Deep Learning-Based Semantic Segmentation and 3D Reconstruction Techniques for Automatic Detection and Localization of Thermal Defects in Building Envelopes

Building age and extreme weather can cause thermal defects in the building envelope. Failure to inspect and fix these defects can threaten safety, increase energy consumption, and negatively impact the environment. Infrared thermal imaging (IRT) is a popular, non-destructive technique for building diagnostics due to its safety, practicality, and energy efficiency. However, manual IRT detection is time-consuming and imprecise. Therefore, this paper proposes a framework for automatically identifying and localising thermal defects in building envelopes. The outcomes of this study not only offer direction for categorising thermal defects in buildings but also provide a practical approach for automatically detecting and locating them.