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.


Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan

Citation:  Transactions of the ASABE. Vol. 49(4): 1235-1245. (doi: 10.13031/2013.21723) @2006
Authors:   T. C. Pearson, D. T. Wicklow
Keywords:   Aflatoxin, Corn, Detection, Feature selection, Fumonisin, Image, Maize pathogens, Near-infrared

Single-kernel reflectance spectra (550 to 1700 nm), visible color reflectance images, x-ray images, multi-spectral transmittance images (visible and NIR), and physical properties (mass, length, width, thickness, and cross-sectional area) were analyzed to determine if they could be used to detect fungal-infected corn kernels. Kernels were collected from corn ears inoculated with one of several different common fungi several weeks before harvest, and then collected at harvest time. It was found that two NIR reflectance spectral bands centered at 715 nm and 965 nm could correctly identify 98.1% of asymptomatic kernels and 96.6% of kernels showing extensive discoloration and infected with Aspergillus flavus, Aspergillus niger, Diplodia maydis, Fusarium graminearum, Fusarium verticillioides, or Trichoderma viride. These two spectral bands can easily be implemented on high-speed sorting machines for removal of fungal-damaged grain. Histogram features from three transmittance images (blue and red components of color images and another at 960 nm) can distinguish 91.9% of infected kernels with extensive discoloration from 96.2% of asymptomatic kernels. Similar classification accuracies were achieved using x-ray images and physical properties (kernel thickness, weight, length). A neural network was trained to identify infecting fungal species on single kernels using principle components of the reflectance spectra as input features.

(Download PDF)    (Export to EndNotes)