Prediction of seed vigour and storability
Seeds are an unequivocal part of crop and food production for human consumption. Seed vigour is both a critical aspect and complex trait to the successful establishment of crops in the field or greenhouse, thereby indirectly linked to crops yields and food security. Seed vigour can be roughly defined as ‘the potency of seed to establish quickly into a healthy seedling across a diverse set of environmental conditions’.
Seed vigour will decline during seed storage. Currently, seed companies cannot predict which seed lots, or which seeds in the population will deteriorate faster or slower during storage. Neither can they distinguish between individual high or low vigour seeds from a stored seed lot.
In this project, we aim to find a correlation between the properties of a young seed and the seedling that originates from it after a long period of storage. This will help to predict the seed storability. Furthermore, we aim to design a method based on this correlation to sort seeds into classes with uniform seed vigour.
To obtain these goals, we measure the seed quality using a broad range of sensors and use machine learning to correlate the seed quality data of individual seeds with the seedlings that originate from the artificially resp. naturally aged seeds.
The project contributes to the ST1 Smart Technologies in Agri-Horti-Water-Food by developing non-destructive measurements to determine seed quality of individual seeds, to maintain the identity of the seed during storage and to develop an AI-based model that correlates the resulting seedling quality with the original young seed quality. This model will fuel a decision support system that determines which seed batches are ready for market and which can be stored even longer.
The project fits well with the goals formulated in the MMIP because of its contribution to (i) a combination of a broad range of non-destructive measurements on individual seeds in a batch (NIR, VIS, X-ray, RGB,...). This allows for a sorting of seed batches on seed vigour prediction based on individual measurements and (ii) a smart decision support system based on (iii) AI-based models that identifies the remaining storability of a pre-sorted batch. As a consequence, seeds are stored more efficiently, until exactly the right moment. Therefore, less seeds need to be discarded and less energy needs to be wasted on growing seedlings of insufficient quality and uniformity.
The intended set of deliverables is as follows:
(i) Pilot experiments:
a. three SOPs (standard operating procedures) to capture the logistics of the individual seed/seedling processing in the seed-seedling chain.
b. Estimation of the number of seeds necessary for the full experiments
c. A dataset of seeds and corresponding seedlings to develop the machine learning algorithms on
(ii) Artificial ageing experiment: three seed lots will be artificially aged using an EPPO treatment to have aged seeds available halfway the project period
a. The EPPO procedure for individual seeds will be developed
b. Report describing the seed vigour of artificially aged seeds
(iii) Long storage experiments: three seed lots will be stored at the beginning of the project to have naturally aged seeds available at the end of the project
a. Report describing the seed vigour of artificially aged seeds
(iv) Volatile Organic Compounds (VOCs) experiment: determine which VOCs are created by the seeds and can be measured using different measurement methods.
a. Report describing the head space measurements of VOCs
(v) Model forming
a. Report describing the ML method that gives the best correlation for ageing after storage between seeds and seedlings
b. Database containing all collected data