Fusarium Head Blight (FHB) or head
scab (caused by Fusarium graminearum), is one of the most devastating
diseases affecting small grain cereals including wheat; and causes severe
reduction in the quality and quantity of grain yield. The traditional screening
of large number of breeding lines for FHB resistance is still dependent on
low-throughput visual inspections in the FHB inoculated field nursery followed by
visual assessment of percent Fusarium Damaged Kernels (FDKs). Further, the
visual assessments are a labor-intensive and time-consuming process and is
biased due to the subjectivity and may have lower reproducibility. Some earlier
studies have evaluated the effectiveness and preciseness of image analysis
approaches to predict FDKs in wheat grain samples; however, only a few focused
at the final application, taking into consideration the association between
cost and benefit, resolution, and accuracy. In this study, for the first time,
we worked with QualySense on a feasibility study aimed at using their
proprietary QSorter® Explorer robotic system for the automatic
prediction, sorting and final quality inspection
assessment of Fusarium damaged wheat kernels. QSorter® Explorer is powered
by advanced mechatronics and artificial intelligence and equipped with a 3D
camera and a high resolution Near Infrared Spectrometer active in the 900-1700
nm spectral range. More than 7,200 single kernel images and NIR spectra were
collected from healthy and Fusarium damaged kernels to develop machine learning
models to predict FDKs and sort samples at 30 kernels/sec. Independent
validations of the FDKs module was conducted using 200 advanced breeding lines
(100 winter wheat and 100 spring wheat genotypes) that were visually rated by
three persons separately. The same samples were analyzed by the QSorter® Explorer by mean of two classification
models FDK-Vision and FDK-NIR which are based respectively on the images and NIR
spectra, to predict and sort the FDKs. While the visual evaluations of FDKs by
the individuals varied significantly, the QSorter® Explorer consistently
produced better results in a time efficient manner. Additionally, compared to
visual assessments, QSorter® predictions were more accurate. This automation will certainly make
screening for breeding lines more accurate and less prone to personal bias
while delivering results faster and streamlining the laboratory efficiency.
ACKNOWLEDGEMENT AND DISCLAIMER
This material is based upon work supported by the
U.S. Department of Agriculture, under Agreement 59-0206-2-153. This is a cooperative project with the U.S. Wheat &
Barley Scab Initiative. Any opinions, findings, conclusions, or recommendations
expressed in this publication are those of the author(s) and do not necessarily
reflect the view of the U.S. Department of Agriculture