Supervised contrastive learning-guided prototypes on axle-box accelerations for railway crossing inspections

作者：

Highlights：

• A deep learning system for worn railway crossings detection via axle-box accelerations.

• A novel combination of prototypical inference and supervised contrastive learning.

• Contrastive learning outperforms standard metric learning for few-shot learning.

• The method outperforms by 8% prior methods for worn crossings classification.

摘要

•A deep learning system for worn railway crossings detection via axle-box accelerations.•A novel combination of prototypical inference and supervised contrastive learning.•Contrastive learning outperforms standard metric learning for few-shot learning.•The method outperforms by 8% prior methods for worn crossings classification.

论文关键词：Dynamic railway surveying,Axle-box accelerations,Crossing wear detection,Deep learning,Supervised contrastive learning

论文评审过程：Received 26 January 2022, Revised 11 June 2022, Accepted 20 June 2022, Available online 3 July 2022, Version of Record 6 July 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117946