METHOD FOR PREDICTING A FEED AND / OR FEED RAW MATERIAL
Patent Information
- Application Number
- BR112021026105
- Authority / Receiving Office
- BR · BR
- Patent Type
- Patents
- Current Assignee / Owner
- Publication Date
- 2026-09-08
Smart Images
Figure 00000052_0000 
Figure 00000053_0000 
Figure 00000054_0000
Abstract
Description
1 / 46 METHOD FOR PREDICTING A FEED AND / OR FEED RAW MATERIAL
[001] The present invention relates to a method for predicting an unknown feed raw material and / or feed by means of near-infrared spectroscopy and similarity analysis using a clean database with spectra of known feed raw material and / or feed.
[002] Animal diets typically contain a variety of different feeds and / or feed raw materials. Therefore, it is necessary to know the identity and type of a feed and / or feed raw material as accurately and quickly as possible. This is particularly relevant when different feeds and / or feed raw materials must be mixed to produce a diet with a specific composition for a particular species. Qualitative analysis methods for feeds and feed raw materials allow, in principle, the accurate identification of feeds and / or feed raw materials of unknown type, i.e., unknown identity and origin, etc. However, these methods require expensive and high-maintenance laboratory equipment. Other disadvantages of these methods are their high time standards and the specialization and experience of the operational staff.In principle, near-infrared spectroscopy would be a suitable means for the identification and determination of feed and / or feed raw materials. According to EP 3361248 A1, the use of near-infrared spectroscopy also allows predicting the influence of processing on the nutritional value of feed and / or feed. This document discloses a method for evaluating the influences of processing on the nutritional value of feed raw materials and / or feed. Petition 870210119564, dated 12 / 22 / 2021, p. 15 / 84 2 / 46 The method comprises the steps of i) subjecting a sample of a feed raw material and / or feed to near-infrared spectroscopy, ii) matching the absorption intensities at the respective wavelengths or wavenumbers in the near-infrared spectrum with the corresponding parameters and their values obtained from the chemical analysis of the same sample and generating a calibration chart and / or calibration equation, iii) subjecting another sample of a feed raw material and / or feed to near-infrared spectroscopy and iv) obtaining the values of specific parameters for this sample from the calibration chart and / or calibration equation.
[003] When used as a routine method, however, near-infrared spectroscopy requires knowledge of the identity and type of feed and / or feed raw materials. However, human errors in the selection of feed and / or feed raw materials can lead to incorrect classification of a feed and / or feed raw material regarding its identity and present form. Based on the incorrect classification, the incorrect calibration method would be chosen for the near-infrared analysis of the ingredients and their specific quantities in the feed and / or feed raw material. Therefore, the data obtained from the incorrectly calibrated NIR spectrometer would be erroneous. Consequently, this data would be misleading for any further operational steps in which the respective feed raw material and / or feed is involved.
[004] One option to overcome this problem is to record a near-infrared spectrum of a feed sample. Petition 870210119564, dated 12 / 22 / 2021, p. 16 / 84 3 / 46 and / or unknown feed raw material and conducting a similarity search for the recorded spectrum. This approach is described in the published international application WO 2016 / 141198 A1 and in the article Algorithms, Strategies and Application Progress of Spectral Searching Methods (Chu X.-L., Li J.-Y., Chen P., Xu Y.-P., Chinese Journal of Analytical Chemistry, 2014, 42 (9), 1379-1386). In detail, a similarity search comprises the step of analyzing the similarity of the recorded spectrum of an unknown feed raw material and / or feed with the near-infrared spectrum of a population of known feed raw materials and / or feeds.The basis for similarity analysis is the transformation of relevant information from a spectrum, i.e., absorption intensities at their wavelengths or wavenumbers, into the corresponding vector, both for the spectrum of an unknown feed raw material and / or feed and for each spectrum in the population of known feed raw material and / or feed spectra. In the next step, the vector thus obtained from the spectrum of an unknown feed raw material and / or feed, which is hereinafter also referred to as the query vector, and the vectors from a population of known feed raw material and / or feed spectra, hereinafter also referred to as database vectors, are subjected to a similarity analysis. The multiplicity of database vectors is hereinafter also referred to as the set of database vectors.Similarity analysis involves calculating the similarity measure and / or the distance measure between the query vector of the registered spectrum of known feed raw material and / or feed and each database vector. Petition 870210119564, dated 12 / 22 / 2021, page 17 / 84 4 / 46 data from the population of known feed raw material spectra and / or feeds. A similarity analysis involving a similarity measure is, in principle, a search for the nearest neighbor of the query system, here the query vector. In this case, a high similarity value for a database vector indicates a high similarity of a database vector to the query vector. Therefore, the similarity values for all database vectors are sorted in descending order, with the highest values at the top. By comparison, when the similarity analysis involves a distance measure, a low similarity value for a database vector indicates a high similarity of a database vector to the query vector. Here, the similarity values for all database vectors are sorted in ascending order, with the lowest values at the top.In any case, the highest-ranked database vector has the greatest similarity to the query vector, regardless of whether the similarity analysis involves a similarity measure or a distance measure. In principle, a general similarity search is always based on the assumption that the top-ranked database vectors are likely the relevant vectors for the query vector. However, state-of-the-art methods cannot solve the problems that arise when there are false positives at the top of the ranking of database vectors. In the worst case, even the highest-ranked vector can be a false positive. The reasons for false positive entries in the ranking of database vectors can be... Petition 870210119564, dated 12 / 22 / 2021, page 18 / 84 5 / 46 Misassignment of a database vector or NIR spectrum corresponding to a non-matching class of feed raw materials and / or feeds, heterogeneity or disorder of the class of feed raw materials and / or feeds whose NIR spectra were recorded, or the similarity of some feed raw materials and / or feed classes to each other. Any of these cases complicates an accurate and reliable assignment of a database vector to the query vector.
[005] An alternative method is disclosed in US 2011 / 0153226 A1. This document discloses a method for spectral research of an unknown mixture, comprising: obtaining one or more candidate mixture combinations by comparing the spectrum of the unknown mixture with the spectrum of each of a first plurality of compounds from the library; generating a model for each of the candidate mixture combinations based, at least in part, on a modeling metric; calculating a residual spectrum corresponding to each of the candidate mixture combinations by removing the spectrum of each of the compounds in the candidate mixture combination from the spectrum of the unknown mixture; identifying one or more potential compounds by comparing each residual spectrum with the spectrum of each of a second plurality of compounds from the library; adding the potential compounds to the candidate mixture combinations to generate an updated list of candidate mixture combinations;and repeat the model generation, calculate the residual spectrum, identify potential compounds, and add potential compounds until a first termination condition is met. Petition 870210119564, dated 12 / 22 / 2021, page 19 / 84 6 / 46
[006] Document EP 0807809 A2 discloses another method for combining an unknown product with one from a library of known products comprising the steps: 1) measuring a near-infrared absorbance spectrum for each of said known products, 2) generating vectors of known products extending into hyperspace representing the absorbance spectra determined for each of said known products, 3) dividing said vectors of known products into groups of vectors extending into hyperspace, wherein the vectors within each group are closer to each other in hyperspace than the vectors outside such group, 4) dividing at least some of said groups of vectors into subgroups of vectors extending into hyperspace, 5) repeating said step 4) in at least some of said subgroups until all said subgroups have less than a predetermined number of vectors,6) Enclose each of said groups and subgroups with a defined envelope in the corresponding hyperspace, 7) measure the absorption spectrum of said unknown product, 8) determine in which of said envelopes enclosing said groups divided in step 3) a vector, representing said unknown product and extending into the hyperspace of said groups, falls, 9) if the vector representing said unknown product falls into an envelope enclosing a group that is divided into subgroups, then determine in which envelope enclosing a subgroup a vector representing said unknown product and extending into the hyperspace of such subgroup falls, 10) repeat step 9) in other divided subgroups until a vector representing said unknown product is determined, Petition 870210119564, dated 12 / 22 / 2021, page 20 / 84 7 / 46 falling into an envelope containing a subgroup that is no longer defined, and 11) then determine which known product represented by a vector within said envelope with the last name corresponds to said unknown product.
[007] CN 109459409 A discloses a method for anomalous spectral recognition in the near-infrared. In this method, the sample space is generally linearized by the Hilbert space filling curve. Then, a hyperparameter must be selected. In the study of outlier identification, the determination of the value of said hyperparameter must be determined according to experience. However, this requires an experienced and trained team. Specifically, this document discloses the identification of an abnormal spectrum involving the spatial distance metric of the principal component. However, any method involving a principal component analysis sets high demands on computational power and time. Therefore, it is not suitable for large volumes of data, as is the case with population spectra.
[008] The article Evaluation of Local Approaches to Obtain Accurate Near-Infrared (NIR) Equations for Prediction of Ingredient Composition of Compound Feeds (Fernández-Ahumada E. et al., Applied Spectroscopy, vol. 67, no. 8, 2013, pages 924-929) refers to a method for improving the accuracy of intact feed calibrations for near-infrared (NIR) prediction of ingredient composition. This article discloses that, prior to the development of the calibration, an outlier elimination routine was used to detect samples with Petition 870210119564, dated 12 / 22 / 2021, p. 21 / 84 8 / 46 atypical spectra were identified by extreme Hotelling T2 and Q residual values. Approximately 10% of the overall database were considered spectral outliers and removed, leaving 20,320 samples. Specifically, this document teaches the CARNAC (Comparison Analysis Using Restructured Near-Infrared and Constituent Data) method using PLS (Partial Least Squares) factors as input variables. This approach, however, is not suitable for small data volumes because, in this case, it is difficult to divide the data into a training set and a test set.
[009] Consequently, there is a need for a method that allows for a less complicated and at the same time very accurate prediction of unknown feed and / or feed raw materials.
[0010] It has been found that this problem is solved by removing outliers from each set of database vectors before using the set of database vectors in the similarity analysis with the query vector of an unknown feed raw material and / or feed. An outlier can be the result of human and / or instrumental error. A human error is, for example, the erroneous assignment of a vector from a near-infrared spectrum of a specific class of feed raw materials and / or feed to a set of vectors (database) of a different class of feed raw materials and / or feed. An example of instrumental error is the measurement of a sample of a feed raw material and / or feed with an infrared spectrometer that is not correctly calibrated or has not been calibrated. Petition 870210119564, dated 12 / 22 / 2021, page 22 / 84 9 / 46 completely. Typically, an outlier is an (observed) value, i.e., a database vector, that is unusual and implausible in the context of other values, i.e., the set of database vectors. Removing outliers, therefore, leads to a homogenization of a set of database vectors. Consequently, the probability of a misassignment is significantly reduced. This increases the accuracy in predicting a feed and / or feed raw material.
[0011] The object of the present invention is, therefore, a computer-implemented method for predicting a feed and / or feed raw material comprising the steps of: a) To provide a near-infrared spectrum of a sample of an unknown feed raw material and / or feed ingredient, b) transform absorption intensities of wavelengths or wavenumbers in the spectrum from step a) to provide a query vector, c) provide a set of database vectors from a population of feed raw material spectra and / or known feeds, wherein an outlier is removed from the set of database vectors, wherein step c) further comprises one or more of options c1) to c4): c1) remove a pair of database vectors that are the most different from each other in a set of database vectors from said set of database vectors, c2) remove a database vector that is, on average, the most different from the other database vectors in a set of database vectors from Petition 870210119564, dated 12 / 22 / 2021, page 23 / 84 10 / 46 referred to as a set of database vectors, c3) remove a database vector being the most different from all other database vectors in a set of database vectors from the referred set of database vectors, c4) remove a database vector being the most different from the centroid of a set of database vectors from the referred set of database vectors, d) Calculate a similarity measure and / or a distance measure between the query vector from step b) and each database vector from step c) to provide a similarity value for each database vector with the query vector. (e) classify the similarity values obtained in step d) in descending order when a similarity measure is calculated in step d), or in ascending order when a distance measure is calculated in step d), where, in either case, the best-ranked database vector has the highest similarity to the query vector, and f) assign the feed raw material and / or feed from the database vector with the highest similarity in step e) to the sample from step a).
[0012] In the context of the present invention, the term unknown feed raw material and / or feed refers to any type of feed and / or feed raw material whose identity, composition, origin and / or form, i.e., whether ground or unground, is unknown. By comparison, in the context of the present invention, the term feed raw material and / or feed Petition 870210119564, dated 12 / 22 / 2021, p. 24 / 84 11 / 46 known refers to any type of feed and / or feed raw material whose identification, composition, origin and / or form, i.e., whether it is ground or unground, is known. Therefore, a population of known feed raw material and / or feed spectra is a number or multiplicity of spectra that are known to belong to a specific feed and / or feed raw material of known identity, composition, origin and / or form.
[0013] According to the present invention, a near-infrared spectrum of a sample of an unknown feed raw material and / or feed is provided in step a). In the context of the present invention, this means that the location where the spectrum to be provided is recorded and the location where the computer-implemented method according to the present invention is performed may be different or identical. For example, it is possible that a near-infrared spectrum of a sample of an unknown feed raw material and / or feed is recorded at one location and sent in any way to a remote location where the computer-implemented method according to the present invention is performed. Alternatively, both the recording of the spectrum and the prediction of the feed raw material and / or feed based on said spectrum may be performed at the same location.
[0014] In one embodiment, step a) of the computer-implemented method comprises recording a near-infrared spectrum of a sample of an unknown feed raw material and / or feed.
[0015] Option c1) to remove outliers Petition 870210119564, dated 12 / 22 / 2021, page 25 / 84 12 / 46 involves a correlation of outlier pairs. Specifically, this option involves identifying the pair of database vectors that, in terms of similarity, are the most distant neighbors or, as synonyms, the most different in a set of database vectors. Then, the pair of database vectors thus identified is removed from the set of database vectors. This option is illustrated in Figure 1.
[0016] Option c2) for removing outliers involves identifying the database vector that, in terms of similarity, is the furthest neighbor on average, or synonymously the most different on average, from all other database vectors in a set of database vectors. Then, the database vector thus identified is removed from the set of database vectors. This option is illustrated in Figure 2.
[0017] Option c3) for removing outliers involves identifying the database vector that, in terms of similarity, is the furthest neighbor or synonymously the most different from all other database vectors in a set of database vectors. Compared to the previous option, the database thus identified is absolutely the most different database vector in a set of database vectors, whereas the previous option identifies the most different database vector in terms of dissimilarity from the other database vectors. This option is illustrated in Figure 3. By comparison, option c1) requires the removal of a pair of database vectors that are the most different from each other. Consequently, option c1) always requires the removal of Petition 870210119564, dated 12 / 22 / 2021, p. 26 / 84 13 / 46 pairs of database vectors, for example, two, four, or more, that are correlated with each other in terms of their dissimilarity. By comparison, option c3) does not require a pairwise removal of database vectors. Instead, it requires only the removal of one database vector, for example, just one, which is the most different from all the other database vectors. Therefore, in contrast to option c1), where a pair of vectors being correlated with each other is removed, in option c3) only one vector is removed and the other database vectors remain in the set.
[0018] Option c4) for removing outliers involves identifying the database vector that, in terms of similarity, is the furthest neighbor or synonymously the most different from the centroid of a set of database vectors. Then, the database vector thus identified is removed from the set of database vectors. In mathematics and physics, the term centroid, when used in context with a plane figure, denotes the arithmetic mean position of all points in the figure; and therefore, it is also referred to as the geometric center of said figure. Therefore, it is also the point where a cutout of the shape could be perfectly balanced on the tip of a pin. When the figure is extended to an object in a multidimensional space, the term centroid denotes the mean position of all said points in all coordinate directions.In the context of the present invention, the term centroid of a set of database vectors therefore denotes the arithmetic mean position of all points of the database vectors in all directions. Petition 870210119564, dated 12 / 22 / 2021, page 27 / 84 14 / 46 coordinates. This option is illustrated in Figure 4. In other words, the geometric center thus obtained, i.e., the centroid, is the basis for the dissimilarity search. In contrast, option c2) requires that each vector be compared with all other vectors in the database vector set in terms of dissimilarity. This routine is repeated with all vectors in the database vector set. The vector that is, on average, the most different is removed in option c2).
[0019] To ensure the highest possible accuracy in the forecast, it is preferable to apply one or more of options c1) to c4) for each set of database vectors. This has the advantage that all sets of database vectors are homogenized and not just one set of database vectors alone.
[0020] The four options can be used individually or in combination. When two or more options are used, it is possible to subject a set of database vectors sequentially or in parallel to two or more options. In the first case, a set of database vectors is subjected to a first option, and the resulting set of database vectors, free from the removed database vector, is subjected to a second or other option. Alternatively, it is also possible to subject a set of database vectors to two or more options in parallel and compare the results of the resulting sets of database vectors, which are free from the removed database vectors, and continue with the set of database vectors that is considered most suitable for the method according to the present invention, for example, Petition 870210119564, dated 12 / 22 / 2021, page 28 / 84 15 / 46 because most of the outliers were removed from the aforementioned set of database vectors. It is preferable to use all four options in parallel, to compare the results of the four options and to remove a database vector only when at least 2 options, in particular at least 3 options, indicate it as the most different.
[0021] In one embodiment of the computer-implemented method according to the present invention, step c1) comprises the steps of: c1a) Calculate a similarity measure and / or a distance measure from each database vector in a set of database vectors to the other database vectors in said set of database vectors to give similarity values of pairs of database vectors, c1b) Rank the similarity values obtained in step c1a) in descending order when a similarity measure is calculated in step c1a), or in ascending order when a distance measure is calculated in step c1a), wherein, in either case, the ranked similarity value relates to the two database vectors being the most different from each other, and c1c) Remove at least the two database vectors with the lowest ranking in step c1b) from the set of database vectors.
[0022] In another embodiment of the computer-implemented method according to the present invention, step c2) comprises the steps of: c2a) Calculate a similarity measure and / or a distance measure for each database vector in a set. Petition 870210119564, dated 12 / 22 / 2021, page 29 / 84 16 / 46 of database vectors to the other database vectors in the aforementioned set of database vectors to give similarity values from one database vector to the other database vectors, c2b) form the sum of the similarity values obtained for each database vector in step c2a) and calculate the average similarity value for each database vector, c2c) rank the average similarity values obtained in step c2b) in descending order, when a similarity measure is calculated in step c2b), or in ascending order, when a distance measure is calculated in step c2b), wherein, in either case, the average similarity value ranked at the end relates to the database vector being the most different, on average, from all other database vectors, and c2d) remove at least the database vector with the lowest ranking in step c2c) from the set of database vectors.
[0023] In another embodiment of the computer-implemented method according to the present invention, step c3) comprises the steps of: c3a) Calculate a similarity measure and / or a distance measure from each database vector in a set of database vectors to the other database vectors in that set of database vectors to give similarity values from one database vector to the other database vectors; c3b) Sort the similarity values obtained in step c3a) in descending order, when the measure of Petition 870210119564, dated 12 / 22 / 2021, page 30 / 84 17 / 46 similarity is calculated in step c3a), or in ascending order, when a distance measure is calculated in step c3a), where, in any case, the similarity value ranked at the end relates to the database vector being the most different from all other database vectors, c3c) count the occurrence of a feed raw material and / or feed among the database vectors in the ranking of step c3b), c3d) weight each feed raw material and / or feed according to its ranking in step c3b) and according to the frequency of its occurrence in step c3c) to give weighted ranking positions of the feed raw materials and / or feeds, and c3e) form the sum of the weighted ranking positions of a raw material of the ingredient and / or feed from step c3d) and,c3f) remove at least the feed raw material and / or feed database vector with the lowest weighted ranking position in step c3e) from the set of database vectors.
[0024] In yet another embodiment of the computer-implemented method according to the present invention, step c4) comprises the steps of: c4a) determine the centroid of all database vectors in a set of database vectors, c4b) calculate a similarity measure and / or a distance measure from each database vector to the centroid from step c4a) to give a similarity value for each database vector to the centroid, Petition 870210119564, dated 12 / 22 / 2021, page 31 / 84 18 / 46 c4c) sort the similarity values obtained in step c4b) in descending order when a similarity measure is calculated in step c4b), or in ascending order when a distance measure is calculated in step c4b), wherein, in either case, the sorted similarity value at the end relates to the database vector being the most different from the centroid, and c4d) remove at least the database vector with the lowest sorting in step c4c) from the set of database vectors.
[0025] The centroid can be calculated according to any suitable procedure known in the art. For example, the centroid can be calculated by taking all the database vectors in a database set with n vectors, summing all the 1-n positions over all the vectors, and dividing each position according to the number of vectors.
[0026] In the context of the present invention, the term cleaning a set of database vectors or cleaning a dataset is used equivalently to the expression removing a spectral outlier from a set of database vectors or dataset or removing a database vector meeting any of the requirements in step c1), c2), c3) and / or c4).
[0027] The method according to the present invention is not limited with respect to a specific distance or similarity measure for analyzing the similarity between the query vector of step b) and the database vectors of step c) and for analyzing the similarity within a set of database vectors in steps c1a), c2a), c3a) and / or Petition 870210119564, dated 12 / 22 / 2021, page 32 / 84 19 / 46 c4b). Therefore, any distance or similarity measure that is suitable for determining the similarity of the vectors in step b) with the vectors in step c) can be used in the method according to the present invention. In principle, a similarity analysis is based on a nearest neighbor search. It has been found that the cosine coefficient is a particularly suitable similarity measure for a nearest neighbor search in the method according to the present invention. For example, the cosine coefficient, which allows the calculation of the similarity between two vectors extremely quickly and with high precision, is particularly suitable in the method according to the present invention. The cosine coefficient of two vectors A and B is represented by the following formula. SA,B = [l^jA^] [^j:;(wJu:?(M2where xjA and xjB are components of vectors A and B, respectively, en is the number of spaces, here the number of absorption intensities at specific wavelengths or wavenumbers. Similarity values range from 1, meaning exactly opposite of each other, to 1, meaning exactly the same, with 0 indicating orthogonality (decorrelation) and intermediate values indicating intermediate similarity or dissimilarity.
[0028] Alternatively, the similarity between vectors can also be calculated by means of a distance measure. For example, the Euclidean distance, which allows the calculation of the similarity between two vectors in an extremely fast and precise way, is particularly suitable in the method according to the present invention. The distance Petition 870210119564, dated 12 / 22 / 2021, page 33 / 84 20 / 46 The Euclidean curve of two vectors A and B is represented by the following formula: Da,b—-%7b)2where xjA and xjB are components of vectors A and B, respectively, and en is the number of spaces, here the number of absorption intensities at specific wavelengths or wavenumbers.
[0029] Advantageously, any database vectors with a similarity value of 0 are directly removed from the database vector set, which allows for more efficient removal of outliers and thus even more accurate prediction of feed raw materials and / or feeds.
[0030] In a preferred embodiment of the computer-implemented method according to the present invention, a database vector with a similarity value of 0 is removed from the set of database vectors in steps c1b), c2c), c3b), and / or c4c).
[0031] It is preferred that in step c) of the computer-implemented method according to the present invention, the similarity measure is the cosine coefficient and the distance measure is the Euclidean distance.
[0032] In its broadest meaning, a vector is a geometric object that has magnitude (or length) and direction. In a Cartesian coordinate system, a vector can be represented by identifying the coordinates of its starting and ending points. Therefore, a vector is suitable for representing an absorption intensity at a specific wavelength or wavenumber in a spectrum. Petition 870210119564, dated 12 / 22 / 2021, page 34 / 84 21 / 46 two-dimensional near-infrared. Furthermore, a vector is not limited to describing a two-dimensional system. Instead, a vector can describe multidimensional spaces, such as a near-infrared spectrum with an infinite number of absorption intensities at an infinite number of different wavelengths or wavenumbers. In this case, each dimension of said vector corresponds to a single absorption intensity at a specific wavelength or wavenumber.
[0033] In one embodiment of the computer-implemented method according to the present invention, the vector in steps b) and c) is a multidimensional vector, with each dimension corresponding to an absorption intensity of a specific wavelength or wavenumber.
[0034] As the query vector for the spectrum of an unknown feed raw material and / or feed, the set of database vectors provided in step c) of the computer-implemented method according to the present invention is also obtained by transforming each spectrum from a population of spectra of known feed raw materials and / or feed into the corresponding vector. If the set of database vectors is not yet present, step c) also comprises transforming each spectrum from a population of spectra of known feed raw materials and / or feed into the corresponding vector to give the set of database vectors. In this case, step c) of the computer-implemented method according to the present invention comprises the steps of transforming each spectrum from a population of spectra of feed raw materials and / or feed into the Petition 870210119564, dated 12 / 22 / 2021, page 35 / 84 22 / 46 corresponding vector to give a set of vector data and providing the set thus obtained of database vectors of a population of spectra of feed raw materials and / or known feeds.
[0035] Therefore, it is also possible to remove an outlier value already from the infrared spectrum of feed raw materials and / or known feeds that must be transformed into the database vector set. In this case, steps c1), c2), c3) and / or c4), preferably steps c1a) to c1c), c2a) to c2d), c3a) to c3f) and / or c4a) to c4d) are performed with the infrared spectrum of a population of feed raw materials and / or known feeds. Then, the thus cleaned infrared spectra are transformed into vectors to provide the database vector set.
[0036] In this alternative embodiment of the computer-implemented method according to the present invention, step c) therefore comprises the steps of: c1') remove a pair of infrared spectra being the most different from each other in a population of infrared spectra from said population of infrared spectra, c2') remove an infrared spectrum being, on average, the most different from the other infrared spectra in a population of infrared spectra from said population of infrared spectra, c3') remove an infrared spectrum being the most different from all other infrared spectra in a population of infrared spectra from said population of infrared spectra, and / or c4') remove an infrared spectrum being the most Petition 870210119564, dated 12 / 22 / 2021, page 36 / 84 23 / 46 different from the centroid of a population of infrared spectra of said population of infrared spectra.
[0037] To ensure the highest possible accuracy in the prediction, it is preferable to apply this option to each set of database vectors. This has the advantage that all sets of database vectors are homogenized and not just one set of database vectors alone.
[0038] In a preferred embodiment of the computer-implemented method according to the present invention, step c1') - before removing a pair of infrared spectra being the most different from each other - further comprises the steps of: c1a') calculate a similarity measure and / or a distance measure from each infrared spectrum to the other infrared spectra in a population of infrared spectra to give similarity values for pairs of infrared spectra, c1b') rank the similarity values obtained in step c1a') in descending order when a similarity measure is calculated in step c1a'), or in ascending order when a distance measure is calculated in step c1a'), wherein, in either case, the similarity values ranked at the end refer to the two infrared spectra being the most different from each other, and c1c') indicate at least the two infrared spectra with the lowest ranking in step c1b') as the most different from each other in the population of spectra. Petition 870210119564, dated 12 / 22 / 2021, page 37 / 84 24 / 46 infrared.
[0039] In another preferred embodiment of the computer-implemented method according to the present invention, step c2') - before removing an infrared spectrum being the most different, on average, from the other infrared spectra - further comprises the steps of: c2a') Calculate a similarity measure and / or a distance measure from each infrared spectrum to the other infrared spectra in a population of infrared spectra to give similarity values from one infrared spectrum to the other infrared spectra, c2b) Sum the similarity values obtained for each database vector in step c2a) and calculate the average similarity value for each database vector, c2c') Sort the average similarity values obtained in step c2b') in descending order when a similarity measure is calculated in step c2a'), or in ascending order when a distance measure is calculated in step c2a'), wherein, in either case, the average similarity value sorted at the end refers to the infrared spectrum being the most different from all other infrared spectra,and c2d') indicate at least the infrared spectrum with the lowest ranking in step c2c) as the most different from all other infrared spectra in the population of infrared spectra.
[0040] In another preferred embodiment of the computer-implemented method according to the present Petition 870210119564, dated 12 / 22 / 2021, page 38 / 84 25 / 46 invention, step c3') - before removing an infrared spectrum being the most different from all other infrared spectra - further comprises the steps of: c3a') Calculate a similarity measure and / or a distance measure from each infrared spectrum to the other infrared spectra in a population of infrared spectra to give similarity values from one infrared spectrum to the other infrared spectra; c3b') Rank the similarity values obtained in step c3a') in descending order when the similarity measure is calculated in step c3a'), or in ascending order when a distance measure is calculated in step c3a'), wherein, in either case, the ranked similarity value at the end refers to the infrared spectrum being the most different from all other infrared spectra; c3c') Count the occurrence of a feed raw material and / or feed in the ranking of step c3b').c3d') weight each feed raw material and / or feed according to its classification in step c3b') and according to the frequency of its occurrence in step c3c') to give weighted ranking positions of the feed raw materials and / or feeds, and c3e') form the sum of the weighted ranking positions of a feed raw material and / or feed, and c3f') indicate at least the infrared spectrum of the feed raw material and / or feed as the most different from all other infrared spectra in the population of infrared spectra. Petition 870210119564, dated 12 / 22 / 2021, page 39 / 84 26 / 46
[0041] In yet another preferred embodiment of the computer-implemented method according to the present invention, step c4') – before removing a database vector being the most different from all other database vectors – further comprises the steps of: c4a') determine the centroid of all infrared spectra in a population of infrared spectra, c4b') calculate a similarity measure and / or a distance measure from each infrared spectrum to the centroid from step c4a') to give a similarity value for each infrared spectrum to the centroid, c4c') rank the similarity values obtained in step c4b') in descending order when a similarity measure is calculated in step c4b'), or in ascending order when a distance measure is calculated in step c4b'), where in either case the ranked similarity value relates to the infrared spectrum being the most different from the centroid, and c4d') indicate at least the infrared spectrum with the lowest ranking in step c4c') as the most different from the centroid of the population of infrared spectra.
[0042] Identifying outliers is facilitated by taking a derivative, preferably the first derivative, of the spectra. Before taking the first derivative, the spectrum in question is normally subjected to a standardization procedure, such as standard normal variation (SNV), detrend, multiplicative dispersion correction (MSC), or extended multiplicative signal correction (EMSC). Detrend (baseline corrections) is performed by subtracting a linear fit or Petition 870210119564, dated 12 / 22 / 2021, page 40 / 84 27 / 46 polynomial of the baseline of the original spectrum to remove the slanted baseline variation, commonly found in NIR reflectance spectra of powder samples. The standard normal variable is another frequently used pretreatment method due to its simple algorithm and effectiveness in correcting dispersion. SNV is frequently used in spectra where baseline and path length changes cause differences between otherwise identical spectra. Multiplicative dispersion correction is obtained by regressing a measured spectrum against a reference spectrum and then correcting the measured spectrum using the slope and intercept of this linear fit. This pretreatment method has proven effective in minimizing baseline shifts and the multiplicative effect. The result of MSC, in many cases, is very similar to SNV.However, many spectroscopists prefer SNV over MSC since SNV corrects each spectrum individually and does not require the entire dataset. The extended multiplicative signal correction preprocessing method allows the separation of physical light scattering effects from chemical absorbance effects in spectra of powders or turbid solutions, for example. The model-based method is particularly useful for minimizing wavelength-dependent light scattering variation. After pretreatment, the corrected spectrum becomes insensitive to light scattering variations and responds linearly to analyte concentration. The mathematical description of EMSC is provided below.
[0043] A measurement spectrum can be approximated by Petition 870210119564, dated 12 / 22 / 2021, page 41 / 84 28 / 46 sum of baseline shifts, ideal chemical absorbance according to Beer's law and wavelength-dependent variations, and written as: Xi « ai + biXi,chem + diÀ + eiÀ2where a: baseline shift; b: path length; dee: wavelength-dependent variation. Xi,corrected=(Xi — diÀ - eiÀ2) / bi
[0044] By estimating the EMSC parameter, an EMSC-corrected spectrum can be obtained, with only part of the chemical absorbance remaining after removing baseline shift and wavelength-dependent variations.
[0045] Cases may arise where the position of a signal peak in a spectrum, either in step b) and / or step c) of the method according to the present invention, cannot be located because the maxima and minima of the individual peaks cannot be clearly identified in such a spectrum. Easier location of individual peaks in the spectrum is possible when the minima and maxima of the peaks are more easily identifiable. Taking the first derivative of a spectrum facilitates the identification of peaks in the spectrum because it gives a zero crossing of the maxima or minima of the peak. Taking the second derivative gives a minimum peak exactly at that position where a maximum peak was in the original spectrum and vice versa. Taking the first or second derivative of a spectrum also facilitates the identification of an outlier in the population of spectra of known raw materials and / or feeds.
[0046] In another embodiment of the method according to the present invention, a derivative is formed from the spectrum of Petition 870210119564, dated 12 / 22 / 2021, page 42 / 84 29 / 46 feed raw material and / or unknown feed from step a) and / or spectra of known raw materials of the raw material and / or raw materials before their transformation into a vector for step c).
[0047] Preferably, the first derivative is formed from the spectrum of feed and / or feed raw material of unknown type from step a) and / or from the spectra of feed raw materials and / or feeds before their transformation into a vector for step c).
[0048] According to step b) of the present invention, the absorption intensities of wavelengths or wavenumbers in a spectrum are transformed to provide a query vector. In principle, one could select the strongest and therefore most significant absorption intensities in a spectrum and transfer only those absorption intensities to give a vector. However, this would require a complete analysis of each individual spectrum of a sample substance, which is not only time-consuming but also requires a good understanding of near-infrared spectra. Therefore, this approach would not be suitable for routine analysis. Furthermore, this approach has the disadvantage that significant but relatively weak absorption intensities in a spectrum may be overlooked, so that information is lost. This can lead to incorrect attribution of feed raw material and / or unknown feed in the end.Therefore, it is advantageous to consider as much information as possible in the spectrum without a prior in-depth analysis of the spectrum. It is therefore preferable to transform the absorption intensities into wavelengths or numbers. Petition 870210119564, dated 12 / 22 / 2021, page 43 / 84 30 / 46 equidistant wavelengths in a spectrum, i.e., in step b) and / or c), to give a vector of said spectrum. In order to allow the best possible similarity analysis between the query vector and the database vectors, it is preferable to transform the absorption intensities of equidistant wavelengths or wavenumbers into a spectrum in step b) and step c) of the method according to the present invention to provide a vector of said spectrum or spectra. Preferably, the distances of the absorption intensities transformed into vectors in step b) are identical to the distances of the absorption intensities transformed into vectors in step c). This allows for greater accuracy in the prediction of the computer-implemented method according to the present invention, even without having any specific knowledge of the sample substance and its spectra.
[0049] In one embodiment of the invention, in step b) and / or c) of the computer-implemented method according to the present invention, the absorption intensities of equidistant wavelengths or wavenumbers in a spectrum are transformed to give a spectrum vector in step b) and / or c).
[0050] In another embodiment of the computer-implemented method according to the present invention, the distances of the absorption intensities being transformed into vectors in step b) are identical to the distances of the absorption intensities transformed into vectors in step c).
[0051] Preferably, the absorption intensities of wavelengths or wavenumbers in a spectrum, which are transformed to give a vector of said spectrum, Petition 870210119564, dated 12 / 22 / 2021, page 44 / 84 31 / 46 have small distances between them. This has the advantage that most, if not all, of the relevant absorption intensities, i.e., information, of a spectrum are transformed into a vector of that spectrum. It is believed that this allows a very precise transformation of all relevant information from a spectrum into vectors, even without knowledge of the feed and / or feed raw material whose spectrum was recorded, in particular its identity, composition, origin and / or form. Preferably, the distance between the wavelengths in step b) of the method according to the present invention is from 0.1 + / - 10% to 10 + / - 10% nm, from 0.1 + / - 10% to 5 + / - 10% nm, or from 0.1 + / - 10% to 2 + / - 10% nm. Consequently, the distance between the wavenumbers in step b) of the method according to the present invention is from 108 + / - 10% to 106 + / - 10%, from 108 + / - 10 to 5 * 106 + / - 10% nm, or from 108 + / - 10% to 2 * 106 + / - 10% nm.In the context of the present invention, the term + / - 10% is used in relation to the explicitly mentioned values to indicate that deviations from said explicitly mentioned values are still within the scope of the present invention, provided that they essentially lead to the effects of the present invention. The distances between the wavelengths or wavenumbers in step c) of the method according to the present invention are preferably the same as in step b), in order to provide the best possible comparison between the recorded spectrum of an unknown feed and / or feed raw material and the spectra of known feeds and / or feed raw materials.
[0052] In one embodiment of the computer-implemented method according to the present invention, the distance Petition 870210119564, dated 12 / 22 / 2021, p. 45 / 84 32 / 46 between the wavelengths or wavenumbers in step b) and / or step c) is 0.1 nm + / - 10% to 10 nm + / - 10% or 108 cm-1+ / - 10% to 106 cm-1+ / - 10%.
[0053] In principle, the computer-implemented method according to the present invention is not subject to any limitation regarding the number of absorption intensities to be transformed to give a vector. Instead, the amount of relevant information in a spectrum of a feed and / or feed raw material depends strongly on the individual feed and / or feed raw material and, in particular, on its composition and components. The more complex a feed and / or feed raw material, that is, the more components a feed and / or feed raw material contains, the more information is needed from a near-infrared spectrum to predict an unknown feed and / or feed raw material. Again, it would not be practical to perform an in-depth analysis to discover the absorption intensities that must necessarily be transformed to give a vector.A suitable option for the number of absorption intensities to be transformed into a vector is to correlate them with the distance between the corresponding wavelengths or wavenumbers, for example, from 0.1 nm + / - 10% to 10 nm + / - 10% or 0.1 + / - 10% to 2 + / - 10% nm, and the recording range of the spectrum, for example, from 1,100 to 2,500 nm. Preferably, the number of absorption intensities to be transformed into a vector is at least 100; in particular, this number ranges from 150 + / - 10% to 15,000 + / - 10% or from 700 + / - 10% to 15,000 + / - 10%.
[0054] In another form of the method according to Petition 870210119564, dated 12 / 22 / 2021, p. 46 / 84 33 / 46 present invention, the number of absorption intensities in each spectrum being transformed into a vector is 100 + / - 10% or more.
[0055] In some cases, it is preferable that a similarity analysis be performed first, followed by counting the occurrences of a feed raw material and / or feed in the ranking of similarity values. Then, the number thus determined of similarity values of the feed raw material and / or feed are weighted according to their ranking position to give the weighted ranking position, the sum is formed of the weighted ranking positions to give scores for the feed raw materials and / or feed, and the highest score indicates the feed raw material and / or feed with the greatest similarity to the sample substance.
[0056] In one embodiment, step e) of the method according to the present invention comprises the steps of: e1) count the number of occurrences of each of the feed raw materials and / or feeds among the best-ranked database vectors in the classification of step e), where the number of occurrences is indicated by the variable N, e2) weight the first N similarity values of each of the feed raw materials and / or feeds according to their position in the classification of step e1) to give weighted ranking positions for each of the feed raw materials and / or feeds, and e3) sum the weighted ranking positions of step e2) for each of the feed raw materials and / or feeds to give scores for each of the Petition 870210119564, dated 12 / 22 / 2021, page 47 / 84 34 / 46 raw materials for animal feed and / or animal feed, where the higher the score, the greater the similarity.
[0057] In this case, the raw material of the raw material and / or raw material of the database vector with the highest similarity in step e3) is assigned to the sample from step a).
[0058] The population of feedstock and / or feed spectra used in the method according to the present invention is not limited to specific feedstock and / or feed. Instead, it preferably comprises spectra of all feedstock and / or feed used in animal nutrition, preferably in the nutrition of poultry, pigs, swine and / or animals kept in aquaculture, such as fish and / or crustaceans. The spectra of feedstock and / or feed may differ significantly depending on their form or appearance, for example, when they are present in ground or unground form. Therefore, the population of spectra preferably also comprises spectra that were recorded from the aforementioned feedstock and / or feed in ground and / or unground form.
[0059] In another embodiment of the method according to the present invention, the population of spectra of feed and / or feed raw materials known in step c) of said method comprises spectra of all feed and / or feed raw materials in ground and / or unground form used in animal nutrition.
[0060] In principle, the method according to the present invention is in no way limited with respect to the number and types of feed and / or feed raw materials, whose spectra, recorded in ground and / or unground form, Petition 870210119564, dated 12 / 22 / 2021, p. 48 / 84 35 / 46 constitute the population of spectra. However, it is preferred that the population of spectra of feeds and / or feed raw materials known in step c) of the aforementioned method comprise spectra of all feeds and / or feed raw materials in ground and / or unground form used in animal nutrition, preferably in the nutrition of poultry, swine, pigs and / or animals kept in aquaculture, such as fish and / or crustaceans. Feed raw materials and / or feed raw materials may be of animal and / or vegetable origin. Particularly preferred feeds and / or feed raw materials are unprocessed and / or processed feed raw materials and / or feed raw materials. Processed feed raw materials and / or feed raw materials are those that have been subjected to any type of heat treatment or pressure treatment to remove or detoxify antinutritional factors.Preferred feed ingredients and / or feed raw materials are oilseeds, in particular soybean meal and bagasse, whole soybeans, rapeseed meal and bagasse, cottonseed meal, peanut meal, sunflower meal, coconut meal and / or palm kernel meal; legumes, especially roasted guar flour; brewing and distillation by-products, especially dried distillers grains with solubles (DDGS), grain processing and feed production by-products, especially corn gluten, corn seed meal and / or baking by-products; animal by-products, especially fishmeal, meat meal, poultry meal, blood meal and / or bone meal; and also any type of grain. In particular, the feed raw material is soybean seed. Petition 870210119564, dated 12 / 22 / 2021, page 49 / 84 36 / 46 of soybeans or a soybean product.
[0061] Depending on factors such as climate, soil and plant genetics, feed and / or feed raw materials from different global growing areas may differ in their ingredients and the content of such ingredients. In order to allow reliable and reproducible predictions of a feed and / or feed raw material, it is therefore preferred that the feed and / or feed raw material whose spectra are part of the spectrum population be from all of its global growing areas.
[0062] In another embodiment of the method according to the present invention, the feed and / or feed raw material, whose spectra are part of the spectral population, is from all its global growing areas.
[0063] The number of feed raw material and / or feed spectra in the spectral population used in the method according to the present invention must be representative to allow a reliable and reproducible prediction of the feed raw material and / or feed in question. Therefore, the spectral population comprises at least 50 sample spectra of each feed and / or feed raw material, i.e., each feed and / or feed raw material in ground and / or unground form used in animal nutrition, preferably in the nutrition of poultry, pigs, swine and / or animals kept in aquaculture, such as fish and / or crustaceans, from each of their global cultivation areas. The method according to the present invention is not subject to any limitations regarding the number of sample spectra of any feed and / or feed raw material from any of their global cultivation areas. Therefore, the number of sample spectra of Petition 870210119564, dated 12 / 22 / 2021, pp. 50 / 84 37 / 46 any feed and / or feed raw material from any of its global growing areas may vary from 50 to 10,000, from 50 to 5,000, from 50 to 2,500, from 50 to 2,000, from 50 to 1,500, from 50 to 1,000, from 100 to 1,000, from 50 to 500, from 100 to 500, from 50 to 250, from 100 to 250 or from 50 to 100.
[0064] In yet another embodiment of the method according to the present invention, the population of spectra of feed and / or feed raw materials known from step c) comprises at least 50 sample spectra of each feed and / or feed raw material from each of its growing areas.
[0065] When the population of spectra of feed and / or feed raw materials of known type considers each global growing area of a feed and / or feed raw material and the number of spectra from each global growing area is representative, the method according to the present invention allows not only a reliable and reproducible prediction of the feed raw material and / or feed in question, but also a prediction of the origin of the feed raw material and / or feed in question.
[0066] Therefore, the aforementioned population of feed and / or feed raw material spectra known from step c) preferably comprises at least 50 sample spectra of each feed and / or feed raw material from each of its global growing areas. The number of sample spectra of a feed and / or feed raw material from each global growing area is not subject to any limitations. Thus, the number of sample spectra of any feed and / or feed raw material from each global growing area may vary from 50 to 10,000, from 50 to 5,000, from 50 to 2,500, from 50 to 2,000, from 50 to 1,500, from 50 to 1,000, from Petition 870210119564, dated 12 / 22 / 2021, pp. 51 / 84 38 / 46 100 to 1,000, 50 to 500, 100 to 500, 50 to 250, 100 to 250 or 50 to 100.
[0067] It is preferred to provide the set of database vectors of a population of feed raw material and / or known feed spectra directly in step c) of the computer-implemented method according to the present invention. It is also possible that first only a population of feed raw material and / or known feed spectra is provided, which is transformed in the next step into the set of database vectors for similarity analysis in step d). In this case, step c) of the computer-implemented method according to the present invention also comprises the step of transforming the absorption intensities of wavelengths or wavenumbers in each spectrum of a population of feed raw material and / or feed spectra into vectors.The resulting multiplicity of vectors from the population of spectra of feed raw materials and / or known feeds is then the database vector set, as mentioned above. In any case, it is preferred to store the population of spectra of feed raw materials and / or known feeds or the database vector set of said population of spectra of feed raw materials and / or feeds in a processing unit, such as a computer or a cloud. The processing unit in which the population of spectra or the database vectors are stored may be identical to or different from the processing unit that executes the computer-implemented method according to the present invention. In the latter case, the first unit of... Petition 870210119564, dated 12 / 22 / 2021, page 52 / 84 39 / 46 processing, which performs the computer-implemented method according to the present invention, and the second processing unit, in which the spectrum population or database vectors are stored, form a network. For example, it is also possible that the spectrum population or database vectors are stored in a cloud. In this case, the first processing unit, for example, a computer, which performs the computer-implemented method according to the present invention, and the second processing unit, for example, a cloud, in which the spectrum population or database vectors are stored, form a network.
[0068] Another object of the present invention is, therefore, a system for predicting a feed and / or feed raw material, comprising a processing unit adapted to perform the computer-implemented method according to the present invention.
[0069] The principle of cleaning a set of database vectors or a population of spectra of an outlier also makes it possible to provide an improved database comprising sets of database vectors of a population of spectra of feed raw materials and / or feeds or a population of spectra of known feed raw materials and / or feeds, which are free of outliers and therefore suitable for use in the method and / or system according to the present invention.
[0070] The aforementioned database comprises i) a set of database vectors of a population of spectra of feed raw materials and / or known feeds and / or ii) a population of spectra of materials Petition 870210119564, dated 12 / 22 / 2021, pp. 53 / 84 40 / 46 primary feeds and / or known feeds, where the set of database vectors and / or the population of spectra is free of outliers, which are stored on a computer, a server, a cloud, or any type of computer-readable media, such as a disk, a CD-ROM, a flash drive, or a USB drive.
[0071] The said database can be reached or obtained by the same principles in which an outlier is removed from a set of database vectors or from a population of infrared spectra of said population of infrared spectra.
[0072] Therefore, the database can be obtained or acquired by step c) and one or more of the steps c1) to c4) of the computer-implemented method according to the present invention, preferably by step c) and one or more steps selected from the group consisting of steps c1a) to c1c), steps c2a) to c2d), steps c3a) to c3f) and steps c4a) to c4d).
[0073] If the population of spectra of known feed raw materials is stored on one computer, it is preferable that said population be stored on a second computer, i.e., a computer that does not yet run the computer-implemented method according to the present invention. As a consequence, the workload is evenly distributed between the computers, which can lead to faster performance of the computer-implemented method according to the present invention. This also allows communication between the user and the provider of a database with the database vector for use in the method according to the present invention, for example, a Petition 870210119564, dated 12 / 22 / 2021, pages 54 / 84 41 / 46 updating database vectors.
[0074] In one embodiment of the system according to the present invention, the processing unit adapted to perform the computer-implemented method according to the present invention forms a network with at least one other processing unit, in which the database vectors are stored.
[0075] In another embodiment, the system according to the present invention further comprises a database comprising i) a set of database vectors of a population of feed raw material spectra and / or known feeds and / or ii) a population of feed raw material spectra and / or known feeds, wherein the set of database vectors and / or the population of spectra is free of outliers. Description of the figures: Figure 1 is a schematic representation of option c1) for removing outliers, where the black markers (“bullets”) are a pair of database vectors that are the most different from each other in a set of database vectors, and the checkered markers are the other database vectors in that set; Figure 2 is a schematic representation of option c2) for removing outliers, where the black marker is the database vector that is the most different, on average, from all other database vectors in the set of database vectors, and the checkered markers are the other database vectors in that set; Figure 3 is a schematic representation of option c3) for removing outliers, where the black marker Petition 870210119564, dated 12 / 22 / 2021, pp. 55 / 84 42 / 46 is the database vector that is the most different from all other database vectors in the set of database vectors, and the checkered markers are the other database vectors in that set; Figure 4 is a schematic representation of option c4) for removing outliers, where the black marker is the database vector that is most different from the centroid of a set of database vectors, the gray marker is the centroid, and the checkered markers are the other database vectors in the set. Example: In the example, four different types of filters, indicated as Filter 1 to Filter 4, were compared regarding their suitability for cleaning a dataset to predict a material class, i.e., a feed raw material and / or feed. The results for these filters were compared with the case where no filter was used, indicated as Filter 0. Therefore, the example with Filter 0 was a comparison example not in accordance with the invention, and the examples with Filters 1 to 4 were in accordance with the invention. In detail, the four different types of filters were: Filter 0: no filter, Filter 1: The two most distant spectra per class have been removed. Filter 2: The average distance for each spectrum of a class was calculated, and the most distant spectrum was removed. Filter 3: The score for each spectrum within a class was calculated using a combination of majority voting and weighting, and the spectrum with the lowest score was removed. Petition 870210119564, dated 12 / 22 / 2021, pp. 56 / 84 43 / 46 - Filter 4: the spectrum with the greatest distance to the centroid was removed.
[0076] Filters were used to clean two datasets of NIR spectra to predict a material class, specifically a feed and / or feed raw material. The two datasets contained spectra measured on two different infrared spectrometers, a Foss NIRS™ DS2500 Feed Analyzer and a Bruker MPA FT-NIR Analyzer or TANGO FTNIR Analyzer. Four different filters were applied to the datasets, i.e., criteria were calculated and the most distant spectra identified or the spectra with the lowest score in a majority voting and weighting procedure were removed from the dataset. Then, the criteria for the filter were recalculated and applied again until 20% of the spectra were removed from the datasets.
[0077] After applying the filters, a nearest neighbor search was performed to predict the material code for a set of query spectra. In detail, 20 runs were performed with 200 random query spectra, where the number of times a material code was correctly predicted among the 200 queries was counted.
[0078] The results for the dataset with spectra measured on a Foss NIRS™DS2500 Feed Analyzer are summarized in Table 1 and the results for the dataset with spectra measured on a Bruker MPA FT-NIR Analyzer or TANGO FT-NIR Analyzer are summarized in Table 2. Execution filter 0 filter 1 filter 2 filter 3 filter 4 1 197 198 199 198 199 2 195 196 195 198 195 Petition 870210119564, dated 12 / 22 / 2021, pp. 57 / 84 44 / 46 3 198 198 198 198 198 4 196 197 197 200 197 5 198 199 199 199 199 6 196 196 195 198 195 7 196 197 197 198 197 8 196 196 197 197 197 9 200 200 200 200 200 10 197 197 197 197 197 11 198 199 199 200 199 12 196 197 197 197 197 13 199 200 200 200 200 14 194 194 194 193 194 15 195 197 197 198 197 16 197 198 198 196 198 17 199 199 199 198 199 18 197 198 197 198 197 19 196 197 197 198 197 20 195 197 197 199 197 Average 196.75 197.50 197.45 198 197.45 Table 1: Results for cleaning a dataset with spectra measured on a Foss NIR analyzer.
[0079] Each of the filters 1 through 4 in cleaning the dataset generally leads to an improvement in the prediction of a material code, compared to filter 0. The results for filter 1, filter 2, and filter 4 are nearly identical. The use of filter 3 led to the best improvement in prediction results with the dataset of spectra measured on a Foss NIR analyzer than the other filters. Specifically, the use of filter 3 gave better Petition 870210119564, dated 12 / 22 / 2021, pages 58 / 84 45 / 46 results compared to the rest of the filters in 8 out of 20 runs. Filter execution: 0 filter 1 filter 2 filter 3 filter 4 1 191 193 193 198 193 2 194 195 195 198 195 3 196 196 196 196 196 4 192 194 194 198 194 5 197 198 199 199 199 6 194 194 195 197 196 7 194 194 194 198 194 8 191 194 194 199 194 9 191 194 194 197 194 10 190 190 192 195 192 11 197 198 197 199 197 12 195 195 197 198 197 13 197 197 197 200 197 14 197 198 197 199 197 15 196 197 195 198 196 16 194 196 195 197 195 17 196 197 198 198 198 18 196 196 196 197 196 19 197 197 197 198 197 20 195 195 195 197 195 Average 194.5 195.4 195.5 197.8 195.6 Table 2: Results for cleaning a dataset with spectra measured with a Bruker NIR analyzer.
[0080] Each of the filters 1 through 4 in cleaning the dataset generally leads to an improvement in predicting a material code, compared to filter 0. The results for filter 1, filter 2, and filter 4 are almost Petition 870210119564, dated 12 / 22 / 2021, pp. 59 / 84 46 / 46 identical. Again, the use of filter 3 led to the best improvement in prediction results with the dataset of spectra measured on a Bruker NIR analyzer than the other filters. Here, the improvement with filter 3 was even stronger than with the dataset of spectra measured on a Bruker NIR. Specifically, the use of filter 3 gave better results in improvement than the rest of the filters in 17 out of 20 runs. This is also expressed by the significantly higher average value for filter 3 than in the other cases.
[0081] Summarizing all options for removing spectral outliers from the vector database according to the present invention led to an improvement in the prediction of a material code, compared to the case where no spectral outliers were removed. Generally, the use of filter 3, a combination of majority voting and weighting for removing spectral outliers from the dataset, gave the best improvement in prediction results for a material code. Furthermore, the results are independent of the specific NIR device on which the NIR spectra of the datasets were measured. Petition 870210119564, dated 12 / 22 / 2021, pages 60 / 84
Claims
1 / 5 CLAIMS 1. A computer-implemented method for predicting a feed and / or feed raw material, characterized in that it comprises the steps of: a) providing a near-infrared spectrum of a sample of an unknown feed raw material and / or feed, b) transforming absorption intensities of wavelengths or wavenumbers in the spectrum from step a) to provide a query vector, c) providing a set of database vectors from a population of spectra of known feed raw materials and / or feeds, wherein an outlier is removed from the set of database vectors, wherein step c) further comprises one or more of options c1) to c3): c1) removing a pair of database vectors being the most different from each other in a set of database vectors from said set of database vectors,comprising the steps of: c1a) calculating a similarity measure and / or a distance measure from each database vector in a set of database vectors to the other database vectors in said set of database vectors to give similarity values of pairs of database vectors, c1b) sorting the similarity values obtained in step c1a) in descending order when a similarity measure is calculated in step c1a), or in ascending order when a distance measure is calculated in step c1a), wherein, in either case, the sorted similarity value at the end relates to the two database vectors being the most different from each other, and c1c) removing in pairs at least the two database vectors with the lowest ranking in step c1b) from the set of database vectors, Petition 870240087633, dated 11 / 10 / 2024,pg. 11 / 21 2 / 5 c2) remove a database vector being on average the most different from the other database vectors in a set of database vectors from said set of database vectors, comprising the steps of: c2a) calculating a similarity measure and / or a distance measure of each database vector in a set of database vectors to the other database vectors in said set of database vectors to give similarity values of each of a database vector to the other database vectors, c2b) forming the sum of the similarity values obtained for each database vector in step c2a) and calculating the average similarity value for each database vector, c2c) sorting the average similarity values obtained in step c2b) in descending order, when a similarity measure is calculated in step c2b), or in ascending order,when a distance measure is calculated in step c2b), wherein, in any case, the average similarity value ranked at the end relates to the database vector being the most different, on average, from all other database vectors, and c2d) remove the database vector with the lowest ranking in step c2c) from the set of database vectors, c3) remove a database vector being the most different from the centroid of a set of database vectors from said set of database vectors, comprising the steps of: c3a) determining the centroid of all database vectors in a set of database vectors, c3b) calculating a similarity measure and / or a distance measure from each database vector to the centroid from step c3a) to give a similarity value for each database vector to the centroid, Petition 870240087633, dated 11 / 10 / 2024,pg. 12 / 21 3 / 5 c3c) classify the similarity values obtained in step c3b) in descending order when a similarity measure is calculated in step c3b), or in ascending order when a distance measure is calculated in step c3b), where, in either case, the similarity value ranked at the end relates to the database vector being the most different from the centroid, and c3d) remove at least the database vector with the lowest ranking in step c3c) from the set of database vectors, d) calculate a similarity measure and / or a distance measure between the query vector from step b) and each database vector from step c) to provide a similarity value for each database vector with the query vector, e) classify the similarity values obtained in step d) in descending order when a similarity measure is calculated in step d) or in ascending order,when a distance measure is calculated in step d), where, in any case, the best-ranked database vector has the highest similarity to the query vector, ef) assign the feed raw material and / or feed from the database vector with the highest similarity in step e) to the sample from step a).
2. A computer-implemented method according to claim 1, characterized in that a database vector with a similarity value of 0 is removed from the set of database vectors in steps c1b), c2c) and / or c3c).
3. A computer-implemented method according to claim 1 or 2, characterized in that the vector in steps b) and c) is a multidimensional vector, with each dimension corresponding to an absorption intensity of a specific wavelength or wavenumber.
4. A computer-implemented method, according to any Petition 870240087633, dated 11 / 10 / 2024, page 13 / 21 4 / 5, one of claims 1 to 3, characterized in that a corresponding outlier spectrum is removed from the infrared spectra of known feed raw materials and / or feeds that are to be transformed into the set of database vectors and steps c1), c2) and / or c3), preferably steps c1a) to c1c), c2a) to c2c) and / or c3a) to c3e) are performed with the infrared spectrum of a population of known feed raw materials and / or feeds.
5. A computer-implemented method according to any one of claims 1 to 4, characterized in that in step b) and / or c) the absorption intensities of equidistant wavelengths or wavenumbers in a spectrum are transformed to give a spectrum vector in step b) and / or c).
6. A computer-implemented method, according to any one of claims 1 to 5, characterized in that the distances of the absorption intensities being transformed into vectors in step b) are identical to the distances of the absorption intensities transformed into vectors in step c).
7. A computer-implemented method, according to any one of claims 1 to 6, characterized in that the population of spectra of feed raw materials and / or feed products known from step c) comprises at least 50 sample spectra of each feed raw material and / or feed product from each of its global growing areas.
8. A computer-implemented method, according to any one of claims 1 to 7, characterized in that step e) comprises the steps of: e1) counting the number of occurrences of each of the feed raw materials and / or feed among the best-ranked database vectors in the classification of step e), where said number of occurrences is indicated by the variable N, and e2) weighting the first N similarity values of each of the feed raw materials and / or feed according to their Petition 870240087633, dated 11 / 10 / 2024, page 1. 14 / 21 5 / 5 position in the ranking of stage e1) to give weighted ranking positions for each of the feed raw materials and / or feeds, and e3) to form the sum of the weighted ranking positions of stage e2) for the feed raw materials and / or feeds to give scores for each of the feed raw materials and / or feeds, where the highest score indicates the greatest similarity.
9. A computer-implemented method according to any one of claims 1 to 8, characterized in that step a) comprises recording a near-infrared spectrum of a sample of an unknown feed raw material and / or feed.
10. System for predicting a feed raw material and / or feed, characterized in that it comprises a processing unit adapted to perform the computer-implemented method as defined in any one of claims 1 to 8, and further comprises a database comprising i) a set of database vectors of a population of spectra of known feed raw materials and / or feeds and / or ii) a population of spectra of known feed raw materials and / or feeds, wherein the set of database vectors and / or the population of spectra is free of outliers.
11. System according to claim 10, characterized in that it further comprises a near-infrared spectrometer for carrying out the computer-implemented method as defined in any one of claims 1 to 9. Petition 870240087633, dated 11 / 10 / 2024, pp. 15 / 21