Atmospheric droplet detection and classification with self-mixing interferometry and neural networks

Area of Science:

• Optical Physics
• Atmospheric Science
• Artificial Intelligence

Background:

• Self-mixing interferometry offers advantages like simplicity and robustness for airborne applications.
• Monitoring atmospheric conditions, particularly water droplet characteristics, is crucial for aviation safety and weather forecasting.

Purpose of the Study:

• To design and validate a self-mixing interferometer system capable of probing atmospheric water droplets.
• To develop and train a small neural network for classifying self-mixing signals influenced by different droplet types.

Main Methods:

• An experimental setup was created to simulate various atmospheric conditions based on water droplet sizes.
• A self-mixing interferometer was employed to analyze the interaction with simulated atmospheric droplets.
• A compact, embeddable neural network was designed and trained to classify interferometric signals.

Main Results:

• The system demonstrated the ability to model atmospheric conditions by varying water droplet sizes.
• The trained neural network achieved over 93% discrimination accuracy in classifying self-mixing signals under laboratory conditions.
• The study successfully linked specific interferometric signal patterns to different droplet characteristics.

Conclusions:

• AI-enabled self-mixing sensors show promise for real-time atmospheric condition monitoring.
• The developed system offers potential for early warning systems, particularly for icing conditions in aviation.
• Further validation under high-speed airborne conditions is necessary to confirm practical applicability.