Research Progress on Intelligent Algorithm-Assisted Surface-Enhanced Raman Spectroscopy for Environmental Trace Pollutant Detection

Abstract

Trace pollutants in environmental waters are often present at low levels and in complex mixtures, while matrix interference can further complicate the interpretation of SERS spectra. This paper reviews the use of intelligent algorithms for SERS-based environmental analysis, focusing on pollutant identification, concentration estimation, and multicomponent spectral analysis. Machine-learning methods such as PCA, SVM, RF, and PLSR, together with CNN-based deep learning, have been used to extract information from full spectra and improve qualitative and quantitative analysis. More recent studies have also explored models trained on spectra of individual compounds and then applied to mixture analysis, offering a possible way to reduce the experimental burden of preparing large numbers of mixed samples. Despite these advances, matrix effects, variations among SERS substrates and instruments, data leakage, and limited generalization still restrict practical use. Improving data standardization, external validation, unknown-class rejection, and the connection between portable screening and laboratory confirmation will be important for developing more reliable intelligent SERS methods for environmental detection.