Material Detection and Classification in 2D Architectural Drawings Using Computer Vision for Material Passports

Abstract: Retrofitting Europe’s ageing buildings is critical for sustainable development, yet traditional digitalization of architectural drawings, such as manual drafting, remains costly and inefficient. This paper presents a novel approach using Computer Vision (CV) to automate material detection and classification in drawings for simplified Material Passports (MP). Three convolutional neural network (CNN) architectures, including U-Net, ResNet50, and MobileNetV2, are evaluated, with MobileNetV2 outperforming for mixed-source data set despite some overfitting on well-represented materials. Case studies validate the model’s practical effectiveness, especially for reinforced concrete elements, showing this CV-based method’s potential to aid digitalization and resource-efficient retrofits.