Deep learning-based excitation-emission matrix fluorescence spectroscopic analysis of polycyclic aromatic...

Rapid and accurate analysis of polycyclic aromatic hydrocarbons (PAHs) in coal tar (CT) is essential yet challenging due to spectral overlap and matrix effects. Excitation-emission matrix (EEM) fluorescence spectral analysis using deep learning algorithms can address this challenge.

Coal tar samples were extracted and diluted with cyclohexane. A dataset of 1173 EEM spectra covering 3- to 6-ring PAHs and various concentration gradients was constructed. Seven machine learning algorithms were used for qualitative classification of PAHs, and the deep learning algorithm, ResNet18, achieved the highest accuracy, close to 90%. Quantitative analysis was performed using a convolutional neural network-based variational autoencoder (CNN-VAE) model, yielding Root mean square error (RMSE) and (Mean absolute error) MAE values as low as 0.0036 mg/L and 0.0028 mg/L, respectively. These findings demonstrate the effectiveness of the proposed method for accurate PAH quantification in complex coal-tar matrices.

This new approach provides a cost-effective, rapid, and reliable alternative to conventional methods, mitigating issues such as Rayleigh scattering and spectral overlap. It offers a new strategy for detecting PAH molecules in complex matrices.