Lightweight Road Damage Detection Algorithm based on the Improved YOLO Model

In recent years, the problem of road damage has become increasingly prominent due to the expansion of road transportation. To address the issues of cumbersome traditional road damage detection methods, low recognition accuracy, and high costs, we propose a road pothole damage identification method based on deep learning algorithms. We use YOLOv8n as the base model and introduce the Slim-Neck paradigm and a new lightweight convolution technique called GSConv to replace regular convolutions, effectively reducing the network’s parameter count and model size. Additionally, we use the VoV-GSCSP module to maintain sufficient detection accuracy while reducing the complexity of computation and network structure. We also improve the lightweight detection head with shared parameters and further reduce the model’s computational load using group convolution, achieving lightweight computational effects. Our proposed LRDD-YOLO model achieves an mAP of $\mathbf{8 2. 4 \%}$ on the Pothole dataset, with a 43% reduction in floating-point operations, demonstrating that our improved road damage detection method can better detect road damage morphology while remaining lightweight.