Building-level material stock calculation method based on the room-specific material intensity and floorplan inferred by generative adversarial networks (GANs)

Material intensity (MI) considers an entire building as a single unit or a few large blocks. Therefore, current MI-based material stock analysis (MSA) methods have difficulty capturing detailed material stock changes according to design variations in individual buildings and securing reliability. This study introduces room-specific material intensity (RSMI) and proposes a building-level material stock calculation method based on RSMI and floorplans inferred through generative adversarial networks (GANs). Using the RSMI and room-specific areas extracted from the floorplan, the proposed method can calculate material stocks that vary depending on the building design. However, building floorplans on an urban scale are generally restricted. Therefore, the proposed method uses GANs to infer floorplans by analyzing building exterior information, which is obtainable from street view imagery, offering high accessibility and facilitating spatialized analysis. The results show that the proposed method improves the material stock calculation accuracy for the volume and sum of absolute errors.