High-throughput detection of environmental contaminants in natural and engineered aquatic systems is crucial to safeguard public health, yet the quantitative analysis of complex contaminant mixtures remains a significant challenge. Organic contaminants like polycyclic aromatic hydrocarbons (PAHs), known to pose health risks to humans upon ingestion, often coexist as complex mixtures in the environment. Here, we develop an artificial intelligence (AI)-empowered framework coupled with surface-enhanced Raman spectroscopy (SERS) to quantitatively detect PAHs in mixtures. A spectral preprocessing algorithm, PreDe, is developed to compress SERS spectra by 99.7% while retaining key Raman features. A subsequent two-stage AI framework deploys a discriminator to prescreen PAH spectra and a classifier to demix PAHs quantitatively. The discriminator achieves 100% accuracy in rejecting non-PAH spectra (i.e., representative pesticides), even without prior exposure to these spectra during training. The quantitative performance of the classifier is associated with the compositional balance of the PAH mixtures. Among the four models tested in the classifier, the convolutional neural network (CNN) and random forest (RF) consistently deliver the highest prediction accuracy and lowest error. This SERS-AI pipeline enables the rapid prescreening and quantitative demixing of target contaminants, offering a powerful new strategy for high-throughput monitoring of contaminants in complex matrices.
原文链接
分享
引用
