Abstract:Aiming at the problem of low detection accuracy and large computation due to the small proportion of bolts in subway vehicle roof images, this paper proposes a small target bolt detection algorithm based on lightweight AWD-YOLOv8s. Firstly, an adaptive weight downsampling AWD module is designed to enhance the ability of the algorithm to extract small target feature information by assigning different weight values to local feature information. Secondly, the intra-scale feature interaction (AIFI) module is introduced to replace the SPPF module. By capturing the dependency between the same scale features, the algorithm is lightweight and improved. Finally, the lightweight Ghost-P2 structure is introduced to enhance the ability of the algorithm to extract the feature information of small target bolts. The experimental results show that the average accuracy of the improved algorithm reaches 98.7%, which is 2.2% higher than that of the original algorithm, the number of model parameters is reduced by 46.7%, the dete