Fusarium head blight (FHB), caused
primarily by Fusarium graminearum
Schwabe, is one of the most common and devastating fungal diseases, causing
significant losses in wheat grain productivity and quality by producing harmful
mycotoxins. Traditionally, wheat breeders mainly focus on visual selection for
FHB resistant lines during FHB resistance breeding in wheat. Advanced
techniques that enable high-precision and quick identification are required to
improve its detection, because the current method i.e., visual assessment is
labour- and time-intensive and subject to subjectivity biases. As a result,
methods with a higher degree of objectivity have been investigated, and special
emphasis has been laid on the use of high-resolution imaging (e.g., RGB and
thermal) coupled with advanced deep learning methods as the foundation for more
reliable detection strategies. In this study, first, infected spikes from the
wheat genotypes grown in FHB nursery were visually rated and grouped into four
different classes including resistant (0-25%), moderately resistant (26-50%),
moderately susceptible (51-70%), and susceptible (71-100%) based on the level
of FHB infections. Following this, high-resolution RGB images of FHB infected
wheat spikes were collected from the same visually rated FHB inoculated genotypes.
Commonly used Convolutional Neural Network (CNN) architectures such as VGG and ResNet
etc.) were employed to develop FHB infection level detection models. Additionally,
to achieve enhanced model performance in the context of relatively small sample
size circumstances, deep transfer learning strategy was adopted as well. To examine
and compare the accuracies of different modelling strategies and CNN
architectures, independent testing data was used for model assessment. The
confusion/error matrix, along with the overall accuracy, Kappa coefficient and F1
score matrix were used for model evaluation. This study shows that high-resolution
RGB imaging combined advanced deep learning approach can be used to predict FHB
infection in seeds within the spikelet’s and to identify very small differences
in spike architecture (associated with FHB infection), which can better help
breeders predict FHB infection at field level.
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