Intelligent corrosion detection in reinforced concrete slabs using GPR and YOLOv11: A fully non-destructive and automated deep learning framework

Corrosion of steel bars is a critical concern in reinforced concrete (RC) structures exposed to aggressive environments. Early corrosion detection is essential for timely maintenance and cost-effective rehabilitation. This study proposes a fully automated, non-destructive corrosion detection framework using the state-of-the-art deep learning model, You Only Look Once version 11 (YOLOv11), to analyse 2D B-scan GPR images. Informed by electrochemical measurements, the model classifies corrosion severity into four levels: negligible, mild, moderate and severe. Evaluation results demonstrate that the model achieves high detection accuracy for mild and severe corrosion, with mAP50 exceeding 0.75. However, performance declines for negligible and moderate corrosion, likely due to the limited number of annotated instances compared to mild and severe cases. These findings highlight YOLOv11’s potential for rapid corrosion assessment while underscoring the need for sufficient and balanced datasets across corrosion levels. • Fully non-destructive and automated corrosion detection using GPR and YOLOv11. • A comprehensive experiment involving four RC slabs for simulating varying corrosion levels. • YOLOv11 demonstrates strong performance in detecting mild and severe corrosion. • Class imbalance in dataset causes lower accuracy in detecting negligible and moderate corrosion.