Research on rail surface defect detection algorithm based on improved YOLOv8

Aiming at the problems of low detection accuracy and high error detection rate of small target defects under complex background, a rail surface defect detection algorithm based on YOLOv8 was proposed. First, the model's focus on small and medium targets is enhanced by replacing the original convolution with SPD-Conv building blocks in YOLOv8's backbone network, while preserving the original network structure. Secondly, the integration of the EMA attention mechanism module into the neck component enables the model to utilize information from different layer features and improve feature representation capabilities. The [email protected] of this model can reach 97.4%, 4.2% higher than that of YOLOv8. In addition, the frame rate, parameter number, computation and model volume of model detection are better than most mainstream target detection algorithms. The experimental results show that this model can effectively improve the accuracy of rail defect detection.