A lightweight design method for internal pavement crack recognition models using ground penetrating radar

Transverse cracks are among the most prevalent forms of damage in Semi-rigid asphalt pavements, traditionally the crack identified through manual methods that are time-consuming and suffer from poor consistency. To improve the efficiency of crack disease identification, this study employs a self-developed high-speed and high-precision 3D ground-penetrating radar to characterize internal crack defects. The You Only Look Once-v8n algorithm is enhanced through lightweight design, optimizing its backbone network and loss function while incorporating an attention mechanism to strengthen crack feature extraction, detection accuracy, and training stability. Practical engineering evaluations demonstrate that the algorithm reduces manual identification time by 50%, with over 60% of results achieving high expert validation rates. The missed detection counts for visible cracks are comparable to manual methods. This algorithm enables automated identification of internal cracks in pavement rehabilitation projects.