There is an increasing need for more efficient and cost-effective design methods, as labor shortages emerge and housing demands rise. Prefabricated buildings have become increasingly popular due to their multifaceted advantages, including accelerated construction timelines, improved quality, and a reduction in material waste. For prefabricated buildings, a similar architectural layout usually indicates a similar structural design, particularly within the same geographical area. Using this correlation enables engineers to reference previous designs instead of starting each project from scratch. However, the traditional approach of manually searching for similar designs in the database can be time-consuming. The labor-intensive nature of this process has prompted the exploration of deep learning as a powerful tool to streamline design research. By utilizing deep learning techniques, designers can efficiently and accurately search the database for previously built buildings that share architectural similarities. In the study, segmentation models were used to automatically extract pertinent information from prefabricated building drawings. A similarity assessment method was proposed to find the most similar buildings according to the key information extracted from the drawings. The results demonstrated the potential and high accuracy of this approach for identifying similar projects from the database.