An Enhanced Crack Detection System Using YOLOv11 Model for Underwater Marine Structures

In order to preserve structural integrity and avoid catastrophic collapses, it is imperative that fractures in maritime concrete structures be detected promptly. Conventional techniques including Ultrasonic Pulse Velocity (UPV), diver’s manual visual checks, dye penetration, and radiographic tests frequently fall short in underwater settings because of drawbacks such poor visibility, expensive operations, potential safety hazards, and variable outcomes. Furthermore, the performance of contemporary machine learning models is frequently limited by the accessibility of sizable, varied, and well annotated datasets. This paper proposes a novel software framework that leverages the YOLOv11 object detection model with instance segmentation and incremental learning for automated crack detection in real-time video feeds. The framework achieves high precision even in challenging underwater conditions. Additionally, the integration of incremental learning dynamically expands the training dataset by incorporating newly detected cracks, enabling continuous adaptation to diverse environments and evolving crack patterns. The design addresses the critical issue of inadequate initial datasets, significantly reducing the reliance on manual annotations. By mitigating the challenges posed by traditional inspection methods, the system enhances scalability and accuracy, offering a transformative solution for monitoring the health of structures underwater.