Quantitative Analysis of Nitrides in Water by Raman Spectroscopy Based on Deep Learning and Relative Position Matrix

Abstract

Monitoring the concentration of nitride in water is of great significance for the protection of water quality. Raman spectroscopy can simultaneously detect the characteristic peaks of various substances, but its signal intensity is relatively weak. In order to address the issue of overlapping peaks and weak peaks in mixtures that significantly hinder the direct analysis of Raman spectroscopy and to achieve a simultaneous quantitative analysis of nitrate and nitrite. This study proposes a quantitative detection method for water nitride concentration. The method is based on the relative position matrix (RPM) and a multimodal data fusion model. First, Raman spectra of 12 mixed solutions of potassium nitrate and sodium nitrite with different concentration grades were collected. Then, they were transformed into 2D images using the relative position matrix method to enhance the feature richness. After that, 1D spectral data and 2D image data were extracted by a multimodal data fusion model, the simultaneous prediction of nitrate and nitrite concentrations through two outputs. The results show that the model achieved an average determination coefficient (R2) of 0.9615 in the test set, with an average mean square error (MSE) of 0.0053 and mean absolute error (MAE) of 0.0317. The practical feasibility of the method was verified through quantitative analysis of actual water samples. Overall, this research provides an algorithmic foundation for accurately monitoring nitride concentration in water based on Raman spectroscopy.

Keywords: Raman spectroscopyDeep learningNitrite