人工智能视觉大模型在铁路线路异物入侵场景中的应用  

Application of Large-scale AI Vision Models to Detect Foreign Object Intrusion into Railway Lines

作　　者：杨涛存 史维峰 李国华 代明睿 李文浩 杜文然 YANG Taocun;SHI Weifeng;LI Guohua;DAI Mingrui;LI Wenhao;DU Wenran(Institute of Computing Technology,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)

机构地区：[1]中国铁道科学研究院集团有限公司电子计算技术研究所,北京100081

出　　处：《中国铁路》2025年第1期23-29,48,共8页China Railway

基　　金：中国国家铁路集团有限公司科技研究开发计划项目(P2023S001)。

摘　　要：铁路线路异物入侵是威胁列车运行安全的严重问题之一,现有的智能识别系统在解决数据稀缺和异物种类多样性等问题时面临巨大挑战。针对上述问题,提出一种基于人工智能大模型的铁路线路异物入侵智能识别方法,基于预训练大模型的特征提取能力和泛化性能,通过对大模型结构深度和宽度的扩展,结合迁移学习策略,微调使其适应铁路线路异物识别任务。实验结果表明,基于人工智能大模型的异物入侵检测算法,能够显著减少对标注数据的依赖。在面对训练数据有限和未知异物类别多样的问题时,能实现较高的检测准确率和实时性能,显示出其在复杂环境中处理未知多变异物入侵的强大能力。Foreign object intrusion into railway lines poses a significant threat to train operational safety.Current intelligent identification systems encounter substantial challenges in addressing issues such as data scarcity and foreign object diversity.To address the aforementioned issues,the paper proposes an intelligent method for the detection of foreign object on railway lines based on large-scale AI model.The study takes into account the feature extraction capabilities and generalization performance of the model during pre-training,expands its structural depth and width,and conducts model fine-tuning in accordance with transfer learning strategies concerned,so as to effectively adapt the model to the task of identifying foreign objects on railway lines.The experimental results demonstrate that the detection algorithm,powered by large-scale AI model,can significantly decrease reliance on annotated data.Even in face of limited training data and a diverse array of unknown foreign object categories alike,the algorithm achieves high detection accuracy and real-time performance.This highlights its robust capability to handle unknown and varied foreign object intrusions in complex environments.

关 键 词：人工智能 计算机视觉 视觉算法 铁路安全 异物入侵监测 深度学习 铁路线路 

分 类 号：U29-39[交通运输工程—交通运输规划与管理] TP391.41[交通运输工程—道路与铁道工程]