Developing Deep Learning Algorithms to Recognize Material and Component for Circular Construction

Rapid urbanization is causing an increase in building material demand worldwide, with construction material use projected to almost double from 2017 to 2060. Cities all over the world are implementing policies to recover demolition and renovation waste in an effort to move from a linear towards a circular system. However, a lack of awareness on which reusable materials will be available at the time of construction remains a significant barrier. Previous research created material resource cadastres by predicting façade materials from Google Street View images. However, there is a desire to extract more specific information from images, such as material recognition for each component, the percentage material composition of the component (if there is more than one material used in the component), verification of the material labels, the expected remaining life of the materials, as well as expanding studied methods to additional datasets. This level of detail necessitates more sophisticated techniques. This thesis aims to develop deep learning techniques for the recognition of building materials and components. For further information, please refer to the attached proposal.