Φ-OTDR Signal Identification Method Based on Multimodal Fusion.

Area of Science:

• Sensor Technology
• Signal Processing
• Artificial Intelligence

Background:

• Distributed Fiber Optic Sensing (DFS) is vital for long-distance pipeline safety.
• Accurate identification of vibration signals is a key challenge in DFS.

Purpose of the Study:

• To develop an end-to-end algorithm for high-accuracy fiber optic vibration signal detection and identification.
• To improve the reliability of DFS in complex environments.

Main Methods:

• Utilized one-dimensional and two-dimensional Convolutional Neural Networks (CNNs) for time and frequency domain feature extraction.
• Integrated a self-attentive mechanism and Long Short-Term Memory (LSTM) for feature fusion.
• Employed a Transformer Encoder with multi-headed self-attention for multimodal feature integration.

Main Results:

• Achieved 98.54% accuracy in vibration signal classification on an urban field dataset.
• Demonstrated effective performance in complex noise conditions.

Conclusions:

• The proposed multimodal deep learning approach significantly enhances vibration signal detection and identification accuracy in DFS.
• This algorithm offers a robust solution for pipeline safety monitoring.