Real-Time Water Quality Monitoring Based on UV-Vis Spectroscopy and an Improved VGG-Net

Ultraviolet–visible (UV-Vis) spectroscopy is widely used for water quality parameter analysis due to its advantages of real-time, rapid measurement, simple equipment, and no secondary pollution. It serves as a critical tool for online water quality monitoring. Water quality parameters exhibit unique absorption characteristics in the UV-Vis spectrum, allowing for accurate measurement of individual parameters using spectral analysis and regression models. However, the accuracy of traditional linear regression models decreases when detecting multiple parameters, as their absorption peaks are concentrated in the 190–320-nm range. This study employs a deep learning algorithm based on convolutional neural networks (CNNs) to perform quantitative analysis on mixed spectra of multiple parameters. By building the spectral model of mixed water quality parameters and training the network with appropriate amount of spectral data, accurate concentrations of multiple water quality parameters can be simultaneously obtained. This research provides a significant theoretical foundation for advancing the application of UV-Vis spectroscopy in online multiparameter water quality monitoring.