April 6-7, 2017 | Lincoln, Nebraska
nsf logo link
2017 NRIC speakers
portrait of Andrew Biewener

 


Jesse Poland
Kansas State University
Deep learning for high-throughput phenotyping of 'complex' traits in wheat
2:25 - 2:55 pm, Thursday 6 April
ABSTRACT:
In the scope of field-based high-throughput phenotyping, the indirect measurement of plant phenology through canopy spectral reflectance is offering insights to the overall status of the crop and opportunities to improve yield prediction models and selection approaches in the breeding program. To advance high-throughput phenotyping for assessment of more complex phenotypes such as plant morphology or developmental stages, new approaches are needed to measure these phenotypes in a manner that is analogous to how we, as humans, score traits in plants. Through training under the direction of an experienced breeder, we learn to rapidly and accurately through visual assessment score complex plant morphology, disease resistance, growth stages and a suite of other important plant characteristics. Analogous to this, we are using deep learning approaches to train convolutional neural networks to directly score phenotypes of interest from field imaged collected by high-throughput phenotyping platforms. These CNNs are developed using the "breeder-trained datasets" consisting thousands of images that are geo-referenced to a single field plot and matched with a breeder defined score. With these type of datasets we have demonstrated 100% accuracy for plot-level scoring of awned / awnless and heading stage in wheat. The scope of using breeder knowledge to train neural networks has the possibility to enable phenotyping for any traits of interest, and to do so in high-throughput. Such tools will enable new possibilities for breeding and genetics to operate on a scale much larger than is currently possible.
BIO:
Jesse Poland is an assistant professor at Kansas State University, director of the Feed the Future Innovation Lab for Applied Wheat Genomics, and associate director of the Wheat Genetics Resource Center. Research in his group is focused on wheat genetics, genomics and germplasm improvement. They are currently developing new approaches in quantitative genetics, genomics and high-throughput phenotyping for use in breeding, diversity studies, and association genetics. In collaboration with public breeding programs, Dr. Poland is implementing the use of genomic selection methods to accelerate wheat breeding. In the area of germplasm development, his group is focused on developing new breeding lines with resistance to the major pests of wheat including stem rust, stripe rust, leaf rust and Hessian Fly, as well as understanding the genetic basis of these traits. To complement advances in genomics, his lab is developing high-throughput phenotyping approaches for field-based evaluation of breeding lines with the primary focus being genetic characterization of heat and drought tolerance and development of improved germplasm. Visit his lab’s website at www.wheatgenetics.org.