YOLOv8n-SBP: A Lightweight and Efficient Model for Pavement Distress Detection

A key challenge in road infrastructure assessment and maintenance pertains to the limited detection accuracy, high computational demands, complex model architectures, and restricted adaptability of existing algorithms for identifying pavement distresses on mobile devices. To address these, this paper proposes an optimized pavement distress detection framework, YOLOv8n with SENet, BiFPN, and PSSDetect (YOLOv8n-SBP), built upon enhancements to the YOLOv8n model. First, the squeeze-and-excitation networks (SENet) block is incorporated following the integration of a spatial pyramid pooling-fast (SPPF) module, thereby amplifying the algorithm’s capacity to discern critical distress features while mitigating environmental interference. Second, the path aggregation network (PANet) is replaced by a bidirectional feature pyramid network (BiFPN) configuration, optimizing multiscale feature fusion to enhance both detection accuracy and operational robustness. Finally, through the implementation of parameter-sharing principles, a spatial channel reconstruction convolutional (SCConv) module is introduced into the detection process, achieving substantial reductions in both model parameter count and computational overhead. Experimental results demonstrated that the proposed YOLOv8n-SBP framework achieved a 0.3% improvement in mean average precision at 50% intersection over union (mAP50) compared with the original architecture, while simultaneously reducing the parameter count by 55.6% and computational complexity by 36.7%. These advancements enable efficient real-time pavement distress detection with enhanced precision, confirming the model’s practical applicability. Comprehensive comparative evaluations with state-of-the-art object detection models further substantiated the framework’s superiority in terms of detection accuracy, architectural efficiency, and adaptability for real-world deployment scenarios.