Machine Learning-Assisted Near- and Mid-Infrared Spectroscopy for Rapid Discrimination of Wild and Farmed Mediterranean Mussels (Mytilus galloprovincialis)

Abstract

The objective of this study was to investigate the ability to discriminate between wild and farmed Mediterranean mussels (Mytilus galloprovincialis) using machine learning-assisted near-infrared (NIR) and mid-infrared (MIR) spectroscopy. Mussels are of significant global importance in aquaculture due to their nutritional characteristics, encompassing a rich source of protein, essential fatty acids, various vitamins, and abundant minerals. Additionally, their ease of farming adds to their value as a desirable aquaculture species. The mussels' capacity to reflect environmental quality attributes makes them valuable as biomonitoring agents. However, differences in nutritional composition may arise between wild mussels harvested from natural marine hard-bottoms and those farmed in open artificial systems in the sea. In this study aimed at distinguishing between the two types of mussels, the classification models were created, and the most accurate results were achieved using the FT-MIR spectral data extracted from the interior part of the mussels, while the performance of FT-MIR data obtained from the mussels' shells was slightly lower, with the accuracy of 92% and R2 of 0.87. Still, the accuracies of all the classification models were over 90%. The Ensemble model, trained using FT-MIR spectra from the interior part of the mussel, achieved an accuracy of 98.4%, surpassing the performance of other variable sets. In both NIR and MIR models, spectra from the mussels' interior provide better discrimination than spectra from the outer shell.