Classification and Inditification of Different Level of Aflatoxin B1 on Maize Kernels Surface Using Visible/Near Infrared Reflectance (VIS/NIR) Hyperspectral Imaging

Abstract

Abstract. A visible/near infrared (VIS/NIR) hyperspectral imaging system with wavelength range between 400 and 1000 nm was used to assess the potential to detect low levels of Aflatoxin B1 (AFB1) contaminants on the surface of healthy maize kernels. Principal components analysis (PCA) was used to reduce the dimensionality of the spectral data, and then stepwise factorial discriminant analysis (FDA) was performed on latent variables provided by the PCA's, a 94% classification accuracy was achieved. The results indicated that hyperspectral imaging technology accompanied by a PCA-FDA method, can detect AFB1 that is applied directly on the maize surface. Furthermore, the optimal wavelengths were selected depending on Variables Importance in Projection (VIP) scores extracted from Partial Least Square (PLS) analysis. Then, the FDA method was used based on those 10 VIP wavelengths to identify different classes. The result demonstrated that, a good classification result could also be obtained using only those wavelengths selected by VIP procedure rather than using the overall wavelengths between 400 and 1000nm, which indicated that the practical value of this method.