Menu Close
 

Using Computer Vision to Relieve the Crop Phenotyping Bottleneck

Project Team

  • Dr. Andrew Leakey, Professor, Dept. of Plant Biology
  • Dr. Narendra Ahuja, Research Professor, Dept. of Electrical and Computer Engineering
  • John M. Hart, Principal Research Engineer, Coordinated Science Laboratory

Abstract

Improving the productivity, sustainability and resilience to climate change of agriculture depends on optimizing crop phenotype by understanding and manipulating the interactions among genotype, management and environment. Statistically resolving these complex interactions requires large volumes of data, which—in the case of phenotypes that range from microscopic to macroscopic in scale—are difficult to collect rapidly, accurately and precisely across thousands of plants and field plots. Therefore, research is critically needed to break the “phenotyping bottleneck.” Computer vision is one of the most potentially powerful tools to achieve this goal because it allows traits to be measured from images. This project will develop a set of computer vision tools to generate trait data from images that scale from microscopic patterns of cells to growth of plants in large field trials. Initially, we plan to focus on traits underpinning the genetics, physiology and agronomy of water use efficiency in major crops. Water use efficiency describes the ratio of crop productivity to crop water use, which is arguably the most important trade-off limiting the profitability, sustainability and resilience of agriculture in the U.S. and worldwide. The computational tools developed in this project will facilitate a quantum leap in the ability of crop scientists to capture phenotypic data across hundreds to thousands of crop genotypes, as needed to overcome key constraints to the application of quantitative genetics, breeding, molecular genetics and biotechnology to improve crop performance. The challenges of the large range of scales the phenotypes of interest span will require rethinking current computer vision approaches and algorithms. The work made possible by seed feeding would provide the necessary preliminary data to allow us to submit a collaborative proposal to the NSF Information and Intelligent Systems core program on Robust Intelligence in Fall 2019. The work would also feed into the development of the $20M NSF Mid-Scale Infrastructure grant that Leakey is leading with a large CDA team.