Dr. Marcus Jansen: Artificial intelligence enables new ways in seed quality testing

Abstract
Seed assessments at various levels traditionally rely on visual inspections, but optical technologies increasingly take over rating tasks. While seed counting with computer vision is already established, optical inspections of germination processes, seedling quality, or seed purity are gaining importance. The implication of machine learning enables training the image processing algorithms in a way that the dedicatedly recognize features that are important for specific ratings. Thereby, it becomes possible to move automatic imaging-based inspection of seeds and seedlings closer to the demand of the rating experts. Beyond application-specific image processing, key advantages of such methods are consistent documentation of the samples by storing the original images and measuring options for all recognized items. These measuring options enable determining the size of each and every detected seed, seedling, root, shoot, or leaf. They can be measured for a range of parameters, including length, width, area, or information on colors and morphology. Many of these data give added value compared to visual scoring. Moreover, artificial intelligence enables recognizing sample- or user-specific quality traits in seeds and seedlings. Thereby, algorithms can be trained to discriminate between normal and non-normal seeds and seedlings. All imaging can be combined with automation technology that increases the throughput of samples. Thus, imaging and image processing is applicable at all scales from small labs to large factories.

More information: https://www.plant-phenotyping.org/ind...