Novel deep learning-based methods for high-throughput image-based plant phenotyping and large scale crop yield prediction
Date
2022-12
Authors
Khaki, Saeed
Major Professor
Advisor
Wang, Lizhi
Hu, Guiping
Olafsson, Sigurdur
Beavis, William D
Nettleton, Dan
Wang, Zhengdao
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Farm crops provide food, feed grain, oil, and fiber for consumption and are a major component of export trades. Success of modern farming and plant breeding relies on accurate and efficient collection of data, and collecting accurate and consistent data is a bottleneck. Due to limited time and labor, accurately phenotyping crops to record color, head count, height, weight, etc. is severely limited. However, this information, combined with other genetic and environmental factors, is vital for developing new superior crop species that help feed the world’s growing population. Recent advances in machine learning, in particular deep learning, have shown a great promise in mitigating this data collection bottleneck in plant science.
In this dissertation, we propose novel deep learning and machine learning methods for efficient collection of phenotypic data based on inputs such as images, plant genotypes, weather, soil, and remote sensing data. More specifically, chapters 2-6 of this dissertation presents deep learning based methods for high-throughput image-based plant phenotyping. The proposed methods combine convolutional neural networks with computer vision techniques to accurately measure plant traits from images. Our extensive experiments and comparisons with other state-of-the-art methods demonstrate the superiority and effectiveness of our proposed methods which can be applied for different crops such as wheat, corn, and soybean.
One of the most complex phenotypic traits of the field crops is the crop yield, which is determined by multiple factors such as genotype, environment, management practices, and their interactions. Having accurate models for large scale crop yield prediction is of great importance to global food production. Policy makers rely on accurate predictions to make timely import and export decisions to strengthen national food security. Seed companies need to predict the performances of new hybrids in various environments to breed for better varieties. As such, chapters 7-9 of this dissertation presents neural network based methods for large-scale crop yield estimation and monitoring crop performances under environmental stresses such as heat and drought. Our proposed methods were used to forecast corn and soybean yield across the entire Corn Belt (including 13 states) in the United States. The proposed models achieved prediction error of around 8% of average yields, substantially outperforming all other methods.
Series Number
Journal Issue
Is Version Of
Versions
Series
Academic or Administrative Unit
Type
dissertation