Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
Filters applied. Clear all
. 2020 Dec 22;16(1):157.
doi: 10.1186/s13007-020-00699-x.

Accurate machine learning-based germination detection, prediction and quality assessment of three grain crops

Affiliations
Free PMC article

Accurate machine learning-based germination detection, prediction and quality assessment of three grain crops

Nikita Genze et al. Plant Methods. .
Free PMC article

Abstract

Background: Assessment of seed germination is an essential task for seed researchers to measure the quality and performance of seeds. Usually, seed assessments are done manually, which is a cumbersome, time consuming and error-prone process. Classical image analyses methods are not well suited for large-scale germination experiments, because they often rely on manual adjustments of color-based thresholds. We here propose a machine learning approach using modern artificial neural networks with region proposals for accurate seed germination detection and high-throughput seed germination experiments.

Results: We generated labeled imaging data of the germination process of more than 2400 seeds for three different crops, Zea mays (maize), Secale cereale (rye) and Pennisetum glaucum (pearl millet), with a total of more than 23,000 images. Different state-of-the-art convolutional neural network (CNN) architectures with region proposals have been trained using transfer learning to automatically identify seeds within petri dishes and to predict whether the seeds germinated or not. Our proposed models achieved a high mean average precision (mAP) on a hold-out test data set of approximately 97.9%, 94.2% and 94.3% for Zea mays, Secale cereale and Pennisetum glaucum respectively. Further, various single-value germination indices, such as Mean Germination Time and Germination Uncertainty, can be computed more accurately with the predictions of our proposed model compared to manual countings.

Conclusion: Our proposed machine learning-based method can help to speed up the assessment of seed germination experiments for different seed cultivars. It has lower error rates and a higher performance compared to conventional and manual methods, leading to more accurate germination indices and quality assessments of seeds.

Keywords: Faster R-CNN; Germination indices; Germination prediction; Machine learning; Seed germination.

Conflict of interest statement

SJS is Managing Director at Computomics GmbH. The other authors declare that they have no competing interests.

Figures

Fig. 1
Fig. 1
Illustration of image collection, annotation and dataset generation module. a Setup for capturing images of the germination process of seeds within petri dishes. Subsequently, images have been cropped to only contain one petri dish per image. b Example of annotated images, where seeds have been marked with a bounding box and a class label (non-germinated in orange, germinated in blue). c Longitudinal images of a custom seed for 48 h. Orange frames around the images indicate that the seed is not germinated, gray indicates a difficult to label transition phase and blue indicates that the seed is clearly germinated. d The dataset was randomly split into a training, validation and test set, stratified by petri dishes. This ensures that seeds of the same petri dish are either in the training, validation or test set. In addition, it also ensures that a petri dish at different time points only appears in one of the sets
Fig. 2
Fig. 2
Key steps in mAP calculation for each model. a The IOU is the area of the overlap between the bounding box of the ground truth (GT) and the prediction (PD) divided by their union. b A typical Precision-Recall-Curve is shown in blue, the interpolated curve is shown in orange. The AP is the area under the curve, indicated in yellow
Fig. 3
Fig. 3
Normalized confusion matrix of test sets in percent for Inception-ResNet v2. True germination state as rows, predicted state as columns for the respective Inception-ResNet v2 model. Germinated seeds are denoted as “g”, non-germinated as “ng” and seeds which are not localized or classified by the model are denoted as “bg” (background). Green: Correct classification of the seed germination state. Yellow: Misclassification of the germination state. Orange: Incorrect localization of background as a seed (incorrect region proposal) resulting in seeds being detected multiple times. Red: Incorrect detection of a seed as background resulting in less detections than seeds present in the petri dish. a Zea mays (8809 detected instances) with a classification error of 4.1% and localization error of 0.9%. b Secale cereale (8564 detected instances) with a classification error of 12.1% and a localization error of 0.4%. c Pennisetum glaucum (8826 detected instances) with a classification error of 9.1% and a localization error of 2.2%
Fig. 4
Fig. 4
Examples of predictions on test datasets. Ground Truth is shown in dark colors (orange: non-germinated, blue: germinated) and predictions are shown in bright colors (yellow: non-germinated, cyan: germinated) a All seeds were correctly detected and predicted. b Four seeds were misclassified as germinated, as indicated by the green arrows. These errors can be rectified in the post processing step c Failed detection of one seed. as indicated by the green arrow. These time series were omitted when calculating germination indices
Fig. 5
Fig. 5
Description of different prediction errors. Predicted bounding box shown in yellow a Misclassification of a germinated seed. b Detection of one seed multiple times. c Failed detection of one seed: multiple seeds are detected as one
Fig. 6
Fig. 6
Germination curves for all three testsets. a Zea Mays, where the shape of the predicted curve resembles the ground truth curve, but is shifted because of misclassification of non-germinated seeds b ,  c The model does not only misclassify non-germinated, but also germinated seeds in Secale cereale (b) and Pennisetum glaucum (c)
Fig. 7
Fig. 7
Relative error of assessment compared to the ground truth for 90 Zea mays seeds. Calculations based on predicted germination curves are shown in blue. Interpolations based on manual assessments are shown for 6, 12 and 24 h between each assessment and are colored in orange, green and red respectively

Similar articles

References

    1. King T, Cole M, Farber JM, Eisenbrand G, Zabaras D, Fox EM, et al. Food safety for food security: Relationship between global megatrends and developments in food safety. Trends Food Sci Technol. 2017;68:160–175. doi: 10.1016/j.tifs.2017.08.014. - DOI
    1. Ray DK, Mueller ND, West PC, Foley JA. Yield trends are insufficient to double global crop production by 2050. PLoS ONE. 2013;8:e66428. doi: 10.1371/journal.pone.0066428. - DOI - PMC - PubMed
    1. Marcos Filho J, Marcos FJ. Seed vigor testing: an overview of the past, present and future perspective. Sci Agric Scientia Agricola. 2015;72:363–374. doi: 10.1590/0103-9016-2015-0007. - DOI
    1. ISTA The germination test. Int Rules Seed Test. 2015 doi: 10.15258/istarules.2015.05. - DOI
    1. Chaugule A. Application of image processing in seed technology: a survey. Int J Emerg Technol Adv Eng. 2012;2(4):153–159.

LinkOut - more resources