A Lightweight YOLOv12n Model for Real-Time Crack Detection on Pavement and Structural Surfaces

Infrastructure deterioration through crack formation poses significant challenges to structural integrity and public safety across civil engineering applications. Conventional crack detection methodologies frequently demonstrate limitations in computational efficiency and deployment flexibility for edge computing scenarios. This research presents an innovative framework utilizing a specifically fine-tuned YOLOv12n architecture designed for efficient crack identification on computationally constrained hardware platforms. Our methodology employs a specialized dataset sourced from Roboflow, achieving remarkable performance metrics including 98.5\% precision, 99.2\% recall, 99.5\% mAP@0.5, and 73.5\% mAP@0.5:0.95, while sustaining an ultra-compact model footprint of merely 2.6 million parameters. The proposed system integrates sophisticated data augmentation strategies and attention-based mechanisms to optimize detection reliability across heterogeneous environmental conditions, including variable illumination scenarios and intricate background textures. Our architecture's resource-efficient design enables rapid inference execution, facilitating deployment on embedded systems and edge devices without compromising detection accuracy. Extensive validation through comprehensive performance metrics, encompassing F1-score analysis, precision-recall characterization, and curve-based evaluations, validates the model's operational robustness. The contribution establishes a practical, scalable framework for autonomous infrastructure monitoring with significant implications for preventive maintenance strategies in civil engineering. \textbf{Keywords:} Structural crack identification, YOLOv12n architecture, surface defect detection, resource-efficient models, edge computing, automated inspection systems