Dissertation

Plant Phenotyping with Limited Annotation: Doing More with Less

Nagasubramanian, Koushik

2022

Overview

Deep learning (DL) methods have transformed the way we extract plant traits – both under laboratory as well as field conditions. Extraction of plant phenotypes with DL can be a game changer in understanding and adapting crops for climate change. These extracted phenotypes can be linked with genotype, environment, management and their interactions to help farmers and breeders. Recent advancements in computation and sensor technology have enabled the cheap collection of high-resolution phenotype data across a large geographical area with high temporal resolution. Continuous increase in the amount of data collected and annotated has made it possible to apply deep learning algorithms successfully in a wide variety of challenging plant phenotyping tasks. Training a DL model typically requires the availability of copious amounts of annotated data; however, creating large-scale annotated dataset requires non-trivial efforts, time, and resources. This limitation has become a major bottleneck in deploying DL tools in practice. In this thesis, we will present a variety of ways in which one can circumvent the need for large annotated datasets for plant phenotyping. Using plant phenotyping project examples, we will illustrate transfer learning, active learning, self-supervised learning and domain adaption approaches. Utilizing these approaches can lower the barrier to deploying ML tools for plant science applications.

• Publisher: ProQuest Dissertations & Theses
• Subject: Electrical engineering
• ISBN: 9798351411248