A Deep Learning Approach to Semantic Segmentation for Architectural Blueprint Interpretation and Geographic-based Material Recommendation

The planning and building of houses is financially-consuming and time-consuming, causing housing development to struggle to keep up with rapidly growing populations. In the rapid expansion of residences, the environmental implications of such developments are frequently disregarded. Though few recent scientists have attempted to leverage machine learning to expedite the house modeling process, these models do not take into consideration environmental factors such as geographic location, temperature, and landscaping. There is a pressing need for developing an automated and accurate system that offers comprehensive interpretation of architectural blueprints, providing essential guidance and insights for various aspects of the construction process. This study incorporates these factors by leveraging machine learning to develop software models for cost-effective and environmentally sustainable housing. To address the aforementioned problem, I propose a machine learning-based semantic segmentation network for architectural blueprint interpretation. This method is designed to address the complex task of understanding blueprints, discerning room types and their spatial arrangement. To achieve this, I employ convolutional neural networks widely recognized for their effectiveness in image analysis tasks. I also introduce a material recommendation method that aligns with the geographical context of the construction site. Extensive experiments, prove that the proposed method outperforms the previous state-of-the-art method in accurately generating semantic segmentation maps for the inputted blueprint images. I expect that these superior segmentation results will significantly enhance the architectural planning process by providing architects and designers with a more detailed and informative representation of blueprint layouts, thus aiding in better decision-making and design refinement.