Identification for the Raman spectra of edible oil mixture with convolutional neural network

• Combine Raman spectroscopy and CNN to achieve componential analysis of mixed edible oil. • Propose the comparison between CNN and five other machine learning models. • Consider the impact of spectral preprocessing on identification results. • The accuracy of CNN model for component analysis reaches 0.9935. As a common condiment in daily life and an essential nutrient for the human body, edible oil has attracted increasing attention due to the serious problem of oil adulteration. The methods for detecting the composition of edible oil are also constantly developing. Spectroscopic technique has gained extensive application in analytical chemistry, particularly through representative methods like Raman spectroscopy. This prominence stems from its capacity to provide spectral fingerprints characteristic of molecular composition while maintaining non-destructive analytical capabilities. Due to the similar composition and Raman characteristic peaks of different types of edible oils, traditional methods of manual identification have significant errors. This research develops an analytical framework integrating Raman spectroscopy and a convolutional neural network (CNN) architecture to achieve precise discrimination of a mixture containing five distinct edible oil varieties. To confirm the advantages of the CNN model, other models are also used for comparison, including K-Nearest Neighbor (KNN), Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), and Decision Tree (DT). In contrast to conventional approaches necessitating sequential preprocessing steps (e.g., baseline calibration, noise filtration, signal smoothing), the CNN-driven framework achieves 99.35 % classification accuracy for edible oil blends using original spectral inputs, surpassing conventional machine learning performance. This breakthrough demonstrates CNN’s capability for rapid mixture analysis with unprocessed Raman spectra.