A Deep-Learning Phenotypic Model to Estimate Key Yield related Traits in Field Conditions for UK Bread Wheat

Alkhudaydi, Tahani (2020) A Deep-Learning Phenotypic Model to Estimate Key Yield related Traits in Field Conditions for UK Bread Wheat. Doctoral thesis, University of East Anglia.

[thumbnail of 2020AlkhudaydiAPhD.pdf]
Download (139MB) | Preview


Wheat is one of the three major crops in the world with a total demand expected to exceed 850 million tons by 2050. One of the key challenges for wheat is to stabilise the yield and quality in wheat production. The application of the internet of things (IoT) in agriculture has enabled us to continuously monitor crop growth through networked remote sensors and non-invasive imaging devices. Analysis of the output of such systems with machine-learning algorithms and image processing techniques can help to extract meaningful information to assisting crop management.
Counting wheat spikelets from infield images is considered one of the challenges related to estimating yield traits of wheat crops. For this challenging problem, we first propose a model for semantic segmentation, SpikeSEG, to isolate the spike regions from noisy and redundant background. Segmentation results can then be used to improve phenotypic counting approaches. In-field spikelet counting is very challenging because of spikelet self-similarity, high volume in one image, and severe occlusion. For this, we propose a density estimation approach related to crowd counting, SpikeCount. Our proposed segmentation/counting methods are based on deep learning architectures as those have the advantage of being able to identify features automatically.
Annotation of images with the ground truth are required for machine learning approaches, but those are expensive in terms of time and resources. We use Transfer Learning in both tasks, segmentation and counting. We also investigated the effect of multi-task learning by using SpikeMulti and compared it with conventional workflow of applying segmentation and the density estimation.
Our results indicate the segmentation is beneficial as focusing only on the regions of interest improves counting accuracy in most scenarios. In particular, a combination of Transfer Learning and Multi-task learning produced excellent results for the counting task for most of the stages of wheat development.

Item Type: Thesis (Doctoral)
Faculty \ School: Faculty of Science > School of Computing Sciences
Depositing User: Nicola Veasy
Date Deposited: 17 Mar 2021 13:28
Last Modified: 17 Mar 2021 13:28
URI: https://ueaeprints.uea.ac.uk/id/eprint/79491


Downloads per month over past year

Actions (login required)

View Item View Item