Spatial Tracking of Water Pollutions Based on SERS Spectra with Machine Learning

Water pollution detection technology is key to safeguarding the ecological security of river basins. However, traditional water pollution tracking methods have shown limitations with the development of industry and urbanization. Surface-enhanced Raman spectroscopy (SERS) has long been recognized as an optical method that can provide fingerprint-like chemical information at the single-molecule level. In this work, a new attention neural network (aNN) combined with random forest (RF) approach was constructed to be applied to accurately identify the SERS spectral signature of wastewater for spatial tracking of water pollutions. A total of 34,520 SERS spectra were collected from 96 combinations of three types of wastewater after data augmentation. The data set was used to train the aNN-RF model, with the identification accuracy of over 88% and predicted mixing ratios of wastewater types showed the same step-down trend as actual mixing ratios. As a proof-of-concept, a one-dimensional watershed scenario was simulated, and wastewater outlet and pollution pathway were identified by the SERS-aNN-RF platform. Results suggested that even in complex and unknown watersheds, the SERS-aNN-RF platform can accurately track pollutions, exhibiting excellent potential in spatial tracking of water pollutions.