Application of NIR Reflectance Spectroscopy Combined with Chemometrics for Non-destructive Detection of Aflatoxin B₁ in Individual Maize Kernels

Maize is a potential host for several fungi, some of which produce mycotoxins. Maximum mycotoxin levels in food vary depending on the type of food, consumer sensitivity, and country regulations. Aflatoxin-B₁ (AFB₁) is a harmful mycotoxin that causes illness and economic losses in humans, animals, and crops. A visible-near infrared spectroradiometer (350-2500 nm) was used in this study to detect maize kernels with varying AFB₁ contamination levels (control, 25, 40, 70, and 200 ppb). Spectral reflectance and chemometrics were used to predict the AFB₁ content in maize kernels. Principal component analysis (PCA) and forward selection method were utilized for data exploration; linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA) were employed to develop classification models; and partial least square regression (PLSR) was used for AFB₁ quantification. The Kernels germ orientation influenced the classification performance. In training, cross-validation, and testing, the LDA model preceded with PCA had the lowest error rates when the kernels were kept at germ-up orientation. Throughout training, cross-validation and testing of the PLSR model, R² values were around 0.85. Forward selection method has provided seven optimal wavelengths from the group of 2051 spectral data. The R² and RMSE of the PLSR model using optimal wavelength data were also close to the results obtained from full spectrum data. The RPD value for the optimal wavelength model during testing was around 2.5, indicating good predictive performance. Results showed that AFB₁ can be identified and quantified using spectral data. Spectroscopy is a feasible solution for automatic, non-destructive detection without affecting the products.