A Study of High-Frequency Noise for Microplastics Classification Using Raman Spectroscopy and Machine Learning.

Area of Science:

• Environmental Science
• Analytical Chemistry
• Materials Science

Background:

• Growing global demand for effective plastic management and regulation.
• Machine learning (ML) techniques show promise for microplastic classification using Raman spectroscopy.
• Accurate microplastic identification is crucial for proper disposal and environmental protection.

Purpose of the Study:

• To investigate the impact of high-frequency noise on ML-based microplastic classification using Raman signals.
• To explore the trade-off between signal smoothing (noise reduction) and peak preservation in Raman spectroscopy.
• To evaluate the effectiveness of different ML models and Principal Component Analysis (PCA) in handling noisy data.

Main Methods:

• Analysis of microplastic classification performance using Raman spectroscopy under varying levels of high-frequency noise.
• Comparison of Linear Discriminant Analysis (LDA) and Error-Correcting Output Codes (ECOC) models.
• Evaluation of Principal Component Analysis (PCA) as a noise reduction and dimensionality reduction technique.

Main Results:

• Linear Discriminant Analysis (LDA) models exhibit poor generalization with noisy Raman signals.
• Error-Correcting Output Codes (ECOC) models demonstrate better resilience to inherent noise in spectral data.
• Principal Component Analysis (PCA) effectively reduces noise and enhances the robustness of classification models.

Conclusions:

• High-frequency noise significantly affects the performance of microplastic classification from Raman spectra.
• ECOC models offer a more robust approach compared to LDA for noisy spectral data.
• PCA is a valuable pre-processing step for improving the reliability of ML-based microplastic identification.