AI-Driven Workflow for Chemical Compounds Classification from IR Spectra of Solutions

This work presents an AI-driven workflow for the classification of chemical compounds using near-infrared (NIR) spectroscopy of solutions. This approach addresses the challenge of identifying specific substances in complex spectral data, particularly at low concentrations. By leveraging data augmentation techniques such as noise addition and spectral transformations, the study enhances the robustness and accuracy of AI models in spectral analysis. Experimental validation was performed using substances like Progesterone, Indol-3 acetic acid (IAA), $\beta$-Estradiol and Bisphenol A, dissolved in methanol and measured with Fourier-transform NIR spectroscopy. The proposed workflow provides promising results for substance identification.