Anomaly monitoring of distributed fiber optic sensors based on spatiotemporal features

In the field of railway transportation, distributed fiber optic sensors are widely used to detect external vibrations. The distributed structure causes external vibrations to simultaneously affect adjacent sensors, with varying degrees of impact. To address the signal overlap introduced by the distributed topology, we propose a hybrid model based on Deeper Graph Convolutional Networks (DeeperGCN) and CNN-Transformer for extracting spatiotemporal features. DeeperGCN is employed to analyze linear topological relationships, effectively capturing the interactions between neighboring sensors. The CNN-Transformer network is used to capture the dynamic changes of vibration signals in the temporal domain. Experimental results from 234,200 samples across five categories demonstrate the feasibility and effectiveness of our approach, improving the accuracy of vibration event classification.