Spectral data processing method and device of tea pile and electronic equipment

Abstract

The invention provides a spectral data processing method and device of a tea pile and electronic equipment, and is applied to the technical field of rapid detection of field tea bud quality, the method comprises the following steps: inputting a tea pile mixed spectrum into a bilinear spectrum model to obtain a front leaf spectrum, the bilinear spectrum model being used for correcting the tea pile mixed spectrum; derivation is carried out on the front leaf spectrum of the tea pile to obtain a derivative set of the front leaf spectrum, and derivation is used for eliminating tea bud stacking structure information of the tea pile; the derivative set is input into a Gaussian process regression model adopting a radial basis kernel function, the front leaf quality is obtained, the radial basis kernel function in the Gaussian process regression model is used for eliminating front leaf proportion information of the tea pile, and the Gaussian process regression model is used for predicting the front leaf quality; according to the method, the influence of the tea bud stacking structure in the tea pile mixed spectrum and the proportion difference of the front leaves on the accurate inversion of the tea bud quality can be eliminated.

Description

Technical Field

[0001] The invention relates to the technical field of rapid detection of tea bud quality in the field, and in particular to a method, a device and an electronic device for processing spectral data of a tea pile. Background Art

[0002] Traditional chemical methods for tea bud quality testing are not only time-consuming but also highly destructive. With the rapid development of hyperspectral technology in recent years, narrow bands can capture subtle changes in absorption characteristics and achieve accurate estimation of the content of physiological and biochemical components of vegetation, but the establishment of these estimation models relies on the measurement of leaf spectra.

[0003] The tea buds of small-leaf tea trees are very small and have curled leaf edges. The spectral information of a single tea bud is weak and existing non-imaging spectrometers find it difficult to measure the spectrum of a single tea bud leaf. Therefore, only the mixed spectrum of a tea pile composed of tea buds can be measured; however, the mixed spectrum of a tea pile will be affected by the stacking structure of the tea buds in the tea pile, making it difficult for the mixed spectrum of the tea pile to represent the spectrum of the leaves.

[0004] It can be seen that the tea bud quality detection method in the related art has a technical problem that the quality of the tea bud leaves in the tea pile is difficult to measure. Summary of the invention

[0005] The present invention provides a spectral data processing method, device and electronic equipment for a tea pile, so as to solve the defect that the quality of tea bud leaves in a tea pile is difficult to measure in the prior art tea bud quality detection method, and eliminate the influence of the tea bud stacking structure and the difference in the proportion of front leaves in the mixed spectrum of the tea pile on the accurate inversion of the tea bud quality.

[0006] The present invention provides a method for processing spectral data of a tea pile, comprising the following steps.

[0007] A tea pile mixed spectrum of a tea pile is obtained, wherein the tea pile mixed spectrum is obtained by collecting spectral data of the tea pile based on a measuring wavelength, the tea pile is composed of randomly stacked tea buds and leaves, and the tea buds and leaves include: front leaves and back leaves; the tea pile mixed spectrum is input into a bilinear spectral derivative Gaussian process regression model to obtain a target front leaf quality output by the bilinear spectral derivative Gaussian process regression model, the bilinear spectral derivative Gaussian process regression model includes: a bilinear spectral model and a Gaussian process regression model using a radial basis kernel function, wherein the tea pile mixed spectrum is input into the bilinear spectral model to obtain a target front leaf quality output by the bilinear spectral derivative Gaussian process regression model. The front leaf spectrum of the tea pile is output by the bilinear spectral model, wherein the bilinear spectral model is used to correct the mixed spectrum of the tea pile; the front leaf spectrum of the tea pile is differentiated to obtain a derivative set of the front leaf spectrum, wherein the differentiation is used to eliminate the tea bud stacking structure information of the tea pile; the derivative set is input into the Gaussian process regression model using the radial basis kernel function to obtain the target front leaf quality output by the Gaussian process regression model using the radial basis kernel function, wherein the radial basis kernel function in the Gaussian process regression model is used to eliminate the front leaf proportion information of the tea pile, and the Gaussian process regression model is used to predict the target front leaf quality.

[0008] According to a spectral data processing method for a tea pile provided by the present invention, before inputting the mixed spectrum of the tea pile into a bilinear spectral derivative Gaussian process regression model to obtain the target front leaf quality output by the bilinear spectral derivative Gaussian process regression model, the method also includes: obtaining front leaf spectrum samples of the front tea bud leaf samples and back leaf spectrum samples of the back tea bud leaf samples; determining a first linear relationship between the front leaf spectrum samples and the back leaf spectrum samples based on the front leaf spectrum samples and the back leaf spectrum samples; obtaining a tea pile mixed spectrum sample of the tea pile sample, tea bud stacking structure information of the tea pile sample and front leaf proportion information of the tea pile sample, wherein the tea pile sample is obtained by stacking the front tea bud leaf sample and the back tea bud leaf sample; determining a second linear relationship between the front leaf spectrum sample, the back leaf spectrum sample and the tea pile mixed spectrum sample based on the tea bud stacking structure information and the front leaf proportion information.

[0009] According to a method for processing spectral data of a tea pile provided by the present invention, the method further comprises: constructing a bilinear spectral model based on the first linear relationship and the second linear relationship; wherein the bilinear spectral model refers to the following formula:

[0010] in, represents the front leaf spectrum, Indicates the proportion of the front leaves of the tea pile, represents the first constant, represents the tea pile mixed spectrum, represents the second constant, Represents the tea bud stacking structure information of the tea pile.

[0011] According to a spectral data processing method of a tea pile provided by the present invention, the derivative of the front leaf spectrum of the tea pile is derived to obtain a derivative set of the front leaf spectrum, including: according to the bilinear spectral model, the measurement wavelength is derived based on the front leaf spectrum to obtain a derivative set of the front leaf spectrum, referring to the following formula: ;

[0012] in, represents the set of derivatives of the front leaf spectrum, represents the diagonal matrix of the front leaf proportion, represents the measurement wavelength, represents the set of derivatives of the tea pile mixture spectrum, represents the frontal leaf spectrum, Indicates the proportion of the front leaves of the tea pile, represents the first constant, represents the tea pile mixed spectrum, Represents the tea bud stacking structure information of the tea pile.

[0013] According to a method for processing spectral data of a tea pile provided by the present invention, before inputting the derivative set into the Gaussian process regression model using a radial basis kernel function to obtain the target front leaf quality output by the Gaussian process regression model using a radial basis kernel function, the method further includes: Acquire a verification spectrum and quality data set and a training spectrum and quality data set, wherein the verification spectrum and quality data set includes a derivative set of the front leaf spectrum of the verification tea pile and a quality data set of the verification tea pile; the training spectrum and quality data set includes a derivative set of the front leaf spectrum of the training tea pile and a quality data set of the training tea pile; Based on the derivative set of the front leaf spectra of the verification tea pile and the derivative set of the front leaf spectra of the training tea pile, the similarity covariance function between the derivative set of the front leaf spectra of the verification tea pile and the derivative set of the front leaf spectra of the training tea pile is determined, referring to the following formula:

[0014] in, represents the similarity covariance function, represents the set of derivatives of the front leaf spectra of the validation tea bunch, represents the set of derivatives of the front leaf spectra of the training tea bunches, represents the scaling parameter, represents the diagonal matrix of the front leaf proportion, It represents the set of derivatives of the mixed spectrum of tea piles. represents the set of spectral derivatives of the training tea pile mixture, represents the length scale parameter of the kernel function; represents the noise standard deviation, Represents the Kronecker symbol.

[0015] According to a method for processing spectral data of a tea pile provided by the present invention, the method also includes: determining a mean function of a derivative set of the front leaf spectra; and constructing a Gaussian process regression model based on the mean function and the similarity covariance function.

[0016] The present invention also provides a spectral data processing device for tea piles, comprising the following modules: A spectral acquisition module is used to obtain a mixed spectrum of a tea pile, wherein the mixed spectrum of the tea pile is obtained by collecting spectral data of the tea pile based on a measurement wavelength, and the tea pile is composed of randomly stacked tea buds and leaves, and the tea buds and leaves include: front leaves and back leaves; a spectral processing module is used to input the mixed spectrum of the tea pile into a bilinear spectral derivative Gaussian process regression model to obtain a target front leaf quality output by the bilinear spectral derivative Gaussian process regression model, and the bilinear spectral derivative Gaussian process regression model includes: a bilinear spectral model and a Gaussian process regression model using a radial basis kernel function, wherein, the mixed spectrum of the tea pile is input into the bilinear spectral model to obtain The front leaf spectrum of the tea pile is output by the bilinear spectral model, wherein the bilinear spectral model is used to correct the mixed spectrum of the tea pile; the front leaf spectrum of the tea pile is differentiated to obtain a set of derivatives of the front leaf spectrum, wherein the differentiation is used to eliminate the tea bud stacking structure information of the tea pile; the derivative set is input into the Gaussian process regression model using the radial basis kernel function to obtain the target front leaf quality output by the Gaussian process regression model using the radial basis kernel function, wherein the radial basis kernel function in the Gaussian process regression model is used to eliminate the front leaf proportion information of the tea pile, and the Gaussian process regression model is used to predict the target front leaf quality.

[0017] The present invention also provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the spectral data processing method of any one of the tea piles described above is implemented.

[0018] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the spectral data processing method of any one of the tea piles described above is implemented.

[0019] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for processing spectral data of a tea pile as described above is implemented.

[0020] The spectral data processing method, device and electronic device of a tea pile provided by the present invention can collect spectral data of a tea pile composed of randomly stacked tea buds and leaves (including front leaves and back leaves) by measuring wavelength, so as to obtain a mixed spectrum containing the overall spectral information of the tea pile; the mixed spectrum of the tea pile is input into a bilinear spectral model, and the bilinear spectral model can convert the mixed spectrum of the tea pile into a front leaf spectrum by correcting the mixed spectrum; a derivative set of the front leaf spectrum is obtained by deriving the front leaf spectrum, thereby eliminating the influence of the tea bud stacking structure information in the tea pile on the spectral data; the derivative set is input into a Gaussian process regression model, thereby further eliminating the influence of the front leaf proportion information in the tea pile on the spectral data; through the prediction ability of Gaussian process regression, a more accurate target front leaf quality can be obtained, thereby solving the technical problem that the quality of tea buds and leaves in a tea pile is difficult to measure in a tea bud quality detection method in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 It is a schematic diagram comparing tea buds of small-leaf tea trees and large-leaf tea trees provided by the present invention.

[0023] Figure 2 It is a schematic diagram of the influencing factors of the mixed spectrum of tea piles provided by the present invention.

[0024] Figure 3 It is a schematic flow chart of the spectral data processing method of tea pile provided by the present invention.

[0025] Figure 4 It is a schematic diagram of the accuracy evaluation of the bilinear spectral derivative Gaussian process regression model provided by the present invention.

[0026] Figure 5 It is a schematic structural diagram of the spectral data processing device for tea piles provided by the present invention.

[0027] Figure 6 It is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] Green tea, one of the most popular beverages, is made from fresh tea buds with one bud and one leaf picked from tea trees. The quality of tea depends on the content of tea polyphenols, amino acids and soluble sugars, so the quantitative evaluation of quality parameters in tea buds is a key factor in guiding the picking of tea buds and producing high-quality green tea. Traditional chemical methods for tea bud quality detection are not only time-consuming but also highly destructive. With the rapid development of hyperspectral technology in recent years, narrow bands can capture subtle changes in absorption characteristics and achieve accurate estimation of the content of physiological and biochemical components of vegetation, but the establishment of these estimation models relies on the measurement of leaf spectra.

[0030] refer to Figure 1 , Figure 1 The present invention provides a comparative schematic diagram of tea buds of small-leaf tea trees and large-leaf tea trees, including: (a) large-leaf tea trees and (b) small-leaf tea trees.

[0031] like Figure 1 As shown, the tea buds of small-leaf tea trees are very small and have curled leaf edges. The spectral information of a single tea bud is weak and it is difficult for existing non-imaging spectrometers to measure the spectrum of a single tea bud leaf. Therefore, only the mixed spectrum of the tea pile composed of tea buds can be measured.

[0032] refer to Figure 2 , Figure 2 Schematic diagram of factors affecting the mixed spectrum of tea piles provided by the present invention, including: (a) stacking structure and (b) field of view ratio.

[0033] like Figure 2As shown in the figure, the accuracy of tea polyphenols, amino acids and soluble sugar content in tea buds inverted based on the mixed spectrum of tea piles is affected by the stacking structure of tea buds in the tea pile and the proportion of the front and back of tea buds in the field of view. Linear Spectral Mixing Model (LSMM) is a model widely used in remote sensing data analysis. This model assumes that the reflectance spectrum of the entire canopy is a linear combination of the vegetation and soil spectra in the field of view. It has made significant progress in many aspects such as land feature classification, vegetation cover inversion, water quality, mineral content and aerosol monitoring. In addition, machine learning algorithms have achieved remarkable success in estimating physiological and biochemical parameters of vegetation using hyperspectral data. In particular, kernel function machine learning algorithms can map data to high-dimensional space, thereby converting problems that were originally nonlinearly separable in the input space into linearly separable forms. This enables it to effectively process data with complex and nonlinear relationships.

[0034] In summary, the development of LSMM and machine learning methods provides a possibility for eliminating the influence of the stacking structure and the difference in the proportion of the front and back sides of the tea buds in the field of view on the accurate inversion of the tea bud quality in the mixed spectrum of the tea pile.

[0035] In the related art, the tea buds of small-leaf tea trees are very small and have curled leaf edges. The average leaf length is 2.7 cm and the average leaf width is 1.3 cm. The spectral information of a single tea bud is weak and it is difficult for existing non-imaging spectrometers to measure the spectrum of a single tea bud leaf. The mixed spectrum of a tea pile will be affected by the stacking structure of the tea buds in the tea pile, especially in the near-infrared region where the high spectral transmittance is more susceptible to the stacking structure (such as stacking thickness), resulting in the mixed spectrum of the tea pile being difficult to represent the spectrum of the leaves. The mixed spectrum of a tea pile will be affected by the proportion of front and back leaves in the field of view, especially in the visible light region where the reflectivity of the back of the tea buds is higher than the reflectivity of the front because the back of the tea buds are covered with white hairs.

[0036] According to a spectral data processing method for a tea pile provided by the present invention, a bilinear spectral derivative Gaussian process regression model (BSDGPR) is constructed by combining a linear spectral mixture model (LSMM) and a Gaussian process regression model (GPR). The influence of the stacking structure information and the proportion of the front and back sides of the tea buds in the field of view in the mixed spectrum of the tea pile is eliminated, and the mixed spectrum of the tea pile is corrected to a leaf spectrum, thereby realizing rapid, non-destructive and high-precision tea bud quality monitoring in the field.

[0037] The method of the present invention is to directly and quickly detect the quality of fresh tea buds by using a bilinear spectral derivative Gaussian process regression model for the spectrum of fresh tea buds picked directly in the field, which can help determine the most suitable time for picking high-quality tea buds in the field so that they can be picked in time.

[0038] The present invention is different from the method in the related art: in the related art, after the tea buds are picked, there will be multiple processes in the tea processing factory to process the fresh tea buds, and it is necessary to monitor the impact of each process on the quality of the tea buds, and finally determine the process for producing high-quality tea products (such as precise control of fermentation time, temperature, etc.).

[0039] The coupled linear spectral mixture model and the first-order and second-order derivative methods eliminated the influence of the stacking structure of tea buds in the tea pile on the mixed spectrum of the tea pile. By scaling the length scale parameter in the square exponential covariance kernel function of the GPR model, the influence of the proportion of the front and back of the tea buds in the field of view on the mixed spectrum of the tea pile was eliminated. The mixed spectrum of the tea pile was corrected to the target front leaf quality (i.e., the tea bud spectrum), which improved the accuracy of rapid and non-destructive monitoring of fresh tea buds in the field.

[0040] Optionally, the spectral data processing method of the tea pile in the embodiment of the present application can be executed by a server, or by a terminal device, or jointly by a server and a terminal device, taking the spectral data processing method of the tea pile in the present embodiment executed by a terminal device as an example.

[0041] Figure 3 Schematic diagram of the process of the spectral data processing method of tea pile provided by the present invention, such as Figure 3 As shown, the method includes the following steps.

[0042] Step 301, obtaining a tea pile mixed spectrum of a tea pile.

[0043] The mixed spectrum of the tea pile is obtained by collecting spectral data of the tea pile based on the measurement wavelength, and the tea pile is composed of randomly stacked tea buds and leaves, and the tea buds and leaves include: front leaves and back leaves.

[0044] In an embodiment of the present invention, during the tea bud picking period each year (usually March-April), tea buds with one bud and one leaf are picked in the field, and the collected tea buds are randomly piled on a pure black non-reflective cloth (the diameter of the tea pile is about 5 cm and the thickness is about 2 cm), and the tea pile mixed spectrum of the tea pile is recorded 11-12 cm above the tea pile using an ASD handheld spectrometer calibrated with a white board within a target time period (e.g., 30 minutes) at a measurement wavelength of a target frequency.

[0045] In some embodiments, the tea pile needs to be tested for tea bud quality. The tea bud quality test first requires that each tea bud sample be withered after measuring the spectral data, and then placed in an oven at 80°C to dry to constant weight, and then weighed and ground into powder using an electric grinder for quality testing. The content of tea polyphenols is determined by ferrous tartrate colorimetry and spectrophotometry, the content of free amino acids is determined by ninhydrin colorimetry at 570 nanometers, and the content of soluble sugars is determined by anthrone reagent method. Each sample is tested three times, and the average value is taken as the final value.

[0046] Step 302, inputting the mixed spectrum of the tea pile into a bilinear spectral derivative Gaussian process regression model to obtain the target front leaf quality output by the bilinear spectral derivative Gaussian process regression model.

[0047] The bilinear spectral derivative Gaussian process regression model includes: a bilinear spectral model and a Gaussian process regression model using a radial basis kernel function.

[0048] In an embodiment of the present invention, a bilinear spectral derivative Gaussian process regression model is constructed by combining a linear spectral mixture model and a Gaussian process regression model. The bilinear spectral derivative Gaussian process regression model is used to eliminate the influence of the stacking structure information and the proportion of the front and back sides of the tea buds in the field of view in the mixed spectrum of the tea pile, and correct the mixed spectrum of the tea pile to the target front leaf spectrum, thereby realizing rapid, lossless and high-precision tea bud quality monitoring in the field.

[0049] The above step 302 specifically includes the following steps.

[0050] Step 3021, inputting the mixed spectrum of the tea pile into the bilinear spectrum model to obtain the front leaf spectrum of the tea pile output by the bilinear spectrum model, wherein the bilinear spectrum model is used to correct the mixed spectrum of the tea pile.

[0051] Since the physiological and biochemical parameters of the same tea bud are the same and the leaf cell structure is similar, there is a linear relationship between the spectrum of the front leaf and the spectrum of the back leaf. Therefore, the mixed spectrum of the tea pile can be converted into the spectrum of the front leaf of the tea pile through linear conversion.

[0052] According to a spectral data processing method for tea piles provided by the present invention, before inputting the mixed spectrum of the tea pile into a bilinear spectral derivative Gaussian process regression model to obtain the target front leaf quality output by the bilinear spectral derivative Gaussian process regression model, the method further includes: Obtaining a front leaf spectrum sample of a front tea bud leaf sample and a back leaf spectrum sample of a back tea bud leaf sample; Based on the front leaf spectrum sample and the back leaf spectrum sample, determining a first linear relationship between the front leaf spectrum sample and the back leaf spectrum sample; Obtaining a tea pile mixed spectrum sample of a tea pile sample, tea bud stacking structure information of the tea pile sample, and front leaf ratio information of the tea pile sample, wherein the tea pile sample is obtained by stacking a front tea bud leaf sample and a back tea bud leaf sample; Based on the stacking structure information of tea buds and the proportion of front leaves, the second linear relationship between the front leaf spectral samples, the back leaf spectral samples and the tea pile mixed spectral samples is determined.

[0053] In some embodiments, from the front tea bud leaf sample, a spectrometer or other related equipment is used to measure and record its spectral data. These data will constitute the front leaf spectrum sample of the front leaf. Similarly, from the back tea bud leaf sample, the same equipment or method is used to measure and record its spectral data. These data will constitute the back leaf spectrum sample of the back leaf.

[0054] A statistical method (such as linear regression analysis) is used to analyze the linear relationship between the front leaf spectrum sample and the back leaf spectrum sample as the first linear relationship.

[0055] From the tea pile sample obtained by stacking the front tea bud leaf sample and the back tea bud leaf sample, the mixed spectrum data are measured and recorded using a spectrometer, which will constitute the tea pile mixed spectrum sample.

[0056] Through physical observation or image analysis of tea pile samples, the arrangement of tea buds, stacking density and other structural characteristics can be determined as tea bud stacking structure information. Through sampling analysis of tea pile samples or using image recognition technology to estimate the proportion of front leaves and back leaves in the tea pile, the proportion of front leaves can be used as information.

[0057] It should be noted that the tea bud stacking structure information and the front leaf ratio information can also be used as preset unknown variables and eliminated through subsequent mathematical operations. In actual application, the tea bud stacking structure information and the front leaf ratio information of the tea pile are unknown variables.

[0058] Based on the stacking structure information of tea buds and the proportion of front leaves, the second linear relationship between the front leaf spectral samples, the back leaf spectral samples and the tea pile mixed spectral samples is determined.

[0059] In the embodiment of the present invention, the bilinear spectral mixture model assumes that the reflectance spectrum of the entire canopy is a linear combination of the vegetation and soil in the field of view, and the tea piles are randomly arranged with the front side of the fresh tea buds facing the leaves and the back side facing the leaves, so the mixed spectrum of the tea pile ( ) can be obtained from the front leaf spectrum ( ) and the back leaf spectrum ( ), the formula is shown in (1).

[0060] (1) In the formula, represents the tea pile mixed spectrum (sample), , represent the spectrum (sample) of the front leaf and the spectrum (sample) of the back leaf, respectively. Indicates the proportion of the front leaves in the sensor field of view (the proportion of the front of the tea bud to the total area of ​​the tea pile, such as Figure 2 As shown in (b) in the figure, therefore, , Indicates the tea bud stacking structure information of the tea pile (such as Figure 2 As shown in (a) in the figure), Represents noise or random error.

[0061] Since the physiological and biochemical parameters of the same tea buds are the same and the leaf cell structure is similar, there is a linear relationship between the spectra of the front and back of the tea buds, as shown in the formula (2).

[0062] (2) in, , represent the front leaf spectrum (sample) and the back leaf spectrum (sample), respectively, and m and n are the slope and intercept of the linear model, respectively.

[0063] Through the embodiment of the present invention, multiple linear equations are fitted by statistical methods, which can describe the relationship between the mixed spectrum sample of the tea pile and the spectrum samples of the front leaf and the spectrum samples of the back leaf, so as to facilitate the subsequent correction of the mixed spectrum of the tea pile to the spectrum of the front leaf.

[0064] According to a method for processing spectral data of a tea pile provided by the present invention, the method further comprises: Based on the first linear relationship and the second linear relationship, a bilinear spectral model is constructed; The bilinear spectral model refers to the following formula:

[0065] in, represents the front leaf spectrum, Indicates the proportion of leaves on the front side of the tea pile. represents the first constant, Indicates the tea pile mixed spectrum, represents the second constant, Represents the tea bud stacking structure information of the tea pile.

[0066] In the embodiment of the present invention, the above formula (2) and formula (1) are used to construct a bilinear spectral model, which is used to convert the mixed spectrum of tea piles into Corrected to the front leaf spectrum , the calculation formula is shown in (3).

[0067] (3) It can be seen from formula (3) that the spectrum of the front leaf is affected by the stacking structure information (SI) of the tea buds and the proportion information of the front leaf (f).

[0068] Through the embodiments of the present invention, the mixed spectrum of the tea pile can be corrected to the front leaf spectrum through a bilinear spectral model.

[0069] Step 3022, deriving the spectrum of the front leaves of the tea pile to obtain a set of derivatives of the spectrum of the front leaves, wherein the derivation is used to eliminate the tea bud stacking structure information of the tea pile.

[0070] According to a spectral data processing method of a tea pile provided by the present invention, the spectrum of the front leaves of the tea pile is derived to obtain a derivative set of the spectrum of the front leaves, including: According to the bilinear spectral model, the measured wavelength is derived based on the front leaf spectrum to obtain the derivative set of the front leaf spectrum, referring to the following formula:

[0071] in, represents the set of derivatives of the front leaf spectrum, represents the diagonal matrix of the front leaf proportion, represents the measurement wavelength, represents the set of derivatives of the mixed spectrum of the tea pile, represents the front leaf spectrum, Indicates the proportion of leaves on the front side of the tea pile. represents the first constant, Indicates the tea pile mixed spectrum, Represents the tea bud stacking structure information of the tea pile.

[0072] Because the proportion of positive leaves in each tea pile It has nothing to do with the measurement wavelength, so the wavelength After derivation, it can be expressed by formula (4).

[0073] (4) Among them, a is the first constant, Not dependent on measurement wavelength Since most of the light energy in the visible region is absorbed and the transmission is very small, it is almost unaffected by the structure. In the near-infrared region, although it is affected by the structure, the SI changes very little with the measured wavelength because the transmittance is consistent.

[0074] Therefore, the present invention calculates the first-order derivative (FD) and the second-order derivative (SD) so that and Very small and negligible, thus eliminating the tea bud stacking structure information of the tea pile ( ) on the spectrum of the front leaf. Formula (4) is simplified to formula (5): (5) In the formula, represents the set of derivatives of the front leaf spectrum; The set of derivatives representing the mixed spectrum of tea piles includes the first-order derivative ( ) and the second-order derivative ( ).

[0075] in, , , m=1,2,…,n, represents the derivative of the spectrum of the mth front leaf, represents the derivative of the mixed spectrum of the mth tea pile, represents the set of real numbers, Represents the dimension, represents the total number of derivatives, a is the first constant, Represents the front leaf ratio parameter (used to adjust the weight), represents the proportion parameter of the front leaves of the mth tea pile, The front leaf ratio information of regulation, Indicates the proportion of front leaves in the mth tea pile.

[0076] Through the embodiment of the present invention, the front leaf spectrum is affected by the tea bud stacking structure information and the front leaf ratio information, and the front leaf ratio information of each tea pile is independent of the measurement wavelength. Therefore, by derivatizing the front leaf spectrum with respect to the measurement wavelength and calculating the first-order derivative and the second-order derivative, the influence of the tea bud stacking structure information on the front leaf spectrum is eliminated.

[0077] Step 3023, input the derivative set into a Gaussian process regression model using a radial basis kernel function, and obtain the target front leaf quality output by the Gaussian process regression model using a radial basis kernel function, wherein the radial basis kernel function in the Gaussian process regression model is used to eliminate the front leaf ratio information of the tea pile, and the Gaussian process regression model is used to predict the target front leaf quality.

[0078] The target front leaf quality is used for subsequent evaluation of the tea bud quality of a tea pile (tea tree), where multiple tea bud leaves in the tea pile all belong to the same tea tree.

[0079] According to a method for processing spectral data of a tea pile provided by the present invention, the method further comprises: determining the mean function of the set of derivatives of the obverse leaf spectra; A Gaussian process regression model is constructed based on the mean function and similarity covariance function.

[0080] Gaussian Process Regression (GPR) is a Bayesian method for general regression problems using kernel learning. It can map data to a high-dimensional space, thereby transforming a problem that is originally nonlinearly separable in the input space into a linearly separable form. Specifically, GPR takes the input data as input. and output variables between( represents the i-th input data, Represents the i-th output variable) to establish the form expressed by Formula 6-8, the formula is as follows: (6) in, Representation function Follow Gaussian process , mean function Describes the function At any point Expected value; covariance function (kernel function) Describes the function At any two points The correlation between them.

[0081] In a Gaussian process, the mean function is usually known and can be any function, such as a linear function, a polynomial function, etc. The covariance function determines how the values ​​of the function at different points depend on each other. Common kernel functions include the squared exponential kernel (RBF kernel), the polynomial kernel, the Matern kernel, etc.

[0082] The present invention constructs a bilinear spectral derivative Gaussian process regression model (BSDGPR) based on the bilinear spectral model and the GPR model: (7) in, represents the set of derivatives of the front leaf spectrum, , They respectively represent the set of derivatives of the front leaf spectra of the verification tea pile and the set of derivatives of the front leaf spectra of the training tea pile.

[0083] According to the embodiments of the present invention, the Gaussian process can flexibly model various complex functional relationships through the mean function and the covariance function.

[0084] According to a method for processing spectral data of a tea pile provided by the present invention, before inputting the derivative set into a Gaussian process regression model using a radial basis kernel function to obtain a target front leaf quality output by the Gaussian process regression model using a radial basis kernel function, the method further comprises: Acquire a verification spectrum and quality data set and a training spectrum and quality data set, wherein the verification spectrum and quality data set includes a derivative set of the front leaf spectrum of the verification tea pile and a quality data set of the verification tea pile; the training spectrum and quality data set includes a derivative set of the front leaf spectrum of the training tea pile and a quality data set of the training tea pile; Based on the derivative set of the front leaf spectra of the verification tea pile and the derivative set of the front leaf spectra of the training tea pile, the similarity covariance function between the derivative set of the front leaf spectra of the verification tea pile and the derivative set of the front leaf spectra of the training tea pile is determined, referring to the following formula:

[0085] in, represents the similarity covariance function, represents the set of derivatives of the positive leaf spectra of the verified tea pile, represents the set of derivatives of the front leaf spectra of the training tea piles, represents the scaling parameter, represents the diagonal matrix of the front leaf proportion, It represents the set of derivatives of the mixed spectrum of tea piles. represents the set of spectral derivatives of the training tea pile mixture, represents the length scale parameter of the kernel function; represents the noise standard deviation, Represents the Kronecker symbol.

[0086] The present invention adopts a scaled square exponential covariance function, and the formula is shown in (8).

[0087] (8) In the formula, is the scaling parameter, , is the length scale parameter of the kernel function, which is the way the training information is propagated along the input dimension. is the noise standard deviation, is the Kronecker symbol.

[0088] The diagonal matrix in the present invention is a positive definite matrix, Equivalent to is linearly scaled and does not change its relative position, so the length scale parameter of the covariance function Under the regulation of Not affected by the proportion of front leaves Therefore, the front blade ratio is eliminated. The impact on model accuracy is that the mixed spectrum of the tea pile is finally converted into the spectrum of fresh front leaves.

[0089] and They are respectively a set consisting of the derivatives of the front leaf spectrum and a set consisting of the derivatives of the tea pile mixed spectrum.

[0090] In some embodiments, the GPR model is trained using a set of training tea pile mixed spectral derivatives. And the covariance function defined above, train the Gaussian process regression model During training, the model will learn the covariance function parameters that best fit the data (e.g. , , ).

[0091] Model prediction: Use the trained GPR model ( ) and verify the tea pile mixed spectrum derivative set The model will output the predicted value of the quality of the verified tea pile , thereby providing an accurate prediction of the quality of the tea bunch.

[0092] refer to Figure 4 , Figure 4 It is a schematic diagram of the accuracy evaluation of the bilinear spectral derivative Gaussian process regression model provided by the present invention.

[0093] like Figure 4 As shown, the accuracy of predicting the contents of tea polyphenols, amino acids and soluble sugars by using the bilinear spectral derivative Gaussian process regression model (BSDGPR) for the mixed spectrum of tea piles was improved. The bilinear spectral derivative Gaussian process regression model included a bilinear spectral derivative Gaussian process regression model (BSDGPR-FD) using the first-order derivative (FD) and a bilinear spectral derivative Gaussian process regression model (BSDGPR-SD) using the second-order derivative (SD).

[0094] Figure 4 The horizontal axis represents the measured values ​​of quality parameters, and the vertical axis represents the estimated values ​​of the quality parameters of the target front leaf quality based on the BSDGPR model. The quality parameters shown refer to the content of tea polyphenols, amino acids and soluble sugars. Figure 4 The middle (a) figure shows the prediction accuracy of tea polyphenols content by inputting the mixed spectrum of tea pile into the BSDGPR-FD model, with a determination coefficient (R2) of 0.81 and a root mean square error (RMSE) of 1.61; Figure 4The middle (b) figure shows the prediction accuracy of tea polyphenols content by inputting the mixed spectrum of tea pile into the BSDGPR-SD model, with an R2 of 0.75 and a RMSE of 1.90; Figure 4 The middle (c) figure shows the prediction accuracy of amino acid content by inputting the mixed spectrum of tea pile into the BSDGPR-FD model, with an R2 of 0.70 and a RMSE of 0.30; Figure 4 The middle (d) figure shows the prediction accuracy of amino acid content by inputting the mixed spectrum of tea pile into the BSDGPR-SD model, with an R2 of 0.66 and a RMSE of 0.36; Figure 4 The middle (e) figure shows the prediction accuracy of soluble sugar content by inputting the mixed spectrum of tea pile into the BSDGPR-FD model, with an R2 of 0.91 and a RMSE of 0.54; Figure 4 The middle figure (f) shows the prediction accuracy of soluble sugar content by inputting the mixed spectrum of tea pile into the BSDGPR-SD model, with an R2 of 0.90 and an RMSE of 0.57.

[0095] In summary, this paper proposes a model for rapid non-destructive field measurement of fresh leaf quality of small-leaf tea trees (BSDGPR). First, the mixed spectrum of the tea pile is converted into the front leaf spectrum based on the bilinear spectral model, and the tea bud stacking structure information is eliminated by the derivative and GPR methods. and the proportion of front leaves The impact of this has enabled rapid, non-destructive and accurate monitoring of tea bud quality in the field.

[0096] Through the above-mentioned embodiments of the present invention, by coupling the linear spectral mixture model and the Gaussian process regression model, the mixed spectrum of the randomly stacked tea pile of fresh tea buds of small-leaf tea trees is corrected to the spectrum of the front leaves, eliminating the influence of the stacking structure and the proportion of the front and back sides of the tea buds in the field of view, and has the following advantages.

[0097] Eliminated the influence of stacking structure: The influence of stacking structure on the mixed spectrum of tea piles caused by high spectral transmittance and low absorptivity in the near-infrared region when the tea buds are stacked is solved. Eliminated the influence of the proportion of front and back tea buds in the field of view: The kernel function eliminates the influence of high reflectivity of the fluff on the back of the tea buds in the visible light region on the mixed spectrum of the tea pile. Rapid and non-destructive field monitoring: The fusion of the linear spectral mixture model and the Gaussian process regression model corrects the spectrum of the tea pile to the spectrum of tea buds of equal size, realizing rapid and non-destructive monitoring of the quality of fresh tea buds in the field.

[0098] The bilinear spectral derivative Gaussian process regression model of the embodiment of the present invention overcomes the difficulty of existing non-imaging spectrometers in acquiring the spectrum of a single tea bud due to the small tea buds of small-leaf tea trees, and solves the problem that the mixed spectrum of the tea pile is disturbed by the stacking structure and the proportion of the front and back sides of the tea leaves in the field of view, thereby providing a new method for rapid, non-destructive and accurate monitoring of small-leaf tea trees in the field.

[0099] The spectral data processing device for a tea pile provided by the present invention is described below. The spectral data processing device for a tea pile described below and the spectral data processing method for a tea pile described above can be referred to each other.

[0100] refer to Figure 5 , Figure 5 It is a schematic structural diagram of the spectral data processing device for tea piles provided by the present invention.

[0101] The spectrum acquisition module 501 is used to obtain a mixed spectrum of a tea pile, wherein the mixed spectrum of the tea pile is obtained by collecting spectrum data of the tea pile based on the measurement wavelength, and the tea pile is composed of randomly stacked tea buds and leaves, and the tea buds and leaves include: front leaves and back leaves; The spectrum processing module 502 is used to input the mixed spectrum of the tea pile into the bilinear spectrum derivative Gaussian process regression model to obtain the target front leaf quality output by the bilinear spectrum derivative Gaussian process regression model. The bilinear spectrum derivative Gaussian process regression model includes: a bilinear spectrum model and a Gaussian process regression model using a radial basis kernel function, which includes: Inputting the mixed spectrum of the tea pile into the bilinear spectrum model to obtain the front leaf spectrum of the tea pile output by the bilinear spectrum model, wherein the bilinear spectrum model is used to correct the mixed spectrum of the tea pile; The front leaf spectrum of the tea pile is derivatized to obtain a derivative set of the front leaf spectrum, wherein the derivation is used to eliminate the tea bud stacking structure information of the tea pile; The derivative set is input into a Gaussian process regression model using a radial basis kernel function to obtain the target front leaf quality output by the Gaussian process regression model using a radial basis kernel function, wherein the radial basis kernel function in the Gaussian process regression model is used to eliminate the front leaf proportion information of the tea pile, and the Gaussian process regression model is used to predict the target front leaf quality.

[0102] Specifically, the spectral data processing device for the above-mentioned tea pile provided by the present invention can implement all the method steps implemented by the above-mentioned spectral data processing method embodiment for the tea pile, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as those of the method embodiment will not be described in detail here.

[0103] Figure 6 is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as Figure 6As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the spectral data processing method of the tea pile, the method comprising: obtaining a tea pile mixed spectrum of the tea pile, wherein the tea pile mixed spectrum is obtained by collecting spectral data of the tea pile based on the measurement wavelength, the tea pile is composed of randomly stacked tea buds and leaves, and the tea buds and leaves include: front leaves and back leaves; inputting the tea pile mixed spectrum into a bilinear spectral derivative Gaussian process regression model to obtain the target front leaf quality output by the bilinear spectral derivative Gaussian process regression model, the bilinear spectral derivative Gaussian process regression model includes: a bilinear spectral model and a Gaussian process regression model using a radial basis kernel function, wherein, including: The mixed spectrum of the tea pile is input into a bilinear spectral model to obtain the front leaf spectrum of the tea pile output by the bilinear spectral model, wherein the bilinear spectral model is used to correct the mixed spectrum of the tea pile; the front leaf spectrum of the tea pile is differentiated to obtain a set of derivatives of the front leaf spectrum, wherein the differentiation is used to eliminate the stacking structure information of the tea buds in the tea pile; the derivative set is input into a Gaussian process regression model using a radial basis kernel function to obtain the target front leaf quality output by the Gaussian process regression model using a radial basis kernel function, wherein the radial basis kernel function in the Gaussian process regression model is used to eliminate the front leaf proportion information of the tea pile, and the Gaussian process regression model is used to predict the target front leaf quality.

[0104] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0105] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the spectral data processing method of the tea pile provided by the above-mentioned methods, the method including: obtaining a tea pile mixed spectrum of the tea pile, wherein the tea pile mixed spectrum is obtained by collecting spectral data of the tea pile based on the measurement wavelength, and the tea pile is composed of randomly stacked tea buds and leaves, and the tea buds and leaves include: front leaves and back leaves; inputting the tea pile mixed spectrum into a bilinear spectral derivative Gaussian process regression model to obtain a target front leaf quality output by the bilinear spectral derivative Gaussian process regression model, and the bilinear spectral derivative Gaussian process regression model includes : A bilinear spectral model and a Gaussian process regression model using a radial basis kernel function, which includes: inputting the mixed spectrum of a tea pile into the bilinear spectral model to obtain the front leaf spectrum of the tea pile output by the bilinear spectral model, wherein the bilinear spectral model is used to correct the mixed spectrum of the tea pile; deriving the front leaf spectrum of the tea pile to obtain a set of derivatives of the front leaf spectrum, wherein the derivative is used to eliminate the tea bud stacking structure information of the tea pile; inputting the set of derivatives into a Gaussian process regression model using a radial basis kernel function to obtain the target front leaf quality output by the Gaussian process regression model using a radial basis kernel function, wherein the radial basis kernel function in the Gaussian process regression model is used to eliminate the front leaf proportion information of the tea pile, and the Gaussian process regression model is used to predict the target front leaf quality.

[0106] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the spectral data processing method of the tea pile provided by the above-mentioned methods, the method comprising: obtaining a tea pile mixed spectrum of the tea pile, wherein the tea pile mixed spectrum is obtained by collecting spectral data of the tea pile based on a measurement wavelength, the tea pile is composed of randomly stacked tea buds and leaves, and the tea buds and leaves include: front leaves and back leaves; inputting the tea pile mixed spectrum into a bilinear spectral derivative Gaussian process regression model to obtain a target front leaf quality output by the bilinear spectral derivative Gaussian process regression model, the bilinear spectral derivative Gaussian process regression model comprising: a bilinear spectral model and a radial basis kernel A Gaussian process regression model of a function, which includes: inputting a mixed spectrum of a tea pile into a bilinear spectrum model to obtain a front leaf spectrum of the tea pile output by the bilinear spectrum model, wherein the bilinear spectrum model is used to correct the mixed spectrum of the tea pile; derivatizing the front leaf spectrum of the tea pile to obtain a derivative set of the front leaf spectrum, wherein the derivative is used to eliminate the tea bud stacking structure information of the tea pile; inputting the derivative set into a Gaussian process regression model using a radial basis kernel function to obtain a target front leaf quality output by the Gaussian process regression model using a radial basis kernel function, wherein the radial basis kernel function in the Gaussian process regression model is used to eliminate the front leaf proportion information of the tea pile, and the Gaussian process regression model is used to predict the target front leaf quality.

[0107] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0108] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for processing spectral data of tea piles, characterized in that: include: Acquire a tea pile mixed spectrum of a tea pile, wherein the tea pile mixed spectrum is obtained by collecting spectral data of the tea pile based on a measurement wavelength, the tea pile is composed of randomly stacked tea buds and leaves, and the tea buds and leaves include: front leaves and back leaves; The mixed spectrum of the tea pile is input into a bilinear spectral derivative Gaussian process regression model to obtain the target front leaf quality output by the bilinear spectral derivative Gaussian process regression model, wherein the bilinear spectral derivative Gaussian process regression model includes: a bilinear spectral model and a Gaussian process regression model using a radial basis kernel function, wherein: Inputting the mixed spectrum of the tea pile into the bilinear spectrum model to obtain the spectrum of the front leaves of the tea pile output by the bilinear spectrum model, wherein the bilinear spectrum model is used to correct the mixed spectrum of the tea pile; Derivative the front leaf spectrum of the tea pile to obtain a derivative set of the front leaf spectrum, wherein the derivation is used to eliminate the tea bud stacking structure information of the tea pile; The derivative set is input into the Gaussian process regression model using the radial basis kernel function to obtain the target front leaf quality output by the Gaussian process regression model using the radial basis kernel function, wherein the radial basis kernel function in the Gaussian process regression model is used to eliminate the front leaf proportion information of the tea pile, and the Gaussian process regression model is used to predict the target front leaf quality.

2. The method for processing spectral data of tea pile according to claim 1, characterized in that: Before inputting the mixed spectrum of the tea pile into the bilinear spectral derivative Gaussian process regression model to obtain the target front leaf quality output by the bilinear spectral derivative Gaussian process regression model, the method further includes: Obtaining a front leaf spectrum sample of a front tea bud leaf sample and a back leaf spectrum sample of a back tea bud leaf sample; Based on the front blade spectrum sample and the back blade spectrum sample, determining a first linear relationship between the front blade spectrum sample and the back blade spectrum sample; Obtaining a tea pile mixed spectrum sample of a tea pile sample, tea bud stacking structure information of the tea pile sample, and front leaf ratio information of the tea pile sample, wherein the tea pile sample is obtained by stacking the front tea bud leaf sample and the back tea bud leaf sample; Based on the tea bud stacking structure information and the front leaf proportion information, a second linear relationship between the front leaf spectrum sample, the back leaf spectrum sample and the tea pile mixed spectrum sample is determined.

3. The method for processing spectral data of tea pile according to claim 2, characterized in that: The method further comprises: Based on the first linear relationship and the second linear relationship, construct a bilinear spectral model; The bilinear spectral model refers to the following formula: ； in, represents the front leaf spectrum, Indicates the proportion of the front leaves of the tea pile, represents the first constant, represents the tea pile mixed spectrum, represents the second constant, Represents the tea bud stacking structure information of the tea pile.

4. The method for processing spectral data of tea pile according to claim 3, characterized in that: The step of deriving the spectrum of the front leaves of the tea pile to obtain a set of derivatives of the spectrum of the front leaves includes: According to the bilinear spectral model, the measurement wavelength is derived based on the front blade spectrum to obtain a derivative set of the front blade spectrum, referring to the following formula: ； in, represents the set of derivatives of the front leaf spectrum, represents the diagonal matrix of the front leaf proportion, represents the measurement wavelength, represents the set of derivatives of the tea pile mixture spectrum, represents the frontal leaf spectrum, Indicates the proportion of the front leaves of the tea pile, represents the first constant, represents the tea pile mixed spectrum, Represents the tea bud stacking structure information of the tea pile.

5. The method for processing spectral data of tea pile according to claim 1, characterized in that: Before inputting the derivative set into the Gaussian process regression model using the radial basis kernel function to obtain the target front blade quality output by the Gaussian process regression model using the radial basis kernel function, the method further includes: Acquire a verification spectrum and quality data set and a training spectrum and quality data set, wherein the verification spectrum and quality data set includes a derivative set of the front leaf spectrum of the verification tea pile and a quality data set of the verification tea pile; the training spectrum and quality data set includes a derivative set of the front leaf spectrum of the training tea pile and a quality data set of the training tea pile; Based on the derivative set of the front leaf spectra of the verification tea pile and the derivative set of the front leaf spectra of the training tea pile, the similarity covariance function between the derivative set of the front leaf spectra of the verification tea pile and the derivative set of the front leaf spectra of the training tea pile is determined, referring to the following formula: ； in, represents the similarity covariance function, represents the set of derivatives of the front leaf spectra of the validation tea bunch, represents the set of derivatives of the front leaf spectra of the training tea bunches, represents the scaling parameter, represents the diagonal matrix of the front leaf proportion, It represents the set of derivatives of the mixed spectrum of tea piles. represents the set of spectral derivatives of the training tea pile mixture, represents the length scale parameter of the kernel function; represents the noise standard deviation, Represents the Kronecker symbol.

6. The method for processing spectral data of a tea pile according to claim 5, characterized in that: The method further comprises: determining a mean function of a set of derivatives of the front leaf spectrum; A Gaussian process regression model is constructed based on the mean function and the similarity covariance function.

7. A spectral data processing device for tea piles, characterized in that: include: A spectrum acquisition module is used to obtain a tea pile mixed spectrum of a tea pile, wherein the tea pile mixed spectrum is obtained by collecting spectral data of the tea pile based on a measurement wavelength, the tea pile is composed of randomly stacked tea buds and leaves, and the tea buds and leaves include: front leaves and back leaves; The spectrum processing module is used to input the mixed spectrum of the tea pile into a bilinear spectrum derivative Gaussian process regression model to obtain the target front leaf quality output by the bilinear spectrum derivative Gaussian process regression model, wherein the bilinear spectrum derivative Gaussian process regression model includes: a bilinear spectrum model and a Gaussian process regression model using a radial basis kernel function, which includes: Inputting the mixed spectrum of the tea pile into the bilinear spectrum model to obtain the spectrum of the front leaves of the tea pile output by the bilinear spectrum model, wherein the bilinear spectrum model is used to correct the mixed spectrum of the tea pile; Derivative the front leaf spectrum of the tea pile to obtain a derivative set of the front leaf spectrum, wherein the derivation is used to eliminate the tea bud stacking structure information of the tea pile; The derivative set is input into the Gaussian process regression model using the radial basis kernel function to obtain the target front leaf quality output by the Gaussian process regression model using the radial basis kernel function, wherein the radial basis kernel function in the Gaussian process regression model is used to eliminate the front leaf proportion information of the tea pile, and the Gaussian process regression model is used to predict the target front leaf quality.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method for processing spectral data of a tea pile according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the spectral data processing method of the tea pile according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the spectral data processing method of the tea pile according to any one of claims 1 to 6 is implemented.

Images