A U-Net Convolutional Neural Network Deep Learning Model Application for Identification of Energy Loss of Infrared Thermographic Exterior Building Envelope Images

This study presents a novel U-NET convolution neural network (CNN) deep learning (DL) model, developed in a Python environment for the identification of envelope deficiencies on a data set of infrared (IR) thermographic images of building envelopes. A data set of 142 IR images acquired with an unmanned aerial vehicle (UAV) were used with supplementary segmentation masks created for appropriate U-NET modelling application. This data preparation process is presented followed by an in-depth review of the CNN architecture used for the segmentation process. The Python3 code developed for this thesis is reviewed in depth, for an easy application in future work performed by non-data-science researchers. The results of this research presented roughly 43% accuracy and very promising novel outputs from the analytical system. The available data used for this study was noted to be the key limitation with this research.