A SWIR hyperspectral imaging method for classifying Aflatoxin B1 contaminated maize kernels

Abstract.

A short wave-infrared (SWIR) hyperspectral imaging system (1000-2500 nm) was used to assess the feasibility of detecting aflatoxin B₁ (AFB₁) on surfaces of 150 kernels of yellow dent maize variety from Indiana State of the U.S.A. Four AFB₁ solutions i.e. 10, 20, 100 and 500 ppb were artificially applied on kernel surfaces. Similarly, a control group was generated from 30 kernels treated with a solution of methanol. Principal component analysis (PCA) was used to reduce dimensionality of the HSI data followed by the application of Mahalanobis distance classifier on the regions of interest (ROI) classes. It was possible to find clusters representing non-contaminated and contaminated maize kernels by interactively tracing clusters in the score plot of the 11^th and 12^th principal components and projecting the results on the score image space. Mahalanobis distance classifier could not only identify aflatoxin contaminated from non-contaminated, but also discriminate between the concentrations applied to kernels. Chemical interpretation of the loading line plots showed key spectral bands such as 1146, 1729, 2274 and 2344 nm which contribute the most to the observed differences. The study suggests that hyperspectral imaging technique accompanied by the PCA-Mahalanobis distance classifier can be used to identify directly coated pure AFB₁ concentrations as low as 10 ppb on maize kernels.