摘要：针对铁路入侵检测中海量视频数据传输效率低、检测精度低的问题，提出了一种基于视频Transformer的自适应语义通信框架。该框架通过传输与任务高度相关的语义特征并集成信道自适应模块，实现面向特定任务的高效且鲁棒的语义信息传输。在构建的铁路真实场景视频数据集上进行实验，将其并与传统的视频分离编码及视频联合编码方法进行对比。结果表明，所提出的自适应语义通信框架在不同的高斯白噪声及多种衰落信道条件下均能取得更高的入侵检测精度，并在低信噪比和复杂信道环境中展现了出优越的鲁棒性和性能增益，为提升铁路智能监控系统的视频分析与传输效能提供了新的技术途径。

关键词：入侵检测；语义通信；深度联合信源信道编码；信道自适应

Abstract: To address the challenges of low efficiency in transmitting massive video data and suboptimal detection accuracy under complex channel conditions for railway intrusion detection, an adaptive semantic communication framework based on Video Transformer is proposed. This framework achieves efficient and robust task-oriented semantic information transmission by conveying highly task-relevant semantic features and integrating a channel-adaptive module. Experiments conducted on a constructed real-world railway scenario video dataset show that, compared with traditional separate video coding and joint video coding methods, the proposed framework attains higher intrusion detection accuracy under various additive white gaussian noise and diverse fading channel conditions. Furthermore, it exhibits superior robustness and performance gains in low signal-to-noise ratio and complex channel environments, offering a novel technical approach for enhancing the video analysis and transmission efficacy of intelligent railway monitoring systems.

Keywords: intrusion detection; semantic communication; deep joint source-channel coding; channel adaptation