In this study, we explored a new method of determining the three forms (rTG, EE, TG) of omega-3; a method based on artificial intelligence analysis of FT-IR spectra. To this end, 90 omega-3 samples (30 rTG, 30 EE and 30 TG) were subjected to analysis....
In this study, we explored a new method of determining the three forms (rTG, EE, TG) of omega-3; a method based on artificial intelligence analysis of FT-IR spectra. To this end, 90 omega-3 samples (30 rTG, 30 EE and 30 TG) were subjected to analysis. This analysis consisted of employing principal component analysis-linear discriminant analysis (PCA-LDA), a support vector machine (SVM), a 1D convolutional neural network (CNN), and 2D convolutional neural network (CNN), and FT-IR spectral data. The absorbance and second derivative spectra, which had been obtained using FT-IR, were then applied to each algorithm. Techniques such as cross-validation and batch normalization were performed to prevent overfitting. Additionally, Bayesian optimization and random search were used for hyperparameter optimization.
The result of discriminant analysis was that all four algorithms demonstrated 100% classification accuracy. Thus, the effectiveness of PCA-LDA, SVM, 1D CNN, and 2D CNN-based FT-IR spectrum classification technology in distinguishing distinct types of omega-3 was proven. In addition, by utilizing the SHAP technique and the Grad-CAM technique, it was that confirmed that the artificial intelligence learning model identifies the discrete types of omega-3 based on the accurate signal. Given these findings, the expectation is that the determination method of this study will be used in the future to determine the content of each type of omega-3 in products that contain a mixture of two or more types.