A transformer-based approach for similar layout retrieval and difference detection in architectural drawings of wood frame buildings

With labor shortages and increasing housing demands, efficient design methods are essential. Prefabricated buildings offer faster construction, better quality, and reduced waste. Typically, similar architectural layouts in prefabricated buildings imply comparable structural designs. Leveraging this correlation enables builders to reference previous designs, thereby reducing design time. However, manually searching databases can be time-consuming. The use of deep learning techniques can expedite this process. Through deep learning, designers can efficiently and accurately search databases for buildings with similar layouts. In this study, a drawing segmentation model was used to extract wall information from layout drawings. Buildings were then grouped into clusters based on this information. Subsequently, a pixel-wise difference method was developed to identify buildings with similar features and to highlight the differences between drawings. Finally, the proposed method was evaluated through two case studies. The results showed the proposed method can achieve acceptable accuracy in finding similar projects and identifying differences.