(AFS) Developing a deep learning-based 3D imaging platform for tracking and modelling whole plant growth responses to environmental &chemical stresses

There is an urgent need for novel technologies that allow scientists and agronomists to monitor and predict the combined impacts of environmental stresses (abiotic and biotic) on plant growth and performance. The capacity to track relevant growth traits at
the whole-plant level is significantly increased if 3D data is available, which can be used to understand and model the response of a given genotype to a given environment. Recently, photometric stereo (PS) 3D imaging has been shown to be able to resolve changes in leaf surface textures and morphology induced by different stresses that are not detectable using 2D imaging methods (ongoing BBSRC funded work in SBS and CMV). PS offers unprecedented spatial resolution for 3D data and is inherently low-cost. Working in a
multidisciplinary environment, the candidate will develop deep-learning approaches to automate the extraction of trait characteristics from PS data and build a unique, field-level image analysis platform to accurately determine the impact of different chemical treatments on different genotypes based on their growth environment.

Project plan.
Methods will be validated initially in the model species Arabidopsis thaliana, after which the candidate will investigate a range of larger grass and broadleaf plants. A specific focus will be on wheat cultivars under glasshouse and field conditions. The Williams' group (GeoSciences) will provide support for conducting wheat field analyses from the BBSRC-NERC funded SARIC project BB/P004628/1. The availability of chemically treated and untreated wheat field trials will facilitate field testing of PS systems, and PS data will be complemented by available drone-based image datasets (RGB, thermal and multispectral) where "structure from motion" approaches have been used to reveal 3D crop canopy traits.
The candidate will investigate if combining trait data from these different datasets will help to improve robustness in detecting key response functions, which then could be added to available crop models to better predict the sensitivity of wheat to combined stresses and optimise the timing of chemical treatments.

Intended impact.
The PS system will be developed as a high-throughput tool to analyse the impact of different chemical treatments on Arabidopsis grown in differing temperature, light or droughted environments. Work in Arabidopsis will also allow investigation of treatments on vulnerable genotypes known to affect growth. Subsequent work in wheat trails will test the real-world capacity of the PS system to provide farmers with useful data, such as early stage detection of stress and/or disease and vulnerability of particular genotypes, and improving the efficiency and sustainability of chemical treatment applications.