Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
Integration of Fluorescence and Reflectance Visible Near-Infrared (VNIR) Hyperspectral Images for Detection of Aflatoxins in Corn Kernels
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.orgCitation: Transactions of the ASABE. 59(3): 785-794. (doi: 10.13031/trans.59.11365) @2016
Authors: Fengle Zhu, Haibo Yao, Zuzana Hruska, Russell Kincaid, Robert L. Brown, Deepak Bhatnagar, Thomas E. Cleveland
Keywords: Aflatoxins, Corn kernel, Fluorescence, Hyperspectral imaging, Integration, Reflectance.
Abstract. Aflatoxin contamination in agricultural products has been an important and long-standing problem around the world. Produced by certain fungal species of the genus, aflatoxins are highly toxic and carcinogenic. This study investigated the integration of fluorescence and reflectance visible near-infrared (VNIR) hyperspectral images to detect aflatoxins in whole corn kernels. Field-inoculated corn ears were harvested, and kernels having different aflatoxin contamination levels were collected. Both fluorescence hyperspectral images under ultraviolet (UV) excitation and reflectance hyperspectral images under halogen illumination were recorded on the two sides of the kernels (endosperm and germ). Subsequent chemical analysis was performed on each kernel to provide reference aflatoxin concentration. Threshold values of 20 and 100 ppb were adopted separately to group kernels as contaminated or healthy. Contaminated kernels exhibited different fluorescence and reflectance spectral features compared with healthy kernels. Spectral datasets were compressed and interpreted using principal component analysis (PCA). Least squares support vector machines (LS-SVM) and k-nearest neighbor (KNN) classifiers were used on the fluorescence PC, reflectance PC, and integrated fluorescence and reflectance PC variables for classifying both sides of kernels as contaminated or healthy. The best overall prediction accuracy was 95.33% for the LS-SVM model with the 100 ppb threshold on the germ side in the integrated analysis. Overall, the germ side performed better than the endosperm side, especially for the true positive rate (TPR). Fluorescence and reflectance image data generally achieved similar classification accuracy. The integrated analysis achieved better results than separate fluorescence or reflectance analysis on the germ side, and conspicuous improvement in the TPR of the germ side was observed after integration. The mean aflatoxin concentration in the prediction samples was reduced from 2662.01 ppb to 64.04, 87.33, and 7.59 ppb after removing samples that were classified as contaminated by fluorescence, reflectance, and integrated analysis, respectively, on the germ side. This study demonstrated the potential of the integrated technique for better screening of aflatoxin-contaminated kernels and could lead to rapid and non-destructive scanning-based detection in the corn industry.