The aim of this EASTBIO studentship is to advance phenotyping in wheat using hyper-spectral sensing data and machine learning. SRUC is at the leading edge in applying spectral sensing in agriculture e.g. prediction of livestock traits and milk quality. The combined technologies of multi- and hyper-spectral imaging and machine learning have already been used by SRUC to create phenotype signatures to inform animal health and predict performance.
This studentship will gather high quality and high volume phenotype signatures from the growing wheat crop in order to generate predictive equations for genotype performance, including yield and quality. An output will be phenotypes for wheat breeders and the crop improvement research community.
Key objectives are to:
(1) Couple spatial variation in crop growth and development to signals from multi- and hyper-spectral imaging.
(2) Optimise spatial resolution in spectral imaging at the trait, plant and crop level for genotype discrimination.
(3) Use crop and field data to predict crop yield and grain quality.
(4) Publish wheat phenotype signatures and link phenotype to genotype.
The project will involve field work over 3 growing seasons to link imaging from multi- and hyper-spectral cameras, from rigs and a UAV (drone), with ground measurements. In years 1 and 2, the student will gather phenotype data in densely-sampled field-grown crop plots from a diverse collection of 600 wheat genotypes in a multi-parent advanced generation inter-cross (MAGIC) population, sourced from NIAB. A third field year will be used for validation and gap-filling in trait or plant information. In order to widen phenotypic expression, agronomic treatments, such as fertilizer and crop protection inputs, will be used manipulate leaf and canopy development, as well as yield and grain quality. Initial work will include testing of sample size and spatial resolution according to the range of phenotypic expression.
After data collection, interrogation of multi-spectral imaging with machine learning will test how different spectral signatures relate to genotype. Data processing embracing state of the art deep learning techniques will be used to examine spectral signatures – from traits, whole plant and crop – for development of predictive tools for crop performance. Statistical inference, whereby properties of the underlying ‘population’ of data are estimated, will explore specificity and sensitivity of the spectral signatures according to genotype.
Crop models will be trained on SRUC’s machine learning server harnessing the computational power supplied by 4 Nvidia Tesla v100 GPUs. Variation in phenotypic expression, including properties of leaf canopy architecture, components of yield, rate of grain filling, rate of grain ripening will be linked to yield and grain quality. This work will establish a set of reference phenotypes that can be used to link traits to genotype.
Compared with existing methods and protocols for crop monitoring and evaluation our integrated spectral imaging and machine learning approach has potential for producing high quality and high volume phenotype signatures. This project, and the student employed, would support research and industry capability for crop characterisation and linking phenotype with genotype, with application to crop improvement and breeding.
Applicants should download the required forms from http://www.eastscotbiodtp.ac.uk/how-apply-0
and send the following documents to [email protected]
a. EASTBIO Application Form
b. EASTBIO DTP Equality Form
d. Academic transcripts (a minimum of an upper second class or first class honours degree or equivalent is required for PhD study
e. Two references should be provided by the deadline using the EASTBIO reference form (http://www.eastscotbiodtp.ac.uk/how-apply-0
). Please advise your referees to return the reference form to [email protected]
f. If you are nominated by the supervisor(s) of the EASTBIO PhD project you wish to apply for, they will provide a Supervisor Support Statement.