Mobile applications for high-throughput seed characterization

Abstract

Kansas State University is a world leader in the study of small grain genetics to develop new varieties which tolerate a wide range of environmental conditions. A phenotype is a composite of a plants observable traits. Several mobile applications, called PhenoApps, have been developed for field-based, high-throughput phenotyping (HTP) to advance plant breeding programs around the world. These applications require novel image analysis algorithms to be developed to model and extract plant phenotypes. Some of the first algorithms developed were focused on using static image analysis to count and characterize a wide variety of seeds in a single image with a static colored background. This thesis describes both a static algorithm and development of a hopper system for a dynamic, real-time algorithm to accurately count and characterize seeds using a modest mobile device. The static algorithm analyzes a single image of a particular seed sample, captured on a mobile device; whereas, the dynamic algorithm analyzes multiple frames from the video input of a mobile device in real time. Novel 3D models are designed and printed to set a steady flow rate for the seeds, but the analysis is also completed to consider seeds flowing at variable rates and to determine the range of allowable flow rates and achievable precision for a wide variety of seeds. Both algorithms have been implemented in user-friendly mobile applications for realistic, field-based use. A plant breeder can use the applications to both count and characterize a smaller sample using the static approach or a larger sample using the dynamic approach, with seeds sampled in real time without the need to analyze multiple static images. There are many directions for future research to enhance the algorithms performance and accuracy.