Method of sorting cereal seeds, sorting device for sorting cereal seeds, according computer progam, according computer-readable storage medium and use of the sortiing device

Abstract

A method of sorting cereal seeds (112) and a sorting device (110) are disclosed. The cereal seeds (112) contain sterile cereal seeds (114) and fertile cereal seeds (116). The method comprises: i. supplying a seed stream (120) to a sorting station (122), the sorting station (122) comprising at least one spectrometer device (130) for determining spectral information of seeds in the seed stream (120); ii. taking, with the spectrometer device (130), at least one near-infrared spectrum of a seed (112) in the seed stream (120); and iii. automatically identifying seeds to be sorted out from the seed stream (120) by using at least one classifier, the classifier being configured for classifying the seed to be a sterile cereal seed (114) or a fertile cereal seed (116) based on the near-infrared spectrum taken in step ii...

Description

[0001] METHOD OF SORTING CEREAL SEEDS, SORTING DEVICE FOR SORTING CEREAL SEEDS, ACCORDING COMPUTER PROGAM, ACCORDING COMPUTER-READABLE STORAGE MEDIUM AND USE OF THE SORTIING DEVICE

[0002] Technical Field

[0003] The present invention relates to a method of sorting cereal seeds, to a sorting device and a use of the sorting device. The invention further relates to a computer program and a computer-read- able storage medium for performing the sorting method. The method and the sorting device may specifically be used in the field of plant breeding for sorting cereal seeds containing a mixture of cereal seeds, such as a mixture of sterile and fertile cereal seeds. However, other fields of application comprising the sorting of cereal seeds are also feasible.

[0004] Background art

[0005] In the field of plant breeding, such as cereal plant breeding, e.g. breeding of wheat, blue aleurone loci and / or genes may be useful color markers for some other genes and / or loci of interest being located on the same chromosome or even on the same chromosome arm as the color marker. Thus, the presence of a blue color in the aleurone of the seed may indicate the presence of the gene of interest. In general, the closer both genes and / or loci, the more reliable the presence of the gene of interest is indicated by the blue color as a higher distance generally increases occurrence of recombination and / or breakage. As an example, seeds may contain a recessive male sterility gene and / or a locus causing male sterility in plants lacking a restorer gene, and the gene of interest is a male fertility restorer gene located on the same chromosome as a blue aleurone color locus. The blue color may be used to distinguish fertile “blue” seeds containing the blue aleurone locus and indicating the presence of the restorer gene from noncolored ’’white” sterile seeds lacking the blue aleurone locus, indicating the absence of the restorer gene. Thus, these seeds containing a male sterility locus and / or gene and containing a fertility restorer gene linked to a blue aleurone locus can be used as maintainer seeds to reproduce male-sterile “white” seeds and fertile “blue” seeds as their selfed progeny contains both types of seeds. The “white” seeds may grow into useful male-sterile (female) lines for hybrid seed production as they can only set seed after cross-pollination by a (male) fertile plant. As another example, the blue aleurone locus may be used in cereal seeds, such as wheat seeds, to identify higher amounts of anthocyanins in the seed, and / or to create new food products that are naturally colored rendering synthetic color product, e.g. in breakfast cereals, superfluous. In principle, the “blue” seeds can be sorted from non-colored “white” seeds in any production field or seed lot for sorting male-sterile seeds. For possible embodiments of male sterility systems based on a male sterility gene / locus, comprising a male fertility restorer gene located on the same chromosome as a blue aleurone color locus, reference may be made to e.g. Whitford et al., 2013, J. Exp. Botany 64 (18): 5411-5428, and Zhou et al., 2006, CropScience 46:250-255, CN 100420368, WO 2019 / 043082 A1, WO 2020 / 056259 A1 and WO 2023 / 005883 A1. The blue aleurone locus and the restorer gene / locus may be located on the same or on different chromosome arms in a monosomic addition chromosome (also named herein additional chromosome), or in one of the 2 chromosomes in a homoeologous chromosome pair.

[0006] WO 2012 / 038350 A1 discloses a method for separating inbred seed from a mixed population of inbred and hybrid seed based on the use of near infrared light, wherein the hybrid seed are obtained from a cross of a male sterile female plant and a male plant, and the inbred seed from self-fertilisation of the male plant. As known in the art, the parent plants used for the production of hybrid seed are genetically diverse.

[0007] WO 2023 / 088892 A1 discloses a method for categorizing / sorting seeds. The method comprises the steps of: providing a sample including at least one seed; obtaining a near infrared, NIR, spectrum of at least a subset of the sample; determining presence of an organic colorant in at least the subset of the sample based on the obtained NIR spectrum; and categorizing / sorting at least the subset of the sample based on the determination. Based on this, it is said that mis-col- ored white seed and blue seed with a fading blue color can be properly categorized. Further, it is said that by using the NIR spectrum, it is possible to detect a difference in concentration between single blue seed and double blue seed, particularly since the level of anthocyanins in double blue seed is approximately twice as high as in single blue seed.

[0008] Despite the advantages achieved by known methods and devices, several technical challenges remain. The sorting of sterile and / or fertile seeds based on the expression of the color associated with the blue aleurone marker may have certain limitations. Specifically, in some genotype and environment combinations, the expression of the blue aleurone (BA) color can be weak. Further, the blue aleurone locus system can generate the occurrence of approx. 2-3 % mis-divi- sion, due to breakage of the addition chromosome, leading to the uncoupling of the gene of interest, such as a fertility restorer gene, and the blue aleurone locus and hence non-colored (white), fertile, seeds can occur (mis-division). These seeds are not color sortable with a visible light color sorter, nor can these seeds be sorted based on anthocyanin concentration detection, and hence contaminate the male sterile plants with fertile plants. Additionally, in weakly blue aleurone-expressing seed lots, there may be some very light blue seeds, which may not be color sortable from sterile white seeds due to an insufficient blue color expression. Hence, the accuracy of color sorting may not be as high as required by seed certification standards due to a combination of low blue aleurone expression and / or mis-division. Thus, there is still a need for a reliable, color-independent and non-destructive seed sorting method. Problem to be solved.

[0009] It is therefore desirable to provide methods and devices which at least partially address aboveidentified technical challenges. Specifically, a method of sorting cereal seeds and a sorting device shall be proposed which provide a reliable, color-independent and non-destructive seed sorting method.

[0010] Summary

[0011] This problem is addressed by a method of sorting cereal seeds, a sorting device, a use of the sorting device, and by a computer program and a computer-readable storage medium for performing the sorting method with the features of the independent claims. Advantageous embodiments which might be realized in an isolated fashion or in any combinations are listed in the dependent claims as well as throughout the specification.

[0012] In a first aspect of the present invention, a method of sorting cereal seeds is disclosed. The cereal seeds contain sterile cereal seeds, specifically male sterile cereal seeds, and fertile cereal seeds, as will be defined in further detail below.

[0013] The term “sorting” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process of differentiation and, optionally, of separation. Specifically, the method of sorting cereal seeds may comprise identifying at least two different populations of cereal seeds which can be used for differentiation and / or separation. For example, the sorting may comprise at least two process steps, which, specifically, may be performed in parallel and / or in a timely overlapping fashion, wherein, in a first step, objects may be differentiated from each other according to one or more sorting criteria, wherein, in a second step, the objects may be separated from each other based on the differentiation. Specifically, the differentiation of objects may comprise assigning objects into at least two different sorting categories. The separation may be performed based on the sorting categories, wherein objects being assigned to a first sorting category may be physically separated from objects being assigned to a second sorting category.

[0014] In accordance with the present invention, a mixed population of seeds shall be sorted. Thus, the seed stream to be analyzed comprises or consists of a mixture of different seed populations which are sorted on the basis of at least one difference between the populations. In particular, the individual seed populations differ in their chromosomal composition. In a preferred embodiment, the chromosomal compositions of the individual seed populations differ in the presence or absence of one additional chromosome, or the presence of absence of a portion of an additional chromosome. Further, the chromosomal composition may differ in the number of additional chromosomes.

[0015] Advantageously, it was shown in the studies underlying the present invention that the presence of an additional chromosome, or portion of it, leads to an NIR detectable alteration of the seeds. Thus, the difference in the chromosomal composition is detectable via the use of near infrared light. Without being bound to theory, it is believed that the detectable alteration is caused by the expression of the additional genes located on the additional chromosome or remaining (mis-di- vided) parts of it. As described herein below, the NIR detectable alteration of the seeds allows for differentiating between seeds which are fertile and seeds which are sterile. In particular, the NIR detectable alteration of the seeds allows for differentiating between white (i.e. non-colored) fertile seeds and white sterile seeds. Thus, the method of the present invention can be used for sorting such seeds, and in particular to obtain a population of seeds which essentially consists of white hybrid seeds. The method of the invention can also be used to screen hybrid cereal (such as wheat) seed for the absence of any additional or homoeologous chromosome, or any inherited part thereof (e.g., to remove impurities, and avoid hybrid seeds that do not meet certain (market / registration / etc.) standards / requests).

[0016] The term “seed” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a unit of reproduction of a flowering plant, capable of developing into another such plant, i.e. by producing functional pollen or male gametes. As such, the term may specifically refer to a “fertile seed”. The term may also refer to a sterile seed. The term “sterile seed” in connection with the present invention may refer to seeds growing into plants failing or partially failing to produce functional pollen or male gametes (also known as male sterility). This can be due to natural or artificially introduced genetic predispositions or to human intervention on the plant in the field. Male sterility / fertility in cereals, such as wheat, can be reflected in seed set upon selfing, e.g. by bagging heads to induce selffertilization or growing male sterile plants in isolation with no pollen from fertile plants being present. Likewise, fertility restoration can also be described in terms of seed set upon crossing a male sterile plant with a plant carrying a functional restorer gene, when compared to seed set resulting from crossing (or selfing) fully fertile plants. A male parent (or pollen parent), is a parent plant that provides the male gametes (pollen) for fertilization, while a female parent or seed parent is the plant that provides the female gametes for fertilization, said female plant being the one bearing the (hybrid) seeds. Male sterility can be restored, for example, by introducing a functional restorer gene into the genome of the sterile plant.

[0017] The seeds to be sorted are plant seeds, preferably cereal seeds. Cereals are members of the monocotyledonous family Poaceae which are cultivated for the edible components of their grain. These grains are composed of endosperm, germ and bran. Maize, wheat and rice together account for more than 80% of the worldwide grain production. Other members of the cereal family comprise rye, oats, barley, triticale, sorghum, wild rice, spelt, einkorn, emmer, and durum wheat. Accordingly, the seeds are typically seeds from a cereal selected from the group consisting of wheat, rice, maize, rye, oats, barley, triticale, sorghum, spelt, einkorn and emmer. In one embodiment, the seeds are from a cereal plant that comprises a wheat-related genome, such as wheat (Triticum aestivum), spelt (Triticum spelt a) durum (T. turgidum), barley (Hordeum vul- gare) and rye (Secale cereale).

[0018] In a preferred embodiment, the cereal seeds are wheat seeds. “Wheat”, as used herein, may refer to plants from the genus Triticum, including but not limited to common / bread wheat (Triticum aestivum or T. aestivum), emmer wheat (T. dicoccum), einkorn wheat (T. monococcum), durum wheat (T. durum), khorasan wheat (T. turanicum), or spelt wheat (T. spelta), specifically hexapioid T. aestivum or T. spelta, and tetrapioid T. durum, including wheat referred to as hard or soft wheat (based on endosperm texture), winter or spring wheat (based on sowing season), and red or white wheat (based on seed coat color). In particular, the wheat is Triticum aestivum).

[0019] As set forth above, the mixture of seeds, typically, shall be sorted based on the difference in the chromosomal composition of the seeds (although they might be presorted based on a further marker, such as a color of the seeds, such as by visible light color sorting).

[0020] In an embodiment, the mixture of seeds to be sorted is a first seed mixture which comprises or essentially consists of seed populations which differ in their chromosomal composition by the presence or absence of an additional chromosome, or a portion thereof (such as one chromosome arm).

[0021] In one embodiment, the first seed mixture to be sorted comprises or essentially consists of the following three populations:

[0022] (1) a population of seeds having at least one additional chromosome,

[0023] (2) a population of seeds lacking the additional chromosome, and (3) a population of seeds comprising only one portion of the additional chromosome (such as one chromosome arm or a major part of the additional chromosome that is inherited to progeny).

[0024] The first seed mixture is, typically, obtained by self-fertilizing a plant which comprises a monoso- mic addition chromosome and harvesting the seeds from the plant. Thus, in an embodiment the seed mixture to be sorted by the method of the present invention has been obtained by the selffertilization of said plant and the harvesting step.

[0025] The thus obtained first seed mixture comprises or essentially consists of the following populations:

[0026] (1) seeds having at least one monosomic addition chromosome,

[0027] (2) seeds lacking the monosomic addition chromosome and

[0028] (3) seeds comprising only a portion of the monosomic addition chromosome.

[0029] The seeds under (3) which comprise only a portion of the monosomic addition chromosome result from a breakage of the monosomic addition chromosome (e.g., during meiosis). Moreover, the seeds under (3) can be further subclassified based on which portion of the chromosome is present (inherited) in the seeds.

[0030] The invention is applicable to any plant seeds comprising an additional chromosome, where a marker locus / gene (such as a color marker locus / gene) is located on another part of the addition chromosome (such as on one chromosome arm) than one or more genes of interest, and where that additional chromosome can break, resulting in progeny seed with only one part of the additional chromosome in one population of seeds, and progeny seed with only the other part of the additional chromosome in another population of seeds (next to seeds lacking entirely the addition chromosome), or in case of an additional chromosome where there is low expression of the marker gene / locus in some situations, so that sorting for the linked genes of interest on that addition chromosome is not easy, but will be described in more detail herein in the production of male-sterile wheat seed, for obtaining hybrid wheat. Seeds derived from self-fertilized plants comprising a monosomic addition chromosome are typically used in the production of hybrid wheat plants, for example, for the maintenance of a male-sterile female parental line or for use in the production of hybrid plants. Typically, the seeds (i.e. the progeny seeds) are sorted based on the color of the seeds. A 2-line male sterility system may be used with a maintainer plant producing white sterile and blue fertile cereal seeds upon selfing. Any regular wheat plant may act as male parent to restore fertility. The self-fertilized plants comprise a homozygous male fertility gene mutation, and a monosomic addition chromosome (42+1 Chr plants) comprising: i) a BLue Aleurone (BLA) locus as a detectable marker locus and ii) a dominant male fertility restorer gene, i.e. a gene that restores the homozygous male fertility gene mutation. The aleurone layer is triploid, as part of the endosperm that is triploid (3n), and as there may be male transmission of the BLA locus, seeds may contain one (1 n), two (2n) or three (3n) copies of the BLA locus in the aleurone layer that will respectively result in light blue seeds, blue seeds, or dark blue seeds. For possible embodiments of such 2-line male sterility systems, reference may be made to e.g. Whitford et al., 2013, J. Exp. Botany 64 (18): 5411-5428, and Zhou et al., 2006, CropScience 46:250-255, CN100420368, WO 2019 / 043082 A1, WO 2020 / 056259 A1 and WO 2023 / 005883 A1. The BLA locus and the restorer gene may preferably be located on the same chromosome arm of a monosomic addition chromosome, or on the same arm of one of the 2 chromosomes in a homoeologous chromosome pair. Thus, these genes may be closely linked, or alternatively the BLA locus and the restorer gene may be each located on another chromosome arm of the same addition or homoeologous chromosome. However, the cereal seeds may also comprise mis-divided cereal seeds, i.e. seeds where breakage of the dominant addition chromosome causes uncoupling of the fertility restorer (e.g., MS1) and blue aleurone locus (BLA) and giving rise to the occurrence of non-colored (white), fertile seeds. As an example, the restorer gene may be a (functional) MS1, MS5, MS9, MS22, MS26, or MS45 gene, specifically depending on what causes the male sterility. For example, if a mutation or inactivation or deletion of a MS1 wheat gene causes male sterility, then an MS1 gene capable of restoring fertility (from wheat or a related species such as Triticum boeoticum, Triticum mono- coccum, Triticum thaouder, or Triticum urartu) may be the restorer gene to use, and if mutation or inactivation or deletion of all MS45 genes (on each wheat sub-genome (A, B and D)) causes male sterility, then an MS45 gene may be the restorer gene to use. Alternatively or additionally, the blue aleurone genes / loci may be one of the genes / loci as described in US 11 ,390,877 B2, WO 2019 / 043082 A1 or WO 2020 / 056259 A1. The blue aleurone locus may be obtainable or obtained, e.g., from Agropyron elongatum, Agropyron trichophorum, Triticum boeoticum, Triticum monococcum, Triticum thaoudar, Triticum aestivum, or Thinopyrum ponticum or from wheat lines having an introgressed BLA locus or may be from known seed accessions Sebesta Blue, Blue Sando, Blue Baart, Blue Onas, Blue 1 , PBB, or Blue Norco.

[0031] For example, WO 2019 / 043082 A1 discloses the selfing of plants which comprises a homozygous male fertility gene mutation, i.e. the deletion of the ms1 gene, and which further comprises a monosomic addition chromosome carrying a dominant male fertility restorer gene and BLue Aleurone (BLA) gene in order to obtain non-colored sterile seeds and blue colored fertile cereal seeds upon selfing. In the art, the non-colored seeds are commonly referred to as “white seeds”, to distinguish from the fertile seeds having a blue aleurone, although they cover the nat- ural / standard spectrum of seed color for the large majority of marketed seeds (not having a blue seed color). Throughout the application, the term “white” seeds is used as well for the standard seed color of a commercial crop (e.g., white seeds, as used herein, for wheat seeds, refers to the known white or red wheat seeds). The aleurone layer is triploid, as being part of the endosperm that is triploid (3n), and as there may be male transmission of the BLA locus, seeds may contain one (1 n), two (2n) or three (3n) copies of the BLA locus in the aleurone layer that will respectively result in light blue seeds, blue seeds, or dark blue seeds. For possible embodiments of such 2-line male sterility systems, reference may be made to e.g. Whitford et al., 2013, J. Exp. Botany 64 (18): 5411-5428, and Zhou et al., 2006, CropScience 46:250-255, CN 100420368, WO 2019 / 043082 A1 , WO 2020 / 056259 A1 and WO 2023 / 005883 A1.

[0032] Accordingly, the plant that is used for obtaining the seed mixture to be sorted by the present invention, preferably, is a plant (in particular a wheat plant) comprising:

[0033] I. a mutation of a male fertility gene, wherein the mutation is homozygous, and

[0034] II. a monosomic addition chromosome carrying a male fertility restorer gene and at least one selection marker gene / locus.

[0035] The homozygous mutation of the male fertility gene, preferably, is a mutation which leads to male sterility of wheat plants (in the absence of a male fertility restorer gene). This can be achieved by inactivating or deleting the wheat MS1 , MS5, MS9, MS22, MS26, and / or MS45 gene(s). In one embodiment, all related sub-genome versions of wheat male fertility genes are inactivated / deleted (i.e. on each wheat sub-genome (A, B and D)). For example, all MS1 genes or all MS45 genes are inactivated / deleted to get male sterility. For MS1 and MS5, male sterility is obtained when only inactivating / deleting one sub-genomic version (on Chr4B and Chr3A, respectively), and monogenic recessive male sterile plants were identified in EMS-treated populations.

[0036] The male fertility restorer gene present on the monosomic addition chromosome, typically, restores the male sterility of the plant caused by the homozygous mutation, such as the deletion or inactivation, of the male fertility gene. Preferably, the male fertility restorer gene is a dominant gene. The male fertility restorer gene may be a MS1 , MS5, MS9, MS22, MS26, or MS45 gene, specifically depending on what causes the male sterility. For example, if the (homozygous) inactivation or deletion of the endogenous MS1 gene(s) causes male sterility, then a MS1 gene typically, is the restorer gene, and if the (homozygous) inactivation or deletion of the endogenous MS45 genes causes male sterility, then a MS45 gene, typically, is the restorer gene. In one embodiment of the invention, the monosomic addition chromosome in the cereal seed contains one or more fragments of chromosomes from one or more other cereal species than the species of said cereal seed.

[0037] The at least one selection marker gene may be any selection marker gene which allows for the selection of progeny seeds expressing the selection marker gene. Preferably, the selection marker gene is a color marker gene able to confer a characteristic coloration of a progeny seed comprising the color marker gene. In particular, the selection marker gene is a blue aleurone gene. Preferably, the blue aleurone gene is from Agropyron elongatum, Agropyron trichopho- rum, Triticum boeoticum, Triticum monococcum, Triticum thaoudar, Triticum aestivum, or Thino- pyrum ponticum.

[0038] In a preferred embodiment, the monosomic addition chromosome may be a monosomic addition chromosome as described in US 11 ,390,877 B2, WO 2019 / 043082 A1 or WO 2020 / 056259 A1.

[0039] In one embodiment, the selection marker gene (such as the BLA gene) and the restorer gene are located on the same chromosome arm of the monosomic addition chromosome, i.e. on the same side of the centromere of the monosomic alien addition chromosome.

[0040] In another embodiment, the selection marker gene (such as the BLA gene) and the restorer gene are located on different chromosome arms of the monosomic addition chromosome, i.e. on different sides of the centromere of the monosomic alien addition chromosome.

[0041] In one embodiment, the selection marker gene and the restorer gene are located on different locations on the additional chromosome and breakage can occur (even if rarely), so that some progeny seeds may inherit only the selection marker gene (and not the restorer gene), some seeds may only inherit the restorer gene (and not the selection marker gene), and some seeds have no additional chromosome (nor a part thereof).

[0042] Further, it is envisaged that, after allowing the plant to self-polli nate, mis-division of the monosomic addition chromosome occurs at a certain frequency leading to two alternative chromosome parts, i.e. two portions of the monosomic addition chromosome, one carrying the blue color marker gene without the fertility restorer and one carrying the fertility restorer without the blue color marker gene. The result of such mis-division is that the seed mixture obtained from selffertilization comprises blue seeds, which are male-sterile, and white seeds, which are fertile.

[0043] The, thus, obtained first seed mixture comprises the following populations: (1) seeds having at least one monosomic addition chromosome (blue fertile seeds),

[0044] (2) seeds lacking the monosomic addition chromosome (white sterile seeds) and

[0045] (3) seeds comprising only a portion of the monosomic addition chromosome.

[0046] The seeds under (3) can be further subclassified in a. seeds comprising an inherited portion of the monosomic addition chromosome, said portion comprising the fertility restorer gene, and lacking the blue color marker gene (white fertile seeds), and / or b. seeds comprising an inherited portion of the monosomic addition chromosome said portion comprising the blue color marker gene, and lacking the fertility restorer gene (blue sterile seeds).

[0047] Accordingly, a seed mixture comprises or consists essentially of (1) white sterile seeds, (2) blue fertile seeds, (3) white fertile seeds, and (4) blue sterile seeds.

[0048] Advantageously, the present invention allows for differentiating between the different seed populations based on their near-infrared spectrum. In particular, the present invention allows for differentiating between (white) seeds lacking the monosomic addition chromosome (white sterile seeds) and (white) seeds comprising a portion of the monosomic addition chromosome, said portion comprising the fertility restorer and lacking the blue color marker gene (white fertile seeds). This is advantageous, as these seeds cannot be differentiated based on the color or anthocyanin content.

[0049] Thus, the present invention allows for sorting a first seed mixture as described herein above, i.e. a seed mixture comprising or consisting essentially of (1) white sterile seeds, (2) blue fertile seeds, (3) white fertile seeds, and (4) blue sterile seeds. Thus, in an embodiment, such seed mixture is sorted.

[0050] Prior to sorting the first seed mixture based on the NIR spectrum, the seed mixture could be subjected to a sorting step based on blue seed color (e.g., using visible light color sorting), thereby obtaining a seed mixture mostly comprising or consisting of the following populations: seeds lacking the monosomic addition chromosome (white sterile seeds) and seeds comprising a portion of the monosomic addition chromosome, said portion comprising the fertility restorer gene, but lacking the blue color marker gene (white fertile seeds).

[0051] Thus, in an embodiment, the first mixture to be sorted comprises or consists essentially of (1) white sterile seeds and (3) white fertile seeds. Typically, said seed mixture does not comprise many blue colored seeds. However, some light blue colored seeds may be present which were not sortable based on their light blue color. Advantageously, such seeds can be identified by the method of the present invention as well.

[0052] As set forth above, the selectable marker gene, such as the BLA gene, and the male restorer gene may be one of the 2 chromosomes in a homoeologous chromosome pair (so called 42 + 2 plants). Accordingly, it is also envisaged that the seed mixture to be analyzed by the method of the present invention is derived from selfing a cereal plant comprising at least one homoeologous chromosome pair, wherein the pair consists of a first and second chromosome, wherein the first chromosome is native to the cereal plant and the second chromosome comprises a heterologous chromosome fragment comprising a dominant male fertility restorer gene and at least one selection marker gene, and wherein the cereal plant comprises a (homozygous) male fertility gene mutation causing male sterility. Such plants are disclosed in US2020 / 0255856A1 which herewith is incorporated by reference in its entirety. The definition for the terms “male fertility restorer gene, “selection marker gene” and the “male fertility mutation” and “cereal” made in connection with the plant comprising at least one monosomic addition chromosome apply to the plant comprising at least one homoeologous chromosome pair as well. In an embodiment, the male fertility restorer gene is the MS1 gene, the selection marker gene is a BLA gene and the “male fertility mutation” is the homozygous inactivation or deletion of the endogenous MS1 gene.

[0053] The ratio of white sterile seeds to white fertile seeds in the seed mixture to be sorted (i.e. in the first or second seed mixture), typically is in range from 10:1 to 100:1. In preferred embodiment, the ratio of white sterile seeds to white fertile seeds in the seed mixture to be sorted is a range from 10:1 to 50:1. By carrying out the method of the present invention, a population of seeds which essentially consists of white hybrid seeds can be obtained. For example, the obtained seed mixture comprises at least 98% white sterile seeds, such as least 99% white sterile seeds, in particular at least 99.5%, 99.6 %, or 99.7 % white sterile seeds.

[0054] In the following, the terms “seed” and “cereal seed” are used interchangeably. Thus, consequently, the terms “sterile seed” and “sterile cereal seed”, and “fertile seed” and “fertile cereal seed”, respectively, are used in a way that refers to the same element.

[0055] The method comprises the following steps that may be performed in the given order. However, a different order may also be possible. In particular, one, more than one or even all of the method steps may be performed once or repeatedly. Further, the method steps may be performed successively or, alternatively, one or more of the method steps may be performed in a timely overlapping fashion or even in a parallel fashion and / or in a combined fashion. The method may further comprise additional method steps that are not listed.

[0056] The method comprises: i. supplying a seed stream to a sorting station, the sorting station comprising at least one spectrometer device for determining spectral information of seeds in the seed stream; ii. taking, with the spectrometer device, at least one near-infrared spectrum of a seed in the seed stream; and iii. automatically identifying seeds to be sorted out from the seed stream by using at least one classifier, the classifier being configured for classifying the chromosomal composition of the seed based on the near-infrared spectrum taken in step ii. In one embodiment, the sorting station is not separating seeds but is differentiating the seeds to identity the percentage of a certain seed type (such as white fertile seeds) in the seed stream. In another embodiment, the sorting station is separating seeds of a certain seed type (such as separating white fertile seeds from sterile white seeds) in the seed stream.

[0057] In particular, the method comprises: i. supplying a seed stream to a sorting station, the sorting station comprising at least one spectrometer device for determining spectral information of seeds in the seed stream; ii. taking, with the spectrometer device, at least one near-infrared spectrum of a seed in the seed stream; and iii. automatically identifying seeds to be sorted out from the seed stream by using at least one classifier, the classifier being configured for classifying the seed to be a sterile cereal seed, specifically a male sterile cereal seed, or a fertile cereal seed based on the nearinfrared spectrum taken in step ii.

[0058] The term “supplying” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process of making available for further processing. The supplying may specifically comprise providing the seed stream to the sorting station. The supplying may be performed in a manner to enable taking near-infrared spectra from the cereal seeds in the seed stream.

[0059] The term “seed stream” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a plurality of seeds. The seed stream may comprise a plurality of seeds, wherein the plurality of seeds may be provided to the sorting station in a directional fashion. Additionally or alternatively, the seed stream may comprise a plurality of seeds arranged in a regular fashion. The seed stream may comprise a plurality of seeds, wherein the seeds of the seed stream may be arranged individually in a row or line. Thus, as an example, the seeds of the seed stream being supplied to the sorting station may comprise a plurality of seed being provided one by one to the sorting station. As will be understood by the skilled person, the seed stream, typically, comprises or consists of a mixture of seeds having different characteristics. Accordingly, the seed stream comprises or consists of the first seed mixture as described above. Alternatively, the seed stream comprises or consists of the second seed mixture as described above.

[0060] The term “sorting station” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a device or combination of devices configured for performing at least one sorting function. Specifically, the sorting station may comprise at least one sensor device, such as the spectrometer device, configured for differentiation or identification of the seeds in the seed stream. The sorting station may further comprise at least one ejector configured for separation of the seeds in the seed stream by ejecting seeds to be sorted out from the seed stream. The sensor device and the ejector of the sorting station may be connected with each other, e.g. via one more controller, such that, upon differentiation of seeds in the seed stream using the sensor device, the ejector ejects the respective seeds from the seed stream.

[0061] The term “spectrometer device” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an optical device configured for acquiring at least one item of spectral information on at least one object. Specifically, the at least one item of spectral information may refer to at least one optical property or optically measurable property which is determined as a function of a wavelength, for one or more different wavelengths. More specifically, the optical property or optically measurable property, as well as the at least one item of spectral information, may relate to at least one property characterizing at least one of a transmission, an absorption, a reflection and an emission of the at least one object, either by itself or after illumination with external light. The at least one optical property may be determined for one or more wavelengths. The spectrometer device specifically may form an apparatus which is capable of recording a signal intensity with respect to the corresponding wavelength of a spectrum or a partition thereof, such as a wavelength interval, wherein the signal intensity may, specifically, be provided as an electrical signal which may be used for further evaluation.

[0062] The spectrometer device, as an example, may be or may comprise a device which allows for a measurement of at least one spectrum, e.g. for the measurement of a spectral flux, specifically as a function of a wavelength or detection wavelength. The spectrum may be acquired, as an example, in absolute units or in relative units, e.g. in relation to at least one reference measurement. Thus, as an example, the acquisition of the at least one spectrum specifically may be performed either for a measurement of the spectral flux (unit W / nm) or for a measurement of a spectrum relative to at least one reference material (unit 1), which may describe the property of a material, e.g., reflectance over wavelength. Additionally or alternatively, the reference measurement may be based on a reference light source, an optical reference path, a calculated reference signal, e.g. a calculated reference signal from literature, and / or on a reference device.

[0063] Specifically, the at least one spectrometer device may be a reflective spectrometer device configured for acquiring spectral information from the light which is reflected by at least one object, e.g. the seeds in the seed stream, specifically diffusively reflected light. In particular, measuring a spectrum with the spectrometer device may comprise measuring absorption in a reflection configuration. Specifically, the spectrometer device may be configured for measuring absorption in a reflection configuration. Additionally or alternatively, the at least one spectrometer device may be or may comprise a transmission spectrometer device. However, other types of spectrometer devices are also feasible.

[0064] The at least one spectrometer device, specifically and as will be outlined in further detail below, may comprise at least one light source which, as an example, may be at least one of a tunable light source, a light source having at least one fixed emission wavelength and a broadband light source. The spectrometer device, as will be outlined in further detail below, further comprises at least one detector configured for detecting light, such as light which is at least one of transmitted, reflected or emitted from the at least one object. The spectrometer device further may comprise, as will be outlined in further detail below, at least one wavelength-selective element, such as at least one of a grating, a prism and a filter, e.g. a length variable filter having varying transmission properties over its lateral extension. The wavelength-selective element may be used for separating incident light into a spectrum of constituent wavelength signals whose respective intensities are determined by employing a detector, such as a detector having a detector array as described below in more detail. The term “spectral information” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an item of information, e.g. on at least one object and / or radiation emitted by at least one object, characterizing at least one optical property of the object, more specifically at least one item of information characterizing, e.g. qualifying and / or quantifying, at least one of a transmission, an absorption, a reflection and an emission of the at least one object. As an example, the at least one item of spectral information may comprise at least one intensity information, e.g. information on an intensity of light being at least one of transmitted, absorbed, reflected or emitted by a seed from the seed stream, e.g. as a function of a wavelength or wavelength sub-range over one or more wavelengths, e.g. over a range of wavelengths. Specifically, the intensity information may correspond to or be derived from the signal intensity, specifically the electrical signal, recorded by the spectrometer device with respect to a wavelength or a range of wavelengths of the spectrum.

[0065] The spectrometer device specifically may be configured for acquiring at least one spectrum or at least a part of a spectrum of detection light propagating from the seed in the seed stream to the spectrometer device. The term “spectrum” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a distribution of spectral information in at least one spectral range. For example, a near-infrared spectrum may comprise a distribution of spectral information in at least part of the near-infrared spectral range. The term “near-infrared” may refer, without limitation, to a spectral range adjacent to the visible spectral range. The near-infrared spectral range may comprise wavelengths of 800 nm to 2780 nm, essentially corresponding to a wavenumber in the range of 3,600 cm-1to 12,500 cm’1.

[0066] The spectrum may specifically describe a radiometric unit of spectral flux, e.g. given in units of watt per nanometer (W / nm), or other units, e.g. as a function of the wavelength of detection light. Thus, the spectrum may describe the optical power of light, e.g. in the NIR spectral range, in a specific wavelength band. The spectrum may contain one or more optical variables as a function of the wavelength, e.g. the power spectral density, electric signals derived by optical measurements and the like. The spectrum may indicate, as an example, the power spectral density and / or the spectral flux of the seed, e.g. relative to a reference sample, such as a transmittance and / or a reflectance of the seed.

[0067] The spectrum, as an example, may comprise at least one measurable optical variable or property of detection light of the seed, specifically as a function of illumination light. As an example, the at least one measurable optical variable or property may comprise at least one radiometric quantity, such as at least one of a spectral density, a power spectral density, a spectral flux, a radiant flux, a radiant intensity, a spectral radiant intensity, an irradiance, a spectral irradiance. Specifically, as an example, the spectrometer device, specifically the detector, may measure the irradiance in Watt per square meter (W / m2), more specifically the spectral irradiance in Watt per square meter per nanometer (W / m2 / nm). Based on the measured quantity the spectral flux in Watt per nanometer (W / nm) and / or the radiant flux in Watt (W) may be determined, e.g. calculated, by taking into account an area of the detector.

[0068] The term “taking at least one near-infrared spectrum” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process of acquiring spectral information from at least one object, specifically from the seed in the seed stream. The taking of the near-infrared spectrum may comprise recording the spectral information from the seed in the seed stream as a function of wavelength or wavenumber. The near-infrared spectrum may be recorded for each seed in the seed stream individually or, alternatively, for a group of seeds in the seed stream.

[0069] The near-infrared spectrum may be obtained in reflection geometry, specifically with reflective background surfaces.

[0070] The near-infrared spectrum may comprise an absorbance of the seed in the seed stream as a function of wavenumber. The near-infrared spectrum may be obtained for a spectral region of 3,600 cm-1to 12,500 cm-1. Optionally, a spectral region of 12,500 cm-1to 11 ,000 cm-1may be excluded in the near-infrared spectrum.

[0071] As outlined above, from the near-infrared spectrum taken in step ii., the seeds to be sorted out are automatically identified and, optionally as will be outlined in further detail below, the seeds identified to be sorted out from the seed stream may be automatically ejected. The term “automatically” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process which is performed completely by means of at least one computing unit, in particular without manual action and / or interaction with a user. The term “automatically” may specifically refer to any process which is performed by means of a controller. The term “identifying” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process of recording a state or a property of an object. The identifying may further comprise at least one assignment of the object into a sorting category based on the recorded state or property of the object. For example, the identifying may comprise recording if the seed in the seed stream is a sterile cereal seed or a fertile cereal seed and assigning the seed into at least one sorting category comprising depending on the recorded sorting criteria of the seed of being sterile or fertile. As an example, fertile seeds may be identified to be sorted out.

[0072] The term “seed to be sorted out” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a seed of the seed stream which is to be separated from the other seeds of the seed stream. The seed to be sorted out may be defined previously to be separated from the other seeds of the seed stream, e.g. by defining a specific phenotype and / or genotype of a seed which is to be separated from other seeds of the seed stream differing from the specific phenotype and / or genotype. For example, the seeds to be sorted out from the seed stream may specifically be the fertile cereal seeds, specifically white-fertile cereal seeds (fertile cereal seeds comprising the fertility restorer gene / locus but lacking the blue aleurone marker gene / locus (in a genetic background comprising a mutation in (such as an inactivation or deletion of) a male fertility gene / locus rendering plants without the fertility restorer gene / locus male-sterile) and / or blue-fertile cereal seeds (fertile cereal seeds comprising the fertility restorer gene / locus and the blue aleurone marker gene / locus, such as blue-fertile cereal seeds wherein the blue aleurone color is light or faint so that they cannot be sorted from white seeds using a standard color sorter (said seeds also have a genetic background comprising a mutation in a male fertility gene / locus rendering plants without the fertility restorer gene / locus male-sterile)) and / or blue-sterile cereal seeds and / or mis-di- vided blue cereal seeds.

[0073] The identifying of seeds to be sorted out comprises using the at least one classifier. The term “classifier” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to at least one mathematical model configured for transforming one or more input values into one or more output values by using one or more parameters which may be adjusted in order to enable the classifier to be trained. The classifier may specifically comprise a trained mathematical model, specifically a mathematical model which was trained on at least one training data set using one or more of machine learning, deep learning, neural networks, or other form of artificial intelligence. The classifier may be at least one artificial intelligence (Al) classifier. The classifier may be configured for processing an input having a high dimensionality into an output of a much lower dimensionality. Such a classifier may be termed “intelligent” because it is capable of being “trained”. The classifier may comprise at least one classification algorithm for performing classification.

[0074] The classifier is configured for classifying the seed to be a sterile cereal seed or a fertile cereal seed based on the near-infrared spectrum taken in step ii.. The term “classifying” or “classification” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process of assigning at least one observation to a set of categories. The at least one observation may comprise a set of quantifiable properties, specifically a set of features, more specifically a set of spectral features, to be used for the assignment. The possible categories may comprise at least two classes, such as a class of sterile cereal seeds and a class of fertile cereal seeds. The classifying may specifically comprise assigning the seed of the seed stream to one of a class of sterile cereal seeds or a class of fertile cereal seeds based the near-infrared spectrum, specifically based on spectral features extracted from the near-infrared spectrum. Thus, the classifying may comprise predicting the seed of the seed stream to be a sterile cereal seed or a fertile cereal seed based the near-infra- red spectrum, specifically based on spectral features extracted from the near-infrared spectrum. The classifier may be configured for the prediction. Alternatively or additionally, the classifier may be configured for classifying the seed of the seed stream to be a sterile cereal seed or a fertile cereal seed based on the near-infrared spectrum irrespective of a presence or content of an organic colorant in the seed, specifically irrespective of a presence or content of anthocyanin in the seed. Alternatively or additionally, the classifier may not be based on the detection of the anthocyanin concentration.

[0075] The classifier may comprise, specifically may be, a trained artificial neural network (ANN). The term “artificial neural network”, also referred to as “neural network”, as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a mathematical model comprising of a plurality of units or nodes which are connected by one or more edges. Each of the units or nodes may receive at least one input signal from one or more connected units or nodes and may be configured for processing the at least one input signal and forward at least one output signal to one or more connected units or nodes. The input signal and / or the output signal may be a real number. The output signal of each unit or node may be computed by at least one mathematical function, specifically at least one non-linear function, taking into account the sum the input signals. The mathematical function of the unit or node may be an activation function of the unit or node. A strength of the input signal at each connection may be determined by using an adjustable weight, wherein the adjustable weight may be adjusted during a learning or training process. The units or nodes may be aggregated into two or more layers. An input signal may be given to an input layer and may be forwarded to an output layer. Optionally, the artificial neural network may comprise one or more hidden layers in between the input layer and the output layer.

[0076] The ANN may be a single-layer neural network. The Matlab fitcnet function may be used to create the ANN. The structure may be a trained model exported from Classification Learner R2023b. The ANN may comprise an input layer corresponding to the predictor data in Tbl or X, wherein X represents the first 39 principle component scores of the normalized near-infrared spectra. The ANN may further comprise a first fully connected layer of layer size 10 and having 100 outputs. Further, a ReLLI activation function may be applied to the first fully connected layer. The ANN may further comprise a final fully connected layer having K = 4 outputs, where K is the number of classes in the response variable. Further, a softmax function (for both binary and multiclass classification) may be applied to the final fully connected layer. The softmax function may take each input x and returns the following, where K is the number of classes in the response variable:

[0077] The results may correspond to the predicted classification scores (or posterior probabilities). The ANN may further comprise an output layer corresponding to the predicted class labels, namely 1 to 4 as proxis for BS, BF, WS, WF.

[0078] The method may further comprise at least one training step. In the training step, the ANN may be trained using labeled near-infrared spectra. The labeled near-infrared spectra may comprise genotyping data and / or provide information on the presence or absence of a color marker and / or fertility restorer gene. The genotyping data may provide information on sterility and / or fertility of seeds under test. The labeled near-infrared spectra may be obtained by using genotyping data of the respective cereal seeds, specifically indicating presence of one or more of the restorer gene, the color gene, a specific locus as detected using markers. Thus, the labeled near-infrared spectra may provide ground truth data for the training step.

[0079] The trained ANN may be trained using records of training data. A record of training data may comprise training input data and corresponding training output data. The training output data of a record of training data may be the result that is expected to be produced by the ANN when being given the training input data of the same record of training data as input. The deviation between this expected result and the actual result produced by the ANN may be observed and rated by means of a “loss function”. This loss function may be used as a feedback for adjusting the parameters of the ANN. For example, the parameters may be adjusted with the optimization goal of minimizing the values of the loss function that result when all training input data is fed into the ANN and the outcome is compared with the corresponding training output data. The result of this training may be that given a relatively small number of records of training data as “ground truth”, the ANN is enabled to perform its job well for a number of records of input data higher by many orders of magnitude. Thus, the ANN may comprise at least one algorithm and model parameters. Parameters of the ANN may be adjusted in the training step.

[0080] Further, the training step may comprise a pretreatment of near-infrared spectra. The pretreatment of near-infrared spectra may comprise at least one of a normalization of the near-infrared spectra and a principal component analysis (PCA) of the near-infrared spectra. The normalization of the near-infrared spectra may comprise at least one normalization selected from the group consisting of: a minimum-maximum normalization of the near-infrared spectra; a mean centering-standard normalization. The minimum-maximum normalization of the near-infrared spectra may comprise setting a minimal value of the near-infrared spectrum to 0 and a maximum value to 1. The mean centering-standard normalization may comprise subtracting an average absorbance of the near-infrared spectrum and dividing the obtained spectrum by a standard deviation of the near-infrared spectrum.

[0081] Further, in the training step, the ANN may be trained using the following hyperparameters: preset: Narrow Neural Network; number of fully connected layers: 1 ; first layer size: 10; activation: rectified linear unit (ReLLI); iteration limit: 1 ,000; regularization strength (Lambda): 0; and standardize data: yes.

[0082] The training step may further comprise a k-fold cross-validation of the ANN, specifically a 20- fold cross validation of the ANN.

[0083] As an example, a training data set comprising near-infrared spectra from 1910 seeds comprising white-sterile cereal seeds, white-fertile cereal seeds, blue-sterile cereal seeds and blue-fertile cereal seeds was used for training the ANN. The near-infrared spectra were obtained on a Multi Purpose Analyzer (MPA) from Bruker® and were annotated according to genotype classes white (W), blue (B), sterile (S) and fertile (F). To counteract a potential overfitting of the classifier, seeds from a set of different BLA genotypes and seeds from BLA plants grown in different environments were analyzed. Normalization of raw spectra was performed for better separation of genotypes (PCA). Hierarchical cluster analysis (HCA) and Kmeans of individual lines enabled unsupervised clustering which followed relatively well the expected ratios between S and F types in WF populations. These data were used as initial ground truth data to create an initial “noisy” Al classifier (ANN). The training was done using 1143 of 1910 individual near-infrared spectra. The spectral data consisted of one or two technical replicate measurements for each seed. The model was first trained using 80% of the near-infrared spectra with 5% of the training set used in cross-validation after each iteration and 20% testing. The ANN was a single-layer neural network as described above.

[0084] Further, in the training step, genotypic validation of the initial NIRS classification results was performed and was enabled by providing the seeds in grid plates allowing for tracking their identity. The model training was repeated using all 1143 individual near-infrared spectra with 5% cross-validation in a second attempt. Both models performed similarly. The second model was used for classification. The calibrating of the classifier may further improve accuracy of the classifier. The True Positive Ratio (TPR) of the prediction of the different classes may be as follows for the calibrated ANN classifier based on four classification groups: white fertile (WF): 91 %; white sterile (WS): 95%; blue fertile (BF): 99%; blue sterile (BS): 67%. It should be noted that the BS group was very small and hence under-represented, which can explain the relatively lower prediction accuracy. Predictions can be made on a new predictor row matrix X, i.e. the NIR spectrum of the seed, wherein X contains exactly 4449 rows corresponding to the 4449 predictors (wavenumbers) used for training of the ANN.

[0085] The classifier may comprise at least one of a F-classifier configured for classifying seeds into fertile cereal seeds and non-fertile cereal seeds, and a S-classifier configured for classifying seeds into sterile cereal seeds and non-sterile cereal seeds.

[0086] Additionally or alternatively, the classifier may be configured for determining a presence and / or an absence of a monosomic addition chromosome, or portion thereof, in the cereal seeds. Accordingly, the classifier may be configured for differentiating between different seed populations within a mixture of seeds as described herein, for the first mixture or the second mixture.

[0087] As an example, the seed stream may comprise, specifically consist of, white-sterile cereal seeds, white-fertile cereal seeds, blue-sterile cereal seeds and / or blue-fertile cereal seeds. The classifier may further be configured, based on the near-infrared spectrum, for classifying the seed into at least one category selected from the group consisting of: a white-sterile cereal seed; a white-fertile cereal seed; a blue-sterile cereal seed; a blue-fertile cereal seed.

[0088] Alternatively or additionally, the seed stream may comprise, specifically consist of, white-sterile cereal seeds and white-fertile cereal seeds. The classifier may further be configured, based on the near-infrared spectrum, for classifying the seed into at least one category selected from the group consisting of: white-sterile cereal seeds; white-fertile cereal seeds. Specifically, as an example, the white-sterile cereal seeds and white-fertile cereal seeds in the seed stream may be cereal seeds previously sorted by a color seed sorting method, specifically a camera-based color seed sorting method, as will be outlined in further detail below. The color seed sorting method may comprise separating the white-sterile cereal seeds and white-fertile cereal seeds from an initial seeds stream comprising white-sterile cereal seeds, white-fertile cereal seeds, blue-sterile cereal seeds and / or blue-fertile cereal seeds.

[0089] The classifier may be configured for classifying the seed based on spectral features from at least one spectral region selected from the group consisting of: 4,000 cm-1to 7,000 cm-1; 9,000 cm’1to 12,000 cm’1; 8260 cm’1to 8430 cm’1.

[0090] As outlined above, the method may comprise using the classifier. The using of the classifier may comprise classifying the seed to be a sterile cereal seed or a fertile cereal seed based on the near-infrared spectrum.

[0091] Alternatively or additionally, the method may further comprise determining a degree of mis-divi- sion of the seed stream. The degree of mis-division may quantify a fraction of fertile cereal seeds in the seed stream. The degree of mis-division in the seed stream, e.g. comprising seed from a particular mutant or breeding line, may be obtained by scanning a couple of hundred of white cereal seeds in order to determine the percentage mis-division. The degree of mis-division may allow to determine if seed sorting to remove mis-divided fertile white seeds is required and may result in a significant time / effort saving for screening a mutant or breeding population, and can reduce efforts in breeding (e.g., one can decide to avoid spending work in breeding on certain lines).

[0092] Alternatively or additionally, the method may further comprise: iv. automatically ejecting seeds identified to be sorted out from the seed stream. The term “ejecting” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process of removing an object from a group of objects. Specifically, the ejecting may comprise removing the seeds to be sorted out from the seed stream. The ejecting may comprise removing the seeds to be sorted out by means of at least one of a mechanical device and a pneumatic device. For example, a pneumatic ejector may be configured for ejecting the seed to be sorted out from the seed stream by using compressed air, such as by using compressed air directed via nozzles to separate the seed to be sorted out from the seed stream. By ejecting the seeds to be sorted out from the seed stream, the seeds in the seed stream may be separated into a first group of seed comprising the seeds to be sorted out and a second group of seeds comprising the non-ejected seeds of the seed stream.

[0093] Further, a batch of seeds may be provided. The term “batch” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an assembly of jointly produced objects or an assembly of objects produced at different times and / or places that were mixed. Specifically, the batch of seeds may comprise a plurality of seed which were jointly produced, specifically with respect to at least one of production place and production time. The batch of seeds may be subjected to method steps i.-iv. repeatedly, specifically at least twice. In each repetition, the batch may be diminished by the seeds ejected in step iv. of the previous run.

[0094] Additionally or alternatively, the method may further comprise at least one additional sorting step. The additional sorting step may comprise a visible light color sorting comprising, by using at least one camera, taking at least one image of the seeds in the seed stream and identifying in the image seeds to be sorted out from the seed stream based on a presence and / or an absence of a color in the image of the seeds in the seeds stream. The visible color sorting may specifically comprise at least one image processing step for determining seeds to be sorted out in the image based on colors of the seed in the seed stream. For example, the identifying of seeds to be sorted out in the visible color sorting may comprise at least one image processing step for identifying seeds in the image and further at least one image processing step for determining if the seeds identified in the image are seeds to be sorted out based on color values of the identified seed. The additional sorting step may be performed prior to step i. and / or after step iv..

[0095] The method, specifically at least step iii., may be computer-implemented. The term “computer- implemented” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a method or method step involving at least one computer and / or at least one computer network. The computer and / or computer network may comprise at least one controller which is configured for performing at least one of the method steps of the method according to the present invention. The computer-implemented method steps may be performed completely automatically, specifically without user interaction.

[0096] In a further aspect of the present invention, a sorting device for sorting cereal seeds is disclosed. The term “sorting device” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a device configured for performing at least one sorting and / or identifying function. Specifically, the sorting device may be configured for differentiating and / or separating the seeds to be sorted out from the other seeds in the seed stream. The sorting device comprises, as will be outlined in further detail below, at least one seed feeder for supplying a seed stream. The sorting device may be or may comprise an optical sorting device configured for sorting based on light information in the visible and / or near-infrared spectral range. Specifically, the sorting device may comprise at least one optical system which is configured, in conjunction with processing software, for identifying seeds to be sorted out in the seed stream. For example, the optical system may comprise at least one spectrometer device and, optionally, at least one camera. Thus, as an example, the sorting device may be configured for visible light color sorting next to NIR sorting, wherein, specifically, different shoot-outs can be combined in one single sorting device, wherein some shoot- outs may use visual light and some shoot-outs may use the NIR sorting of the present invention. The sorting device may further comprise at least one separation system for performing separation of the seeds to be sorted out from the other seeds in the seed stream. For example, the ejector may be part of the separation system.

[0097] The sorting device comprises: a. at least one seed feeder for supplying a seed stream to at least one sorting station; and b. the at least one sorting station, the sorting station comprising at least one spectrometer device for determining spectral information of seeds in the seed stream, wherein the sorting station further comprises at least one ejector for ejecting seeds from the seed stream.

[0098] In a preferred embodiment, the seed feeder comprises a mixture of seeds as described herein above. The sorting device is configured for performing the method according to the present invention, such as according to any one of the embodiments disclosed above and / or according to any one of the embodiments disclosed in further detail below. Thus, for definitions of terms and / or description of possible embodiments, reference is made to the description of the method of sorting cereal seeds above.

[0099] The term “seed feeder” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a device configured for supplying a seed stream. Specifically, the seed feeder may comprise at least one feed hopper configured for receiving a plurality of seeds and for providing the seeds to at least one chute in a controllable fashion. The seed feeder may further comprise at least one vibratory feeder configured for applying vibrations to the feed hopper such that seeds comprised therein may leave the feed hopper to the at least one chute. The seed feeder may be configured for supplying the plurality of seeds arranged in a regular fashion. The seed feeder may be configured for supplying the plurality of seeds arranged individually in a row or line.

[0100] The spectrometer device may comprise, specifically may be, a near-infrared spectrometer device. The spectrometer device may be a reflective spectrometer device, specifically a reflective absorption spectrometer device. The spectrometer device may comprise at least one light source for generating illumination light in the near-infrared spectral range, the illumination light illuminating the seeds in the seed stream.

[0101] As used herein, the term “light” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to electromagnetic radiation in one or more of the infrared, the visible and the ultraviolet spectral range. Herein, the term “ultraviolet spectral range”, generally, refers to electromagnetic radiation having a wavelength of 1 nm to 380 nm, preferably of 100 nm to 380 nm. Further, in partial accordance with standard ISO-21348 in a valid version at the date of this document, the term “visible spectral range”, generally, refers to a spectral range of 380 nm to 760 nm. The term “infrared spectral range” (IR) generally refers to electromagnetic radiation of 760 nm to 1000 pm.

[0102] Consequently, the term “light source”, also referred to as an “illumination source”, as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an arbitrary device configured for generating or providing light in the sense of the above-mentioned definition. The light source specifically may be or may comprise at least one electrical light source, such as an electrically driven light source.

[0103] The spectrometer device may comprise at least one detector for detecting detection light from the seeds in the seed stream. The term “detector” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an arbitrary device configured for detecting, i.e. for at least one of determining, measuring and monitoring, at least one parameter, qualitatively and / or quantitatively, such as at least one of a physical parameter, a chemical parameter and a biological parameter. The detector may be configured for generating at least one detector signal, more specifically at least one electrical detector signal, such as an analogue and / or a digital detector signal, the detector signal providing information on the at least one parameter measured by the detector. The detector may comprise one single optically sensitive element or area or a plurality of optically sensitive elements or areas. Specifically, the detector may be or may comprise at least one detector array, more specifically an array of photosensitive elements. Each of the photosensitive elements may comprise at least a photosensitive area which may be adapted for generating an electrical signal depending on the intensity of the incident light. The detector may comprise a pixelated sensor, such as a CCD chip or a CMOS chip, which are, commonly, used in various cameras nowadays. As an alternative, the detector generally may be or comprise a photoconductor, in particular an inorganic photoconductor, especially PbS, PbSe, Ge, InGaAs, ext. InGaAs, InSb, or HgCdTe.

[0104] Further, in spectroscopy, various sources and paths of light are to be distinguished. In the context of the present invention, a nomenclature is used which, firstly, denotes light propagating from the light source to the seed as “illuminating light” or “illumination light”. Secondly, light propagating from the seed to the detector is denoted as “detection light”. The detection light may comprise at least one of illumination light reflected by the seed, illumination light scattered by the seed, illumination light transmitted by the seed, luminescence light generated by the seed, e.g. phosphorescence or fluorescence light generated by the seed after optical, electrical or acoustic excitation of the seed by the illumination light or the like. Thus, the detection light may directly or indirectly be generated through the illumination of the seed by the illumination light.

[0105] The spectrometer device may further comprise at least one wavelength-selective element for selecting a wavelength of the detection light detected by the detector. The term “wavelength- selective element” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an arbitrary optical element which interacts with differing spectral portions of incident light in a different manner, e.g. by having at least one wavelength-dependent optical property, such as at least one wavelength-dependent optical property selected from the list consisting of a degree of reflection, a direction of reflection, a degree of refraction, a direction of refraction, an absorption, a transmission, an index of refraction. The wavelength-selective element may comprise at least one of a filter, a grating, a prism, a plasmonic filter, a diffractive optical element and a metamaterial. Additionally or alternatively, the wavelength-selective element may comprise at least one of the following elements: a MEMS-interferometer, a MEMS-Fabry Perot interferometer; a linear variable filter; an array of individual filters, such as bandpass filter or the like. Further elements are feasible.

[0106] The sorting device may further comprise at least one camera. The term “camera” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a device having at least one imaging element configured for recording or capturing spatially resolved one-dimensional, two-dimensional or even three-dimensional optical data or information. As an example, the camera may comprise at least one camera chip, such as at least one CCD chip and / or at least one CMOS chip configured for recording images. As used herein, without limitation, the term “image” specifically may relate to data recorded by using a camera, such as a plurality of electronic readings from the imaging device, such as the pixels of the camera chip.

[0107] The camera, besides the at least one camera chip or imaging chip, may comprise further elements, such as one or more optical elements, e.g. one or more lenses. As an example, the camera may be a fix-focus camera, having at least one lens which is fixedly adjusted with respect to the camera. Alternatively, however, the camera may also comprise one or more variable lenses which may be adjusted, automatically or manually.

[0108] The camera specifically may be a color camera. Thus, such as for each pixel, color information may be provided or generated, such as color coordinates for three colors, e.g. H (hue), S (saturation), L (lightness) and / or R, G, B. A larger number of color values is also feasible, such as four colors for each pixel, for example R, G, G, B. Color cameras are generally known to the skilled person. Thus, as an example, each pixel of the camera chip may have three or more different color sensors, such as color recording pixels like one pixel for red (R), one pixel for green (G) and one pixel for blue (B). For each of the pixels, such as for R, G, B, values may be recorded by the pixels, such as digital values in the range of 0 to 255, depending on the intensity of the respective color. Instead of using color triples such as H, S, L and / or R, G, B, as an example, quadruples may be used, such as R, G, G, B or C, M, Y, K or the like. The color sensitivities of the pixels may be generated by color filters or by appropriate intrinsic sensitivities of the sensor elements used in the camera pixels. These techniques are generally known to the skilled person.

[0109] The sorting device may be configured for performing at least one additional sorting step. The additional sorting step may comprise a visible light color sorting comprising, by using the camera, taking at least one image of the seeds in the seed stream and identifying in the image seeds to be sorted out from the seed stream based on a presence and / or an absence of a color in the image of the seeds in the seeds stream.

[0110] Further, the sorting device may comprise at least one controller. The term “controller” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an arbitrary logic circuitry configured for performing basic operations of a computer or system, and / or, generally, to a device which is configured for performing calculations or logic operations. The controller may specifically be a computing unit comprising one or more processors, one or more memory devices and programmable input and / or output peripherals. The devices of the controller may be integrated on a single integrated circuit. The controller may specifically be a microcontroller. The controller may be configured for processing basic instructions that drive a computer or system, e.g. at least one of the spectrometer device, the camera and the ejector. The controller may comprise at least one arithmetic logic unit (ALU), at least one floating-point unit (FPU), such as a math co-processor or a numeric co-processor, a plurality of registers, specifically registers configured for supplying operands to the ALU and storing results of operations, and a memory, such as an L1 and L2 cache memory. In particular, the controller may comprise a multi-core processor. Specifically, the controller may comprise a central processing unit (CPU). Additionally or alternatively, the controller may be or may comprise a microcontroller, thus specifically the controller’s elements may be contained in one single integrated circuitry (IC) chip. Additionally or alternatively, the controller may be or may comprise one or more application-specific integrated circuits (ASICs) and / or one or more field-programmable gate arrays (FPGAs) and / or one or more tensor processing unit (TPU) and / or one or more chip, such as a dedicated machine learning optimized chip, or the like. The controller specifically may be configured, such as by software programming, for performing one or more controlling and / or evaluation operations. The controller may be configured for performing at least step iii. of the method, and wherein, optionally, the controller further may be configured for controlling at least one of steps ii. and iv. of the method.

[0111] The sorting device may further comprise at least one target chute and at least one sort-out chute. The term “target chute” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a seed passage dedicated to seeds which are not to be sorted out. The term “sort-out chute” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a seed passage dedicated to the seeds to be sorted out. The target chute and the sort-out chute may specifically be different from each other. The seeds ejected in step iv., specifically by the ejector, may be collected by the sort-out chute. The remaining seed stream may be collected by the target chute. The sorting station, specifically the ejector, may be configured for separating the seeds from the seed stream into the sort-out chute in case the seeds are identified as seeds to be sorted out. The sorting device may be configured such that seeds from the seed stream which are not identified as seeds to be sorted out may pass the sorting station towards the target chute.

[0112] The term “ejector” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an arbitrary device configured for ejecting seeds. The ejector may specifically be or may comprise at least one of a mechanical device and a pneumatic device. The ejector may comprise at least one of a mechanical ejector having at least one mechanical actor and a pneumatic ejector having at least one nozzle for directing an air jet at the seed to be sorted out from the seed stream. For example, the ejector may comprise the at least one pneumatic ejector configured for ejecting the seed to be sorted out from the seed stream by using compressed air directed via one or more nozzles to separate the seed to be sorted out from the seed stream.

[0113] In a further aspect of the present invention, a computer program is disclosed, comprising instructions which, when the program is executed by the sorting device according to the present invention, such as according to any one of the embodiments disclosed above and / or according to any one of the embodiments disclosed in further detail below, cause the sorting device to per- form the method according to the present invention, such as according to any one of the embodiments disclosed above and / or according to any one of the embodiments disclosed in further detail below.

[0114] Thus, specifically, one, more than one or even all of method steps i. to iii., and optionally step iv. , as indicated above may be performed and / or controlled by using a computer or a computer network, preferably by using a computer program.

[0115] Similarly, a computer-readable storage medium, specifically a non-transient computer-readable medium, is disclosed, comprising instructions which, when the instructions are executed by the sorting device according to the present invention, such as according to any one of the embodiments disclosed above and / or according to any one of the embodiments disclosed in further detail below, cause the sorting device to perform the method according to the present invention, such as according to any one of the embodiments disclosed above and / or according to any one of the embodiments disclosed in further detail below.

[0116] As used herein, the term “computer-readable data medium” specifically may refer to non-transi- tory data storage means, such as a hardware storage medium having stored thereon computerexecutable instructions. The computer-readable storage medium specifically may be or may comprise a storage medium such as a random-access memory (RAM) and / or a read-only memory (ROM). The computer-readable storage medium may be or may comprise at least one computer-readable data carrier.

[0117] Referring to the computer-implemented aspects of the invention, one or more of the method steps or even all of the method steps of the method according to one or more of the embodiments disclosed herein may be performed and / or controlled by using a computer or computer network. Thus, generally, any of the method steps including provision and / or manipulation of data may be performed by using a computer or computer network. Generally, these method steps may include any of the method steps, typically except for method steps requiring manual work, such as providing the samples and / or certain aspects of performing the actual measurements.

[0118] In a further aspect of the present invention, a use of the sorting device according to the present invention, such as according to any one of the embodiments disclosed above and / or according to any one of the embodiments disclosed in further detail below, for sorting of cereal seeds of a seed stream containing sterile cereal seeds and fertile cereal seeds is disclosed. The sterile cereal seeds comprise male sterile seeds. The method of sorting cereal seeds and the sorting device according to the present invention may provide a large number of advantages over known methods and devices. Specifically, the method and the sorting device may enable a reliable, color-independent and non-destructive seed sorting method of sorting sterile and fertile seeds. Thus, as an example, the present invention may be used in any former 2-line male sterility system having mis-division (i.e. , separation of the restorer gene / locus from the linked color marker locus / gene), so as to significantly reduce fertile seeds in the male sterile female seed used for commercial hybrid seed production (useful everywhere but specifically in regions like Europe, where specific purity standards have to be met for authorization and / or registration). The method may improve use and / or acceptability of such 2-line systems, in particular systems involving the separation of fertile white or blue seeds from white sterile seeds in any 2-line system with e.g. a BLA locus as marker for the restorer gene (e.g., with an addition chromosome or homoeologous chromosome pair comprising a fertility restorer gene / locus and a color marker on the same chromosome). This may be of particular importance to meet regulatory requirements, such as hybrid purity for a hybrid wheat launch. Further, the method and the sorting device may be used in any commercialization of 2-line systems with some mis-division, in improved seed sorting of white sterile and white or blue fertile seeds in high throughput, and in improved screening of mutants or breeding lines for (presence or absence or % of) mis-division.

[0119] The method may specifically provide a near-infrared spectroscopy (NIRS)-based classifier for differentiation between sterile and fertile seed, in particular between white sterile seeds and white fertile seeds and / or blue fertile seeds. The classifier may specifically detect the presence of the many genes (or the gene products produced by said genes, or the consequences of these gene products (e.g. a change in cell wall, or starch crystallinity) which are present on the addition chromosome or on a part of the addition chromosome via N IRS-detectable differences allowing to correctly classify white seeds without addition chromosome from seeds containing the addition chromosome or part of the addition chromosome independent on the presence or absence of the blue aleurone. The classifier may specifically be an Al-based classifier for differentiating between white sterile seeds, white fertile seeds, blue fertile seeds and blue sterile seeds (across several genotypes and growing environments). The classifier may identify the mis-divided blue seeds with an apparent accuracy of above 80 % accuracy. The classifier may be further trained using genotyping data comprising information on the presence or absence of the MS1 restorer gene and the BLA color locus.

[0120] The classifier may be configured for determining the presence of the addition chromosome or at least one part of the addition chromosome, leading to a detectable alteration of seeds due to the expression of thousands of additional genes located on the addition chromosome or remaining mis-divided parts of it. The resulting changes in grain composition or structural features may be detected by the NIRS. The method may be used for segregating sterile white seeds lacking an additional chromosome from fertile blue seeds containing the complete additional chromosome with the MS1 restorer gene and BA / BLA color marker locus as well as for segregating white sterile seed (lacking an additional chromosome) from mis-divided fertile white seeds, which are only carrying the MS1 restorer gene (and not the BLA color marker locus), using the single seed NIRS inspection.

[0121] The NIRS spectra of the seeds may comprise spectra obtained in reflection or transmission mode. The N IRS-based classifier may be specific for the blue seeds versus the white seeds of the normal BLA seed lots. The N IRS-based classifier may further identify the correct percentage of mis-divided seeds within the entire white seed lot. The NIRS-based classifier may specifically be configured for identifying spectral features in restricted and / or partial regions of the entire NIRS-spectrum enabling a shorter scanning and / or processing time in high-throughput seed sorting.

[0122] The classifier may specifically be configured for screening BLA lines for identification of lines with reduced or zero occurrence of mis-divided seeds via NIRS, and may serve as a non-de- structive alternative to phenotypic sterility screening or molecular seed bulk Ms1 screening. Thus, the method may save time since the seeds can be directly subjected to mis-division analysis. Further, the classifier may be configured for counter-selecting breeding lines prone to a high degree of mis-division. Thus, the method may be used as a final step in seed sorting to get pure white sterile seeds, essentially depleted for any (1) light blue (1n BA) fertile seeds or (2) mis-divided white but fertile seeds, in order to ensure meeting the criteria for registration and / or certification of cereal seeds. Specifically, the method can be implemented on any seed sorting device equipped with a NIRS sensor.

[0123] The classifier may specifically be an Al-based classifier for differentiating between white sterile seed and white fertile seed or blue seeds with an apparent accuracy of above 80 %, such as at least 85 %, 90 %, 95 %, 99 %, or 99,7 % accuracy. The accuracy can be fine-tuned by performing a genotypic analysis of the N IRS-predicted phenotype to validate and further train the model underlying the classifier. The training may increase the accuracy of the method. The classifier may be configured for identifying spectral regions of interest in the near-infrared spectrum. This may specifically be implemented in an easy way on an existing seed sorter equipped with a suitable NIRS. The method and the sorting device may be able to distinguish white sterile from white fertile seeds using a non-destructive method by the near-infrared-based seed sorting. Thus, the method and the sorting device may be used in commercial hybrid breeding platforms using the currently available 2-line systems showing mis-division. In addition, the method can be used in combination with visible light color sorting to allow further improving the purity of the white sterile seeds in seed lots which are weakly colored. The NIRS-separation also allows for the fast identification of lines with no or less mis-division / breakage of the BLA addition chromosome, by directly analyzing the white seed fraction of such lines and determining the percent of mis-di- vided seeds amongst them.

[0124] The present invention further relates to a method for providing a seed mixture consisting essentially of white sterile seeds, said method comprising : a) subjecting a mixture of seeds as described herein above to the method of sorting of the present invention, b) identifying white sterile seeds by said method, c) providing a seed batch consisting essentially of the white sterile seeds identified in step b). In one embodiment, this method comprises the additional step of producing hybrid seeds from said (purified) white sterile seeds.

[0125] In a preferred embodiment, the method comprises the following steps: a) allowing a plant as described herein above to self-pollinate, b) harvesting seeds from said plant, thereby obtaining a mixture of seeds as described herein above, c) subjecting the mixture of seeds to the method of sorting of the present invention, d) identifying white sterile seeds by said method, e) providing a seed batch consisting essentially of the white sterile seeds identified in step d). In one embodiment, this method comprises the additional step of producing hybrid seeds from said (purified) white sterile seeds.

[0126] In an embodiment, the seed batch provided in the last step c) or d) is a seed batch that comprises at least 98% white sterile seeds, such as least 99% white sterile seeds, in particular at least 99.5%, 99.6 % or 99.7 % white sterile seeds.

[0127] The present invention further relates to a method for producing male sterile female plants, said method comprising carrying out the steps of the method of the present invention for providing a seed batch consisting essentially of white sterile seeds and allowing the white sterile seeds to grow into the male sterile female plants. In a further step, the grown male sterile female plants are crossed with genetically diverse (male fertile, pollen-providing) plants to obtain hybrid seed.

[0128] As used herein, the terms “have”, “comprise” or “include” or any arbitrary grammatical variations thereof are used in a non-exclusive way. Thus, these terms may both refer to a situation in which, besides the feature introduced by these terms, no further features are present in the entity described in this context and to a situation in which one or more further features are present. As an example, the expressions “A has B”, “A comprises B” and “A includes B” may both refer to a situation in which, besides B, no other element is present in A (i.e. a situation in which A solely and exclusively consists of B) and to a situation in which, besides B, one or more further elements are present in entity A, such as element C, elements C and D or even further elements.

[0129] Further, it shall be noted that the terms “at least one”, “one or more” or similar expressions indicating that a feature or element may be present once or more than once typically are used only once when introducing the respective feature or element. In most cases, when referring to the respective feature or element, the expressions “at least one” or “one or more” are not repeated, notwithstanding the fact that the respective feature or element may be present once or more than once.

[0130] Further, as used herein, the terms "preferably", "more preferably", "particularly", "more particularly", "specifically", "more specifically" or similar terms are used in conjunction with optional features, without restricting alternative possibilities. Thus, features introduced by these terms are optional features and are not intended to restrict the scope of the claims in any way. The invention may, as the skilled person will recognize, be performed by using alternative features. Similarly, features introduced by "in an embodiment of the invention" or similar expressions are intended to be optional features, without any restriction regarding alternative embodiments of the invention, without any restrictions regarding the scope of the invention and without any restriction regarding the possibility of combining the features introduced in such way with other optional or non-optional features of the invention.

[0131] The non-destructive NIR based separation method described herein which allows to sort white fertile (WF) seeds from white sterile (WS) would enable the development of a commercial hybrid breeding platform in the most stringent countries, from the currently available 2-line system that has some mis-division. In addition, such method, in combination with visible light color sorting, would allow to further improve the purity of the white sterile seed fractions in seed lots with a very faint blue aleurone expression. Such NIR based separation also allows the identification of the proportion (or absence) of mis-divided fertile white seeds in (sterile) white seed fractions. In one embodiment of the invention said NIR based separation method is not based on determining the presence or concentration of an organic colorant, such as anthocyanin, e.g., when separating white fertile seeds from white sterile seeds (such as such seeds produced in a cereal hybrid system based on a (monosomic) addition chromosome or on a homoeologous chromosome pair comprising a fertility restorer gene / locus and a color marker gene / locus, in a genetic background comprising a homozygous mutation in a male fertility gene causing male sterility, in absence of said fertility restorer gene / locus).

[0132] Summarizing and without excluding further possible embodiments, the following embodiments may be envisaged:

[0133] Embodiment 1 : A method of sorting cereal seeds, wherein the cereal seeds contain sterile cereal seeds, specifically male sterile cereal seeds, and fertile cereal seeds, the method comprising: i. supplying a seed stream to a sorting station, the sorting station comprising at least one spectrometer device for determining spectral information of seeds in the seed stream; ii. taking, with the spectrometer device, at least one near-infrared spectrum of a seed in the seed stream; and iii. automatically identifying seeds to be sorted out from the seed stream by using at least one classifier, the classifier being configured for classifying the seed to be a sterile cereal seed, specifically a male sterile cereal seed, or a fertile cereal seed based on the near-infrared spectrum taken in step ii..

[0134] Embodiment 2: The method according to the preceding embodiment, wherein the classifier is configured for determining a presence and / or an absence of at least a part of an addition chromosome in the cereal seeds (such as a part inherited to progeny).

[0135] Embodiment 3: The method according to the preceding embodiment, wherein the addition chromosome comprises a fertility restorer gene / locus, such as MS1 , MS5, MS45, or MS9, and a color marker gene / locus, such as a blue aleurone gene / locus, and wherein the addition chromosome is monosomic or is part of a homoeologous chromosome pair.

[0136] Embodiment 4: The method according to any one of the preceding embodiments, wherein the classifier is further configured, based on the near-infrared spectrum, for classifying the seed into at least one category selected from the group consisting of: a white-sterile cereal seed; a white-fertile cereal seed; a blue-sterile cereal seed; a blue-fertile cereal seed.

[0137] Embodiment 5: The method according to any one of the preceding embodiments, wherein the seed stream comprises, specifically consist of, white-sterile cereal seeds, white-fertile cereal seeds, blue-sterile cereal seeds and / or blue-fertile cereal seeds.

[0138] Embodiment 6: The method according to any one of the preceding embodiments, wherein the seed stream comprises, specifically consist of, white-sterile cereal seeds and white-fertile cereal seeds.

[0139] Embodiment 7: The method according to any one of the preceding embodiments, wherein seeds comprising a portion of a monosomic addition chromosome comprising a fertility restorer gene / locus, but lacking the portion of said monosomic addition chromosome comprising a color marker gene / locus (white fertile seeds) are sorted from seeds lacking (a portion of) the monosomic addition chromosome (white sterile seeds).

[0140] Embodiment 8: The method according to any one of the two preceding embodiments, wherein the white-sterile cereal seeds and white-fertile cereal seeds in the seed stream are cereal seeds previously sorted by a color seed sorting method, specifically a camera-based color seed sorting method, the color seed sorting method separating the white-sterile cereal seeds and white-fertile cereal seeds from an initial seeds stream comprising white-sterile cereal seeds, white-fertile cereal seeds, blue-sterile cereal seeds and / or blue-fertile cereal seeds.

[0141] Embodiment 9: The method according to any one of the preceding embodiments, wherein the seeds to be sorted out from the seed stream are the fertile cereal seeds, specifically white-fertile cereal seeds and / or blue-fertile cereal seeds.

[0142] Embodiment 10: The method according to any one of the preceding embodiments, further comprising determining a degree of mis-division of the seed stream, wherein the degree of mis-division quantifies a fraction of fertile cereal seeds in the seed stream.

[0143] Embodiment 11 : The method according to any one of the preceding embodiments, wherein the near-infrared spectrum is obtained in reflection geometry, specifically with reflective background surfaces. Embodiment 12: The method according to any one of the preceding embodiments, wherein the near-infrared spectrum comprises an absorbance of the seed in the seed stream as a function of wavenumber.

[0144] Embodiment 13: The method according to any one of the preceding embodiments, wherein the near-infrared spectrum is obtained for a spectral region of 3,600 cm-1to 12,500 cm-1.

[0145] Embodiment 14: The method according to the preceding embodiment, wherein a spectral region of 12,500 cm-1to 11 ,000 cm-1is excluded in the near-infrared spectrum, or wherein the classifier is not based on the detection of the anthocyanin concentration.

[0146] Embodiment 15: The method according to any one of the preceding embodiments, wherein the classifier is configured for classifying the seed based on spectral features from at least one spectral region selected from the group consisting of: 4,000 cm-1to 7,000 cm-1; 9,000 cm’1to 12,000 cm’1; 8260 erm1to 8430 erm1.

[0147] Embodiment 16: The method according to any one of the preceding embodiments, wherein the classifier comprises at least one of a F-classifier configured for classifying seeds into fertile cereal seeds and non-fertile cereal seeds, and a S-classifier configured for classifying seeds into sterile cereal seeds and non-sterile cereal seeds.

[0148] Embodiment 17: The method according to any one of the preceding embodiments wherein the classifier comprises, specifically is, a trained artificial neural network (ANN).

[0149] Embodiment 18: The method according to the preceding embodiment, further comprising at least one training step, wherein, in the training step, the ANN is trained using labeled near-infrared spectra.

[0150] Embodiment 19: The method according to the preceding embodiment, wherein the labeled near-infrared spectra comprise genotyping data, wherein the genotyping data provide information on sterility and / or fertility of seeds under test and / or provide information on the presence or absence of a fertility restorer gene.

[0151] Embodiment 20: The method according to any one of the two preceding embodiments, wherein the training step comprises a pretreatment of near-infrared spectra. Embodiment 21 : The method according to the preceding embodiment, wherein the pretreatment of near-infrared spectra comprises at least one of a normalization of the near-infra- red spectra and a principal component analysis (PCA) of the near-infrared spectra.

[0152] Embodiment 22: The method according to the preceding embodiment, wherein the normalization of the near-infrared spectra comprises at least one normalization selected from the group consisting of: a minimum-maximum normalization of the near-infrared spectra; a mean centering-standard normalization.

[0153] Embodiment 23: The method according to any one of the five preceding embodiments, wherein, in the training step, the ANN is trained using the following hyperparameters: preset: Narrow Neural Network; number of fully connected layers: 1 ; first layer size: 10; activation: rectified linear unit (ReLLI); iteration limit: 1 ,000; regularization strength (Lambda): 0; and standardize data: yes.

[0154] Embodiment 24: The method according to any one of the six preceding embodiments, wherein the training step comprises a k-fold cross-validation of the ANN, specifically a 20-fold cross validation of the ANN.

[0155] Embodiment 25: The method according to any one of the preceding embodiments, wherein the using of the classifier comprises classifying the seed to be a sterile cereal seed or a fertile cereal seed based on the near-infrared spectrum, such as wherein said sterile cereal seed is white.

[0156] Embodiment 26: The method according to any one of the preceding embodiments, further comprising: iv. automatically ejecting seeds identified to be sorted out from the seed stream.

[0157] Embodiment 27: The method according to the preceding embodiment, wherein a batch of seeds is provided, wherein the batch of seeds is subjected to method steps i.-iv. repeatedly, specifically at least twice, wherein, in each repetition, the batch is diminished by the seeds ejected in step iv. of the previous run. Embodiment 28: The method according to any one of the two preceding embodiments, further comprising at least one additional sorting step, wherein the additional sorting step comprises a visible light color sorting comprising, by using at least one camera, taking at least one image of the seeds in the seed stream and identifying in the image seeds to be sorted out from the seed stream based on a presence and / or an absence of a color in the image of the seeds in the seeds stream.

[0158] Embodiment 29: The method according to the preceding embodiment, wherein the additional sorting step is performed prior to step i. and / or after step iv..

[0159] Embodiment 30: The method according to any one of the preceding method embodiments, wherein the method is computer-implemented.

[0160] Embodiment 31 : A sorting device for sorting cereal seeds, comprising: a. at least one seed feeder for supplying a seed stream to at least one sorting station; and b. the at least one sorting station, the sorting station comprising at least one spectrometer device for determining spectral information of seeds in the seed stream, wherein the sorting station further comprises at least one ejector for ejecting seeds from the seed stream, wherein the sorting device is configured for performing the method according to any one of the preceding embodiments.

[0161] Embodiment 32: The sorting device according to the preceding embodiment, wherein the spectrometer device comprises, specifically is, a near-infrared spectrometer device.

[0162] Embodiment 33: The sorting device according to any one of the preceding embodiments referring to a sorting device, wherein the spectrometer device is a reflective spectrometer device, specifically a reflective absorption spectrometer device.

[0163] Embodiment 34: The sorting device according to any one of the preceding embodiments referring to a sorting device, wherein the spectrometer device comprises at least one light source for generating illumination light in the near-infrared spectral range, the illumination light illuminating the seeds in the seed stream. Embodiment 35: The sorting device according to any one of the preceding embodiments referring to a sorting device, wherein the spectrometer device comprises at least one detector for detecting detection light from the seeds in the seed stream.

[0164] Embodiment 36: The sorting device according to the preceding embodiment, wherein the spectrometer device further comprises at least one wavelength-selective element for selecting a wavelength of the detection light detected by the detector.

[0165] Embodiment 37: The sorting device according to any one of the preceding embodiments referring to a sorting device, further comprising at least one camera, wherein the sorting device is configured for performing at least one additional sorting step, wherein the additional sorting step comprises a visible light color sorting comprising, by using the camera, taking at least one image of the seeds in the seed stream and identifying in the image seeds to be sorted out from the seed stream based on a presence and / or an absence of a color in the image of the seeds in the seeds stream.

[0166] Embodiment 38: The sorting device according to any one of the preceding embodiments referring to a sorting device, further comprising at least one controller, wherein the controller is configured for performing at least step iii. of the method, and wherein, optionally, the controller further is configured for controlling at least one of steps ii. and iv. of the method.

[0167] Embodiment 39: The sorting device according to any one of the preceding embodiments referring to a sorting device, further comprising at least one target chute and at least one sort- out chute, wherein the seeds ejected in step iv. are collected by the sort-out chute and wherein the remaining seed stream is collected by the target chute.

[0168] Embodiment 40: The sorting device according to any one of the preceding embodiments referring to a sorting device, wherein the ejector comprises at least one of a mechanical ejector having at least one mechanical actor and a pneumatic ejector having at least one nozzle for directing an air jet at the seed to be sorted out from the seed stream.

[0169] Embodiment 41 : A computer program comprising instructions which, when the program is executed by the sorting device according to any one of the preceding embodiments referring to a sorting device, cause the sorting device to perform the method according to any one of the preceding embodiments referring to a method. Embodiment 42: A computer-readable storage medium, specifically a non-transient computer- readable medium, comprising instructions which, when the instructions are executed by the sorting device according to any one of the preceding embodiments referring to a sorting device, cause the sorting device to perform the method according to any one of the preceding embodiments referring to a method.

[0170] Embodiment 43: A use of the sorting device according to any one of the preceding embodiments referring to a sorting device for sorting of cereal seeds of a seed stream containing sterile cereal seeds and fertile cereal seeds, the sterile cereal seeds comprising male sterile seeds (lacking an addition chromosome).

[0171] Embodiment 44: The method or use according to any one of the preceding embodiments referring to a method or use, wherein a spectral region of 12,500 cm-1to 11 ,000 cm-1is excluded in the near-infrared spectrum, or wherein the classifier is not based on the detection of the anthocyanin concentration.

[0172] Embodiment 45: The method or use according to any one of the preceding embodiments referring to a method or use, wherein the monosomic addition chromosome contains one or more fragments of chromosomes from one or more other cereal species than the species of said cereal seed.

[0173] Embodiment 46: The method or use according to any one of the preceding embodiments referring to a method or use, wherein said sterile and fertile cereal seeds are white cereal seeds (lacking a blue aleurone), or wherein said sterile cereal seed is white and said fertile cereal seed has a very light / faint blue aleurone color or has a low anthocyanin concentration so that it is not efficiently distinguished from sterile white seed in standard color sorting or in NIR-based sorting determining the presence or concentration of anthocyanin.

[0174] Short description of the Figures

[0175] Further optional features and embodiments will be disclosed in more detail in the subsequent description of embodiments, preferably in conjunction with the dependent claims. Therein, the respective optional features may be realized in an isolated fashion as well as in any arbitrary feasible combination, as the skilled person will realize. The scope of the invention is not restricted by the preferred embodiments. The embodiments are schematically depicted in the Figures. Therein, identical reference numbers in these Figures refer to identical or functionally comparable elements. In the Figures:

[0176] Figure 1 shows an embodiment of a sorting device for sorting cereal seeds in a schematic side view;

[0177] Figure 2 shows an embodiment of a method of sorting cereal seeds;

[0178] Figure 3 shows exemplary near-infrared spectra of cereal seeds in reflection and transmission geometry;

[0179] Figure 4A to 4C show examples of pretreated near-infrared spectra of cereal seeds;

[0180] Figure 5 shows a variance diagram of near-infrared spectra of cereal seeds;

[0181] Figures 6A to 6C show diagrams of regions of interest in the near-infrared spectra; and

[0182] Figures 7A to 7C show classification results for different classifiers.

[0183] Detailed description of the embodiments

[0184] Figure 1 shows an exemplary embodiment of a sorting device 110 for sorting cereal seeds 112 in a schematic side view. The cereal seeds 112 contain sterile cereal seeds 114 and fertile cereal seeds 116. For example, the cereal seeds 112 may be wheat seeds.

[0185] The sorting device 110 comprises at least one seed feeder 118 for supplying a seed stream 120 to at least one sorting station 122. As shown in Figure 1, in this exemplary embodiment, the seed feeder 118 may comprise at least one feed hopper 124 configured for receiving a plurality of seeds 112 and for providing the seeds 112 to at least one chute 126 in a controllable fashion. The seed feeder 118 may further comprise at least one vibratory feeder 128 configured for applying vibrations to the feed hopper 124 such that seeds 112 comprised therein may leave the feed hopper 124 to the at least one chute 126.

[0186] Further, the sorting device 110 comprises the at least one sorting station 122. The sorting station 122 comprises at least one spectrometer device 130 for determining spectral information of seeds 112 in the seed stream 120. The spectrometer device 130 may specifically be a near-infrared spectrometer device. The spectrometer device 130 may be configured for taking near- infrared spectra of the seeds 112 in the seed stream 120. As shown in Figure 1 , the spectrometer device 130 may be a reflective spectrometer device. The spectrometer device 130 may comprise at least one light source 132 for generating illumination light 134 in the near-infrared spectral range, the illumination light 134 illuminating the seeds 112 in the seed stream 120. Further, the spectrometer device 130 may comprise at least one detector 136 for detecting detection light 138 from the seeds 112 in the seed stream 120. The spectrometer device 130 may further comprise at least one wavelength-selective element (not shown in Figure 1) for selecting a wavelength of the detection light 138 detected by the detector 136. As can be seen in Figure 1 , the light source 132 and the detector 136 may be arranged on the same side of the seed stream 120. Thus, the spectrometer device 130 may operate in reflection geometry. The sorting station 122, as an example, may comprise two spectrometer devices 130 arranged on opposing sides of the seed stream 120.

[0187] The sorting station 122 further comprises at least one ejector 140 for ejecting seeds 112 from the seed stream 120. The ejector may comprise a pneumatic ejector 142 having at least one nozzle for directing an air jet at the seed to be sorted out from the seed stream 120. For example, the ejector 142 may comprise the at least one pneumatic ejector 142 configured for ejecting the seed to be sorted out from the seed stream 120 by using compressed air directed via nozzles to separate the seed to be sorted out from the seed stream 120.

[0188] The sorting device 110 may further comprise at least one target chute 144 and at least one sort- out chute 146. In this example, the seeds to be sorted out may comprise the fertile cereal seeds 116. The seeds ejected by the ejector 140 may be collected by the sort-out chute 146. The remaining seed stream may be collected by the target chute 144. The sorting station 122, specifically the ejector 140, may be configured for separating the seeds 112 from the seed stream 120 into the sort-out chute 146 in case the seeds 112 are identified as seeds to be sorted out. The sorting device 110 may be configured such that seeds 112 from the seed stream 120 which are not identified as seeds to be sorted out may pass the sorting station 122 towards the target chute 144.

[0189] The sorting device 110 may further comprise at least one controller 148. As shown in Figure, 1 , the controller 148 may be configured for communicating, e.g. via wireless and / or wired means, with other devices of the sorting device 110, e.g. with the seed feeder 118 and / or with the sorting station 122.

[0190] The sorting device 110 is configured for performing the method according to the present invention, such as according to the exemplary embodiment of Figure 2 and / or according to any other embodiment disclosed herein. Thus, for a description of the method, reference is made to the description of Figure 2.

[0191] Figure 2 shows an embodiment of a method of sorting cereal seeds 112. The cereal seeds 112 contain sterile cereal seeds 114 , specifically male sterile cereal seeds 114, and fertile cereal seeds 116.

[0192] The method comprises the following steps that may be performed in the given order. However, a different order may also be possible. In particular, one, more than one or even all of the method steps may be performed once or repeatedly. Further, the method steps may be performed successively or, alternatively, one or more of the method steps may be performed in a timely overlapping fashion or even in a parallel fashion and / or in a combined fashion. The method may further comprise additional method steps that are not listed.

[0193] The method comprises: i. (denoted by reference number 150) supplying a seed stream 120 to a sorting station 122, the sorting station 122 comprising at least one spectrometer device 130 for determining spectral information of seeds 112 in the seed stream 120; ii. (denoted by reference number 152) taking, with the spectrometer device 130, at least one near-infrared spectrum of a seed 112 in the seed stream 120; and iii. (denoted by reference number 154) automatically identifying seeds to be sorted out from the seed stream 120 by using at least one classifier, the classifier being configured for classifying the seed 112 to be a sterile cereal seed 114 , specifically a male sterile cereal seed 114 or a fertile cereal seed 116 based on the near-infrared spectrum taken in step ii..

[0194] As shown in Figure 2, the method optionally further comprises: iv. (denoted by reference number 156) automatically ejecting seeds identified to be sorted out from the seed stream 120.

[0195] In the exemplary embodiment of Figure 2, the classifier may comprise, specifically may be, a trained artificial neural network (ANN). The method may further comprise at least one training step. In the training step, the ANN may be trained using labeled near-infrared spectra. The labeled near-infrared spectra may comprise genotyping data. The genotyping data may provide information on sterility and / or fertility of seeds under test. The labeled near-infrared spectra may be obtained by using genotyping data of the respective cereal seeds, specifically indicating presence of one or more of the restorer gene, the color gene, a specific locus as detected using markers. Thus, the labeled near-infrared spectra may provide ground truth data for the training step. In the following Figures 3 to 7C, details on the training step of the ANN will be described.

[0196] Thus, for details on the trained ANN, reference is made to the description of Figures 3 to 7C.

[0197] Figure 3 shows exemplary near-infrared spectra of cereal seeds in reflection and transmission geometry. Specifically, in the diagram of Figure 3, the absorbance 158 is shown as a function of wavenumber 160. Near-infrared spectra of different samples in reflection and transmission geometry are shown in Figure 3. As can be seen in Figure 3, near-infrared spectra obtained in reflection geometry may have distinct spectral features and, thus, may be used for the present sorting method.

[0198] The training step may comprise a pretreatment of near-infrared spectra. Figure 4A to 4C show examples of pretreated near-infrared spectra of cereal seeds. Figure 4A shows raw near-infra- red spectra as obtained from the spectrometer device 130. The pretreatment of near-infrared spectra, as an example, may comprise at least one of a normalization of the near-infrared spectra and a principal component analysis (PCA) of the near-infrared spectra. The normalization of the near-infrared spectra may comprise at least one normalization selected from the group consisting of: a minimum-maximum normalization of the near-infrared spectra; a mean centeringstandard normalization. The minimum-maximum normalization of the near-infrared spectra is shown in Figure 4B and may comprise setting a minimal value of the near-infrared spectrum to 0 and a maximum value to 1. The mean centering-standard normalization is shown in Figure 4C and may comprise subtracting an average absorbance of the near-infrared spectrum and dividing the obtained spectrum by a standard deviation of the near-infrared spectrum.

[0199] Figure 5 shows a variance diagram of near-infrared spectra of cereal seeds. In particular, Figure 5 shows the variance of the pretreated (standard normalized) near-infrared spectra and their first and second derivatives. Specifically, Figure 5 shows the variance 162 of the absorbance 158 as a function of wavenumber 160. Additionally, the variance of the first and second derivative are shown in the diagram of Figure 5 (denoted by reference number 164). As can be seen in Figure 5, the near-infrared spectra of the cereal seeds 112 may comprise two major regions in the spectral range of 4,000 cm-1to 7,500 cm-1and 8,500 cm-1to 12,000 cm-1.

[0200] This can be further seen in Figures 6A to 6C. Figures 6A to 6C show diagrams of regions of interest in the near-infrared spectra. The diagrams of Figures 6A to 6C show a heat map of an intensity of the near-infrared spectra as a function of wavenumber 160 and for different seed lots 166. Figure 6A shows the heat map for the mean centering-standard normalization of the nearinfrared spectra, Figure 6B shows the heat map for the minimum-maximum normalization of the near-infrared spectra and Figure 6C shows the heat map for the raw near-infrared spectra. In particular, the diagrams show the correlation (Pearson) of wavenumber regions with the expected general grouping into sterile and fertile genotype. This correlation study was done separately on each seed lot to see if general patterns between seed lots emerge. The correlation coefficient is shown as gray scale intensity (dark is RA2 close to 1 , values below 0.3 are depicted as white due to insufficient correlation). The Y-axis displays the spectral range from 3600 to 12500 wavenumbers while the X axis displays the different seed lot numbers 166. As can be seen in Figures 6A to 6C, three different regions of interest can be identified in the near-infrared spectral range: A first region from 4,000 cm-1to 7,000 cm-1(corresponding to approx. 1400 nm to 2500 nm), a second region around 8,000 cm-1(corresponding to approx. 1200 nm) and a third region from 9,000 cm-1towards the visible spectral range (corresponding to approx, below 1100 nm). Thus, in this example, the classifier may be configured for classifying the seed 112 based on spectral features from at least one spectral region selected from the group consisting of: 4,000 cm’1to 7,000 cm’1; 9,000 cm’1to 12,000 cm’1; 8260 cm’1to 8430 cm’1.

[0201] As an example, in the method of Figure 2, a training data set comprising near-infrared spectra from 1910 seeds comprising white-sterile cereal seeds, white-fertile cereal seeds, blue-sterile cereal seeds and blue-fertile cereal seeds was used for training the ANN. The near-infrared spectra were obtained on a Multi Purpose Analyzer (MPA) from Bruker® and were annotated according to genotype classes white (W), blue (B), sterile (S) and fertile (F). To counteract a potential overfitting of the classifier, seeds from a set of different BLA genotypes and seeds from BLA plants grown in different environments were analyzed. Normalization of raw spectra was performed for better separation of genotypes (PCA). Hierarchical cluster analysis (HCA) and Kmeans of individual lines enabled unsupervised clustering which followed relatively well the expected ratios between S and F types in WF populations. These data were used as initial ground truth data to create an initial “noisy” Al classifier (ANN). The training was done using 1143 of 1910 individual near-infrared spectra. The spectral data consisted of one or two technical replicate measurements for each seed. The model was first trained using 80% of the nearinfrared spectra with 5% of the training set used in cross-validation after each iteration and 20% testing. The ANN was a single-layer neural network as described above.

[0202] Further, in the training step, genotypic validation of the initial NIRS classification results was performed and was enabled by providing the seeds in grid plates allowing for tracking their identity. The model training was repeated using all 1143 individual near-infrared spectra with 5% cross-validation in a second attempt. Both models performed similarly. The second model was used for classification. The calibrating of the classifier may further improve accuracy of the classifier. The True Positive Ratio (TPR) of the prediction of the different classes may be as follows for the calibrated ANN classifier based on four classification groups: white fertile (WF): 91 %; white sterile (WS): 95%; blue fertile (BF): 99%; blue sterile (BS): 67%. It should be noted that the BS group was very small and hence under-represented, which can explain the relatively lower prediction accuracy. Predictions can be made on a new predictor row matrix X, i.e. the NIR spectrum of the seed, wherein X contains exactly 4449 rows corresponding to the 4449 predictors (wavenumbers) used for training of the ANN.

[0203] Figures 7A to 7C show classification results for different trained classifiers. Specifically, the diagrams of Figures 7A to 7C show the genotype prediction 168 for different seed lots 166. In the example of Figures 7A to 7C, the classification results of unsupervised HCA in combination with Kmeans clustering (denoted by reference number 170) and the trained ANN (denoted by reference number 172) is shown together with a reference ratio (denoted by reference number 174). Figure 7A shows the classification results for the mean centering-standard normalization of the near-infrared spectra, Figure 7B shows the classification results for the minimum-maximum normalization of the near-infrared spectra and Figure 7C shows the heat map for the raw near-infrared spectra. The training was performed on 1910 near-infrared spectra in total obtained on a Multi Purpose Analyzer (MPA) from Bruker® and annotated according to genotype classes white (W), blue (B), sterile (S) and fertile (F). As can be seen in Figure 7A to 7C, the classification is mostly consistent across data normalization and corresponds to the expected reference ratio. Mixed genotype lines could be predicted both by the unsupervised HCA I Kmean clustering classifier and the trained ANN. However, the trained ANN generally followed the expected ration to a higher degree compared with the HCA / Kmean clustering classifier.

[0204] Example 1 :

[0205] In a first example, a training data set comprising near-infrared spectra from 1910 seeds comprising white-sterile cereal seeds, white-fertile cereal seeds, blue-sterile cereal seeds and blue- fertile cereal seeds was used for training the ANN. The ANN was a single-layer neural network. The near-infrared spectra were obtained on a Multi Purpose Analyzer (MPA) from Bruker® and were annotated according to genotype classes white (W), blue (B), sterile (S) and fertile (F). To counteract a potential overfitting of the classifier, seeds from a set of different BLA genotypes and seeds from BLA plants grown in different environments were analyzed. Normalization of raw spectra allowed for better separation of genotypes (PCA). HCA and Kmeans of individual lines enabled unsupervised clustering which followed relatively well the expected ratios between S and F types in WF populations, as were previously determined on subsets of those seed batches by genetic marker analysis. These data were be used as initial ground truth data to create initial “noisy” Al classifiers (ANN). The initial ground truth data specifically were obtained without genotyping and only based on the unsupervised clustering data obtained by the HCA and Kmeans clustering. Thus, the first ANN used statistical ground truth data only. The first ANN was based on normalized near-infrared spectra using the mean centering-standard normalization as input. The training was done using 1143 of 1910 individual spectra. The spectral data consisted of one or two technical replicate measurements for each seed. The model was first trained using 80% of the spectra with 5% of the training set used in cross-validation after each iteration and 20% testing. The first ANN used the classification model directly on the entire near-infrared spectrum. The results of this first trained ANN are shown in Table 1

[0206] Table 1: Results of first trained ANN classifier

[0207] The machine learning approach based on general hierarchical cluster analysis and partial knowledge of pure genotype sample sets (100% BF and 100% WS) was sufficient to achieve an overall 70 % and 62 % predictive accuracy (PPV) for sterile and fertile genotypes (regardless of blue versus white) when validating against genotyping data (as provided later). The Al classifier showed a good association with S / F ratios for all of the used seed lots. The prediction accuracy was 85 -88 %. The N IRS-based classifier showed the capacity for seed separation into sterile and fertile seeds. Three spectral regions were identified: 4,000 cm-1to 7,000 cm-1; 9,000 cm-1to 12,000 cm’1; 8260 cm’1to 8430 cm’1. Further, a comparison of genotype predictions with genotyping data was made. The results of the initial machine learning model were compared to genotyping data. Genotyping was conducted using validated DNA markers to detect the presence / absence of the gene restoring male sterility, and the blue aleurone color locus. Specifically, after validation, the near-infrared spectra were retrained with the genotyping data and reference information of control groups white- sterile (WS) or blue-fertile (BF). Models were trained with different settings: with or without prior PCA reduction of parameters; single, bi-, multilayered neural networks and random tree machine learning, 5 % cross- validation, 20 % as an independent test set; trained with 2 or 4 classification groups: sterile (S) / fertile (F) or white-sterile (WS), white- fertile (WF), blue-sterile (BS) and blue-fertile (BF); retrained models were evaluated according to their prediction capabilities of the genotyping data.

[0208] In this first example, a re-teaching of the first trained ANN model was performed using the genotyping data. The model training was repeated using all 1143 individual spectra with 5% cross- validation in this second training attempt. The second trained ANN used a classification with two classification groups as shown in Table 2. Further, the second ANN was based on normalized near-infrared spectra using the minimum-maximum normalization of the near-infrared spectra as input. Specifically, in the near-infrared spectra, the 1%-quantile of absorbance was set to 0 and the 99%-quantile of absorbance was set to 1 . The second ANN transformed the near-infrared spectrum according to the eigenvectors of the PCA of the training set. This reduces the dimensionality of the data which speeds up the classification by the neural network (fewer variables to check, here: 39 instead of 4500). The second ANN used predominantly the genotyping results after initial model training as ground truth data. The results are shown in Table 2.

[0209] Table 2: Results for re-teaching with ground truth data

[0210] After re-teaching with genotyping information, the ANN classifier reached a predictive accuracy of 94% (PPV) across both categories. Thus, the re-teaching of the ANN classifier further improved accuracy of the classifier.

[0211] Example 2:

[0212] Example 2 essentially corresponds to the second ANN of Example 1. However, in this example, a classification with four classification groups is shown. The results of the re-teaching are shown in Table 3.

[0213] Table 3: Results of the re-teaching with four classification groups

[0214] The True Positive Ratio (TPR) of the prediction of the different classes was as follows for the calibrated ANN classifier: white fertile (WF): 91 %; white sterile (WS): 95%; blue fertile (BF): 99%; blue sterile (BS): 67%. It should be noted that the BS group was very small and hence under-represented, which can explain the relatively lower prediction accuracy. Thus, the additional refinement shows the potential of distinguishing both blue / white and sterile / fertile combinations. It should be noted that the group of BS was underrepresented in this example. Further improvement can be expected from a higher representation of the BS class. A lower seed quality seen in some batches was not found to affect the model significantly (drop in accuracy by <5%).

[0215] List of reference numbers sorting device cereal seeds sterile cereal seeds fertile cereal seeds seed feeder seed stream sorting station feed hopper chute vibratory feeder spectrometer device light source illumination light detector detection light ejector pneumatic ejector target chute sort-out chute controller supplying a seed stream taking at least one near-infrared spectrum automatically identifying seeds to be sorted out automatically ejecting seeds absorbance wavenumber variance variance of the first and second derivative seed lots genotype prediction

[0216] HCA I Kmeans clustering ANN reference ratio

Claims

Claims1 . A method of sorting cereal seeds (112), wherein the cereal seeds (112) contain sterile cereal seeds (114) and fertile cereal seeds (116), the method comprising: i. supplying a seed stream (120) to a sorting station (122), the sorting station (122) comprising at least one spectrometer device (130) for determining spectral information of seeds in the seed stream (120); ii. taking, with the spectrometer device (130), at least one near-infrared spectrum of a seed (112) in the seed stream (120); and iii. automatically identifying seeds to be sorted out from the seed stream (120) by using at least one classifier, the classifier being configured for classifying the seed to be a sterile cereal seed (114) or a fertile cereal seed (116) based on the near-infrared spectrum taken in step ii..

2. The method according to the preceding claim, wherein the classifier is configured for determining a presence and / or an absence of at least a part of an addition chromosome in the cereal seeds.

3. The method according to any one of the preceding claims, wherein the sterile seed (116) comprises, specifically is, a male sterile seed.

4. The method according to any one of the preceding claims, wherein the classifier is further configured, based on the near-infrared spectrum, for classifying the seed (112) into at least one category selected from the group consisting of: a white-sterile cereal seed; a white-fertile cereal seed; a blue-sterile cereal seed; a blue-fertile cereal seed.

5. The method according to any one of the preceding claims, wherein the seed stream (120) comprises, specifically consist of, white-sterile cereal seeds and white-fertile cereal seeds, wherein the white-sterile cereal seeds and white-fertile cereal seeds in the seed stream (120) are cereal seeds previously sorted by a color seed sorting method, specifically a camera-based color seed sorting method, the color seed sorting method separating the white-sterile cereal seeds and white-fertile cereal seeds from an initial seeds stream comprising white-sterile cereal seeds, white-fertile cereal seeds, blue-sterile cereal seeds and / or blue-fertile cereal seeds.

6. The method according to any one of the preceding claims, further comprising determining a degree of mis-division of the seed stream (120), wherein the degree of mis-division quantifies a fraction of fertile cereal seeds (116) in the seed stream (120).

7. The method according to any one of the preceding claims, wherein the near-infrared spectrum is obtained in reflection geometry, specifically with reflective background surfaces.

8. The method according to any one of the preceding claims, wherein the classifier is configured for classifying the seed (112) based on spectral features from at least one spectral region selected from the group consisting of: 4,000 cm-1to 7,000 cm-1; 9,000 cm-1to 12,000 cm’1; 8260 cm’1to 8430 cm’1.

9. The method according to any one of the preceding claims wherein the classifier comprises, specifically is, a trained artificial neural network (ANN), wherein the method further comprises at least one training step, wherein, in the training step, the ANN is trained using labeled near-infrared spectra, wherein the labeled near-infrared spectra comprise genotyping data, wherein the genotyping data provide information on sterility and / or fertility of seeds under test.

10. The method according to any one of the preceding claims, further comprising: iv. automatically ejecting seeds identified to be sorted out from the seed stream (120).

11. The method according to the preceding claim, further comprising at least one additional sorting step, wherein the additional sorting step comprises a visible light color sorting comprising, by using at least one camera, taking at least one image of the seeds (112) in the seed stream (120) and identifying in the image seeds to be sorted out from the seed stream (120) based on a presence and / or an absence of a color in the image of the seeds (112) in the seeds stream (120).

12. A sorting device (110) for sorting cereal seeds (112), comprising: a. at least one seed feeder (118) for supplying a seed stream (120) to at least one sorting station (122); and b. the at least one sorting station (122), the sorting station (122) comprising at least one spectrometer device (130) for determining spectral information of seeds (112) in the seed stream (120), wherein the sorting station (122) further comprises at least one ejector (140) for ejecting seeds (112) from the seed stream (120),wherein the sorting device (130) is configured for performing the method according to any one of the preceding claims.

13. A computer program comprising instructions which, when the program is executed by the sorting device (130) according to any one of the preceding claims referring to a sorting device, cause the sorting device (130) to perform the method according to any one of the preceding claims referring to a method.

14. A computer-readable storage medium, specifically a non-transient computer-readable medium, comprising instructions which, when the instructions are executed by the sorting device (130) according to any one of the preceding claims referring to a sorting device, cause the sorting device (130) to perform the method according to any one of the preceding claims referring to a method.

15. A use of the sorting device (130) according to any one of the preceding claims referring to a sorting device (130) for sorting of cereal seeds (112) of a seed stream (120) containing sterile cereal seeds (114) and fertile cereal seeds (116), the sterile cereal seeds (114) comprising male sterile seeds involving at least one addition chromosome.

Images