Ratio Quantification of Geranium and Rose Essential Oil Mixtures via Deep Learning Analysis of Complex Raman Spectra

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Essential oils (EOs) are economically valuable but prone to adulteration.
• Raman spectroscopy offers molecular specificity for EO analysis.
• Quality control of EOs is challenging due to potential adulteration.

Purpose of the Study:

• To develop a robust method for analyzing rose essential oil (REO) and geranium essential oil (GEO) mixtures.
• To identify key spectral markers for distinguishing EO mixing ratios.
• To create a deep learning model for deconstructing mixed EO spectra into pure components.

Main Methods:

• Collection of 2700 Raman spectra from REO, GEO, and their mixtures.
• Construction of a convolutional neural network (CNN) for ratio analysis.
• Development of a channel attention residual feature extraction network (CARFENet) for spectral deconstruction.
• Validation using internal datasets and external test sets.

Main Results:

• Identified critical spectral peaks (e.g., 800, 1000, 1668 cm⁻¹) as discriminative markers.
• The CNN model effectively distinguished varying EO mixing ratios.
• CARFENet achieved high accuracy in predicting pure component spectra, with similarity indices > 0.99.

Conclusions:

• Specific spectral features serve as reliable markers for EO composition.
• CARFENet provides effective quantitative analysis of pure EO components in mixtures.
• This approach enhances quality control for essential oils using Raman spectroscopy.