SeedGerm-VIG: an open and comprehensive pipeline to quantify seed vigor in wheat and other cereal crops using deep...

As one of the most important cereal crops, wheat (Triticum aestivum L.) production and grain quality are essential to many nations in the world. Early developmental phases such as seed germination and seedling establishment are key to wheat's growth and development as they impact directly on a crop's early performance and yield potential. Hence, it is critical to develop varieties with favorable early growth characteristics under various growing conditions.

Here, we present SeedGerm-VIG, an automated and comprehensive pipeline developed for assessing seed vigor in wheat and other cereal crops. Building on the SeedGerm system, we integrated multiple deep learning models (i.e., YOLOv8x-Germ and optimized U-Net) and computer vision algorithms into the automated seed-level analysis pipeline to identify key germination phases and measure seed-, root-, and seedling-level phenotypic traits. Then, by using a time-series directed graph, we not only tracked root tips to reliably measure root emergence during the germination procedure (seed-lot R2 = 84.1%) but also established a new approach to examine speed and uniformity of seed germination. These resulted in the establishment of a vigor scoring matrix, through which 21 commercial genotypes' (n = 494 randomly sampled seeds, with over 29,500 seed-level images) vigor scores were summarized and evaluated at key phases such as protrusion, radicle emergence, and chloroplast biogenesis. These measures largely matched with manual assessment based on the International Seed Testing Association (ISTA) guidelines. Finally, we also demonstrated that the SeedGerm-VIG pipeline could be used to assess seed vigor for other cereal crops, including rice (n = 120 seeds) and barley (n = 240 seeds), reproducibly.

In conclusion, we believe that our work demonstrates a valuable step forward to enable a broader plant and crop research community to examine seed vigor and vigor-related phenotypic features in an automated manner, facilitating effective and scalable plant selection and relevant seed science research for crop improvement.