Theses and Dissertations

Advisor

Zhang, Xin

Committee Member

Liu, Wenbo

Committee Member

Wijewardane, Nuwan

Committee Member

Bheemanahalli, Rangappa Raju

Date of Degree

5-16-2025

Original embargo terms

Immediate Worldwide Access

Document Type

Graduate Thesis - Open Access

Major

Biosystems Engineering

Degree Name

Master of Science (M.S.)

College

James Worth Bagley College of Engineering

Department

Department of Agricultural and Biological Engineering

Abstract

Root phenotyping provides insights into root structure, function, and response to environmental stimuli, aiding researchers in improving crop yields, developing new plant varieties, and enhancing plant health. Traditionally, measuring root traits, such as length and area, is manual, time-consuming, and resource intensive. This study introduces a deep learning-based computer vision method for soybean root segmentation and traits estimation using RGB (red, green, blue) images captured under controlled illumination. A separate web-based application was also developed that can extract the root length and area using RGB images. A diverse set of 3,363 soybean root images was collected using an in-house root imaging platform with a resolution of 5,184×3,456 on a black background. Ground truth data were recorded manually and using well accepted tools such as WinRHIZOTM and RhizoVision. We adopted and fine-tuned a convolutional neural network (CNN)-based model, ITErRoot, to segment roots and developed a pipeline to extract root traits using Python followed by integration as a web application. The results demonstrated that these two phenotypic traits extracted by our approach were highly correlated with WinRHIZOTM (R2=0.79) and RhizoVision (R2=0.82) in case of total root lengths of complex roots with long and thin root hairs while it achieved RMSE of 1.3 cm against manually measured simple roots without hairs. We also compared our predicted values with dry weights and were able to achieve correlation of (R2=0.45) in both cases i.e., total root length and root area. Further we evaluated this approach for root area estimation and were able to achieve correlation (R2) of 0.78 and 0.79 with WinRHIZOTM and RhizoVision respectively.

Download

Share

COinS