Implementation of a bolt defect detection algorithm based on an improved YOLOv8 approach

High-speed train maintenance is a critical component in ensuring railway transportation safety. Addressing challenges in detecting defects on the skirt bolts of high-speed trains—such as minute target sizes, complex environments, and diverse defect types—this paper proposes LMS-YOLO, a lightweight multi-scale detection algorithm based on improved YOLOv8 (You Only Look Once version 8). This algorithm incorporates lightweight multi-scale feature fusion and integrated Global Attention Mechanism (GAM), enabling the model to adaptively focus on critical feature regions within images and effectively enhance the localization accuracy of small defects. Experiments conducted on a self-built dataset of 3,672 high-speed train skirt images demonstrate that LMS-YOLO achieves an average precision of 87.2% and a recall rate of 92.7% on a 536-image test set. outperforming mainstream algorithms such as original YOLOv8l, Faster R-CNN (Faster Region-based Convolutional Neural Network), EfficientNet, and SSD (Single Shot MultiBox Detector). It demonstrated particularly outstanding performance in detecting defects like missing bolts and missing marking lines, providing reliable technical support for the intelligent operation and maintenance of High-speed trains.