Abstract
High-throughput phenotyping is an important means to meet the agricultural needs for future food and energy production. This entails an increasing amount of work in Image-based, non-destructive phenotyping systems. This thesis de-scribes a low-cost phenotype collection system for growth chambers, and methods to segment plants from time series images using temporal information. The system uses a webcam to record plant growth in a top-down view with afixed time interval to create time-lapse images of multiple plants. It has successfully recorded the growth of Arabidopsis thaliana over three months from seedling to flower. The development of plant segmentation methods involves experiments to compare and select the optimal colour space for plant segmentation, and the development of an unsupervised plant segmentation method that is capable of segmenting multiple plant species (e.g. Arabidopsis thaliana, Oats, Oilseed Rape) without relying on knowledge of plant colour. The method is also modified to provide colour-based, superpixel-based and supervoxel-based approaches to the segmentation of plants from time series images.
| Date of Award | 2018 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Hannah Dee (Supervisor) & John Doonan (Supervisor) |