Laser-induced breakdown spectroscopy (LIBS) is an elemental analysis technique. It uses a pulsed laser beam to interact with the sample and has the advantage of simple sample preparation, allowing remote detection, and enabling rapid online multi-element analyses. Therefore, LIBS has been widely used in biomedical, industrial, environmental analyses, and other fields. However, it has poor spectral stability, which needs to be solved. In recent years, many researchers have tried to improve spectral stability by applying data preprocessing methods. Most of these methods are complex to operate or offer limited improvement in quantitative analysis, including one-dimensional wavelet transform denoising. Therefore, we propose a method to improve the detection stability of LIBS by using two-dimensional wavelet denoising. The method is simple to operate and reduces the relative standard deviation (RSD) of the spectrum better than the one-dimensional wavelet transform.

In this study, four elements, Cu, Ni, Mo, and V, are verified and analyzed based on alloy steel samples. We perform multiple measurements on the alloy steel samples and then arrange the multiple LIBS to form two-dimensional data, with each column being the LIBS data for each test and each row being the spectral intensity for different measurement times at each wavelength. The combined data is denoised by a two-dimensional wavelet, and the original data is denoised by a one-dimensional wavelet. The data are processed with different decomposition layers, and the change of the denoising effect of the two-dimensional wavelet with the increase in decomposition layers is studied. The best decomposition layers are confirmed. The RSD and signal-to-noise ratio (SNR) of the processed data and the original data are calculated to confirm the advantages and feasibility of two-dimensional wavelet denoising. Subsequently, we quantitatively analyze the data before and after wavelet denoising based on four elements, Cu, Ni, Mo, and V, so as to evaluate the enhancement of the accuracy of quantitative analysis by two-dimensional wavelet denoising.

By analyzing the characteristic spectral lines of the four elements, the data is denoised by a two-dimensional wavelet with different decomposition layers. The results are as follows: with the increase in the number of decomposition layers, the SNR of the spectrum first increases and then slowly decreases when the third decomposition layer is reached; the RSD of the spectrum continues to decrease, but the amplitude slows down after the third or fourth layer is reached (Fig. 3). After the optimal number of decomposition layers is confirmed, one-dimensional wavelet denoising and two-dimensional wavelet denoising are applied to eight alloy steel samples. SNR and RSD for processed and raw data are calculated. The results show that wavelet denoising improves the SNR of the spectrum (Fig. 5). However, one-dimensional wavelet denoising is not effective in improving the spectral stability, and two-dimensional wavelet denoising has significantly reduced the RSD of the spectrum (Fig. 6). This result shows that two-dimensional wavelet denoising can make up for the deficiency of one-dimensional wavelet denoising while retaining the ability to improve the SNR. Subsequently, we perform a quantitative analysis of the four elements based on alloy steel samples of different concentrations. Because of the self-absorption effect of the spectrum, a quadratic function is used in this study for curve fitting of the quantitative analysis results. The results show that the fitting degree of the quantitative analysis of the four elements is improved after two-dimensional wavelet denoising (Fig. 7). This indicates that two-dimensional wavelet denoising can improve the accuracy of quantitative analysis. Therefore, two-dimensional wavelet denoising has a greater potential for improving the spectral fluctuations of LIBS techniques.

In this paper, LIBS data from multiple measurements are combined into a matrix to convert one-dimensional data into two-dimensional data. Then, the data is processed by using two-dimensional wavelet denoising. This method not only simplifies the processing process of one-dimensional wavelet denoising but also provides a new idea for spectral data processing. The change of two-dimensional wavelet denoising effect with the number of decomposition layers is experimentally investigated. When the number of decomposition layers is increased, the SNR of the spectrum will first increase and then decrease at a slow rate after the optimal number of decomposition layers is reached. However, the RSD of the spectrum decreases all the time, but the amplitude will gradually approach zero. Furthermore, we compare two wavelet denoising methods. The results show that the two-dimensional wavelet has a similar improved effect on the SNR of the spectrum as the one-dimensional wavelet, with a maximum increase of about 9.7%. At the same time, a two-dimensional wavelet can make up for the defect that a one-dimensional wavelet cannot improve the spectral stability, and the RSD of the spectrum is reduced by up to 37%. In addition, we quantitatively analyze alloy steel samples with different concentrations. The results show that the stability of spectral data is enhanced after the two-dimensional wavelet denoising. The accuracy of quantitative analysis is improved, and the curve fitting degree is increased by 0.177 at most. In summary, our research shows that two-dimensional wavelet denoising has unique advantages in the repeatability of LIBS and has great potential for spectral data preprocessing.