Automated post-earthquake structural damage assessment of concrete buildings using a hybrid deep learning and rule-based framework on image datasets

This study proposes a novel intelligent hybrid framework for the automated post-earthquake damage detection and assessment of reinforced concrete buildings to enhance infrastructure resilience and enable rapid emergency response. The framework uniquely integrates rule-based engineering knowledge with cutting-edge deep learning models. Specifically, YOLOv11(You Only Look Once version 11) is utilized as the base model to finetune new models with diverse image datasets, such as PEER Hub ImageNet and new images from the 2023 Türkiye Earthquake labeled by the authors. Data augmentation techniques are applied for robust damage classification and detection. A meta-model classifier, which leverages machine learning and image-aware embeddings, intelligently fuses the YOLOv11 outputs and established engineering rules. This sophisticated, engineering-informed meta-learning fusion approach is key to the system's efficiency and impressive generalization for different earthquake events. When tested on two independent large image datasets, the proposed framework achieves an accuracy of 87.4% at the ±1 level tolerance for one dataset with very diverse images and damage levels. The results indicate that our framework is a strong, reliable, and rapidly deployable method for damage assessment of structures. • An intelligent hybrid framework is developed by combining YOLOv11-based deep learning and rule-based logic for post-earthquake assessment of resilient infrastructure. • Diverse datasets are curated from the 2023 Turkey earthquakes (including inside/outside scenes, components, and damage types) to enable rapid and robust safety evaluation. • A high reliability is demonstrated through a meta-model classifier that intelligently fuses AI outputs with engineering rules, ensuring superior generalization across multiple seismic events.