Image-based Microplot Segmentation/Detection and Deep Learning in Plant Breeding Experiments
Date
2023-10-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
0000-0001-5125-6207
Type
Thesis
Degree Level
Doctoral
Abstract
In the coming years, the agricultural sector will encounter significant challenges from population growth, climate change, and evolving consumer demands. To address these challenges, farmers and plant breeders actively develop advanced plant varieties with enhanced productivity and resilience to harsh environmental conditions. However, the current methods for evaluating plant traits, such as manual operations and visual assessment by breeders, are time-consuming and subjective. A promising solution to this issue is image-based phenotyping, which leverages image-processing and machine-learning techniques to facilitate rapid and objective monitoring of numerous plants, enabling breeders to make more informed decisions. In order to perform per-microplot phenotypic analysis from the imagery and extract phenotypic traits from the field, it is necessary to identify and segment individual microplots (a small subdivided area within a field) in the orthomosaics. Nonetheless, the current procedures for segmenting and identifying microplots within aerial imagery used in agricultural field experiments necessitate manual operations, resulting in considerable time and labour investments. By automating this process, the evaluation of microplot phenotypes, such as physical traits, can be expedited, facilitating automated monitoring and quantification of plant characteristics. Our objective is to develop novel phenotyping algorithms to segment, detect, and classify microplots using image-processing and machine-learning techniques to achieve the goal. The thesis comprises four projects such as a comprehensive review of vegetation and microplot segmentation methods, the development of algorithms for the detection of both rectangular and non-rectangular microplots, and the utilization of deep learning techniques to predict lodging on microplots and highlighting the impact of deep learning on microplot phenotyping. These innovative approaches possess broad applicability in remote sensing field trials, encompassing diverse applications such as weed detection, crop row identification, plant recognition, height estimation, yield prediction, and lodging detection. Moreover, our proposed methods hold great potential for streamlining microplot phenotyping efforts by reducing the need for labour-intensive manual procedures.
Description
Keywords
Segmentation, Plant Phenotyping, Deep Learning, Image Processing, Optimization, Detection
Citation
Degree
Doctor of Philosophy (Ph.D.)
Department
Computer Science
Program
Computer Science