W550 Using Computational Methods to Quantify, Describe, and Retrieve Visual Phenotypes

Date: Sunday, January 15, 2012
Time: 9:30 AM
Room: Golden West
Jason Green , University of Missouri, Columbia, MO
Heidi Appel , University of Missouri
Jaturon Harnsomburana , University of Missouri
Adrian Barb , Penn State Great Valley
Peter Balint-Kurti , North Carolina State University
Chi-Ren Shyu , Univeristy of Missouri, Columbia, MO
In ever increasing ways, biological discovery is facilitated through the utilization of computational methods to automate parts of the experimental process.  With the increased use of imaging to capture phenotypic expression, the development of computational techniques for quantifying traits, which are critical for many plant genetic, breeding, and ecologic experiments, has gained considerable attention.  This is a particularly important topic, as trait measurement has traditionally been a slow, laborious, and at times subjective or imprecise process and is a bottleneck in many large-scale studies.  We first introduce PhenoPhyte, a new online application to assist plant researchers in automatically quantifying area, growth, and herbivory from imagery.  These images must conform to a simple and low-cost protocol that can allow phenotype capture, both destructively and in situ, in the lab, greenhouse, or field.  Our approach can be generalized to more complex phenotypes, and this is shown with methods for measuring disease resistance to Southern Leaf Blight in maize.  Because plant scientists are accustomed to utilizing semantics to describe phenotypic appearances, we also discuss a method, utilizing our ontological structure VPhenoO, for mapping automatically-generated numeric phenotype measurements into qualitative semantics or descriptions.  This work paves the way for automatic semantic annotation of phenotype images through the modeling of individual or group visual perceptions. Finally, we describe a pair of advanced search mechanisms, part of VPhenoDBS, that have been constructed based on automatic phenotype measurements.  These types of search engines could revolutionize the way plant scientists search their phenotype image collections.