Rice germination dynamics of five cultivars – LemnaTec SeedAIxpert case study.
When testing germination of crop seeds, the intended use is the main driver for setting criteria to call a seed batch a “good” one. When sowing the seeds in the field, farmers rely on high germination rates – any gap in the field is undesirable and lowers the potential harvest or could give space for unwanted plants growing in between the crop canopy. After sowing, seeds ideally should germinate at once so that all resulting plants have comparable growth stages. Delayed germination – similar to not germinating seeds – cause irregular plant cover in the field. Non-uniform crop stands can cause problems for harvesting, as early grown plants could become over-ripe while late-grown plants still remain unripe.
Thus, complete and uniform germination properties are important for well-performing crops.
To figure out such germination properties, seed testers place seeds on moist paper and observe the germination by repeatedly counting the number of germinated seeds during a given incubation period. Incubation conditions like temperatures and duration, together with the time points of inspection usually are given in seed testing guidelines, such as the ISTA rules.
The classical job of seed testers has some challenges that the testing personnel faces in their daily work:
Time-consuming and fatiguing counting process
Re-assessments nearly impossible
Personal impressions taking influence on the result
To overcome such challenges, LemnaTec has developed SeedAIxpert, our digital seed testing system. It assists the seed testers by taking images of the sample and by processing the images with a dedicated analysis software. The result of the analysis is a count of germinated seeds per sample. Thus, SeedAIxpert eases the counting process as the operator no longer must count and take notes by himself, but leaves that fatiguing part of the work to the computer. The process is much quicker than the classical counting process, with a few seconds for the image taking and the analysis running in the background. Therefore, throughput can be increased. Whenever a result appears strange, the seed testers can go back to the corresponding image and have a closer look what could be the reason. This image-based documentation is a strong advantage compared to the classical note-taking process, because once the operator becomes aware of the unusual result later, seedlings might have changed their habitus, or the sample was discarded anyway. In seed testing, frequently multiple testers work in parallel and probably in shifts to ensure adequate throughput. Of course, all are well-trained and following the guidelines, but the human factor never can be avoided totally. With clear documentation and consistent image processing, SeedAIxpert makes the testing process more reliable and comparable by reducing the human factor.
In the case study, we compare the germination performance of five rice cultivars, all of them incubated under controlled conditions according the official guidelines for rice germination testing. The rice seeds, 50 seeds per cultivar, were placed onto moist paper. Images were taken between three and eleven days after seed moistening using the LemnaTec SeedAIxpert.
All images were processed using LemnaTec’s machine learning algorithms that were specifically trained for rice seed analysis.
The algorithms detect the seeds, emerging roots, and shoots and separate them from the background. This detection and classification process is visualized by color-coding the detected elements.
With these detection and classification it is possible to count seeds and seedlings, determine the fraction of germinated seeds, and measure the size of the seedlings including their roots and shoots.
With the data output, we can compare the germination status of the cultivars at given time points, e.g. at day 6 of the testing period.
There we see that three cultivars – 1, 3, and 4 had already high germination percentages around 80%, but the other cultivars were still below 30% germination.
Even more interesting than a single-time comparison are time courses, where we can see the timely resolved development of the germination process.
In the case study, seeds of cultivars 1, 3, and 4 germinated quickly and reached high germination rates, with cultivar 4 being the fastest and having nearly 100% germination. Cultivar 2 and 5 were slow in germination and reached low percentages during the observation time.
The image series below exemplarily shows the time-resolved development of germination in cultivar 3. Image series and corresponding analyses for this cultivar and all other cultivars are the data basis for the time-course diagrams showing germination percentages and seedling sizes.
Beyond germination detection, the algorithms measure the length of emerged roots and shoots for each seedling. At this point, the SeedAIxpert’s data output exceeds the capability of the visual inspection typically done by seed testers. When visual inspections are done for 50 seeds on a paper, counting germinated vs not germinated takes some minutes per paper sheet, however it would take hours if one would hold a ruler on each shoot and root growing out of the seeds. With SeedAIxpert, this additional information is given instantly together with the germination percentage. This adds value to the data, because such information quantifies how strong the cultivars grow roots or shoots.
Germination percentage time courses of cultivars 1 and 3 were nearly identical, however taking the add-on information – root- and shoot-length – into account, we could show that the seedlings of cultivar 3 were larger (longer roots and shoots) than those of cultivar 1 towards the end of the observation time. Without having the numbers that underpin the size differences, there might have been a visual impression noted down by a seed testing operator – e.g. “cultivar 3 has large seedlings”. Such notes however are no clearly defined values in contrast to a seedling size measurement.
The cultivars exhibiting low and slow germination, had small seedlings with short roots and shoots, indicating not only impaired germination properties but also retarded growth. They might catch up in size later, however an initial slow and small growth is a clear disadvantage when growing in a crop field.