Tea polyphenol quantitative prediction method based on hyperspectral image

Abstract

The invention belongs to the technical field of tea polyphenol prediction, particularly relates to a tea polyphenol quantitative prediction method based on a hyperspectral image, and aims to solve the problem of low tea polyphenol prediction accuracy caused by single tea polyphenol characterization parameter in a tea polyphenol measurement method in the prior art. According to the method, the tea polyphenol can be predicted by inputting the hyperspectral image of the tea into the prediction model. The step of training the prediction model comprises the steps of collecting a hyperspectral image of tea, screening spectral band features in the hyperspectral image and image fusion features fused by color features and texture features, obtaining a tea polyphenol sample, and training the prediction model according to the spectral band features, the image fusion features and the tea polyphenol sample. According to the method, the hyperspectral image of the tea is collected, and the spectral band characteristics of the tea and the texture and color characteristics of the tea image are extracted, so that the problem of inaccurate tea polyphenol content prediction caused by only single characteristic in the prior art can be solved, and high-precision tea polyphenol content prediction is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of tea polyphenol prediction, and in particular relates to a method for quantitatively predicting tea polyphenols in tea leaves based on hyperspectral images. Background Art

[0002] Tea polyphenols are one of the main bioactive components in tea, playing a crucial role in its growth. They not only participate in photosynthesis and growth regulation, but also enhance the tea's resilience to stress, combating pests and diseases and environmental stress. This allows tea plants to survive and grow better in harsh environments, thereby improving tea yield and quality. Therefore, tea polyphenol content has become a key indicator for assessing tea quality, and understanding tea polyphenol content throughout the tea's growth process has become crucial.

[0003] Existing methods for measuring tea polyphenols in tea include: First, by analyzing the spectrum of captured tea images, using the spectral information to establish a relationship model between the spectral information and the tea polyphenol content, thereby predicting the tea polyphenol content in the tea leaves. Second, by analyzing image information (such as tea color) in captured tea images, using this information to establish a relationship model between the image information and the tea polyphenol content, thereby predicting the tea polyphenol content in the tea leaves. However, these methods all select a single parameter to characterize the tea polyphenol content in tea leaves, while the factors that affect the content of tea polyphenols are multifaceted, which can easily lead to low accuracy in tea polyphenol content prediction. Summary of the Invention

[0004] The present invention provides a method for quantitatively predicting tea polyphenols in tea leaves based on hyperspectral images, aiming to solve the problem that tea polyphenols characterization parameters in the existing tea polyphenols measurement methods are single, resulting in low accuracy in tea polyphenols prediction.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] In a first aspect, the present invention provides a method for quantitatively predicting tea polyphenols in tea leaves based on hyperspectral images, comprising: inputting a collected hyperspectral image of tea leaves into a trained prediction model to obtain a predicted value of the tea polyphenol content of the tea leaves;

[0007] The training steps of the prediction model are as follows:

[0008] S100, collects hyperspectral images of tea leaves;

[0009] S200, screening the spectral band characteristics of tea leaves in the hyperspectral image;

[0010] S300, obtaining image fusion features of the hyperspectral image, including:

[0011] S301, obtaining a tea polyphenol sample from tea leaves;

[0012] S302, segmenting a target area in the hyperspectral image;

[0013] S303, extracting the image texture index parameter set of the target area, and performing correlation analysis between the image texture index parameter set and the tea polyphenol sample, screening out image texture index parameters whose correlation meets the set standard, and obtaining texture features;

[0014] S304, extracting the image color index parameter set of the target area, and performing correlation analysis between the image color index parameter set and the tea polyphenol sample, screening out image color index parameters whose correlation meets a set standard, and obtaining color features;

[0015] S305: Fusing the texture feature and the color feature to obtain an image fusion feature;

[0016] S400 , training a prediction model based on the spectral band features obtained from the hyperspectral images of all selected tea leaves, the image fusion features, and the tea polyphenol samples in step S301 to obtain a trained prediction model.

[0017] A further solution: in the step S100, a tea area to be photographed is selected, and a hyperspectral camera is used to photograph the tea canopy in the selected tea area to obtain the hyperspectral image of the tea.

[0018] Based on this approach, the hyperspectral camera can image the entire tea canopy, capturing its holistic information. This not only provides images with high spatial resolution, facilitating surface analysis of the tea leaves, but also helps capture spectral data reflected from different wavelengths.

[0019] A further solution: in step S301, tea buds are randomly sampled in the selected tea area, and the tea polyphenol content of the buds is measured to obtain the tea polyphenol sample.

[0020] Based on the above scheme, the young buds of tea usually contain higher tea polyphenols, so collecting young bud samples is helpful to obtain representative tea polyphenol sample data.

[0021] A further solution: before performing step S200, the hyperspectral image obtained in step S100 is calibrated using a white reference image and a black reference image; the calibration is performed using the following formula:

[0022]

[0023] Wherein, R0 represents the hyperspectral image obtained in step S100; R c represents the calibrated hyperspectral image; W represents the white reference image; and B represents the black reference image.

[0024] Based on the above scheme, the calibration process can eliminate the effects of uneven illumination and other noise on the hyperspectral image.

[0025] Further solution: In step S200, the step of screening the spectral band characteristics includes:

[0026] S201, obtaining an average spectrum curve of the calibrated hyperspectral image;

[0027] S202, selecting 100 bands from the average spectrum curve using a random forest algorithm;

[0028] S203 , using a continuous projection algorithm to extract a 556 nm band, a 668 nm band, a 710 nm band, and a 949 nm band from the 100 bands as the screened spectral band features.

[0029] Based on the above scheme, the random forest algorithm and the continuous projection algorithm are used in turn to perform step-by-step screening, which can effectively reduce the redundant information of spectral band characteristics and retain the most representative spectral bands with high correlation with tea polyphenols.

[0030] A further solution: in the step S303, an image texture index parameter set of the target area is extracted based on LBP and GLCM; the image texture index parameter set includes a Mean parameter, a Variance parameter, a Std parameter, a Homogeneity parameter, a Contrast parameter, a Dissimilarity parameter, an Entropy parameter, a Second moment parameter, and a Correlation parameter;

[0031] In step S304, the image color index parameter set of the target area is extracted based on the color histogram and the pseudo-color image; the image color index parameter set includes RGB parameters, GSV parameters, L*a*b parameters, GRVI parameters, GLI parameters, MGRVI parameters, NRLI parameters, NGLI parameters, ARVI parameters, ExR parameters, ExG parameters, ExGR parameters and INT parameters.

[0032] Further solution: In the step S303, the method of performing correlation analysis on the image texture index parameter set and the tea polyphenol sample is: constructing a least square regression relationship between the image texture index parameter set and the tea polyphenol sample to perform correlation analysis;

[0033] In step S304, the method of performing correlation analysis on the image color index parameter set and the tea polyphenol sample is to construct a least square regression relationship between the image color index parameter set and the tea polyphenol sample to perform correlation analysis.

[0034] Based on the above scheme, by constructing a least squares regression equation, the regression coefficient of each image texture index parameter or each image color index parameter in predicting the tea polyphenol content can be obtained, thereby evaluating the correlation of each index parameter and obtaining the texture feature or color feature with a high correlation with tea polyphenols, which is relatively simple and convenient.

[0035] Further solution: The setting criteria in step S303 and step S304 are both: the absolute value of the correlation coefficient is greater than 0.8 and the P value is less than 0.05.

[0036] Further solution: In S400, the prediction model adopts a long short-term memory network model.

[0037] A further solution, in said S400, further includes using a validation set to validate the trained prediction model.

[0038] Based on the above scheme, using a validation set for validation can help improve the accuracy, stability, and generalization ability of the prediction model, making the final prediction model more accurate and reliable.

[0039] The beneficial effects of the present invention are:

[0040] The present invention provides a method for quantitatively predicting tea polyphenols in tea leaves based on hyperspectral images, which simultaneously includes the spectral band characteristics, texture and color characteristics of tea leaves. By screening the spectral band texture characteristics, optical features in the hyperspectral images that are highly correlated with tea polyphenols can be captured. By analyzing the texture and color characteristics, the tea polyphenol content can be indirectly characterized. The combination of the two can make up for the problem of inaccurate tea polyphenol content prediction caused by only including a single feature in the existing technology, and achieve high-precision tea polyphenol content prediction.

[0041] The spectral band characteristics can reflect the sensitivity of light to the substances in tea leaves that are most highly correlated with tea polyphenols, thereby establishing a relationship with tea polyphenols. This can compensate for the influence of light when predicting tea polyphenols based solely on texture and color characteristics. Furthermore, the texture and color characteristics can better characterize the growth of tea leaves and have a high correlation with tea polyphenols, compensating for the inaccurate prediction of tea polyphenol content caused by varying tea photography angles or distances between the camera and the tea leaves when relying solely on spectral band characteristics. The prediction model described in the present invention can analyze tea spectral images from multiple aspects and angles, thereby more accurately predicting tea polyphenol content. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For those skilled in the art in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0043] Figure 1 It is a flowchart of the training steps of the prediction model in the present invention;

[0044] Figure 2 Schematic diagram of the process of obtaining fusion features in step S300 of the present invention;

[0045] Figure 3 2 is a flow chart of screening spectral band characteristics in step S200 of the present invention;

[0046] Figure 4 Schematic diagram of the average spectrum curve in step S201 of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0048] This embodiment provides a method for quantitatively predicting tea polyphenols in tea leaves based on hyperspectral images, comprising: inputting the collected hyperspectral images of tea leaves into a trained prediction model to obtain a predicted value of the tea polyphenol content of the tea leaves;

[0049] like Figure 1 and Figure 2 As shown, the training steps of the prediction model are as follows:

[0050] S100, collects hyperspectral images of tea leaves;

[0051] S200, screening the spectral band characteristics of tea leaves in the hyperspectral image;

[0052] S300, obtaining image fusion features of the hyperspectral image, including:

[0053] S301, obtaining a tea polyphenol sample from tea leaves;

[0054] S302, segmenting a target area in the hyperspectral image;

[0055] S303, extracting the image texture index parameter set of the target area, and performing correlation analysis between the image texture index parameter set and the tea polyphenol sample, screening out image texture index parameters whose correlation meets the set standard, and obtaining texture features;

[0056] S304, extracting the image color index parameter set of the target area, and performing correlation analysis between the image color index parameter set and the tea polyphenol sample, screening out image color index parameters whose correlation meets a set standard, and obtaining color features;

[0057] S305: Fusing the texture feature and the color feature to obtain an image fusion feature;

[0058] S400 , training a prediction model based on the spectral band features obtained from the hyperspectral images of all selected tea leaves, the image fusion features, and the tea polyphenol samples in step S301 to obtain a trained prediction model.

[0059] A more specific example of step S100 in the above solution is: in step S100, a tea area to be photographed is selected, and a hyperspectral camera is used to photograph the tea canopy in the selected tea area to obtain the hyperspectral image of the tea.

[0060] The equipment used to photograph the tea canopy can also serve as a sensor for detecting the growth of field crops. The shooting time can be selected between 10:00 and 14:00 in clear, cloudless weather.

[0061] Based on the above solution, before performing step S200, the hyperspectral image obtained in step S100 is first calibrated using a white reference image and a black reference image; the calibration process is performed using the following formula:

[0062]

[0063] Wherein, R0 represents the hyperspectral image obtained in step S100; R crepresents the calibrated hyperspectral image; W represents the white reference image; and B represents the black reference image.

[0064] Specifically, the white reference image serves as a reflectance reference for the entire band, representing the maximum reflectance under ideal conditions; wherein the white reference image can be obtained using a polytetrafluoroethylene plate. The black reference image represents the minimum reflectance (i.e., background noise) under ideal conditions; wherein the black reference image can be obtained by covering the lens of the hyperspectral camera with a completely opaque black cover.

[0065] A more specific example of step S200 in the above solution is: Figure 3 As shown, in step S200, the step of screening the spectral band characteristics includes:

[0066] S201, obtaining an average spectrum curve of the calibrated hyperspectral image;

[0067] S202, selecting 100 bands from the average spectrum curve using a random forest algorithm;

[0068] S203 , using a continuous projection algorithm to extract a 556 nm band, a 668 nm band, a 710 nm band, and a 949 nm band from the 100 bands as the screened spectral band features.

[0069] A further example of the above solution is: in step S201, the average spectrum curve is as follows Figure 4 As shown in the Figure 4 In the figure, the horizontal axis represents the wavelength (in nanometers), and the vertical axis represents the reflectivity. The 556nm, 668nm, 710nm, and 949nm wavelengths are all located where significant spectral changes occur, demonstrating strong correlation or differentiation in spectral reflectance.

[0070] In step S202, the random forest algorithm is an ensemble learning algorithm of decision trees, and the steps include: receiving the average spectral curve and the target variable (for example, the variety of tea, the content of tea polyphenols, etc.); constructing multiple decision trees with the random forest, each tree randomly selecting part of the spectral bands as candidate bands, and scoring the importance of these candidate bands to the target variable; and selecting the top 100 bands according to the ranking of the scores.

[0071] In step S203, the steps of the successive projection algorithm (SPA) include: further screening the 100 bands screened by the random forest algorithm, removing bands containing redundant information, and obtaining bands without redundant information; randomly selecting a band from the bands without redundant information as a starting point, calculating the geometric distance (projection component) between the band without redundant information and the randomly selected band, and selecting the band with the smallest correlation with the randomly selected band; and then continuously iterating to finally screen out the 556nm band, the 668nm band, the 710nm band, and the 949nm band.

[0072] A more specific example of step S301 in the above scheme is: in step S301, tea buds are randomly sampled in a selected tea area, and the tea polyphenol content of the buds is measured to obtain the tea polyphenol sample.

[0073] Specifically, tea buds (fresh leaves of one-bud, two-leaf tea) are randomly picked from a selected tea area and brought back to the laboratory. The tea polyphenol content of the buds can be measured using the national standard method GB / T8305-2013. The steps for measuring tea polyphenols in the buds using the national standard method GB / T8305-2013 include: extracting with a 70% methanol aqueous solution in a 70°C water bath, oxidizing the -OH groups in the tea polyphenols with a folinol reagent to produce a blue color, measuring the absorbance at a wavelength of 765nm, and using gallic acid as a calibration standard to quantify the tea polyphenols, thereby obtaining a tea polyphenol sample.

[0074] The step S302 segments the target area in the hyperspectral image, and a threshold segmentation method can be used to segment the tea leaf area in the hyperspectral image. The operation can be performed on Python.

[0075] A more specific example of step S303 in the above solution is as follows: in step S303, an image texture index parameter set of the target area is extracted based on LBP and GLCM; the image texture index parameter set includes a Mean parameter, a Variance parameter, a Std parameter, a Homogeneity parameter, a Contrast parameter, a Dissimilarity parameter, an Entropy parameter, a Second moment parameter, and a Correlation parameter;

[0076] Among them, the extraction process based on the LBP (local binary pattern) is as follows: randomly select the grayscale value of a central pixel in the target area and compare it with the grayscale values ​​of its surrounding pixels (the grayscale value of the surrounding pixels is counted as 1 if it is higher than the grayscale value of the central pixel, otherwise it is counted as 0), and the comparison results are arranged in a binary number; the binary number is converted into a decimal number to obtain the LBP value of the central pixel; the LBP values ​​of all pixels in the target area are counted to obtain an LBP feature map (a total of 16 feature values), and the image texture index parameter set in the LBP feature map is extracted.

[0077] The extraction process based on the GLCM (Gray Level Co-occurrence Matrix) is as follows: quantize the grayscale image of the target area into N grayscale levels (for example, 16 or 32 levels); define the spatial relationship between pixels, calculate each direction and distance, count the frequency of occurrence of adjacent pixel grayscale value combinations, and generate the corresponding GLCM; extract the image texture index parameter set from the GLCM in each direction.

[0078] The Mean parameter represents the average frequency of each eigenvalue in the LBP feature map. The Variance parameter represents the range of variation in the frequency of different LBP values. The Std (standard deviation) parameter is the square root of the variance. The Homogeneity parameter measures the size of the element values ​​near the main diagonal of the GLCM. The Contrast parameter measures the contrast between grayscale values. The Dissimilarity parameter measures the degree of difference in the grayscale values ​​of adjacent pixels. The Entropy parameter measures the randomness of the grayscale distribution. The Second moment parameter represents the uniformity of the GLCM. The Correlation parameter measures the linear correlation of the grayscale values ​​between pixels.

[0079] In addition, a least squares regression equation is constructed between the image texture index parameter set and the tea polyphenol sample to perform correlation analysis; image texture index parameters whose correlation meets a set standard are screened out to obtain texture features. The set standard is: the absolute value of the correlation coefficient is greater than 0.8 and the P value is less than 0.05.

[0080] Specifically, the least squares regression equation is obtained by fitting a linear regression model to minimize the sum of squared errors between the predicted value and the true value. The main steps include: establishing a linear regression model, extracting regression coefficients, predicting tea polyphenol content, calculating image texture indices with low correlation with the tea polyphenol sample, and retaining image texture indices with significant correlation, which are referred to as the texture features.

[0081] It should be noted that step S303 can be implemented using Python.

[0082] A more specific example of step S304 in the above scheme is: in step S304, the image color index parameter set of the target area is extracted based on the color histogram and the pseudo-color image; the image color index parameter set includes RGB parameters, GSV parameters, L*a*b parameters, GRVI parameters, GLI parameters, MGRVI parameters, NRLI parameters, NGLI parameters, ARVI parameters, ExR parameters, ExG parameters, ExGR parameters and INT parameters.

[0083] Specific wavelengths (e.g., infrared, green, and blue) are selected from the hyperspectral image of the target area and mapped into three channels: R (red), G (green), and B (blue) to form a pseudo-color image. For each pixel in the target area of ​​the pseudo-color image, the frequency distribution of the pixel values ​​in the R, G, and B channels is calculated, forming three color histograms.

[0084] The RGB parameter represents the intensity values ​​of each pixel in the pseudo-color image in the R (red), G (green), and B (blue) channels. The GSV (Grayscale Value) parameter represents the pixel value after converting the RGB color space to a grayscale image. The L*a*b parameter (L*) represents brightness, ranging from [0, 100], with higher values ​​indicating higher brightness; a* represents the color distribution between green and red (positive values ​​indicate red, negative values ​​indicate green); and b* represents the color distribution between blue and yellow (positive values ​​indicate yellow, negative values ​​indicate blue). The GRVI (Green Normalized Vegetation Index) parameter measures the contrast between green and red in tea leaves. The GLI (Green Leaf Index) parameter measures the intensity of the green channel in tea leaves. The MGRVI (Modified Green Normalized Vegetation Index) parameter is used to reduce the effects of noise. The NRLI (Normalized Red-Green Index) parameter measures the contrast between red and green. The NGLI (Normalized Green-Blue Index) parameter measures the contrast between green and blue. The ARVI (Atmospheric Impedance Vegetation Index) parameter is a tea leaf index used to reduce atmospheric influences. The ExR (Extra Red Index) parameter emphasizes the red color in tea leaves. The ExG (Extra Green Index) parameter emphasizes the green color in tea leaves. The ExGR (Extra Green-Red Index) parameter represents the combination of the ExR and ExG parameters. The INT (Intensity) parameter measures the overall brightness of the image pixels.

[0085] In addition, a least squares regression equation is constructed between the image color index parameter set and the tea polyphenol sample to perform correlation analysis; image color index parameters whose correlation meets a set standard are screened out to obtain color features. The set standard is: the absolute value of the correlation coefficient is greater than 0.8 and the P value is less than 0.05.

[0086] It should be noted that the step of obtaining color features by constructing a least squares regression equation for correlation analysis is consistent with the step of obtaining texture features in step S303. Furthermore, step S304 can be implemented using Python.

[0087] A more specific example of step S400 in the above solution is that in S400, the prediction model adopts a long short-term memory network model.

[0088] The long short-term memory network model (LSTM model) is an open source model and is used as a prediction model in this embodiment. The spectral band features and the image fusion features obtained from the hyperspectral images of all selected tea leaves are used as variables, and the tea polyphenol samples in step S301 are used as dependent variables to train the prediction model.

[0089] On the basis of the above solution, in S400, the trained prediction model is further verified using a verification set.

[0090] In the step of training the prediction model, when collecting hyperspectral images of tea leaves in step S100, several hyperspectral images of tea leaves can be collected and divided into a training set and a validation set, with the ratio of 6:4, 7:3, or 8:2, etc. The hyperspectral images of the validation set are processed in the same steps as steps S200 to S305 for processing the hyperspectral images of the training set, namely, screening the spectral band features of tea leaves in the hyperspectral images of the validation set; obtaining image fusion features of the hyperspectral images of the validation set; and verifying the trained prediction model based on the spectral band features, image fusion features, and tea polyphenol samples obtained from the hyperspectral images of all tea leaves selected in the validation set.

[0091] The present invention is not limited to the above-mentioned optional implementation modes. Anyone can derive other forms of products under the inspiration of the present invention. However, no matter what changes are made in the shape or structure, any technical solution that falls within the scope defined by the claims of the present invention falls within the scope of protection of the present invention.

Claims

1. A method for quantitative prediction of tea polyphenols based on hyperspectral images, characterized in that: include: The collected hyperspectral images of tea leaves are input into the trained prediction model to obtain the predicted value of tea polyphenol content of the tea leaves; The training steps of the prediction model are as follows: S100, collects hyperspectral images of tea leaves; S200, screening the spectral band characteristics of tea leaves in the hyperspectral image; S300, obtaining image fusion features of the hyperspectral image, including: S301, obtaining tea polyphenol samples from tea leaves; S302, segmenting a target area in the hyperspectral image; S303, extracting the image texture index parameter set of the target area, and performing correlation analysis between the image texture index parameter set and the tea polyphenol sample, screening out image texture index parameters whose correlation reaches a set standard, and obtaining texture features; S304, extracting the image color index parameter set of the target area, and performing correlation analysis between the image color index parameter set and the tea polyphenol sample, screening out image color index parameters whose correlation reaches a set standard, and obtaining color features; S305, fusing the texture feature and the color feature to obtain an image fusion feature; S400, training a prediction model based on the spectral band features obtained from the hyperspectral images of all selected tea leaves, the image fusion features and the tea polyphenols samples in step S301 to obtain a trained prediction model.

2. The method for quantitative prediction of tea polyphenols based on hyperspectral images according to claim 1, characterized in that: In the step S100, a tea area to be photographed is selected, and a hyperspectral camera is used to photograph the tea canopy in the selected tea area to obtain the hyperspectral image of the tea.

3. The method for quantitative prediction of tea polyphenols based on hyperspectral images according to claim 2, characterized in that: In step S301, tea buds are randomly sampled in the selected tea area, and the tea polyphenol content of the tea buds is measured to obtain the tea polyphenol sample.

4. The method for quantitative prediction of tea polyphenols based on hyperspectral images according to claim 1, characterized in that: Before performing step S200, the hyperspectral image obtained in step S100 is calibrated using a white reference image and a black reference image; the calibration process is performed using the following formula: Wherein, R0 represents the hyperspectral image obtained in step S100; R c represents the calibrated hyperspectral image; W represents the white reference image; and B represents the black reference image.

5. The method for quantitative prediction of tea polyphenols based on hyperspectral images according to claim 4, characterized in that: In step S200, the step of screening the spectral band characteristics includes: S201, obtaining an average spectrum curve of the calibrated hyperspectral image; S202, selecting 100 bands from the average spectrum curve using a random forest algorithm; S203, using a continuous projection algorithm to extract a 556 nm band, a 668 nm band, a 710 nm band and a 949 nm band from the 100 bands as the screened spectral band features.

6. The method for quantitative prediction of tea polyphenols based on hyperspectral images according to claim 1, characterized in that: In the step S303, an image texture index parameter set of the target area is extracted based on LBP and GLCM; the image texture index parameter set includes a Mean parameter, a Variance parameter, a Std parameter, a Homogeneity parameter, a Contrast parameter, a Dissimilarity parameter, an Entropy parameter, a Second moment parameter and a Correlation parameter; In step S304, the image color index parameter set of the target area is extracted based on the color histogram and the pseudo-color image; the image color index parameter set includes RGB parameters, GSV parameters, L*a*b parameters, GRVI parameters, GLI parameters, MGRVI parameters, NRLI parameters, NGLI parameters, ARVI parameters, ExR parameters, ExG parameters, ExGR parameters and INT parameters.

7. The method for quantitative prediction of tea polyphenols based on hyperspectral images according to claim 1, characterized in that: In the step S303, the method of performing correlation analysis between the image texture index parameter set and the tea polyphenol sample is: constructing a least square regression relationship between the image texture index parameter set and the tea polyphenol sample to perform correlation analysis; In the step S304, the method of performing correlation analysis between the image color index parameter set and the tea polyphenol sample is: constructing a least square regression relationship between the image color index parameter set and the tea polyphenol sample to perform correlation analysis.

8. The method for quantitative prediction of tea polyphenols based on hyperspectral images according to claim 7, characterized in that: The setting criteria in step S303 and step S304 are both: the absolute value of the correlation coefficient is greater than 0.8 and the P value is less than 0.

05.

9. The method for quantitative prediction of tea polyphenols based on hyperspectral images according to claim 1, characterized in that: In S400, the prediction model adopts a long short-term memory network model.

10. The method for quantitative prediction of tea polyphenols based on hyperspectral images according to claim 1, characterized in that: In the S400, it also includes using a validation set to validate the trained prediction model.

Images