Sample Generation Method Based on Variational Modal Decomposition and Generative Adversarial Network (VMD-GAN) for...

Area of Science:

• Analytical Chemistry
• Environmental Science
• Spectroscopy

Background:

• Ultraviolet-visible (UV-Vis) spectroscopy is sensitive for detecting chemical oxygen demand (COD), particularly at low concentrations.
• A significant challenge in UV-Vis spectroscopy for water quality monitoring is the scarcity of real-world polluted water samples.
• Existing spectral data generation methods often fail to capture the complexity and characteristics of actual spectral data.

Purpose of the Study:

• To develop an advanced spectral sample generation method to address the data scarcity issue in water pollution analysis.
• To improve the diversity and complexity of generated spectral data for more accurate modeling.
• To enhance the performance of chemical oxygen demand (COD) detection models through data augmentation.

Main Methods:

• Proposed a hybrid method combining Variational Modal Decomposition (VMD) and Generative Adversarial Networks (VMD-GAN).
• Utilized VMD to decompose absorption spectra into principal components and residuals.
• Employed GAN for generating new principal components and a Gaussian fitting function for generating diverse residuals, followed by spectral reconstruction.

Main Results:

• Generated absorption spectra that closely mimic actual water sample characteristics, improving data diversity and complexity.
• Successfully expanded the available dataset of spectral samples for water quality analysis.
• Regression models built with the augmented data showed improved performance in predicting COD for real water samples.

Conclusions:

• The VMD-GAN method effectively generates realistic spectral samples, overcoming the limitations of insufficient real-world data.
• This approach significantly enhances the capability for accurate modeling and analysis of water pollution.
• The proposed method offers a viable solution for improving UV-Vis spectroscopic analysis in environmental monitoring.