Biological Systems Engineering

 

First Advisor

Yufeng Ge

Date of this Version

Spring 4-2023

Citation

Thapa, K.2023.Characterization of physical and biochemical traits in wheat and corn plants using high throughput image analysis (Master's Thesis).Biological Systems Engineering - Dissertations, Theses, and Student Research.

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: Mechanized systems management, Under the Supervision of Professor Yufeng Ge. Lincoln, Nebraska: May 2023

Copyright © 2023 Kantilata Thapa

Abstract

Plant phenotyping has been recognized as a rapidly growing field of research due to the labor-intensive, destructive, and time-consuming nature of traditional phenotyping methods. These phenotyping bottlenecks can be addressed by advancements in image-based phenotyping like RGB and hyperspectral imaging for the assessment of plant traits important for breeding purposes. This study aims (1) to characterize the physical and biochemical traits of wheat and corn plants using RGB and hyperspectral imaging in the greenhouse, and (2) to estimate leaf nitrogen (N), phosphorus (P), and potassium (K) content using hyperspectral imaging and an analytical spectral device (ASD spectrometer) and compare the performance from both datasets. Sixty wheat plants with 24 genotypes and 72 corn plants (a single genotype) with four different treatment combinations were manually measured and imaging was performed at different growth stages. RGB and hyperspectral images were processed to extract plant projected area (pixel count) and spectral reflectance, respectively. Partial Least Squares Regression (PLSR), Random Forest (RF), and Support Vector Regression (SVR) models were built to estimate N, P, and K contents from image-generated hyperspectral data, and from the ASD spectrometer. The results showed higher correlation for leaf area with plant pixel count with R2 of 0.75 for wheat and R2 of 0.68 for corn plants. For wheat plants, N was predicted more accurately with hyperspectral image datasets with R2 of 0.69 but P and K prediction was higher with ASD data using the PLSR model. For hyperspectral image datasets of corn plants, N prediction was higher using PLSR modeling with R2 0.66 whereas P and K prediction was higher using the RF model with R2 of 0.74 and 0.87 respectively. For corn plants using data from ASD, N, P, and K were predicted high by using the RF model with R2 of 0.67,0.41, and 0.69 respectively. RGB and hyperspectral imaging would reduce the need for manual measurement and chemical analysis of leaf tissue, and the technique can be validated in other crops with different architectures for high-throughput macronutrient estimation. The findings from this study can help integrate various disciplines of science, including plant breeding, agronomy, computer vision, mathematics, and engineering, for crop improvement.

Advisor: Yufeng Ge

Share

COinS