Rapid detection of ionic contents in water through sensor fusion and convolutional neural network

Salt contents in soil or groundwater are one of the primary indicators to evaluate contamination levels. Electrical conductivity (EC) or salinity information from the conventional laboratory analysis is typically inefficient in delineating contamination. This study investigated a rapid determination of ionic contents in water through the combination of Ultraviolet Spectroscopy (UVS) and Electrochemical Impedance Spectroscopy (EIS), and the application of convolutional neural network (CNN). Various aqueous salt samples were prepared with Ca²⁺, K⁺, Na⁺, Cl^-, Br^-, SO₄^2-, and HCO₃^- ions. Firstly, their spectral data obtained from UVS and EIS were analyzed. The spectral analysis showed that the data fusion of both spectroscopies provided more evidence to distinguish the ionic contents, consequently enhancing prediction performance of CNN. In turn, the fused spectra were handled with CNN to predict ionic contents. The result suggested the validity of the proposed method in detecting ionic contents by showing 48.6 mmol/kg RMSE and 0.95 R² between actual and predicted ionic concentrations, which outperformed Partial Least Squares Regression (PLSR) and Random Forest. The detection of ionic contents beyond EC or salinity is advantageous since it provides more information on the soil and water contamination, and it facilitates tracking the contaminant sources. The proposed method has the potential to become more accurate with increased datasets and further optimization of CNN, which will further improve the practicability.