Deep Learning–Based Automated Crack Detection for Post‐Earthquake Damage Assessment in Reinforced Concrete Structures

Abstract

This study explores the integration of deep learning technologies, specifically U‐Net based segmentation methods, for evaluating earthquake‐induced damages. The study leverages a dataset derived from the Kahramanmaraş earthquake to train and test deep learning models capable of identifying and quantifying structural damages such as concrete cracks. The 2023 Kahramanmaraş earthquakes underscored the critical need for rapid and accurate post‐earthquake structural assessments to ensure the safety and timely rehabilitation of affected reinforced concrete (RC) structures. The research reveals that deep learning models, particularly those employing U‐Net architectures, offer substantial improvements over traditional visual inspection methods by providing faster, more consistent, and more accurate damage assessments. Intersection over union (IoU) scores with 0.737 for concrete cracks, highlight the models’ effectiveness in identifying distinct damage patterns. These capabilities are crucial for the rapid assessment of structural integrity and the prioritization of repair and rehabilitation efforts.