• Research

SeedGerm - The next generation phenotyping platform to quantify crop seed germination and seedling vigour

A novel system that automates reliable, scalable and cost-effective seed phenotyping to assess seed performance.

Project summary.

Led by: Zhou Group

Start date: December 2016

End date: March 2019

Duration: 30 months

Grants: Syngenta and EI industrial collaboration funding; NRP translational fund; BBSRC follow-on pathfinder fund

Value: £101,575 (cumulative)

SeedGerm is a novel stand-alone software (machine-learning based image analysis) and hardware (automated mobile imaging control) solution, capable of assessing key commercial traits (e.g. seed germination and seed vigour) for a wide range of seed varieties. The identified target industry segments are seed technology, seed testing, agri-tech and research. This project is critical to define opportunities within each of these segments and outputs will enable us to deliver a feasible plan to introduce the technology to industry and academia within 2-3 years.

Seed germination frequency and seed vigour are the two most important measurements for seed performance. Germination is used for certification, developing guidance on sowing density and insurance models. Seed vigour, describing rates and uniformity of seed emergence, is closely linked to faster canopy closure, weed suppression and crop yields. Associated measurements are heavily reliant on manual inspection by in-house experts or external testing labs.

There have been some attempts towards digitalising seed tests but most available software solutions are low-throughput, inaccurate, dependent on other external software, compatible for a narrow range of plant species and have limited experimental setup. Hence, an advanced seed scoring platform such as SeedGerm will enable high throughput germination analysis together with our industrial partners (including Syngenta) to facilitate the high throughput phenotyping of germinating seed lots. 

Details.

Tackling the seed testing bottleneck, the Zhou laboratory at EI, the Penfield group at JIC, and the Rene laboratory at Syngenta jointly developed SeedGerm. SeedGerm software utilises novel machine-learning image analysis to train and classify seed performance traits, including seed germination frequency and seed vigour measurements, for different crops such as corn, brassica, pepper and tomato. 

The Zhou laboratory at EI developed a standalone SeedGerm scoring hardware and a tailored software application to automate image capture, quality control and a GUI interface for various experimental setup, e.g. plant species, seed number, panel number and measurement parameters.

SeedGerm hardware development has run parallel to the analytic software and has reached the final prototyping stage. It is designed as a scalable and low-cost benchtop device capable of image capture and microenvironment control (temperature and humidity). Currently, the small box version can handle testing of approx. 360 seeds (brassica) and has been successfully trialled at JIC for high throughput germination experiments on the Brassica napus ASSYST panel. The big box version can handle different over 1200 seeds, with different crop species and varied experimental settings. 

Outputs of the project will be reviewed and support an application for a standard BBSRC FoF. SeedGerm technology (software and hardware) will be developed to an industrial standard and include improvements based on the Market Survey outputs. A comprehensive commercial strategy and business plan will be developed to fully commercialise SeedGerm within 2-3 years. A deep learning based software solution, convolutional neural networks (CNNs) and recurrent neural networks (CNNs), will be developed. So we will be able to train and learn more features from seed germination image series to:

1) Provide more accurate germination scores and vigour quantifications.

2) Utilise increasing training datasets to expand SeedGerm software to analyse different plant species for new revenues. 

3) Monitor imbibition, tracking of root growth, as well as produce predictive models using environmental factors (temperature and humidity) to simulate seed germination under the real field environment.

Collaborators.

Professor Steven Penfield

John Innes Centre

Dr Rene Benjamins

Syngenta

Impact statement.

Since 2016, both Syngenta (Syngenta Seeds, Netherlands) and JIC have trialled the software as part of their existing research programmes for testing tomato, corn, pepper and brassica. Highlighting commercial potential, Syngenta purchased a non-exclusive licence for the early-stage Phase Two software in 2016.

Related reading.