Multi-class defects in railway tracks
Description
This dataset contains 7,585 real-world railway images with object-level annotations for research on automated railway track and rail defect detection using artificial intelligence, computer vision, machine learning, and deep learning. The dataset is related to the research article “Multi-Class Rail Defect Detection using Advanced Artificial Intelligence”, published in High-speed Railway (2026), DOI: 10.1016/j.hspr.2026.03.002. The associated study investigated advanced object-detection frameworks, including YOLOv11, YOLO-NAS, and Roboflow 3.0, for multi-class railway defect detection. The published study reports a dataset collected from North Western Railways, India, comprising 10 defect categories. The present data record provides 7,585 annotated railway images intended to facilitate reproducible research, independent validation, model development, and benchmarking of AI-based railway inspection systems. The annotations enable supervised object-detection tasks involving the localization and identification of railway defects and track conditions. The dataset can be used for research in: Railway track and rail defect detection Multi-class object detection Computer vision and deep learning Artificial intelligence-based railway inspection Railway infrastructure condition assessment Automated railway monitoring and safety Predictive and condition-based maintenance Development and benchmarking of object-detection models This dataset forms part of an ongoing research programme on intelligent railway inspection and complements the authors' related research on machine learning, deep learning, nondestructive testing, and automated railway defect identification. Researchers using this dataset are requested to cite the dataset and the relevant associated publications. The dataset is provided to support open, reproducible, and further research in AI-enabled railway inspection and infrastructure safety.