Multi-Modal Weak Signal Analysis for Buried Optical Cable DVS Using Combined Multi-Head Attention Mechanism

Distributed optical fiber sensing technologies, represented by Phase-sensitive Optical Time-Domain Reflectometry (Φ-OTDR), have gained widespread application in vibration monitoring for infrastructures like roads and pipelines due to advantages such as long monitoring distances, high spatiotemporal coverage, and convenient deployment. However, the monitoring data are simultaneously affected by complex and uncertain couplings involving optical cable structural response, vibration propagation medium characteristics, initial vibration states, and multi-modal environmental excitations, requiring proper decoupling for effective application. Most existing methods focus on extracting and processing only one or a few specific types of features, resulting in high computational load and insufficient mining of effective information from the sensing signals. Among these, decoupling and extracting multi-modal weak signals under unknown excitation scenarios remains a key technical bottleneck. To address this, this paper, based on the fiber-environment coupled vibration observation model for Φ-OTDR Distributed Fiber-optic Vibration Sensor (DVS), innovatively proposes an analysis framework combining a scenario-guided combined multi-head attention mechanism, achieving efficient decoupling, separation, and extraction of multi-modal weak signals in complex environments. This method can effectively focus on key signals for application scenarios, significantly improving the real-time performance of signal analysis, and provides a universal real-time solution for DVS signal processing in complex scenarios. Measured data from a 30km optical cable deployed on a city ring expressway validate the effectiveness of the proposed method, demonstrating excellent performance and real-time processing capabilities in comprehensive application scenarios including road roughness detection, construction machinery detection and localization, and vehicle trajectory recognition.