Methods to determine ploidy of plants and uses

Abstract

The present invention relates to the technical field of image detection. In particular, the invention relates to a computer-implemented method, an apparatus, a use, and a program element to determine the ploidy of a vascular plant. The present invention also relates to a computer-implemented method for training an image classifier to determine the ploidy of a vascular plant.

Description

[0001] METHODS TO DETERMINE PLOIDY OF PLANTS AND USES

[0002] TECHNICAL FIELD

[0003] The present invention relates to the technical field of image detection. In particular, the invention relates to a computer-implemented method, an apparatus, a use, and a program element to determine the ploidy of a vascular plant. The present invention also relates to a computer-implemented method for training an image classifier to determine the ploidy of a vascular plant.

[0004] TECHNICAL BACKGROUND

[0005] Understanding and manipulating ploidy levels in vascular plants is crucial for plant breeding, genetic research, and the development of new crop varieties. Ploidy levels can have significant impacts on plant characteristics, including size, fertility, vigor, disease resistance, and even the ability to reproduce. Breeders often work with plants of a specific ploidy level to manipulate and enhance desirable traits. The term ploidy level is meant herein to designate the number of complete sets of chromosomes in a cell. Organisms can be described according to the number of sets of chromosomes: haploid (1 set), diploid (2 sets), triploid (3 sets), etc.

[0006] On the one hand, the ploidy level of a vascular plant may be determined using techniques focusing on the plant’s DNA, including chromosome counting and flow cytometry. These time-consuming methods require a sample of the plant which needs to be analyzed in a lab environment.

[0007] On the other hand, the ploidy level may be determined by analyzing the size of the stomata, by techniques such as stomatai size analysis. Stomatai size analysis involves measuring and examining the size and number of stomata, which are small pores on the surface of leaves and stems that regulate the exchange of gases and water vapor in plants. Stomatai counts and sizes can serve as an indicator of the ploidy levels in plants, where the stomata in an image are counted and measured to determine the plant's ploidy.

[0008] The ploidy level correlates with the size of the stomata and therefore also with the number of stomata in a vascular plant. Detecting and measuring stomata in a stomata image is very challenging due to the faint contours of the stomata. Several computational methods were developed to address these limitations using techniques such as semantic segmentation or object detection, among others. These methods focus on measuring individual stomata, which is computationally cost-intensive because the computational methods need to localize and delineate individual stomata in the stomata image. Furthermore, these methods remain error-prone, due to the faint contours of the stomata. Additionally, variations in image quality can make accurate stomatai counting and measuring cumbersome. Therefore, there is a need for an improved and more accurate method to determine the ploidy level of a vascular plant in stomata images.

[0009] SUMMARY OF THE INVENTION

[0010] In one aspect of the present disclosure, a computer-implemented method to determine the ploidy level using a stomata image of a vascular plant is provided, the method comprising:

[0011] Providing a stomata image of the vascular plant;

[0012] Applying an image classifier which is adapted to predict the ploidy level of the vascular plant in the stomata image based on a stomata pattern caused by the ploidy level;

[0013] Outputting the ploidy level of the vascular plant in the stomata image.

[0014] In a further aspect of the present disclosure, a computer-implemented method for training an image classifier to determine the ploidy level of a vascular plant is provided, comprising:

[0015] Providing a dataset of stomata images of the vascular plant with corresponding annotations indicating the ploidy level, wherein the dataset of stomata images is further divided into a training subset and an evaluation subset;

[0016] Training the image classifier using the training subset of stomata images and corresponding annotations indicating the ploidy level to learn a stomata pattern caused by the ploidy level;

[0017] Evaluating the performance of the image classifier based on the evaluation subset to assess the accuracy of the ploidy level determinations.

[0018] A further aspect of the present disclosure relates to a use of the indicated ploidy level by an image classifier for selecting a plant for breeding. The indicated ploidy level may be provided as explained in the present disclosure. A still further aspect of the present disclosure relates to an apparatus to carry out the method according to the present invention.

[0019] A further aspect of the present disclosure relates to a program element with instructions, which when executed on a processor, is configured to carry out the steps of the method according to the present invention.

[0020] Any disclosure and embodiments described herein relate to the methods, the use, the apparatus and the program element and vice versa. Advantageously, the benefits provided by any of the embodiments and examples equally apply to all other embodiments and examples and vice versa.

[0021] An object of the present disclosure is to provide a method for accurately determining the ploidy level of a vascular plant. In particular, it is an object of the present disclosure to provide a non-destructive and efficient method to determine the ploidy level of a vascular plant. The method utilizes stomata images to accurately determine the ploidy level of a vascular plant. Additionally, it ensures robustness to variations in the quality of the stomata images.

[0022] The term vascular plant is meant herein to designate any plant that possesses specialized tissues for the conduction of fluids and which contains stomata.

[0023] The term stomata image is meant herein to designate an image wherein stomata are discernable. In other words, a stomata image is a visual representation of plant tissue that allows for the identification and observation of stomata. The stomata image may be created through various methods, including but not limited to microscopy, clearing and mounting techniques, epidermal peeling, nail polish impression, stomatai replica, glycerin method, leaf imprint method, or transmitted light method. These methods may involve the use of staining agents, clearing agents, or impression materials to enhance the visibility of the stomata. The stomata image may be created using different types of microscopes, including but not limited to light microscopes, digital microscopes, smartphone microscopes or portable microscopes.

[0024] In an embodiment of the method for determining the ploidy level of a vascular plant, the image classifier may calculate a confidence score of each of the ploidy levels of the vascular plant in the stomata image based on a stomata pattern caused by the ploidy level. In another embodiment of the method for determining the ploidy level of a vascular plant, the method comprises providing metadata of the stomata image. This enables the image classifier to select an individual classifier that has been trained on a specific dataset comprising metadata of the stomata images. As a result, this approach can significantly improve the accuracy of determining the ploidy level of a vascular plant, as the classifier can take into account metadata including but not limited to plant species, environment, and imaging technique.

[0025] In another embodiment of the method for determining the ploidy level of a vascular plant, the vascular plant may be a crop plant.

[0026] In another embodiment of the method for determining the ploidy level of a vascular plant, the stomata image is produced with a magnification of from 100 to 1000, preferably from 200 to 800, more preferably from 300 to 600.

[0027] In another embodiment of the method for determining the ploidy level of a vascular plant, the stomata image is provided with a predefined magnification. This enables the image classifier to select an individual classifier which was trained on a specific dataset of the same magnification. As a result, this approach can significantly improve the accuracy of determining the ploidy level of a vascular plant.

[0028] In another embodiment of the method for determining the ploidy level of a vascular plant, the image classifier is a convolutional neural network, Support Vector Machine, or Random Forest.

[0029] In another embodiment of training an image classifier to determine the ploidy level of a vascular plant, the method comprises providing metadata of the stomata image. The inclusion of metadata in stomata images facilitates the training of an individual image classifier that incorporates both the stomata image and the associated metadata. For instance, an image classifier may be trained using stomata images and the corresponding metadata, such as the plant species name. Various combinations can be envisioned, such as training the image classifier using stomata images, the name of the plant species, and the magnification of the microscopic image, among others. This approach allows for a comprehensive and customizable training of the image classifier by incorporating relevant information from both the stomata images and their metadata. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In the following, the present disclosure is further described with reference to the enclosed figures:

[0031] Figure 1 illustrates an exemplary method to determine the ploidy level of a vascular plant according to the present disclosure.

[0032] Figure 2 illustrates an exemplary method for training an image classifier according to the present disclosure.

[0033] Figure 3 illustrates an exemplary computing system to carry out the method of the present disclosure.

[0034] DETAILED DESCRIPTION

[0035] Figure 1 illustrates schematically an exemplary method to determine the ploidy level of a vascular plant according to the present disclosure.

[0036] In step 110, a stomata image of a vascular plant is provided. The stomata image may be provided by various means, including but not limited to:

[0037] Scanning the stomata image using a flatbed scanner or a specialized microscope scanner. This process involves placing the stomata image wherein stomata are discernable on the scanner bed and utilizing scanning software to capture and save the stomata image as a digital file on the computing system.

[0038] In the case where the stomata image was captured using a digital camera, a microscope or a digital imaging system, it may be transferred to a computing system through a USB cable or wirelessly. This enables real-time assessment as the stomata image may be captured and analyzed instantly.

[0039] The stomata image may be captured through a mobile device, such as smartphone microscope. The stomata image may be transferred wirelessly to a computing device to apply an image classifier. Alternatively, the image classifier may be installed on the smartphone. This enables remote real-time assessment of the ploidy of the vascular plant, such as in the field or in a greenhouse. In step 120, an image classifier is applied to predict the ploidy level of the vascular plant in the stomata image based on a stomata pattern caused by the ploidy level. Step 120 may involve the image classifier to calculate a confidence score of each of the ploidy levels of the vascular plant in the stomata image based on a stomata pattern caused by the ploidy level. The confidence score is meant herein to designate a quantitative measure or value that represents the level of certainty or reliability associated with each class assignment made by the image classifier.

[0040] An image classifier is meant herein to designate a type of classifier that is specifically designed to categorize or classify digital images based on their visual features, such as colors, textures, shapes and patterns. The image classifier may encompass various algorithms such as Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), or Random Forests, among others.

[0041] Due to the correlation of the size of the stomata with the ploidy level, stomata may exhibit different patterns due to their varying sizes. Smaller stomata appear more closely packed together, resulting in a denser stomata pattern, while larger stomata are more widely spaced, leading to a more scattered stomata pattern. An image classifier is applied to assign a ploidy level to a stomata image based on a comprehensive understanding of the content in the stomata image, rather than focusing on measuring individual stomata. In other words, an image classifier is applied to determine the ploidy level of a vascular plant in a stomata image based on a stomata pattern in the image. The image classifier may assign the ploidy level based on the arrangement of the stomata, analyzing how the stomata are organized and positioned relative to each other. In other words, the method involves evaluating the spatial relationship of the stomata, considering the distances and spatial distribution between stomata to predict the ploidy level. This approach is faster compared to techniques to measure individual stomata, because there is no annotation of the stomata needed. Furthermore, this method is more precise because the image classifier is not dependent on faint contours of the stomata. Additionally, this approach is less sensitive to variations of image quality because it does not need to identify individual stomata. The image classifier may also consider additional features in the stomata image, such as the shape or arrangement of stomata. By examining the spatial relationships between stomata, the classifier can extract further patterns that contribute to the determination of ploidy level. The image classifier is capable of accurately identifying meaningful patterns associated with specific ploidy levels. This enables the ploidy level detection apparatus to provide reliable and precise results for a given stomata image. The image classifier may be a convolutional neural network (CNN). The CNN may comprise multiple layers, including but not limited to convolutional layers, pooling layers, and fully connected layers. The CNN may comprise one or more parallel convolutional layers in sequence. Parallel convolutional layers in sequence allows the CNN to learn multiple sets of filters in parallel, which may capture different aspects of the input image. The CNN may comprise one or more parallel convolutional layers where each convolutional layer within the parallel arrangement possesses a distinct kernel size in comparison to the other convolutional layers in the same parallel configuration. A convolutional layer with a small kernel size will be activated by images with fine-grained patterns, while a convolutional layer with a large kernel size will activate images with course-grained patterns. The CNN may comprise one or more pooling layers in sequence, including but not limited to average pooling layers and max pooling layers. Pooling layers keep the number of parameters of the CNN relatively low. The CNN may comprise an average pooling layer sequential to a max pooling layer, which causes the CNN to average its activations all over the image. This helps the CNN to recognize patterns instead of single objects in the image. The CNN may comprise a dropout layer to prevent overfitting. In another embodiment, the image classifier may comprise a Monte Carlo dropout layer. Each iteration, the model of the image classifier will exhibit slight variations due to the Monte Carlo dropout layer. By conducting multiple iterations of the image classifier on the identical stomata image, the output will comprise a distribution of confidence scores. As a result, this enables an evaluation of the variability linked to the confidence scores, thereby revealing the degree of confidence the model has concerning the predictions.

[0042] Finally, in step 130, the ploidy level of the vascular plant in the stomata image is outputted. The image classifier may output the confidence scores of each ploidy level of the vascular plant.

[0043] In optional step 110a, metadata of the stomata image may be provided. Metadata of the stomata image is meant herein to comprise the name of the plant species, the used method to capture the stomata image, the magnification of the microscopic image, the date when the stomata image was captured, among others. This enables the image classifier to select an individual classifier that has been trained on a specific dataset comprising metadata of the stomata images. As a result, this approach can significantly improve the accuracy of determining the ploidy level of a vascular plant, as the classifier can take the metadata into account.

[0044] The vascular plant may be a crop plant. A crop plant encompasses a variety of plants including, but not limited to, root vegetables such as asparagus, sugar beets, parsnips, carrots, potatoes, and celery, among others; leafy green vegetables such as lettuce, spinach, kale, and arugula, among others; cruciferous vegetables such as broccoli, cauliflower, cabbage, Brussels sprouts, rapeseed and canola, among others; allium vegetables such as onions, garlic, leeks, and shallots, among others; cucurbit vegetables such as cucumber, zucchini, pumpkin, and squash, among others; legumes such as soybean, chickpeas, lentils, beans, and peas, among others; fruit-bearing plants such as oil-palms, coffee, cocoa, melons, strawberries, tomatoes, peanuts, and grapes, among others; cereal crops such as wheat, barley, oats, sorghum, rice, corn, rye, and flax, among others; and other crop plants such as sunflower, cotton, alfalfa, tobacco, and sugarcane.

[0045] The stomata image may be a microscopic image. The stomata image may have a magnification of from 100 to 1000, preferably from 200 to 800, more preferably from 300 to 600.

[0046] In another embodiment of the method to determine the ploidy level of a vascular plant, the received stomata image is captured at a predefined magnification. The image classifier may be trained on stomata images with this predefined magnification. Training the image classifier on stomata images with the same magnification as the received stomata image improves the accuracy of the classification process as the classifier has been optimized to recognize and interpret patterns at the desired magnification level.

[0047] The invention also provides a computer-implemented method for training an image classifier to determine the ploidy level of a vascular plant, comprising:

[0048] Providing a dataset of stomata images of the vascular plant with corresponding annotations indicating the ploidy level, wherein the dataset of stomata images is further divided into a training subset and an evaluation subset;

[0049] Training the image classifier using the training subset of stomata images and corresponding annotations indicating the ploidy level to learn a stomata pattern caused by the ploidy level;

[0050] Evaluating the performance of the image classifier based on the evaluation subset to assess the accuracy of the ploidy level determinations.

[0051] The dataset of stomata images may comprise augmented stomata images using various algorithms, including but not limited to geometric transformations, color space augmentations, kernel filters, mixing images, random erasing, feature space augmentation, adversarial training, generative adversarial networks, neural style transfer, and meta-learning. The image classifier may be trained by iteratively adjusting the model’s weights and biases to minimize the difference between predicted and actual ploidy levels. This may involve a stochastic gradient descent (SGD), where the image classifier is trained by inputting the stomata images from the training set as input and updating the weights and biases based on the calculated error during each iteration. Furthermore, the accuracy and generalization of the image classifier may be improved by adjusting hyperparameters, such as learning rate and batch size, among others.

[0052] The inclusion of metadata in stomata images facilitates the training of an individual image classifier that incorporates both the stomata image and the associated metadata. For instance, an image classifier may be trained using stomata images and the corresponding metadata, such as the plant species name. Various combinations can be envisioned, such as training the image classifier using stomata images, the name of the plant species, and the magnification of the microscopic image, among others. This approach allows for a comprehensive and customizable training of the image classifier by incorporating relevant information from both the stomata images and their metadata.

[0053] Figure 2 illustrates an exemplary method for training an image classifier according to the present disclosure. The image classifier may be trained on a training dataset comprising at least one stomata image of a vascular plant, along with corresponding annotations indicating the ploidy level of each image. The method comprises training an image classifier, in step S5, to determine the ploidy level of the vascular plant in the stomata image based on a stomata pattern caused by the ploidy level. In step SI, a sample of a vascular plant, such as a leaf sample, is provided. Step S2 involves determining the ploidy level of the vascular plant, through a range of methods, including but not limited to chromosome counting, flow cytometry, pollen size analysis or stomatai size analysis. In step S3, at least one stomata image of the vascular plant is captured using one of the aforementioned methods. In Step S4, the image classifier receives a training dataset comprising at least one stomata image of a vascular plant, along with corresponding annotations indicating the ploidy level of each image which is then used in step S5 to train the image classifier.

[0054] In optional step S3a, metadata of the stomata image may be provided. This may include the name of the plant species, the used method to capture the stomata image, the magnification of the microscopic image, the date when the stomata images were captured, among others.

[0055] In optional step S3b, the stomata images and corresponding annotations of the ploidy level are stored in a database. Optionally, the metadata of the stomata image is also stored in the database. This allows to provide the image classifier with a training dataset comprising at least one stomata image along with corresponding annotations indicating the ploidy level of each stomata image and metadata associated with the stomata image. Additionally, storing metadata of the stomata image allows to create a specific training dataset, for instance a training dataset of a specific magnification of the stomata images, or a training dataset of a specific plant species, among others. The image classifier may consist of multiple individual image classifiers, each trained on a specific training dataset.

[0056] In optional step S3c, the stomata images may be augmented using various algorithms, including but not limited to geometric transformations, color space augmentations, kernel filters, mixing images, random erasing, feature space augmentation, adversarial training, generative adversarial networks, neural style transfer, and meta-learning.

[0057] The invention provides a use of the determined ploidy level of an image classifier for selecting a plant for breeding. Determining the ploidy level may be provided as explained in the present disclosure to select a plant for further breeding.

[0058] An automated greenhouse may be used to automatically select plants for further breeding. An automated greenhouse according to the present invention is a greenhouse fitted with means for automatically monitoring plants growing in separate containers comprising a reader. The reader may scan the identification tags and the resulting determined ploidy level is employed to automatically select plants for further breeding. The determined ploidy level of a vascular plant may be recorded and stored in a database for further breeding. To facilitate the process, each plant may be equipped with an identification tag, such as a QR code, an RFID tag or an NFC tag, among others. When the tag is read by a reader, the determined ploidy level stored in the database is utilized to select a plant for breeding.

[0059] The invention provides a ploidy level detection apparatus, comprising:

[0060] An input unit for receiving a stomata image of a vascular plant;

[0061] An image classification unit for generating the ploidy level of the vascular plant in the stomata image based on a stomata pattern caused by the ploidy level;

[0062] An output unit for outputting the ploidy level of the vascular plant in the stomata image.

[0063] The input unit of the ploidy level detection apparatus is designed to receive stomata images of a vascular plant. The input unit has the capability to receive stomata images of various types of devices which can capture these images. Examples of such devices comprise a microscope, smartphone, scanner, among others.

[0064] An image classification unit is specifically designed or adapted to analyze and classify images based on certain patterns or features. Examples of image classification units may include machine learning models, neural networks, or specialized software algorithms that are trained to recognize and classify the ploidy level of a vascular plant based on a stomata pattern observed in the stomata image.

[0065] The output unit of the ploidy level detection apparatus is designed to present the detected ploidy level of the vascular plant in the stomata image in a readable format. Examples of output units comprise a computer monitor, a printer, and a display panel, among others. Figure 3 illustrates an exemplary computing system for carrying out the method according to the present invention. The computing system 300 comprises a memory component 320 configured to store a training dataset comprising at least one stomata image of a vascular plant, along with corresponding annotations indicating the ploidy level of each image. The processor 330 is configured to train an image classifier using the training dataset, thereby developing an algorithm capable of classifying the images of the training dataset into different ploidy levels based on an stomata pattern. The input unit 310 is configured to receive a stomata image of a vascular plant, for the purpose of determining its ploidy level. The processor 330 is configured to apply the trained image classifier to analyze the provided image, generating a confidence score indicating the ploidy level of the vascular plant. The output unit 340 is configured to output the confidence score, providing information about the specific ploidy level of the vascular plant displayed in the image.

[0066] A computing system is meant herein to designate an electronic device or system that is specifically designed and programmed to perform various operations on data. It encompasses hardware components, such as processors, memory units, input / output interfaces, and storage devices, as well as software programs or algorithms that enable the device to execute specific tasks. A computing system is capable of receiving, analyzing, manipulating, and outputting data, either in real-time or through batch processing. A mobile device, such as a smartphone, may be used to determine the ploidy of a vascular plant. This allows to determine the ploidy of a plant in remote settings, such as in the field or in a greenhouse. Preferably, the mobile device comprises a protective cover. For example, the mobile device is protected by a case made from durable materials such as polycarbonate or aluminium. This case protects the mobile device against impacts, dust, and moisture. Preferably, if the mobile device is exposed to a wet or humid environment, a waterproof or water-resistant case with an IP rating of at least IP67 is applied.

[0067] A program element is meant herein to designate a specific portion or component of a program that performs a distinct function or carries out a particular task. It is a logical or physical unit of code that is designed to be executed by a computer or computing device. A program element may include individual instructions, subroutines, functions, methods, libraries, modules, or any other discrete part of a program. It may be written in various programming languages and may be stored in different formats, such as source code, object code, or executable files.

[0068] Any disclosure, embodiments and examples described herein relate to the methods, the computing system, and program element. Advantageously, the benefits provided by any of the embodiments and examples equally apply to all other embodiments and examples.

[0069] In the claims as well as in the description the word “comprising” or “including” or similar wording does not exclude other elements or steps and shall not be construed limiting to the elements or steps lined out. The indefinite article “a” or “an” does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation or further elements may be included.

[0070] The following embodiments are intended to be illustrative of the present disclosure and not limiting.

[0071] Embodiment 1 is a computer-implemented method to determine the ploidy level of a vascular plant, comprising:

[0072] Providing a stomata image of the vascular plant;

[0073] Applying an image classifier which is adapted to predict the ploidy level of the vascular plant in the stomata image based on a stomata pattern caused by the ploidy level;

[0074] Outputting the ploidy level of the vascular plant in the stomata image. Embodiment 2 is the computer-implemented method of embodiment 1, further comprising:

[0075] Calculating a confidence score of each of the ploidy levels of the vascular plant in the stomata image based on a stomata pattern caused by the ploidy level; and

[0076] Outputting the confidence scores indicating the ploidy levels of the vascular plant in the stomata image.

[0077] Embodiment 3 is the method according to embodiment 1 or 2, wherein the image classifier does not recognize individual stomata.

[0078] Embodiment 4 is the method according to any one of embodiments 1 to 3, the method further comprising:

[0079] Providing metadata of the stomata image.

[0080] Embodiment 5 is the method according to any one of embodiments 1 to 4, wherein the vascular plant is a crop plant.

[0081] Embodiment 6 is the method according to any one of embodiments 1 to 5, wherein the stomata image is produced with a magnification of from 100 to 1000, preferably from 200 to 800, more preferably from 300 to 600.

[0082] Embodiment 7 is the method according to any one of embodiments 1 to 6, wherein a stomata image is provided with a predefined magnification.

[0083] Embodiment 8 is the method according to any one of embodiments 1 to 7, wherein the image classifier is a convolutional neural network, Support Vector Machine or Random Forest.

[0084] Embodiment 9 is a computer-implemented method for training an image classifier to determine the ploidy level of a vascular plant, comprising:

[0085] Providing a dataset of stomata images of the vascular plant with corresponding annotations indicating the ploidy level, wherein the dataset of stomata images is further divided into a training subset and an evaluation subset;

[0086] Training the image classifier using the training subset of stomata images and corresponding annotations indicating the ploidy level to learn a stomata pattern caused by the ploidy level; Evaluating the performance of the image classifier based on the evaluation subset to assess the accuracy of the ploidy level determinations.

[0087] Embodiment 10 is the method according to embodiment 9, the method further comprising:

[0088] Providing metadata of the stomata image. Embodiment 11 is the use of the determined ploidy level as generated according to the method of any one of embodiments 1 to 8 for selecting a plant for breeding.

[0089] Embodiment 12 is a ploidy level detection apparatus, comprising:

[0090] An input unit for receiving a stomata image of a vascular plant;

[0091] An image classification unit for generating the ploidy level of the vascular plant in the stomata image based on a stomata pattern caused by the ploidy level;

[0092] An output unit for outputting the ploidy level of the vascular plant in the stomata image.

[0093] Embodiment 13 is a program element with instructions, which when executed on a processor, is configured to carry out the steps of the method of any one of embodiments 1 to 10.

Claims

CLAIMS1. A computer-implemented method to determine the ploidy level of a vascular plant, comprising:- Providing an image of the vascular plant wherein stomata are discernable;- Applying an image classifier which is adapted to predict the ploidy level of the vascular plant in the stomata image based on a stomata pattern caused by the ploidy level;- Outputting the ploidy level of the vascular plant in the image.

2. The method according to claim 1, further comprising:- Calculating a confidence score of each of the ploidy levels of the vascular plant in the image based on a stomata pattern caused by the ploidy level; and- Outputting the confidence scores indicating the ploidy levels of the vascular plant in the image.

3. The method according to claim 1 or 2, wherein the image classifier does not identify individual stomata.

4. The method according to claim 1 to 3, the method further comprising:- Providing metadata of the image.

5. The method according to any of claims 1 to 4, wherein the vascular plant is a crop plant.

6. The method according to any of claims 1 to 5, wherein the image is produced with a magnification of from 100 to 1000, preferably from 200 to 800, more preferably from 300 to 600.

7. The method according to any of claims 1 to 6, wherein an image is provided with a predefined magnification.

8. The method according to any of claims 1 to 7, wherein the image classifier is a convolutional neural network, Support Vector Machine or Random Forest.

9. A computer-implemented method for training an image classifier to determine the ploidy level of a vascular plant, comprising:- Providing a dataset of images of the vascular plant wherein stomata are discernable with corresponding annotations indicating the ploidy level, wherein the dataset of images is further divided into a training subset and an evaluation subset;- Training the image classifier using the training subset of images and corresponding annotations indicating the ploidy level to learn a stomata pattern caused by the ploidy level;- Evaluating the performance of the image classifier based on the evaluation subset to assess the accuracy of the ploidy level determinations.

10. The method according to claim 9, the method further comprising:- Providing metadata of the image.

11. Use of the determined ploidy level as generated according to the method of any of claims 1 to 8 for selecting a plant for breeding.

12. A ploidy level detection apparatus, comprising:- An input unit for receiving an image of a vascular plant wherein stomata are discernable;- An image classification unit for generating the ploidy level of the vascular plant in the image based on a stomata pattern caused by the ploidy level;- An output unit for outputting the ploidy level of the vascular plant in the image.

13. A program element with instructions, which when executed on a processor, is configured to carry out the steps of the method of any of claims 1 to 10.

Images