DeepGPR: Learning to Identify Moisture Defects in Building Envelope Assemblies from Ground Penetrating Radar

Conventionally used moisture detection equipment such as infrared scanners and capacitance meters require a trained interpreter to understand moisture issues on rooftops. Additionally, conventional sensors can only provide reliable results in specific environmental conditions. In this paper, we will discuss the various methods used for roof moisture scans and their limitations. We will then provide an in-depth analysis of GPR paired with deep segmentation neural networks for roof moisture scans, including its advantages, limitations, and potential applications. We will also present a case study demonstrating the effectiveness of this approach in detecting moisture damage in a real-world scenario. Our preliminary experiments find that deep neural networks are effective in segmenting GPR radargrams and finding moisture, with particular neural networks more effective than others.