Lecture ‘Automated thresholding for seed processing to maximize seed lot value‘
How to define the borderline between usable and low-quality seeds: The phenoCheck and
phenoTest as enabling tools to automatically generate annotated phenotypic training data for seed processing. The key challenge in seed processing is to determine the optimum balance between maximising output in seeds while ensuring a sufficient seed quality. Even a low percentage of low-quality seeds can decide between sale of a certified seed lot or discarding and financial loss. Likewise, imprecise processing can lead to the loss of potentially usable seeds or unnecessary and costly additional processing regimes. Whether the seed sorting is based on form, size, specific weight, colour, spectroscopy, X-ray or other technologies, all depends on finding the right thresholds to classify between the best and usable seeds and those with low germination or vigour. Due to the lack of quantitative, high-throughput and reproducible tools to phenotype seeds and the resulting seedlings, these thresholds remain essentially a black box.
Here we present two technologies, the phenoCheck and phenoTest which alleviate these shortcomings of conventional phenotyping and present a powerful new avenue to optimise seed quality and processing through unprecedented annotated phenotypic big data. The phenoCheck allows to determine seed quality by allowing to phenotype seeds, including their internal structures, through X-ray in 3D at high-throughput. This technology may be used in-line or out-of-line of seed processing to guide the processing steps based on the 3D seed quality data. The phenoTest is a fully-automated X-ray-based technology that allows to 3D phenotype individual seedlings, including the AI-based segmentation and measurement of their individual organs, over their development, i.e. in 4D at high-throughput.
The combination of both technologies allows 1:1 correlation of 3D-internal seed features, or other seed parameters such as colour or weight determined with other technologies, to the germination capacity, vigour and plant phenotype of the resulting seedlings. Thus, the technologies, for the first time, allow to produce highly standardised, quantitative annotated big data relating the seed phenotype to seed quality and plant development. Thus, they enable seed producers, seed processors and breeders in making maximum use of the produced seed with the best possible quality for an economically-optimised seed production and a sustainable crop production by providing seeds with optimal field emergence, vigorous plant development and high yield for the farmer.