Note: Gaussian mixture model for event recognition in optical time-domain reflectometry based sensing systems.

Area of Science:

• Optoelectronics and Sensor Technology
• Signal Processing
• Machine Learning

Background:

• Phase-sensitive optical time-domain reflectometry (PhOTDR) is crucial for distributed sensing.
• Recognizing non-conventional events in PhOTDR signals presents challenges.
• Existing methods may lack robustness in event classification.

Purpose of the Study:

• To develop a novel algorithmic approach for recognizing specific non-conventional events in PhOTDR sensor signals.
• To enhance the reliability and accuracy of event detection in optical sensing systems.
• To address limitations in current signal analysis techniques for PhOTDR.

Main Methods:

• Signal denoising using advanced filtering techniques.
• Clustering of signal features with a Gaussian mixture model.
• Experimental validation using real-world measured signals.

Main Results:

• The proposed algorithm successfully distinguishes between two classes of non-conventional events.
• A best-case recognition probability approaching 0.9 was achieved.
• Performance is dependent on sufficient training data samples.

Conclusions:

• The novel algorithm offers a robust solution for event recognition in PhOTDR.
• The combination of filtering and Gaussian mixture models improves classification accuracy.
• This approach has potential applications in advanced optical sensing and monitoring.