Denoising and Baseline Correction Methods for Raman Spectroscopy Based on Convolutional Autoencoder: A Unified Solution

Ming Han,Yu Dang,Jianda Han

DOI: https://doi.org/10.3390/s24103161

IF: 3.9

2024-05-17

Sensors

Abstract:Preprocessing plays a key role in Raman spectral analysis. However, classical preprocessing algorithms often have issues with reducing Raman peak intensities and changing the peak shape when processing spectra. This paper introduces a unified solution for preprocessing based on a convolutional autoencoder to enhance Raman spectroscopy data. One is a denoising algorithm that uses a convolutional denoising autoencoder (CDAE model), and the other is a baseline correction algorithm based on a convolutional autoencoder (CAE+ model). The CDAE model incorporates two additional convolutional layers in its bottleneck layer for enhanced noise reduction. The CAE+ model not only adds convolutional layers at the bottleneck but also includes a comparison function after the decoding for effective baseline correction. The proposed models were validated using both simulated spectra and experimental spectra measured with a Raman spectrometer system. Comparing their performance with that of traditional signal processing techniques, the results of the CDAE-CAE+ model show improvements in noise reduction and Raman peak preservation.

engineering, electrical & electronic,instruments & instrumentation,chemistry, analytical

What problem does this paper attempt to address?

This paper attempts to solve two main problems in Raman spectroscopy pre - processing: **denoising** and **baseline correction**.

1. **Denoising problem**: Raman spectra are affected by various noises during the acquisition process, such as Gaussian noise, fluorescence background and cosmic rays. These noises will reduce the quality of the spectra and affect the accuracy of subsequent analysis. Although traditional denoising algorithms (such as wavelet - threshold denoising, Savitzky - Golay filtering, etc.) can effectively remove some noises, they often reduce the intensity of Raman peaks and change the peak shape, thus affecting the integrity of spectral features.

2. **Baseline correction problem**: There is usually a complex baseline in Raman spectra, which will interfere with the identification of spectral features. Although traditional baseline correction methods (such as piecewise linear fitting, adaptive iteratively reweighted penalized least squares, etc.) can correct the baseline to a certain extent, they often reduce the intensity of Raman peaks, especially in cases where the baseline is complex and the peak width is wide.

To solve the above problems, the paper proposes a unified solution based on the Convolutional Autoencoder (CAE), specifically including:

- **Convolutional Denoising Autoencoder (CDAE)**: Used for denoising, the denoising effect is enhanced by adding two additional convolutional layers at the bottleneck layer.
- **Convolutional Autoencoder (CAE + )**: Used for baseline correction. Not only convolutional layers are added at the bottleneck layer, but also a comparison function is introduced after decoding to achieve effective baseline correction.

These two methods, through deep - learning techniques, can more effectively remove noise and correct the baseline while maintaining the intensity and shape of Raman peaks, thereby improving the pre - processing quality of Raman spectroscopy data.