Random forest microplastic classification using spectral subsamples of FT-IR hyperspectral images.

Area of Science:

• Analytical Chemistry
• Environmental Science
• Machine Learning

Background:

• Microplastic pollution is a significant environmental concern.
• Accurate identification of microplastic types is crucial for environmental monitoring.
• Current identification methods can be time-consuming and complex.

Purpose of the Study:

• To develop a fast and efficient model for identifying common microplastics.
• To reduce the complexity of spectral data for microplastic analysis.
• To enable microplastic identification using simplified spectral inputs.

Main Methods:

• A random decision forest model was developed for spectral identification.
• Input data was reduced to key wavenumbers using a machine learning classifier.
• Spectra were extracted from Fourier-transform infrared hyperspectral images.
• Automated background correction and identification algorithms were employed.
• Classification accuracy was validated against procedurally generated ground truth.

Main Results:

• The random decision forest model achieved fast identification of microplastic spectra.
• Dimension reduction enabled the use of individual wavenumber measurements.
• Prediction time was significantly decreased.
• High classification accuracy was demonstrated on pure-type microplastic samples.

Conclusions:

• The developed model offers a rapid approach for microplastic identification.
• The method's efficiency is suitable for systems with limited spectral data.
• Further validation is needed for complex environmental samples containing mixed materials.