Rapid and label-free detection of aflatoxin B1 in peanut oil using surface-enhanced Raman spectroscopy combined with deep learning models

This study presents a novel approach for rapid, label-free and sensitive detection of Aflatoxin B 1 (AFB 1 ) in peanut oil using Surface-Enhanced Raman Spectroscopy (SERS) combined with deep learning models. Silver-coated gold nanoparticles (Au@Ag NPs) were synthesized as SERS substrate. A confocal Raman spectrometer was used to acquire the Raman spectra of the AFB 1 spiked peanut oil samples. Then, the collected SERS spectral data were augmented and preprocessed to improve the regression model's generalization capabilities. A total of six regression models including Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), Random Forest Regression (RFR), Convolutional Neural Networks (CNN), Long Short-Term Memory Networks (LSTM), and a combined CNN-LSTM model were developed. The results revealed that CNN-LSTM model efficiently captured complex non-linear relationships, reduced reliance on parameter adjustment and minimized overfitting. It also handled large-scale datasets effectively, reducing the computational load. CNN-LSTM model achieved excellent predictive performance with determination of coefficient (R2 P) = 0.9892, root mean square error of prediction (RMSEP) = 0.2104, ratio of performance to deviation (RPD) = 6.8723 and excellent sensitivity (LOD = 0.31 μg/kg). These findings demonstrate the proposed method provides rapid, label-free, and efficient AFB 1 detection in peanut oil, with significant potential for real-time monitoring application. • SERS combined with deep learning enables AFB 1 quantification in peanut oil. • Au@Ag NPs developed as efficient SERS substrate. • Comparative analysis of conventional vs. deep learning models. • CNN-LSTM model achieves superior predictive performance.