Computer Science and Engineering, Department of

 

First Advisor

Ashok Samal

Second Advisor

James C. Schnable

Date of this Version

5-2017

Comments

A THESIS Presented to the Faculty of The Graduate College at the University of Nebraska In Partial Fulfillment of Requirements For the Degree of Master of Science, Major: Computer Science, Under the Supervision of Ashok Samal and James Schnable. Lincoln, Nebraska: May, 2017

Copyright (c) 2017 Bhushit Agarwal

Abstract

The phenome of a plant is the sum of all observable phenotypes for that plant. Phenotypes are observable characteristics or traits of a plant. These traits generally reflect a combination of influences from the genotype of the plant and the environment in which the plant has grown and developed. Collecting phenotypic data has traditionally been a slow and costly manual process, albeit one highly necessary for plant breeding and the development of improved agronomic practices. As a result automated methods for plant phenotyping analysis have become an active research field in recent years. Image-based plant phenotyping analysis facilitates extraction of meaningful phenotypes by analyzing a large volume of plants in a relatively short period of time with little or no manual intervention. This thesis introduces a study of specific phenotype: the timing of emergence from the soil for new seedlings.

We define emergence to be the time when the plant is first visible above the surface of the soil. Emergence is an important phenotype which not only reflects variation in seed quality and incomplete breaking of dormancy for seeds of different genotypes or seeds produced under different conditions but also helps determine various aspects of the plant growth at an early stage. Uneven emergence timing is associated with lower yields and poor farmer acceptance.

Current methods for detected emergence require manual inspection and generally have a precision of only approximately 24 hrs. Several different definitions of emergence are employed, some of which define emergence to be the time when first leaf of the plant is visible to provide ease of detection. In this thesis, we present an automated approach to determine the timing of emergence, presented in form of a time window from a sequence of images. The upper bound of the window is estimated by using a backward (in time) examination of the images starting at a time when there is confirmed emergence. This is determined by first segmenting the plant from background soil. Then we determine the time when the plant is green enough and big enough to be differentiated from the soil. Spatial information of the plant is used to segment the plant at night to accommodate illumination and color variations. The lower bound of the emergence window is determined by analyzing the movement in soil before emergence of the plant using optical flow. Adding information on variation in the amount of green signal included in each individual time-lapse image complements the accuracy of the lower bound placement.

The emergence detection algorithm is evaluated with a set of 76 plants captured at 5 mins interval. Emergence was observed in 40 plants and results show that the predicted time of emergence for 95% of the plants was within 10 hours and for 70% of the plants was under 4 hours of the ground truth.

Adviser: Ashok Samal and James Schnable

Share

COinS