High-Accuracy Event Classification of Distributed Optical Fiber Vibration Sensing Based on Time-Space Analysis

Area of Science:

• Optoelectronics
• Machine Learning
• Signal Processing

Background:

• Distributed optical fiber sensing (DVS) enables vibration measurement along optical fibers.
• Accurate classification of vibration events is crucial for DVS practical deployment.

Purpose of the Study:

• To develop a high-accuracy event recognition method for DVS data using a convolutional neural network (CNN).

Main Methods:

• Collected over 10,000 outdoor vibration datasets.
• Trained a CNN model on raw DVS time-space data.
• Evaluated classification accuracy using various data domains (time, frequency, time-frequency).

Main Results:

• Achieved 99.9% recognition accuracy using the CNN on time-space DVS data.
• CNN outperformed traditional time-domain, frequency-domain, and time-frequency domain analyses.
• Demonstrated strong generalization performance with 99.2% accuracy after one week.

Conclusions:

• The proposed CNN method offers highly accurate and robust vibration event classification for DVS.
• The approach shows significant potential for practical DVS applications due to its high accuracy and generalization.
• Time-space data analysis with CNN is superior for DVS event recognition.