Classification of laser modality for a self-mixing interferometric sensor.

Area of Science:

• Optoelectronics
• Laser Physics
• Signal Processing

Background:

• Robust detection and classification of multimodal self-mixing (SM) signals are crucial for accurate information retrieval from optical feedback-based interferometric laser sensors.
• Abrupt shifts in SM signal modality, caused by varying operating conditions, can lead to significant measurement errors if not identified.
• Proactive identification of SM signal modality is essential for implementing necessary adjustments in sensor setup or signal processing to mitigate errors.

Purpose of the Study:

• To propose and evaluate machine learning-based techniques for identifying and classifying multimodal self-mixing (SM) signals.
• To address the challenge of unidentified modality shifts in SM interferometric laser sensor data.
• To enhance the reliability and accuracy of sensing information retrieved from SM sensors.

Main Methods:

• Feature extraction from SM signals using Principal Component Analysis (PCA), peak width analysis, and Linear Discriminant Analysis (LDA).
• Training and testing of machine learning classifiers, including Linear Regression, XGBoost Regressor, and Decision Tree Regressor.
• Experimental acquisition of a dataset comprising 45 unseen SM signals from an SM sensor for method validation.

Main Results:

• The Decision Tree Regressor achieved 100% accuracy in identifying and classifying SM signal modalities.
• The XGBoost Regressor demonstrated a high accuracy of 96% for SM signal modality classification.
• Linear Regression achieved 76% accuracy, indicating varying performance among the tested machine learning algorithms.

Conclusions:

• Machine learning classifiers, particularly Decision Tree and XGBoost Regressors, are highly effective for robust SM signal modality identification and classification.
• The proposed feature extraction and classification methods successfully address the challenge of modality shifts, significantly reducing potential measurement errors.
• Accurate SM signal classification enables adaptive adjustments in sensor systems and signal processing, leading to improved sensing performance and data reliability.