Synthetic minority oversampling and iterative fluorescence-suppression integrated algorithm for Raman spectrum...

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Raman spectroscopy is crucial for chemical analysis but is often hindered by noise and fluorescent background signals.
• Imbalanced datasets pose challenges in developing robust predictive models for trace analytes like pesticides.

Purpose of the Study:

• To develop an automated Raman spectrum preprocessing method for enhanced pesticide detection.
• To improve the accuracy and reliability of pesticide identification, especially for low-concentration or hard-to-detect compounds.

Main Methods:

• Wavelet transform for noise reduction.
• Modified polynomial curve fitting for automated fluorescent background suppression.
• Synthetic minority oversampling technique (SMOTE) for addressing imbalanced datasets.
• Convolutional Neural Network (CNN) for pesticide identification.

Main Results:

• Effective denoising and fluorescent background suppression were achieved.
• SMOTE enabled accurate prediction of pesticides from imbalanced datasets, comparable to large data volumes.
• The CNN model accurately identified single pesticides and accurately predicted mixed pesticide compositions (99.1% accuracy).

Conclusions:

• The proposed method offers a robust solution for Raman spectrum preprocessing, improving pesticide detection capabilities.
• This approach enhances the identification of challenging pesticide samples, demonstrating high accuracy in complex mixtures.