Filtering of Traditional Wooden Architecture Drawings

This paper presents an AI-driven approach to the classification and filtering of traditional wooden architectural drawings, which are crucial for the preservation and management of cultural heritage. We developed a classifier to categorize hand-drawn architectural plans into six primary types, extending to eight classes by incorporating background elements such as reports and other types of drawings. Various models were trained and evaluated on different configurations, including original and rotated datasets, to enhance robustness against image rotation. Additionally, the Rotation Ensemble Inference (REI) method was proposed and tested to further improve classification accuracy, particularly for rotated images. The experimental results demonstrate the effectiveness of these approaches in accurately classifying and filtering large datasets of architectural drawings, significantly reducing manual verification efforts.