An Improved YOLOv8 Model for Pavement Distress Detection Under Low-Computing-Power Conditions

Area of Science:

• Computer Vision
• Artificial Intelligence
• Civil Engineering

Background:

• Automated pavement distress detection (PDD) is vital for structural health monitoring (SHM) of transportation infrastructure.
• Existing PDD methods face challenges in real-time multi-target detection under resource constraints.

Purpose of the Study:

• To develop an efficient and accurate automated pavement distress detection model for real-time applications.
• To improve the accuracy-efficiency trade-off in pavement distress detection systems.

Main Methods:

• Constructed YOLOv8-PDD by integrating the large separable kernel attention (LSKA) mechanism into the Spatial Pyramid Pooling-Fast (SPPF) module.
• Replaced Complete-IoU (CIoU) loss with Distance-IoU (DIOU) loss and adopted Soft-Non-Maximum Suppression (NMS) over the original NMS algorithm.
• Evaluated the model's performance on pavement distress detection tasks.

Main Results:

• YOLOv8-PDD achieved 78.3% mean average precision (mAP@0.5), an 8.1% improvement over the baseline YOLOv8n, with a minor 0.2 GFLOPs complexity increase.
• The model demonstrated a negligible latency increase (+0.09 ms) while significantly enhancing detection accuracy.
• Achieved 68 FPS on consumer-grade CPUs (OpenVINO-optimized), outperforming 10 other models in accuracy-speed balance, with notable improvements in pothole and strip patch detection.

Conclusions:

• YOLOv8-PDD offers a superior accuracy-efficiency trade-off for real-time pavement distress detection.
• The model exhibits strong adaptability for detecting small targets and complex backgrounds, crucial for infrastructure monitoring.
• The proposed enhancements provide a promising solution for efficient and accurate structural health monitoring of transportation infrastructure.