Efficient Mobile Road Defect Detection with Parameter-Reduced AL-YOLOv8s

Road damage detection is crucial for timely maintenance and ensuring traffic safety. However, existing models often struggle to meet the required accuracy for defect detection operations and have large parameter counts, hindering deployment on mobile devices. This paper proposes an improved YOLOv8s-based road damage detection model, named AL-YOLOv8s, which incorporates several enhancements inspired Bi-directional Feature Pyramid Network (BiFPN) architectures. The Spatial Pyramid Pooling - Fast (SPPF) pooling layer is enhanced with a spatial pyramid pooling with edge localization and attention network (SPPELAN) module, while the network structure is reconstructed using a bidirectional feature fusion mechanism. Additionally, an exponential moving average (EMA) attention mechanism is integrated into the backbone, and a lightweight downsampling operation (Adown) is employed to improve Conv modules. Experiments on the RDD2022 road defect dataset demonstrate that our improved model achieves a mean average precision (mAP) of 93.8%, which is 3.1% higher than the current state-of-the-art model, while reducing the number of parameters by 12% to 6.88M. The proposed AL-YOLOv8s model addresses the critical need for accurate and lightweight road damage detection, enabling efficient deployment on mobile devices for real-time inspections and improved traffic safety.