A Latent Variable Model For Plant Stress Phenotyping Using Deep Learning

Abstract: Plant phenotyping involves the characterization of physical traits in plants, often with the goal of describing physical differences between genotypes or differentiating their responses to the environment. Previously done by hand, phenotyping today is largely done via image analysis. Image-based phenotyping provides a high-throughput, non-destructive way to analyze populations which would otherwise be prohibitively large to phenotype manually. In recent years, deep learning has emerged as a common tool for image-based plant phenotyping. In this presentation I will briefly outline deep learning and describe the different types of plant phenotyping tasks which have been approached in the literature using deep learning. I will present some of the work I have done in the discipline including an open-source software platform for deep learning in plant phenotyping, as well as the use of plant models to generate synthetic training data for deep neural networks. Finally, I will introduce Latent Space Phenotyping, the first latent variable model for performing plant stress phenotyping directly from images. Five validation case studies using both natural as well as synthetic image data will be presented.

Biography: Jordan Ubbens is a PhD candidate in Computer Science at the Plant Phenotyping and Imaging Research Center (P2IRC) at the University of Saskatchewan, under the supervision of Dr. Ian Stavess. His primary research interest is in the application of deep learning and computer vision to plant phenotyping. As a PhD member of P2IRC he has published three first-author journal papers and in several international conferences. Jordan completed his M.Sc. in Computer Science at the University of Regina in 2015.

Friday March 27, 2020 at 2:30 PM in Biology 106

Doors Open at 2:00 PM