Method and apparatus for determining whether an oilseed, a nut, in particular a hazelnut, or a seed is rotten

Abstract

The invention relates to a method for determining whether an oilseed, a nut, in particular a hazelnut (2), or a seed is rancid, which consists of irradiating a sample, detecting an absorption or reflection spectrum, determining the content of at least one fatty acid breakdown product in the sample, and establishing a degree of rancidity using the content of said product. The method also consists of irradiating an oilseed, a nut, in particular a hazelnut (2), or a seed, projecting reflected and / or transmitted light onto a detection photosensor (4'), detecting absorption or reflection spectra in a wavelength range of 300 to 2500 nm using said photosensor (4'), which has several pixels, and assigning a degree of rancidity to each pixel of the photosensor (4').

Description

Method and apparatus for determining whether an oilseed, a nut, in particular a hazelnut, or a seed is rotten

[0001] The invention relates to a method for determining whether an oilseed, a nut, in particular a hazelnut, or a seed is rotten, according to claim 1.

[0002] The invention further relates to an apparatus for determining whether an oilseed, a nut, in particular a hazelnut, or a seed is rotten, according to claim 12. Technological background

[0003] The detection and subsequent classification of bulk materials using photosensors is a common method applied on an industrial scale. In such known prior art methods, seeds are individually examined spectroscopically by irradiating them with a light source. A photosensor then records an absorption or reflection spectrum. Subsequently, a computer unit analyzes the absorption or reflection spectrum of each seed in a region of interest and calculates, based on a calibration curve, the content of a specific component in the seed.

[0004] The detection of different components in individual elements of a bulk material is of interest, for example, in the food industry, to distinguish between deteriorated and undeteriorated elements of the bulk material. According to the prior art, such methods generally operate in the near-infrared region. To enable the use of these methods in production facilities, the photosensors employed must have high image acquisition frequencies, typically 500 Hz or higher. This ensures high processing throughput while also providing reliable analysis of the components of each individual element examined. Conventionally, the data recorded by the photosensors are analyzed using standard statistical classification methods, such as Partial Least Squares, Principal Component Regression, or similar methods.This qualitative analysis yields excellent results when clear differences exist between deteriorated and undeteriorated elements in the absorption or reflection spectrum. Hazelnuts are processed into numerous products, including whole hazelnuts. Hazelnuts that appear to be of impeccable quality on the outside have often already begun to rot internally. Consequently, they cannot be distinguished by conventional optical methods in the visible region. The causes are varied and related to improper storage, insect infestations, or mold growth. Frequently, the rotting process is also associated with the formation of dangerous mold (toxins, for example, aflatoxin).Mold originates from a filamentous fungus: filamentous fungi constitute a systematically heterogeneous group of fungal growths, mostly belonging to the taxonomic groups Ascomycetes and Zygomycetes. Filamentous fungi such as Aspergillus flavus are found in soil, decaying vegetation, and on hay and grains exposed to microbial spoilage. A filamentous fungus infects the fruit from the outside and establishes itself on the surface.

[0005] In ecology and thanatology, putrefaction is defined as the decomposition of biotic substances by microorganisms under oxygen-deficient conditions. The term putrefaction is often used specifically to describe decomposition accompanied by the formation of unpleasant odors. Putrefaction is a natural form of fermentation and is referred to here as the putrefied state. Therefore, the development of putrefaction differs substantially from the development of mold. In this case, in addition to taste alterations, detrimental health effects can also occur. In any case, rotten hazelnuts, whether raw or roasted, must be removed from the marketable product. However, since this is not a simple yes-or-no decision, but rather requires the application of a graduated quality criterion, a precise determination and evaluation of the putrefied state is necessary.

[0006] From the prior art, document CN 113420614 A is known. This document discloses a method for identifying moldy peanuts, in particular a method for identifying moldy peanuts based on a near-infrared hyperspectral image using a deep learning algorithm with a neural network. The described neural network comprises 265 neurons at the input and two neurons at the output of the neural network. Reference is also made to a characteristic wavelength of 1343 nm used to perform preliminary segmentation. In that procedure, the mold product aflatoxin is detected.A map of the distribution of mold information of the peanuts is generated, in which, by means of the number of moldy pixels detected and a threshold value (β), each peanut particle is incorporated into the map of the distribution of mold information to identify a moldy peanut and to generate an image of the result of the identification of moldy peanuts.

[0007] One drawback of this known solution is that mold formation occurs on the surface of the peanut, and therefore only information from the peanut surface is evaluated. The apparently necessary deep learning algorithm described in detail may, due to its parameterization, possibly achieve the desired result in detecting mold on peanuts, but it is inadequate for the early detection of rot or the rotten state.

[0008] Prior art patents AT 519918 A1 and WO 2018 / 191768 A1 are known. These documents disclose a method for detecting rancidity in oilseeds, seeds, and nuts. Rancidity is the state in which fats and other lipids decompose through oxidation or the action of lipolytic enzymes (lipases). The deterioration of vegetable and animal fats, which can be perceived at an initial stage by a change in odor and taste (rancidity), is due, in the case of fats with water content, to hydrolysis and the consequent breakdown of longer-chain fats, and to the action of atmospheric oxygen (oxidation). The patent application comprises irradiating a sample of an oilseed, nut, or seed with a light source and recording the absorption or reflection spectrum of the reflected or transmitted light.Subsequently, the absorption or reflection spectrum is compared with an external chemical analysis of the sample to establish a relationship between the absorption or reflection spectrum and the rancidity of the sample.

[0009] A disadvantage of known prior art methods is that they can only determine the rancidity of the oilseed or nut, and that as soon as a marker classifying a product as defective is recognized in the absorption or reflection spectrum, the product is immediately discarded. However, when the differences between the spectrum of a good product and that of a defective product are very small, this small difference is amplified in the external process to such an extent that the signal-to-noise ratio increases considerably, and the uncertainty in the decision between a good and a defective product also increases considerably. Furthermore, the differences in the spectrum are determined using a reference sample of good and defective specimens.However, since these are natural products with spectral dispersion, this comparison is associated with high uncertainty and can vary from batch to batch. Furthermore, contamination or product defects are often unevenly distributed. The result of misclassification is a poor separation between good and spoiled food.

[0010] From the state of the art, CN104544334 A is known. This document discloses the performance of a color classification using a color classifier to separate mealy and rotten nuts.

[0011] In summary, conventional determination methods used in the food industry during the processing of oilseeds, nuts (particularly hazelnuts), or seeds result in a large amount of waste material, although at least some of the separated products could still be reused or further processed. At the same time, excessive separation also leads to economic disadvantages. Description of the invention

[0012] An objective of the invention is to avoid at least one of the disadvantages of the prior art. In particular, one objective is achieved by providing a method and apparatus for determining whether an oilseed, a nut, particularly a hazelnut, or a seed is rotten, in order to reduce waste material in the determination procedure compared to known methods. This is especially critical when the products contain substances that may cause harm to the consumer's health. In any case, when the quality of the product is affected, an alteration of the flavor is to be expected. This occurs, in particular, in the processing of hazelnuts.

[0013] This objective is achieved through the features of the independent claims. Advantageous embodiments are set forth in the figures and in the dependent claims.

[0014] The method according to the invention for determining whether an oilseed, a nut, in particular a hazelnut, or a seed is rotten comprises at least the following calibration steps: a) irradiate a sample of an oilseed, a nut, in particular a hazelnut, or a seed with a calibration light source; b) record an absorption or reflection spectrum in a wavelength range of 300 to 2500 nm using a calibration photosensor; c) assign at least one characteristic wavelength or at least one characteristic wavelength range from the recorded absorption or reflection spectrum of the sample to a degree of putrefaction; d) repeat the above steps for a representative number of samples and establish a correlation between the degree of putrefaction and the assigned wavelengths or wavelength ranges from the recorded absorption or reflection spectra.

[0015] Through calibration steps, a correlation is established between a value of the absorption or reflection spectrum at a characteristic wavelength, or at least at a characteristic wavelength range, and the degree of putrefaction of the sample. A correlation describes a relationship between two or more characteristics, states, or functions, preferably a causal correlation being applicable in this case.

[0016] In this way, it is possible to determine the degree of putrefaction of the oilseed fruit, of a nut, in particular a hazelnut, or of a seed only by evaluating the absorption or reflection spectrum, without the need for additional independent analysis, such as a laboratory analysis.

[0017] Preferably, the calibration steps are carried out in the order mentioned above to allow reproducible calibration, particularly for hazelnuts.

[0018] Furthermore, the method according to the invention comprises the following detection steps: e) irradiating an oilseed, a nut, in particular a hazelnut, or a seed with a detection light source; f) recording absorption or reflection spectra in a wavelength range of 300 to 2500 nm by means of a detection photosensor, the detection photosensor having a plurality of pixels and each pixel recording an absorption or reflection spectrum; g) assign a degree of putrefaction to each pixel of the detection photosensor by applying the correlation according to step d) to the absorption or reflection spectra recorded by the pixels of the detection photosensor; h) classify the oilseed fruit, the nut, in particular the hazelnut, or the seed as rotten when at least a certain number of pixels present a degree of rot that exceeds a first threshold value and / or when the degree of rot assigned to at least one pixel exceeds a second threshold value.

[0019] The detection stages allow a reproducible evaluation of the recorded absorption or reflection spectra based on the calibration stages previously performed.

[0020] Preferably, the detection steps are carried out in the above-mentioned order to allow reproducible detection of the state of putrefaction, particularly in hazelnuts.

[0021] According to the invention, absorption or reflection spectra are recorded for each individual pixel of the detection photosensor, and each of these spectra is assigned a degree of rot using the correlation established during the calibration steps. This provides the advantage that a single absorption or reflection spectrum is not used for each irradiated oilseed, nut (particularly hazelnut), or seed to determine whether the corresponding oilseed, nut (particularly hazelnut), or seed is rotten. An additional criterion can thus be introduced: a predetermined number of pixels must exhibit a degree of rot exceeding a first threshold value. The detection photosensor has a plurality of pixels and records an absorption or reflection spectrum for each pixel. Therefore, the first threshold value is verified for each pixel of the detection photosensor.

[0022] The measured values ​​or pixel spectra of the detection photosensor are defined as a degree of putrefaction, for example, on a scale from 0 to 100%. Subsequently, classification can be carried out, for example, in a sorting facility, based on the first threshold value or the second threshold value.

[0023] Alternatively or additionally, the corresponding oilseed, nut, or seed may also be separated when the degree of decay assigned to at least one pixel exceeds a second threshold value. By defining the first threshold value and the number of pixels whose assigned degree of decay must exceed the first threshold value, as well as the second threshold value, the method according to the invention allows for the consideration of inhomogeneities in the decay process of the natural agricultural products examined by the method according to the invention. In this way, a separation between good and defective products can be carried out considerably more precisely than was previously possible using prior art methods.

[0024] The method according to the invention allows the determination of the state of putrefaction of the oilseed fruit, nut (particularly hazelnuts), or seed by means of a spatially resolved determination of the degree of putrefaction. Furthermore, by means of the first and second threshold values, an inhomogeneous distribution of fatty acid degradation products in the oilseed fruit (particularly hazelnuts) or seed can be incorporated into the assessment of the state of putrefaction. In this way, a better differentiation can be achieved between, for example, a non-rotten hazelnut and a rotten hazelnut.

[0025] Preferably, the sample of the oilseed, nut (particularly hazelnut), or seed is moved in step a) and / or step e) by passing in front of the corresponding light source. This makes the method described above applicable to moving oilseeds, simplifying the classification of rotten and good oilseeds during the process. Large mass flows of oilseeds, for example, 6 t / h, can be classified very quickly using the method described above.

[0026] Preferably, before step b), the light reflected and / or transmitted by the sample is projected onto a calibration photosensor. This allows for an improved recording of the spectra.

[0027] Preferably, in step b) a wavelength range of 900 to 1700 nm is used. This allows for more accurate measurement of the infrared region of the light and the associated information for calibrating the degree of decay.

[0028] Preferably, after step b), a degree of putrefaction is established based on the content of at least one fatty acid degradation product in the sample. This is done by searching for fatty acid degradation products in the spectral region used to provide reproducible detection of rotten oilseeds.

[0029] In particular, after step b), a degree of putrefaction is established based on the content of at least one fatty acid degradation product in the sample and, additionally, on the content of acetic acid in the sample. In this case, the content of the at least one fatty acid degradation product and the acetic acid content are mathematically combined in the evaluation, for example, by calculating a mean, a median, or a sum. Alternatively or additionally, the contents may be weighted. This increases the accuracy of the classification in the method according to the invention.

[0030] Preferably, the at least one fatty acid degradation product comprises at least one component from the group consisting of butyrolactone, diacetyl, 2-methylbutanal, 3-methylbutanal, acetylacetone, filbertone, or 2,3-butanediol. A model may be provided that is not limited to a single chemical substance, but rather to those components that exhibit a significant concentration difference between good and rotten oilseeds, nuts, particularly hazelnuts, or seeds. This means that, in the relevant spectral region, the quantitative degree of flavor-altering substances can be automatically deduced from the measured values ​​of various components or substances associated with the state of putrefaction, based on the relative or absolute amplitudes of the spectrum.For example, the values ​​can be normalized and then a derivative applied to the spectra and a comparison made.

[0031] Preferably, the degree of putrefaction is established based on the contents of at least two fatty acid degradation products in the sample. For example, the two fatty acid degradation products that exhibit the greatest concentration difference between a rotten and a good oilseed are used. The contents are combined mathematically in the evaluation, for example, by calculating a mean, a median, or a sum. Alternatively or additionally, the contents may be weighted. By using at least two characteristic components of the putrefaction process to establish the degree of putrefaction, the accuracy of the method according to the invention is further increased.

[0032] Preferably, several components of fatty acid degradation products are sought that exhibit the greatest concentration difference between good and rotten oilseeds, nuts (particularly hazelnuts), or seeds. In this case, the model is reduced, for example, to five components that exhibit the greatest concentration difference between the spectra of good and rotten oilseeds, nuts (particularly hazelnuts), or seeds. This means that, in the relevant spectral region, the quantitative degree of flavor-altering substances can be automatically deduced more accurately from the measured values ​​of various substances and the absolute amplitudes of the spectrum.

[0033] Preferably, the correlation is at least a correlation function, or at least a table of indexes, or a table of values, or comprises at least a comparison of spectral information that includes at least the degree of decay and the corresponding wavelengths or wavelength ranges of the recorded absorption or reflection spectra. A correlation may comprise a correlation function, a table of indexes, a table of values, or the like, so that a causal relationship can be established between the degree of decay and the corresponding wavelengths or wavelength ranges of the recorded absorption or reflection spectra.A correlation function allows for a continuous relationship between the degree of rot and the corresponding wavelengths or wavelength ranges of the recorded absorption or reflection spectra, and thus a complete and accurate causal relationship. Although index tables or tables of values ​​do not establish a continuous causal relationship between the degree of rot and the corresponding wavelengths or wavelength ranges of the recorded absorption or reflection spectra, such a relationship is sufficiently adequate for some oilseeds to reproduceably distinguish between rotten and good oilseeds.

[0034] Preferably, prior to step e), the light reflected and / or transmitted by the oilseed, nut (particularly hazelnut), or seed is projected onto a detection photosensor. This allows for an improved assignment of the measured values ​​to the degree of rot.

[0035] Preferably, in step e) a wavelength range of 900 to 1700 nm is used. This allows for more accurate measurement of the infrared region of the light and the associated information for detecting the degree of putrefaction.

[0036] The assignment of at least one characteristic wavelength or at least one characteristic wavelength range of the recorded absorption or reflection spectrum of the sample to the degree of putrefaction is carried out, according to a preferred embodiment of the method according to the invention, by means of at least one average or median, bandwidth, or individual frequency bands of the recorded absorption or reflection spectra. This provides the advantage of establishing a precise correspondence between the recorded absorption or reflection spectra and the degree of putrefaction.

[0037] Preferably, correlation is used in a mode of operation characterized by detection stages to assign to the recorded spectra a corresponding measured value or reference value for the degree of putrefaction.

[0038] Preferably, the determination of the content of at least one fatty acid degradation product in the sample comprises a measurement in an analytical laboratory, in particular gas chromatography, followed by analysis by mass spectrometry. In this context, good separation between good and defective product is important in the laboratory analysis of the chemicals, which can be clearly seen in a scatter plot of the chemicals used with respect to the analyzed samples, for example, hazelnut samples. Such separation is a necessary, though not sufficient, requirement for a good quantitative model. The possibility of quantitatively using a substance based on the NIR spectrum depends on the position of other additional chemicals that may overlap the absorption band.

[0039] Preferably, each pixel is assigned a reference value based on calibration through the correlation established between the spectral data and the analytical laboratory. The correlation obtained, i.e., the spectral information that can be related to the information in the reference dataset, is subsequently used in an operating mode characterized by detection stages to assign to the recorded spectra a corresponding measured value for the degree of putrefaction.

[0040] According to a preferred embodiment of the method according to the invention, a common sensor is used as both a calibration photosensor and a detection photosensor. This provides the advantage that only one sensor is required for both the calibration and detection steps. Additionally, a single light source can be used as both the calibration and detection light sources. This reduces the costs of implementing the method according to the invention.

[0041] Preferably, the absorption or reflection spectra are recorded by the calibration photosensor and / or the detection photosensor using hyperspectral acquisition. This allows for the recording of a particularly wide wavelength range. Hyperspectral cameras are preferably used for this purpose. In particular, the absorption or reflection spectra are recorded by the calibration photosensor and / or the detection photosensor using a color camera. This allows for the additional recording of a range from 380 to 780 nm. For example, discolored oilseeds can be detected and easily separated using the color camera. In addition to rotten oilseeds, discolored oilseeds can also be separated, thus improving the quality of the sorting. Discoloration can indicate other undesirable reductions in the quality of the oilseeds.It is particularly preferable that the information measured by the color camera be evaluated using the artificial intelligence module described below. This can raise the desired detection rate to 99%.

[0042] The determination of the content of at least one fatty acid degradation product in the sample, according to the preferred embodiment of the method according to the invention, comprises performing gas chromatography and subsequent analysis by mass spectrometry. In this way, a particularly precise determination of the fatty acid degradation product content can be achieved.

[0043] The apparatus according to the invention for determining whether an oilseed, a nut, in particular a hazelnut, or a seed is rotten comprises a calibration light source, a detection light source, a calibration photosensor, and a detection photosensor, and is configured to carry out at least one of the methods according to the invention disclosed herein.

[0044] Preferably, the apparatus comprises a sorting unit, the sorting unit being configured to separate an oilseed, a nut, in particular a hazelnut, or a seed from a product stream when the oilseed, nut, in particular a hazelnut, or seed is classified as rotten. This provides the advantage that the method according to the invention can be applied on a large industrial scale.

[0045] A preferred embodiment of the above-mentioned method for determining whether an oilseed, nut, in particular a hazelnut, or seed is rotten comprises the calibration steps: - irradiating a sample of an oilseed, nut, in particular a hazelnut, or seed with a calibration light source; - project the light reflected and / or transmitted by the sample onto a calibration photosensor (4); - record an absorption or reflection spectrum in a wavelength range of 300 to 2500 nm, preferably 900 to 1700 nm, using the calibration photosensor; - determine the content of at least one fatty acid degradation product in the sample; - establish a degree of putrefaction based on the content of at least one fatty acid degradation product in the sample; - assigning at least one characteristic wavelength or at least one characteristic wavelength range from the recorded absorption or reflection spectrum of the sample to the degree of putrefaction; - repeating the above steps for a representative number of samples and establishing a correlation function between the degree of putrefaction and the assigned wavelengths or wavelength ranges from the recorded absorption or reflection spectra, characterized in that the method comprises the detection steps: - irradiate an oilseed, a nut, in particular a hazelnut, or a seed with a detection light source; - projecting the light reflected and / or transmitted by the oilseed fruit, the nut, in particular the hazelnut, or the seed onto a detection photosensor; - recording absorption or reflection spectra in a wavelength range of 300 to 2500 nm, preferably 900 to 1700 nm, by means of the detection photosensor, the detection photosensor having a plurality of pixels and each pixel recording an absorption or reflection spectrum; - assign a degree of putrefaction to each pixel of the detection photosensor by applying the correlation function to the absorption or reflection spectra recorded by the pixels of the detection photosensor; - classify the oilseed fruit, the nut, in particular the hazelnut, or the seed as rotten when at least a certain number of pixels present a degree of rot that exceeds a first threshold value and / or when the degree of rot assigned to at least one pixel exceeds a second threshold value.

[0046] For this purpose, a common sensor can be used as a calibration photosensor and as a detection photosensor.

[0047] A single light source can also be used as both a calibration light source and a detection light source.

[0048] In particular, the degree of putrefaction is established on the basis of the contents of at least two fatty acid degradation products in the sample.

[0049] In particular, the assignment of at least one characteristic wavelength or at least one characteristic wavelength range of the recorded absorption or reflection spectrum of the sample to the degree of putrefaction is made by at least one of an average, a bandwidth, or individual frequency bands of the recorded absorption or reflection spectra.

[0050] Preferably, the recording of absorption or reflection spectra by the calibration photosensor and / or the detection photosensor is carried out by hyperspectral acquisition.

[0051] In particular, the determination of the content of at least one fatty acid degradation product in the sample comprises performing gas chromatography and subsequent analysis by mass spectrometry.

[0052] A preferred embodiment of the apparatus for determining whether an oilseed, a nut, in particular a hazelnut, or a seed is rotten comprises a calibration light source, a detection light source, a calibration photosensor and a detection photosensor, and is configured to carry out one of the methods described above.

[0053] Preferably, the apparatus comprises a sorting unit, the sorting unit being configured to separate an oilseed, a nut, in particular a hazelnut, or a seed from a stream of products when the oilseed, nut, in particular a hazelnut, or seed is sorted as rotten.

[0054] Preferably, the apparatus comprises a control device. The control device can control at least one component of the group consisting of the calibration photosensor, the detection photosensor, the calibration light source, and the detection light source, so that at least one of the methods described above can be executed automatically and reproducibly. In particular, the control device causes an output device to emit at least one measured value. The output device is provided, for example, as a display showing at least one piece of information indicative of the state of rottenness of the oilseed. For example, a distribution function or statistical data relating to the rotten oilseeds present in the product stream or mass flow is displayed. In this way, the process can be monitored more effectively.

[0055] In particular, the control device controls the sorting unit. To this end, the control device has control data that governs the sorting unit. For example, the sorting unit may include a pneumatic diverter unit connected to the detection photosensor. The pneumatic diverter unit, using the control data, separates oilseeds, nuts (particularly hazelnuts), or seeds classified as rotten from an essentially continuous product flow. In this context, for example, the intensity and, in particular, the direction of the compressed air in the pneumatic diverter unit can be controlled.

[0056] Preferably, the classification unit operates with at least two adjustable threshold values, a first adjustable threshold value comprising an amount of substance for a pixel and a second adjustable threshold value comprising an inhomogeneous distribution of the substance, taking into account the second adjustable threshold value, for classification, the number of pixels with the attribute: rotten.

[0057] Preferably, the apparatus comprises a calculation unit that evaluates spectral data to perform a classification of oilseeds. The spectral data may be measurement data and may be used to generate control data for the control device, the control data being subsequently used, in particular, to control the classification unit.

[0058] In particular, the calculation unit is configured to calculate, on the basis of measurement data or spectral data, at least one of the two threshold values ​​and generate control data for the control device from said calculation.

[0059] Preferably, an AI (artificial intelligence) module connected to the computing unit is present. The aforementioned solution requires specialized knowledge of the absorption bands of oilseeds and fatty acid degradation products in the infrared region to select the correct spectral region and thus perform optimal calibration to distinguish between good products and rotten oilseeds. Such specialized knowledge of a physicochemical nature can be dispensed with using deep learning algorithms with the aid of neural networks. Historical measurement data or spectra of predominantly rotten and predominantly good samples can be provided to the neural network as training data. These neural networks independently find the optimal region of interest in the measurement data or spectra.This further improves calibration using the recorded spectral data, particularly the hyperspectral data. "Poor" training data are automatically weighted less heavily than more representative data. Poor training data are those in which the differentiation between the good and defective regions within an oilseed is less pronounced, and therefore a representative training model cannot be generated. This method can further reduce the parameter known as "rejection," erroneously defined as waste material, by a factor of 5 to 10. Instead of approximately 5% erroneous rejections in the sorting unit, this percentage can be reduced to less than 1%. This allows the model, trained in this way, to predict whether the waste material will fall within the desired range.Therefore, the AI ​​module can provide essential parameters for the evaluation of rotten oilseeds.

[0060] The artificial intelligence used in this document does not require prior segmentation of the spectra, nor are specific wavelengths supplied as preferred wavelengths to the neural network. Thus, the AI ​​can learn without any prior information. This offers the advantage that this method allows for a more flexible response to different putrefaction products or fatty acid degradation products. The trained AI autonomously selects the most appropriate range of values ​​to achieve an optimal selection result during the evaluation.

[0061] In general terms, the AI ​​module serves to build a statistical model based on training data and validated using test data, which is then applied to the data of the operational product flow. Among the usable algorithms are, in particular, supervised machine learning algorithms, in which a model is trained on a set of training data and subsequently applied to additional evaluation data to calculate a classification. Among the approaches used to train such models is deep learning (artificial neural networks), in which several layers of artificial neurons link the input variables (feature vector) with the output variable (classification, regression, etc.).In addition to numerous other machine learning methods, Random Forest (randomized decision trees) or Support Vector Machines (estimation using support vectors in the vector space of feature vectors) algorithms can also be used, particularly to limit computational effort.

[0062] A computer program product according to the invention comprises program instructions configured to execute at least one of the aforementioned methods. The computer program product may comprise instruction data, calculated data, and parameters, as described above.

[0063] A computer-readable medium according to the invention comprises at least one computer program product that, when executed by at least one computing unit, causes the computing unit to perform at least one of the methods mentioned above. The computer-readable medium may comprise control data, measurement data, calculated data, measured values, and parameters, as described above.

[0064] The method according to the invention and the apparatus according to the invention, as well as preferred and alternative embodiments, are explained in further detail below with reference to the figures.

[0065] Other advantages, features and details of the invention are derived from the following description, which describes embodiments of the invention with reference to the drawings.

[0066] The list of reference symbols, as well as the technical content of the patent claims and figures, form part of the disclosure. The figures are described jointly and in relation to each other. The same reference symbols designate the same components, and reference symbols with different indices designate functionally equivalent or similar components.

[0067] The invention is explained in further detail by the following figures using exemplary embodiments. The list of reference symbols forms part of the disclosure.

[0068] Position indications such as "up", "down", "right" or "left", refer in each case to the corresponding representations and should not be interpreted as limiting.

[0069] Although the invention is represented and described in detail by the figures and the corresponding description, such representation and detailed description are to be understood as illustrative and exemplary and not as limiting the invention. It is understood that those skilled in the art may make modifications and variations without departing from the scope of the following claims. In particular, the invention also comprises embodiments with any combination of features mentioned or shown above in relation to different aspects and / or embodiments.

[0070] The invention also comprises individual features shown in the figures, even if such features are shown therein in relation to other features and / or have not been previously mentioned. Furthermore, the expression "comprising" and its derivatives do not exclude other elements or steps. Likewise, the indefinite article "a" and its derivatives do not exclude a plurality. The functions of several features indicated in the claims may be performed by a single unit. The terms "essentially," "approximately," "about," and the like, in relation to a feature or a value, also define precisely the feature or the value. None of the reference signs included in the claims shall be construed as limiting the scope of the claims.Terms such as "first" or "second" serve only to distinguish subsequent nouns and in no case define an order or evaluation of subsequent nouns. Description of the figures

[0071] The figures are described jointly and in a correlated manner. The same reference signs designate the same components. Figure 1 shows an apparatus for determining whether an oilseed, a nut, in particular a hazelnut, or a seed is rotten; Figure 2 shows an alternative embodiment of the apparatus according to the invention; Figure 3 shows another alternative embodiment of the apparatus according to the invention; and Figure 4 shows a standardized concentration of fatty acid degradation products in rotten hazelnuts and those fatty acid degradation products that exhibit the greatest differences between a rotten hazelnut (Q) and a non-rotten hazelnut (FF). Realization of the invention

[0072] Figure 1 shows, in a preferred embodiment, an apparatus 1 according to the invention for determining whether an oilseed, a nut, in particular a hazelnut 2, or a seed is rotten, the apparatus being configured to execute a method according to the invention. The apparatus 1 according to the invention comprises a calibration light source 3, a detection light source 3, a calibration photosensor 4, and a detection photosensor 4. In the embodiment of the apparatus 1 according to the invention shown in Figure 1, the calibration photosensor 4 and the detection photosensor 4 are implemented as a single photosensor, although they can also be separate photosensors, as shown in Figure 3. The method according to the invention for determining whether an oilseed, a nut, in particular a hazelnut 2, or a seed is rotten comprises a series of calibration steps and a series of detection steps.The calibration steps of the method according to the invention comprise irradiating a sample of the oilseed fruit, the nut, in particular the hazelnut 2, or the seed with the calibration light source 3. Subsequently, the light reflected and / or transmitted by the sample is projected onto the calibration photosensor 4, the calibration photosensor 4 recording an absorption or reflection spectrum in a wavelength range of 300 to 2500 nm, preferably 900 to 1700 nm. Furthermore, the calibration steps comprise determining the content of at least one fatty acid degradation product in the sample, establishing a degree of putrefaction based on the content of the at least one fatty acid degradation product in the sample, and assigning at least one characteristic wavelength or at least one characteristic wavelength range from the recorded absorption or reflection spectrum of the sample to the degree of putrefaction.These preceding steps are repeated in the method according to the invention for a representative number of samples, and a correlation function is established between the degree of putrefaction and the assigned wavelengths or wavelength ranges of the recorded absorption or reflection spectra. The putrefaction process, for example of a hazelnut, is associated with the formation or degradation of fatty acids. The method according to the invention allows, through calibration steps, the linking of the content of at least one fatty acid degradation product in the sample with a degree of putrefaction that enables the evaluation of the quality of the oilseed, the nut, in particular the hazelnut 2, or the seed.Furthermore, this degree of putrefaction is assigned to a characteristic wavelength or at least a characteristic wavelength range of the recorded absorption or reflection spectrum of the sample, allowing the degree of putrefaction to be inferred from the absorption or reflection spectrum. By repeating the process for a representative number of samples, a correlation function is established, for example, thus providing a direct inference of the degree of putrefaction by evaluating the absorption or reflection spectrum using the correlation function. A conventional oilseed, a nut (particularly a hazelnut), or a seed can be used as a sample.The apparatus 1 comprises a control device 7 that controls at least one component of the group consisting of the calibration photosensor, the detection photosensor, the calibration light source, and the detection light source, so that at least one of the methods described above can be executed automatically and reproducibly. The control device 7 controls the sorting unit 5. For this purpose, the control device 7 has control data that controls the sorting unit 5. The apparatus 1 comprises a calculation unit 8 that evaluates spectral data to perform a sorting of the oilseeds. The spectral data can be measurement data and can be used to generate control data for the control device 7.

[0073] Subsequently, this correlation function can be applied within the detection steps of the method according to the invention. The detection steps comprise irradiating an oilseed, a nut, in particular a hazelnut 2, or a seed with the detection light source 3. Both the calibration light source 3 and the detection light source 3 can be the same light source, which is why only a single light source is provided in the apparatus 1 according to the invention shown in Figure 1.

[0074] In an alternative or complementary embodiment, an index table or a table of values ​​may be used, providing a direct inference of the degree of putrefaction by evaluating the absorption or reflection spectrum using the index table or the table of values.

[0075] According to the embodiment of the apparatus 1 according to the invention shown in Figure 2, separate light sources can also be used as a calibration light source 3 and as a detection light source 3. According to the embodiment shown in Figure 3, separate light sources are used as a calibration light source 3 and as a detection light source 3, and separate photosensors are used as a calibration photosensor 4 and as a detection photosensor 4. The detection steps further comprise projecting the light reflected and / or transmitted by the oilseed, the nut, in particular the hazelnut 2, or the seed onto the detection photosensor 4. As explained with regard to the calibration light source 3 and the detection light source 3, a common sensor can also be used as a calibration photosensor 4 and as a detection photosensor 4.This reduces the complexity of apparatus 1 according to the invention. Apparatus 1 according to Figure 2 also comprises a calculation unit and a control device (not shown), as described above.

[0076] According to the embodiment of Device 1 according to the invention shown in Figure 3, different sensors can also be used as the calibration photosensor 4 and the detection photosensor 4. According to this embodiment, both the detection photosensor 4 and the calibration photosensor 4 can operate in reflective mode. This means that the light reflected by the sample is projected onto the calibration photosensor 4 and that the light reflected by the seed 2 is projected onto the detection photosensor 4. As can be seen in Figures 1 and 2, the calibration photosensor 4 and / or the detection photosensor 4 can also operate in transmissive mode, with the light transmitted by the sample being projected onto the calibration photosensor 4 and / or the light transmitted by the seed 2 being projected onto the detection photosensor 4.Within the scope of the invention, by oilseed fruit, nut, in particular hazelnut 2, or seed, we understand both a roasted oilseed fruit, a nut, in particular a hazelnut, or a seed and an unroasted oilseed fruit, a nut, in particular a hazelnut, or a seed.

[0077] As part of the detection steps, absorption or reflection spectra are also acquired in a wavelength range of 300-2500 nm, preferably 900-1700 nm, using the detection photosensor 4. The detection photosensor 4 of the device 1 according to the invention for carrying out the process according to the invention has a plurality of pixels and records an absorption or reflection spectrum for each pixel. A degree of decay is then assigned to each pixel of the detection photosensor 4 by applying the correlation function to the absorption or reflection spectra recorded by the pixels of the detection photosensor 4.Furthermore, oilseeds, nuts (particularly hazelnuts 2), or seeds are classified as rotten when at least a certain number of pixels exhibit a degree of rot exceeding a first threshold value and / or when the degree of rot assigned to at least one pixel exceeds a second threshold value. The device 1 according to Figure 3 further comprises a computing unit 8, a control unit 7, and an AI module 9. The AI ​​module 9 may also be present in devices 1 according to Figure 2 or Figure 3. The AI ​​module is connected to or integrated within the computing unit. With the aid of the AI ​​module, the differentiation, and thus a parameter called the erroneously defined rejection rate, can be further reduced by a factor of 5 to 10. Instead of approximately 5% erroneous rejection in the sorting unit 5, this rejection rate can be reduced to less than 1%.

[0078] The method according to the invention allows the degree of putrefaction of the oilseed fruit, the nut, in particular the hazelnut 2, or the seed to be determined by spatially resolving the degree of putrefaction. Furthermore, by means of the first and second threshold values, a non-homogeneous distribution of fatty acid degradation products in the oilseed fruit, the nut, in particular the hazelnut 2, or the seed can be incorporated into the assessment of the degree of putrefaction. In this way, improved differentiation can be achieved between, for example, an unrotten hazelnut 2 and a rotten hazelnut 2.

[0079] It is known that the putrefaction process, for example in hazelnuts, is associated with the formation and / or degradation of fatty acids. The standard reference analysis for determining fatty acids or their volatile degradation products is a chromatographic determination using gas chromatography, also called GC, followed by mass spectrometry. However, this invasive method can only determine a homogenized mixed sample or the content of individual hazelnuts, but it cannot selectively recognize individual hazelnuts or eject them in real time. As a standard non-invasive laboratory method, FTIR analysis should also be mentioned. In this analysis, a spectrum is calculated using Fourier transform based on an interferogram of the surface of the sample or the homogenized sample.However, this method is also unsuitable, compared to the procedure according to the invention, for classifying large material flows, such as 6 t / h. During harvesting, but at the latest during processing, large quantities must be classified in such a way as to exclude health risks and minimize flavor alterations. Even before further processing takes place, the procedure according to the invention allows for the automated detection, at a high processing rate, of oilseeds, nuts (particularly hazelnuts), or seeds that are flavor-altered, toxigenic, or rotten.

[0080] During the hazelnut harvest, but at the latest during the processing stage, large quantities of hazelnuts 2 must be sorted in such a way as to exclude any health risks and to minimize any alteration of flavor. Before the hazelnuts 2 are further processed, for example into chocolates or snacks, rotten, toxigenic, or off-flavor hazelnuts 2 must be removed automatically and at a high processing rate.

[0081] Since flavor-altering substances, such as free fatty acids and their degradation products, are spatially non-homogeneous, for example in a hazelnut 2, in the process according to the invention, homogenized samples are analyzed in the laboratory or measuring laboratory using available laboratory analytical methods, such as gas chromatography or FTIR, and based on a statistically significant number of samples, to determine their chemical content. The determined concentration is then correlated with the corresponding hyperspectral information. By means of the determined correlation between the spectral response, such as wavelength and / or amplitude, and the concentrations of the substances, the detection photosensor 4 can be calibrated. In the case of the rotten hazelnut 2, these alterations are due to free fatty acids.It has also been shown that chemical changes inside, for example, a hazelnut 2, correlate with measurements taken on its surface. The same applies to oilseeds, nuts, or seeds. The method according to the invention is particularly effective for oilseeds, nuts, or seeds where the seed coat is positioned in the optical path.

[0082] The solution according to the invention consists of searching for fatty acid degradation products in the spectral range used. Preferably, the model is not limited to a single chemical substance, but rather to five main components that exhibit the greatest concentration difference between good oilseeds, nuts (particularly hazelnuts 2), or seeds and rotten oilseeds, nuts (particularly hazelnuts 2), or seeds. This means that, in the determining spectral range, the quantitative degree of the flavor-altering substances can be automatically deduced from the measured values ​​of various substances and their absolute amplitudes in the spectrum. According to the invention, the measured value is then defined as the degree of spoilage, for example, in the range of 0-100%.A classification can then be carried out, for example in a classification facility, based on the first threshold value and the second threshold value.

[0083] Preferably, the assignment of at least one characteristic wavelength or at least one characteristic wavelength range from the recorded absorption or reflection spectrum of the sample to the degree of putrefaction is carried out, within the framework of the procedure according to the invention, by means of at least one of the following: an average value, a bandwidth, or individual frequency bands from the recorded absorption or reflection spectra. In this way, wavelengths or wavelength ranges representative of specific fatty acid degradation products in the absorption or reflection spectrum can be used to determine the degree of putrefaction. Furthermore, the acquisition of the absorption or reflection spectra by the calibration photosensor 4 and / or the detection photosensor 4 is preferably carried out by hyperspectral acquisition.Accordingly, the calibration photosensor 4 and / or the detection photosensor 4 of the device 1 according to the invention are preferably configured as a hyperspectral camera or as two independent hyperspectral cameras. The hyperspectral cameras enable the use of the device 1 according to the invention or the process according to the invention in modern sorting facilities in the food industry. Unlike conventional color cameras, these cameras record not only the visible light range but also additional spectral ranges, for example, in the infrared range.Since images obtained in the infrared range provide information about the chemical properties of product surfaces using chemical imaging technology, these cameras are ideally suited for assessing food quality and detecting defects or contamination invisible to the naked eye. Unlike analytical methods performed in a laboratory or testing facility, these photographic methods are suitable for testing high volumes of food products using automated technology and for sorting them in high-speed, real-time sorting systems.

[0084] To determine the content of at least one fatty acid degradation product in the sample, gas chromatography followed by mass spectrometry analysis is preferably performed. This allows for a precise determination of the fatty acid degradation product content in the sample.

[0085] As shown in Figure 1 and Figure 2, the device 1 according to the invention preferably comprises a sorting unit 5. The sorting unit 5 is configured to separate an oilseed, a nut, in particular a hazelnut 2, or a seed from a product stream when the oilseed, nut, in particular a hazelnut 2, or seed is classified as rotten. For this purpose, the sorting unit 5 may comprise, for example, a pneumatic diverter unit 6 connected to the detection photosensor 4. The pneumatic diverter unit separates the oilseeds, nuts, in particular hazelnuts 2, or seeds classified as rotten from the essentially continuous product stream.

[0086] The method according to the invention comprises the sensory acquisition of hazelnut kernels, or in general of oilseeds, nuts, or seeds, recorded in reflectance or transmittance, in a wavelength range of 300-2500 nm or a portion thereof, preferably using a hyperspectral or multispectral camera. The spatially resolved spectral data thus obtained are evaluated by a computing unit 8, such as a PC, an embedded system with an FPGA, CPU, or GPU, so that they are correlated, in a training mode, with measurement values ​​from a reference laboratory for the determination of chemical components in, for example, hazelnuts by GC, for example.The correlation obtained during training—that is, the spectral information that can be related to the information contained in the reference dataset—is subsequently used in an operating mode, characterized by detection stages, to assign a corresponding measurement value of the degree of decay to the newly recorded spectra. Thus, each pixel in the camera's field of view corresponds to a reference value based on calibration through the correlation established between the spectral data and the measurement laboratory.

[0087] Any suitable statistical procedure from the field of multivariate regression analysis may be used as a method for extracting the corresponding correlation information, which, for example, correlates the variance of the spectral dataset with the variance of the measurement laboratory dataset. Such procedures are known without limitation of generality, for example, principal component regression (PCR), partial least squares regression (PLSR), multiple linear regression (MLR), multivariate curve resolution (MCR), independent flexible class analogy modeling (SIMCA), support vector machine (SVR) regression, artificial neural network regression, decision trees, and / or random forests.

[0088] Based on the spatially resolved optical measurement thus obtained of the surfaces of oilseeds, nuts, particularly hazelnuts 2, or seeds, it can be determined by defining the first and second threshold values ​​whether, for example, an inspected almond is rotten or belongs to the good product. The first threshold value defines from how many pixels of rot per object, for example, the almond or a hazelnut 2, must be classified as rotten. The second threshold value defines from what point a pixel value in the camera's field of view must be counted as rotten.

[0089] Figure 4 shows an illustrative evaluation of fatty acid degradation products in rotten hazelnuts 2 within the framework of a selection of those fatty acid degradation products that show the greatest differences between a rotten hazelnut and a non-rotten hazelnut.

[0090] Preferably, within the framework of the process according to the invention, the fatty acid degradation products are identified as butyrolactone, diacetyl, 2-methylbutanal, 3-methylbutanal, acetylacetone, filbertone, and 2,3-butanediol. In particular, the putrefaction is attributed to butyrolactone, a cyclic ester of hydroxycarboxylic acids. In particular, the identification of fatty acid degradation products in the sample by mass spectrometry detection can be carried out by constructing chromatograms for selected mass-to-charge ratios in the range of 20 to 300, preferably for at least a mass-to-charge ratio of 86 for butyrolactone, diacetyl, 2-methylbutanal, 3-methylbutanal, and acetylacetone, 45 for 2,3-butanediol, and 69 for filbertone. The mass-to-charge ratio of acetic acid is typically 60.

[0091] The device 1 according to the invention, preferably comprising a sorting unit 5, as well as the method according to the invention, are characterized in particular by the following advantages: high processing speed and high decision reliability are provided in the inline sorting process, thanks to the evaluation of the correlations between the concentration of at least two fatty acid degradation products and the spectral response. Furthermore, a high-quality offline determination of concentration is achieved based on the current state of offline laboratory art, which can be used inline in the sorting process.According to the invention, the classification of the degree of putrefaction does not constitute a binary magnitude, such as rotten or not rotten, but an analog magnitude that can be determined, by surface measurement, on the basis of the amplitude at certain wavelengths or the average value of the amplitudes in a range of wavelengths and a spatial distribution in, for example, the hazelnut 2. Based on the spectral detection of the degree of putrefaction, a classification with high product quality requirements and a precise adjustment of the classification limit value in the classification installation can be carried out. List of references

[0092] 2 Oil-bearing fruit, particularly hazelnut 3 Calibration light source 3 Detection light source 4 Calibration photosensor 4. Detection Photosensor 5 Classification Unit 6 Deviation Unit 7 Control Unit 8 Calculation Unit 9 AI Module

Claims

1. A method for determining whether an oilseed, a nut, in particular a hazelnut (2), or a seed is rotten, comprising at least the following calibration steps, preferably in the following order: a) irradiating a sample of an oilseed, a nut, in particular a hazelnut (2), or a seed with a calibration light source (3), b) recording an absorption or reflection spectrum in a wavelength range of 300-2500 nm using a calibration photosensor (4), c) associating at least one characteristic wavelength or at least one characteristic wavelength range of the recorded absorption or reflection spectrum of the sample with a degree of rottenness,d) repeating the above steps for a representative number of samples and establishing a correlation between the degree of putrefaction and the associated wavelengths or wavelength ranges of the recorded absorption or reflection spectra, wherein the procedure then comprises the following detection steps, preferably in the following order: e) irradiating an oilseed, a nut, in particular a hazelnut (2), or a seed with a detection light source (3), f) recording absorption or reflection spectra in a wavelength range of 300-2500 nm using a detection photosensor (4), the detection photosensor having a plurality of pixels and recording an absorption or reflection spectrum for each pixel,g) associating a degree of putrefaction with each pixel of the detection photosensor (4) by applying the correlation according to step d) to the absorption or reflection spectra recorded by the pixels of the detection photosensor (4); h) classifying the oilseed, nut, in particular the hazelnut (2), or seed as rotten when at least a predetermined number of pixels exhibit a degree of putrefaction exceeding a first threshold value and / or when a degree of putrefaction associated with at least one pixel exceeds a second threshold value.

2. The method according to any of claim 1, characterized in that, before step b), the light reflected and / or transmitted by the sample is projected onto a calibration photosensor (4).

3. The method according to any of claim 1 or 2, characterized in that, after step b),At least one degree of putrefaction is determined based on the content of at least one fatty acid degradation product in the sample, and in particular additionally based on the acetic acid content in the sample.

4. The process according to claim 3, characterized in that the at least one fatty acid degradation product comprises at least one component from the group consisting of butyrolactone, diacetyl, 2-methylbutanal, 3-methylbutanal, acetylacetone, filbertone, or 2,3-butanediol, and / or in that the degree of putrefaction is determined based on the content of at least two fatty acid degradation products in the oilseed.

5. The process according to any of claims 3 or 4, characterized in that several fatty acid degradation product components exhibiting the greatest concentration difference between oilseeds, nuts, in particular hazelnuts (2), are sought.or good seeds and oilseeds, nuts, in particular hazelnuts (2), or rotten seeds.

6. The method according to any of the preceding claims, characterized in that the correlation is at least a correlation function or at least a table of indices or a table of values, or comprises at least a comparison of spectral information comprising at least the degree of rot of the oilseed and the associated wavelengths or wavelength ranges of the recorded absorption or reflection spectra.

7. The method according to any of the preceding claims, characterized in that a common sensor is used as a calibration photosensor (4) and as a detection photosensor (4) and / or a single light source is used as a calibration light source (3) and as a detection light source (3).

8. The method according to any of the preceding claims,characterized in that at least one characteristic wavelength or at least one characteristic wavelength range of the recorded absorption or reflection spectrum of the sample is associated with the degree of putrefaction by at least one of the following: an average value, a bandwidth, or individual frequency bands of the recorded absorption or reflection spectrum.

9. The method according to any of the preceding claims, characterized in that the absorption or reflection spectra are recorded by the calibration photosensor (4) and / or the detection photosensor (4) by hyperspectral acquisition and, in particular, by a color camera.

10. The method according to any of the preceding claims, characterized in that the correlation, in a mode of operation characterized by the detection steps,is used to associate a corresponding measured value of the degree of putrefaction to newly recorded spectra.

11. The method according to claim 10, characterized in that each pixel corresponds to a reference value based on calibration by means of the correlation determined between the spectral data and a measuring laboratory.

12. A device (1) for determining whether an oilseed, a nut, in particular a hazelnut (2), or a seed is rotten, comprising a calibration light source (3), a detection light source (3), a calibration photosensor (4), a detection photosensor (4), and a calculation unit (8), characterized in that the device (1) is configured to carry out a method according to any one of claims 1 to 11.

13. The device (1) according to claim 12, characterized in that the device (1) comprises a sorting unit (5), the sorting unit (5) operating,In particular, with at least two adjustable threshold values, the first adjustable threshold value comprising an amount of substance for a pixel and the second adjustable threshold value comprising a non-homogeneous distribution of the substance, the second adjustable threshold value taking into account the number of pixels with the attribute: rotten for classification.

14. A computer program product comprising program instructions that, when executed, are configured to cause a device according to claim 12 to carry out at least one procedure according to any one of claims 1 to 11.

15. A computer-readable medium comprising at least one computer program product that, when executed by at least the computing unit (8) of a device according to claim 12, causes the device to carry out at least one procedure according to any one of claims 1 to 11.

Images