Using Image Processing to Classify Roofs based on Damage Level

Roofing systems are considered one of the items that in most need of frequent inspection and rehabilitation due to its ongoing exposure to the elements. Manual roof inspections are time-consuming and subjective. This study uses Convolutional Neural Network (CNN), an image-processing technique, to classify roofs according to their damage level. The proposed model analyzes roofing images and determines whether the roof has sustained low, moderate, or severe damage. The study was applied on more than 200 images of roofs of the University of Waterloo campus. The developed model has shown promising results, achieving approximately 95% accuracy with no major biases. The proposed method aims to provide a fast, objective, and reliable roofing inspection and assessment framework, helping asset managers of large portfolios better assign the allocated rehabilitation funds.