Comparative Evaluation of Fluorescence, NIR, and FT-IR Spectroscopy Combined with Machine Learning for Geographical Origin and Species Identification of Brown Algae

Ensuring the authenticity of origin labeling is a challenge for brown algae products such as kelp, wakame, and hijiki. While conventional DNA analysis has high discriminatory capabilities, it is not necessarily suitable for routine screening of large quantities of samples due to time and cost considerations. This study aimed to identify the species (variety) and geographical origin of brown algae (kelp, wakame, and hijiki), with both variety and origin evaluated for kelp and geographical origin evaluated for wakame and hijiki. To achieve this, classification methods were developed for each of three spectroscopic analysis techniques—excitation emission matrix (EEM), Fourier transform infrared spectroscopy (FT-IR), and near-infrared spectroscopy (NIR)—using machine learning algorithms, and their classification performance was systematically compared and evaluated. Six classification models, k-nearest neighbors (KNN), convolutional neural networks (CNN), LightGBM, XGBoost, random forest, and support vector machines, were constructed to distinguish varieties and origins based on EEM, NIR, and FT-IR data. Depending on the combination of methods, high-precision identification was obtained (>99%), especially for kelp variety identification using NIR + KNN (≈100%). These results suggest that each spectral dataset contains characteristic information specific to each sample and that selecting a model suited to these characteristics is effective for highly accurate identification of variety (in kelp) and geographical origin. The selected method can serve as a rapid and simple identification tool that contributes to verifying the authenticity of brown algae products and improving raw material traceability.