Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.

Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Quantitative Genetics and Phonemics in Crops Using Statistical and Machine Learning Approaches

Chenyong Miao, University of Nebraska - Lincoln


Plant biologists seek to meet the growing food demands in the world by developing high yielding and more resilient crop varieties. Advances in both quantitative genetics and high throughput phenotyping have the potential to facilitate this work to improve crop qualities. Genome-wide association studies (GWAS) are approaches to identify the genes controlling variation in phenotype within a species. While many statistical models exist for GWAS, the relative strengths and weaknesses of these models in crop species were not well elucidated. In the first chapter, current GWAS models were evaluated using real world genetic data from four crop species and different assumptions about genetic architecture and heritability. The second chapter presents a new semantic segmentation approach to measure morphological phenotypes in sorghum. This approach lets researchers measure plant traits using automated phenotyping which previously required time intensive hand measurements of the same plants. Automated phenotyping also makes it easier to measure how the phenotypes of individual plants change over time. The third chapter adopts a statistical approach called functional PCA model for conducting GWAS in sorghum using time series data. The approach presented can help researchers better understand how an individual gene plays in determining plant phenotype over time. Leaf number, and the timing of leaf emergence, is another important agronomic trait of interest and of use to plant breeders and plant biologists. However, work on the computer vision task of leaf counting has focused on Arabidopsis because that is where the training data has been. In the last chapter, a new benchmark image dataset was generated including annotating the number and position of each leave in over 150,000 maize and sorghum images. I show that machine learning models trained using this dataset achieves leaf counting performance comparable to humans in maize. The data, approaches, and conclusions presented in this dissertation provide valuable knowledge to guide the improvement of crop qualities in the future.

Subject Area

Agronomy|Information Technology|Artificial intelligence|Plant sciences|Genetics|Bioinformatics

Recommended Citation

Miao, Chenyong, "Quantitative Genetics and Phonemics in Crops Using Statistical and Machine Learning Approaches" (2020). ETD collection for University of Nebraska - Lincoln. AAI28258967.