Vitis vinifera landmark identification method and system based on hyperspectral ensemble learning

Abstract

The invention discloses a grape landmark identification method and system based on hyperspectral ensemble learning, and relates to the technical field of grape landmark identification and monitoring. The grape landmark identification method based on hyperspectral ensemble learning comprises the following steps: extracting and monitoring spectral information; constructing and splitting a data set; constructing a feature set; and evaluating the prediction accuracy. According to the method, spectral information extraction and monitoring are carried out through grape landmark samples, then data set construction and data set splitting are carried out, then feature recognition and feature set construction are carried out to obtain a feature set, and finally SSC prediction model construction and Bayesian global optimization hyper-parameter are carried out based on a feature training set. According to the method, the effect of improving the accuracy of the SSC prediction and landmark accurate identification result of the Coketang grapes is achieved, and the problem of high deviation degree of the SSC prediction and landmark accurate identification result of the Coketang grapes caused by planting adjustment difference in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of grape landmark recognition and monitoring, and particularly to a method and system for grape landmark recognition based on hyperspectral integrated learning. Background Art

[0002] Keketerek grapes are landmark agricultural products in Xinjiang. The most important indicators of the quality and flavor of Keketerek grapes are SSC (Soluble Solids Content). The SSC of Keketerek grapes is usually higher than 18% (Brix degree), far exceeding that of ordinary grape varieties (12%-15%). However, due to the significant difference in SSC between Keketerek grape landmark products and non-landmark products, it significantly affects the nutritional quality of Keketerek grapes. Therefore, it is necessary to adopt detection technology to achieve the prediction of SSC of Keketerek grapes and the accurate identification of landmarks.

[0003] Existing methods mainly use a refractometer for measurement, and landmark identification usually uses isotope and elemental analysis methods. Fruit quality detection is carried out through hyperspectral imaging technology. Existing methods mainly focus on the single-task analysis of hyperspectral imaging technology, that is, single regression or classification tasks, and rarely perform multi-task analysis.

[0004] For example, the method and system for detecting the soluble solids content of grapes based on hyperspectral disclosed in the patent application with the publication number of CN118111955A includes: collecting hyperspectral images of grape clusters based on a hyperspectral imager, scanning a standard whiteboard for black and white correction to obtain grape hyperspectral images; constructing an instance segmentation model of grapes and performing instance segmentation processing on the grape hyperspectral images; constructing a prediction model for the soluble solids content of different grape varieties based on the grape hyperspectral images; predicting the soluble solids content of the instance-segmented grape hyperspectral images through the soluble solids content prediction model to obtain the prediction of the soluble solids content of grapes and comparing it with the experimental data, and analyzing the reasons for the cutting force error in combination with the model establishment process.

[0005] For example, a method for converting glucose models measured by different near-infrared instruments disclosed in the patent announcement with the announcement number of CN106872396B includes: 1) obtaining spectral data under two near-infrared instruments; 2) mathematical conversion between spectral data; 3) screening common wavelengths of the data of the two instruments; 4) calculating the converted spectral set; 5) constructing the converted model.

[0006] However, in the process of implementing the technical solutions of the present invention in the embodiments of the present application, it is found that the above technologies have at least the following technical problems:

[0007] In the prior art, due to the significant difference in SSC between the landmark products and non-landmark products of Keketerek grapes, the SSC of non-landmark products fluctuates greatly due to the limitations of planting conditions, resulting in differential interference in the SSC prediction and landmark precise identification of Keketerek grapes, and the problem of a high degree of deviation in the results of SSC prediction and landmark precise identification of Keketerek grapes caused by planting adjustment differences. Summary of the Invention

[0008] By providing a method and system for grape landmark identification based on hyperspectral ensemble learning in an embodiment of the present application, the problem of a high degree of deviation in the results of SSC prediction and landmark precise identification of Keketerek grapes caused by planting adjustment differences in the prior art is solved, and the accuracy of SSC prediction and landmark precise identification of Keketerek grapes is improved.

[0009] An embodiment of the present application provides a method for grape landmark identification based on hyperspectral ensemble learning, including the following steps: S1, extracting and monitoring spectral information according to the obtained grape landmark samples to obtain hyperspectral images and measuring SSC, where the grape landmark samples include Xinjiang Keketerek landmark and non-landmark grape samples; S2, constructing a spectral dataset based on the obtained spectral data for dataset construction, splitting the dataset based on the spectral dataset to obtain a spectral set, where the spectral set includes a spectral training set and a spectral test set, and the data for constructing the spectral dataset includes hyperspectral data, SSC, and landmark categories; S3, performing feature identification to obtain shared features, constructing a feature set based on the shared features to obtain a feature set, where the feature set includes a feature training set and a feature test set, feature identification means identifying the features corresponding to SSC and landmark categories through BA, the feature training set is used to improve the accuracy of constructing the SSC prediction model, the feature test set is used to predict and evaluate the XGBoost ensemble learning model, and the sample size of the feature test set is not less than the preset total number of samples; S4, constructing an SSC prediction model based on the feature training set and globally tuning the hyperparameters by Bayesian to improve the prediction accuracy of SSC and landmark categories, the SSC prediction model construction is used to obtain the XGBoost ensemble learning model, the coefficient of determination, and the root mean square error, and the Bayesian global tuning of hyperparameters is used to optimize the XGBoost ensemble learning model to improve the model performance.

[0010] The embodiment of the present application provides a grape landmark recognition system for hyperspectral integrated learning, including a spectral information extraction and monitoring module, a data set construction and splitting module, a feature set construction module, and a prediction accuracy evaluation module: Among them, the spectral information extraction and monitoring module is used to extract and monitor spectral information according to the obtained grape landmark samples to obtain hyperspectral images and measure SSC. The grape landmark samples include Xinjiang Keketerek landmarks and non-landmark grape samples; the data set construction and splitting module is used to construct data based on the obtained spectral data set to obtain a spectral data set, and split the data set based on the spectral data set to obtain a spectral set. The spectral set includes a spectral training set and a spectral test set. The data for constructing the spectral data set includes hyperspectral data, SSC, and landmark categories; the feature set construction module is used to perform feature recognition to obtain shared features, and construct a feature set based on the shared features to obtain a feature set. The feature set includes a feature training set and a feature test set. Feature recognition means identifying the features corresponding to SSC and landmark categories through BA. The feature training set is used to improve the accuracy of constructing the SSC prediction model, and the feature test set is used to predict and evaluate the XGBoost integrated learning model. The sample size of the feature test set is not less than the preset total number of samples; the prediction accuracy evaluation module is used to construct an SSC prediction model and perform Bayesian global tuning of hyperparameters based on the feature training set to improve the prediction accuracy of SSC and landmark categories. The construction of the SSC prediction model is used to obtain the XGBoost integrated learning model, the coefficient of determination, and the root mean square error. Bayesian global tuning of hyperparameters is used to optimize the XGBoost integrated learning model to improve the model performance.

[0011] One or more technical solutions provided in the embodiment of the present application have at least the following technical effects or advantages:

[0012] 1. By extracting and monitoring spectral information through grape landmark samples, then constructing a data set based on the spectral data set to obtain a spectral data set and splitting the data set to obtain a spectral set, then performing feature recognition to obtain shared features and constructing a feature set to obtain a feature set, and finally constructing an SSC prediction model and performing Bayesian global tuning of hyperparameters based on the feature training set, the effectiveness of the prediction results of SSC and landmark categories is improved, and further the prediction of the SSC of Keketerek grapes and the accuracy of landmark precise recognition are improved, effectively solving the problem of high deviation degree of the prediction of the SSC of Keketerek grapes and the result of landmark precise recognition caused by planting adjustment differences in the prior art.

[0013] 2. The extraction accuracy quantization value is obtained through coupled processing by combining hyperspectral extraction data, and then it is determined whether the extraction accuracy quantization value meets the extraction accuracy condition. When the extraction accuracy quantization value meets the extraction accuracy condition, the corresponding hyperspectral data is obtained. When the extraction accuracy quantization value does not meet the extraction accuracy condition, extraction accuracy optimization is performed, thereby realizing the precise evaluation of the extraction accuracy of the hyperspectral data of the fruit equatorial center image, and further realizing the improvement of the accuracy of obtaining hyperspectral data.

[0014] 3. The grape SSC difference value is obtained by performing difference processing on the actual grape SSC value and the predicted grape SSC value. Then, the average grape SSC difference value is obtained by performing difference processing on the actual grape SSC value and the average value of the actual grape SSC values. Finally, after analyzing the degree of approximation between the grape SSC difference value and the average grape SSC difference value, a deviation processing of the fitting degree is performed to obtain the determination coefficient, thereby realizing the precise evaluation of the fitting accuracy of the regression model, and further realizing the improvement of the fitting accuracy of the regression model. Description of the Drawings

[0015] Figure 1 It is a flowchart of a method for grape landmark recognition using hyperspectral ensemble learning provided by an embodiment of the present application;

[0016] Figure 2 It is a hyperspectral image of Keketerek grapes provided by an embodiment of the present application;

[0017] Figure 3 It is the average spectra of Keketerek grape landmarks and non-landmarks provided by an embodiment of the present application;

[0018] Figure 4 It is the SSC and landmark shared features provided by an embodiment of the present application;

[0019] Figure 5 It is a scatter plot of SSC content comparison provided by an embodiment of the present application;

[0020] Figure 6 It is a confusion matrix for Keketerek grape landmark recognition provided by an embodiment of the present application. Detailed Embodiments

[0021] Embodiments of the present application provide a method and system for grape landmark recognition based on hyperspectral integrated learning, which solve the problem of high deviation in the prediction of the SSC of Keketerek grapes and the accurate recognition of landmarks caused by planting adjustment differences in the prior art. By extracting and monitoring spectral information from grape landmark samples to obtain hyperspectral images and measuring the SSC, then constructing a dataset based on the spectral dataset to obtain a spectral dataset, splitting the dataset based on the spectral dataset to obtain spectral sets, then performing feature recognition to obtain shared features, constructing a feature set based on the shared features to obtain a feature set, and finally constructing an SSC prediction model based on the feature training set and globally tuning hyperparameters by Bayesian to improve the prediction accuracy of SSC and landmark categories, thus achieving an improvement in the result accuracy of the SSC prediction of Keketerek grapes and the accurate recognition of landmarks.

[0022] The technical solution in the embodiments of the present application for solving the problem of high deviation in the prediction of the SSC of Keketerek grapes and the accurate recognition of landmarks caused by the above-mentioned planting adjustment differences is generally as follows:

[0023] Extract and monitor spectral information through grape landmark samples, then construct a dataset based on the spectral dataset to obtain a spectral dataset and split the dataset to obtain spectral sets, then perform feature recognition to obtain shared features and construct a feature set to obtain a feature set, and finally construct an SSC prediction model based on the feature training set and globally tune hyperparameters by Bayesian, achieving the effect of improving the result accuracy of the SSC prediction of Keketerek grapes and the accurate recognition of landmarks.

[0024] To better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings of the specification and specific embodiments.

[0025] As Figure 1As shown in the figure, it is a flowchart of a grape landmark recognition method based on hyperspectral integrated learning provided by an embodiment of the present application. The method includes the following steps: S1, spectral information extraction and monitoring: spectral information extraction and monitoring are carried out according to the obtained grape landmark samples to obtain hyperspectral images and measure the SSC. The grape landmark samples include Keketerek landmarks in Xinjiang and non-landmark grape samples; S2, dataset construction and splitting: based on the obtained spectral dataset, data is constructed to obtain the spectral dataset. At the same time, the extraction accuracy quantization value is used to quantify the extraction accuracy of the hyperspectral data of the fruit equator center image and determine whether to optimize the extraction accuracy. Based on the spectral dataset, the dataset is split to obtain spectral sets, including a spectral training set and a spectral test set (such as 290 samples in the spectral training set and 72 samples in the spectral test set). The data for constructing the spectral dataset includes hyperspectral data, SSC, and landmark categories; S3, feature set construction: feature recognition is carried out to obtain shared features, and based on the shared features, a feature set is constructed to obtain a feature set, including a feature training set and a feature test set (the sample size of the feature test set is not less than 20% of the total number of samples). Feature recognition means identifying the features corresponding to the SSC and landmark categories through BA (Base Accuracy). The feature training set is used to improve the accuracy of constructing the SSC prediction model, and the feature test set is used to predict and evaluate the XGBoost integrated learning model. The sample size of the feature test set is not less than the preset total number of samples; S4, prediction accuracy evaluation: based on the feature training set, an SSC prediction model is constructed and Bayesian global tuning of hyperparameters is carried out to improve the prediction accuracy of the SSC and landmark categories. The construction of the SSC prediction model is used to obtain the XGBoost integrated learning model, the coefficient of determination, and the root mean square error. Bayesian global tuning of hyperparameters is used to optimize the XGBoost integrated learning model to improve the model performance. Among them, the XGBoost integrated learning model is constructed using the feature training set. In the Bayes algorithm, the Gaussian process is the surrogate model, the expected improvement is the objective function, the maximum number of iterations is 50, and the hyperparameters are globally optimized. The SSC and landmark categories are used as inputs, and the model predicts the SSC and landmark categories as outputs.

[0026] Among them, as Figure 2 shown, it is the hyperspectral image of Keketerek grapes provided by an embodiment of the present application; as Figure 3 shown, it is the average spectrum of Keketerek grape landmarks and non-landmarks provided by an embodiment of the present application; as Figure 4 shown, it is the shared feature of SSC and landmarks provided by an embodiment of the present application; as Figure 5 shown, it is the scatter plot of SSC content comparison provided by an embodiment of the present application. It can be seen from Figure 2-5 this that there are differences in the spectra of Keketerek grape landmarks and non-landmarks.

[0027] Among them, the prediction accuracy is R2 = 0.943, RMSE = 0.302, and its hyperparameters are mtry = 3; min_n = 6; tree_depth = 7; learn_rate = 0.00191; loss_reduction = 0.001; sample_prop = 0.967.

[0028] As Figure 6 shown, it is the confusion matrix of Korektik grape landmark recognition provided by the embodiment of the present application. Among them, the recall rate is 92.4%, and its optimal hyperparameters are mtry = 3; min_n = 12; tree_depth = 2; learn_rate = 0.0107; loss_reduction = 0.171; sample_prop = 0.922.

[0029] A grape landmark recognition system based on hyperspectral ensemble learning provided by the embodiment of the present application includes a spectral information extraction and monitoring module, a data set construction and splitting module, a feature set construction module, and a prediction accuracy evaluation module: Among them, the spectral information extraction and monitoring module is used to extract and monitor spectral information based on the obtained grape landmark samples to obtain hyperspectral images and measure SSC. The grape landmark samples include Xinjiang Korektik landmarks and non-landmark grape samples; the data set construction and splitting module is used to construct data based on the obtained spectral data set to obtain a spectral data set, and split the data set based on the spectral data set to obtain a spectral set. The spectral set includes a spectral training set and a spectral test set. The spectral data set construction data includes hyperspectral data, SSC, and landmark categories; the feature set construction module is used to perform feature recognition to obtain shared features, and construct a feature set based on the shared features to obtain a feature set. The feature set includes a feature training set and a feature test set. Feature recognition means identifying the features corresponding to SSC and landmark categories through BA. The feature training set is used to improve the accuracy of SSC prediction model construction. The feature test set is used to predict and evaluate the XGBoost ensemble learning model. The sample size of the feature test set is not less than the preset total sample number; the prediction accuracy evaluation module is used to construct an SSC prediction model and perform Bayesian global tuning of hyperparameters based on the feature training set to improve the prediction accuracy of SSC and landmark categories. The SSC prediction model construction is used to obtain the XGBoost ensemble learning model, the coefficient of determination, and the root mean square error. Bayesian global tuning of hyperparameters is used to optimize the XGBoost ensemble learning model to improve the model performance.

[0030] In this embodiment, hyperspectral images are obtained through spectral information extraction monitoring and SSC is measured; hyperspectral data that meets the extraction accuracy conditions is obtained through dataset construction and splitting. At the same time, the spectral dataset construction data obtained is used to construct a spectral dataset and a spectral set; a feature set is constructed to obtain a feature set for predicting and evaluating the XGBoost ensemble learning model; the XGBoost ensemble learning model, the coefficient of determination, and the root mean square error are obtained through prediction accuracy evaluation, and the hyperparameters are globally tuned by Bayesian to improve the prediction accuracy of SSC and landmark categories; through the synergistic effects of spectral information extraction monitoring, dataset construction and splitting, feature set construction, and prediction accuracy evaluation, it helps to provide more accurate hyperspectral data, and thus achieves the effect of improving the result accuracy of SSC prediction and landmark precise identification of Keketie Rek grapes.

[0031] Furthermore, the hyperspectral image is obtained by setting the spectral image of the labeled grape landmark sample information through a preset hyperspectral imaging device; the labeled grape landmark sample information represents grape landmark sample information not less than the preset sample size (generally 100) obtained from the database; the spectral image setting includes band range setting and resolution setting; the band range setting means sending a prompt to a preset person to set the band corresponding to the grape landmark sample information within the preset band range (generally the visible to near-infrared band from 400 to 2500 nm); the resolution setting means sending a prompt to a preset person to set the spectral resolution corresponding to the grape landmark sample information within the preset resolution range (generally 2 - 5 nm).

[0032] It should be added that the spectral dataset is constructed based on the obtained spectral dataset construction data, and the specific process is as follows: the fruit equatorial center image is obtained, and the fruit equatorial center image is obtained by imaging and scanning with a preset hyperspectral imaging device, that is, using the preset hyperspectral imaging device and a whiteboard for whiteboard calibration, turning off the halogen lamp, and performing blackboard calibration on the external environmental light source to eliminate errors, and then performing imaging and scanning on each sample; the hyperspectral data is obtained by extracting data from the fruit equatorial center image, and the data extraction means extracting based on ENVI software from the preset image area (the number of preset image areas is greater than or equal to 3) at the fruit equatorial center; the SSC of the labeled grape landmark sample is measured by a refractometer (the measurement results are independently repeated not less than 2 times); the landmark category is obtained by dividing the landmark category according to the SSC; the spectral dataset construction data is used to construct a spectral dataset, and the dataset construction means summarizing the spectral dataset construction data into a set (such as integrating spectral data, SSC, landmark category, and acquisition time into a unified storage format); the obtained spectral dataset is smoothed; the smoothing process means eliminating the noise of the spectral dataset based on Savitzky-Golay smoothing.

[0033] In this embodiment, by setting the spectral image, it helps to avoid feature loss caused by fixed parameters (such as key absorption peaks not being captured), and improve the correlation between spectral data and chemical components; by using ENVI software to extract spectral data from the preset area of the fruit equatorial center image, it ensures the collaborative optimization of spatial resolution and spectral resolution; based on the Savitzky-Golay smoothing algorithm to eliminate random noise in the spectral data (such as instrument noise, ambient light interference), it helps to retain key spectral features, improve the image quality of the fruit equatorial center image, and thus achieve the effect of improving the accuracy of the SSC prediction and landmark precise identification of Keketerek grapes.

[0034] Furthermore, based on the obtained spectral data set, a spectral data set is constructed by constructing data, and it also includes quantifying the extraction accuracy; the specific process of quantifying the extraction accuracy is as follows: obtaining the hyperspectral extraction accurate parameters during the data extraction process of the fruit equatorial center image; based on the obtained hyperspectral extraction accurate parameters and the preset hyperspectral extraction accurate parameters, quantifying the hyperspectral data extraction accuracy of the fruit equatorial center image to obtain the extraction accuracy quantification value; the specific process of obtaining the extraction accuracy quantification value is as follows:

[0035] AA1, after analyzing the proportion degree of the preset spectral resolution deviation and the spectral resolution deviation, combined with the resolution deviation - extraction accurate factor for weighted operation to obtain the resolution deviation - extraction accurate value, which is used to reflect the effect of the spectral resolution deviation on the hyperspectral data extraction accuracy of the fruit equatorial center image. Specifically, the expression of the resolution deviation - extraction accurate value is F = 1, 2,..., J, where F represents the number of the fruit equatorial center image, J represents the total number of the fruit equatorial center images, GGP1(F) represents the resolution deviation - extraction accurate value of the Fth fruit equatorial center image, ΔGPP(F) represents the average spectral resolution deviation of the Fth fruit equatorial center image, ΔGPP(0) represents the preset average spectral resolution deviation, and M 1 represents the resolution deviation - extraction accurate factor. Among them, the absolute value of the difference between the resolution and the preset spectral resolution (set in advance by preset personnel) at the preset wavelength of the fruit equatorial center image is monitored by a preset hyperspectral imaging device (such as a spectrometer, atomic emission spectrometer, high - resolution camera, etc.), and its average value is used as the average spectral resolution deviation.

[0036] AA2, after analyzing the proportion degree of the average spectral signal - to - noise ratio and the preset average spectral signal - to - noise ratio, combined with the signal - to - noise ratio - extraction accurate factor for weighted operation to obtain the signal - to - noise ratio - extraction accurate value, which is used to reflect the effect of the average spectral signal - to - noise ratio on the hyperspectral data extraction accuracy of the fruit equatorial center image. Specifically, the expression of the signal - to - noise ratio - extraction accurate value is GGP2(F) represents the accurate value of the signal-to-noise ratio of the equatorial center image of the F-th fruit, ΔGPXZ(F) represents the average spectral signal-to-noise ratio of the equatorial center image of the F-th fruit, ΔGPXZ(0) represents the preset average spectral signal-to-noise ratio, and M 2 represents the accurate factor for signal-to-noise ratio extraction. Among them, by monitoring the ratio of the power and noise power of the preset spectral signal of the equatorial center image of the fruit with a preset hyperspectral imaging device and a power meter, and taking its average value as the average spectral signal-to-noise ratio.

[0037] AA3. After analyzing the proportion degree of the preset average wavelength offset and the average wavelength offset, combined with the accurate factor for wavelength offset extraction, a weighted operation is performed to obtain the accurate value for wavelength offset extraction, which is used to reflect the effect of the average wavelength offset on the accuracy of hyperspectral data extraction of the equatorial center image of the fruit. Specifically, the expression of the accurate value for wavelength offset extraction is GGP3(F) represents the accurate value for wavelength offset extraction of the equatorial center image of the F-th fruit, ΔBC(F) represents the average wavelength offset of the equatorial center image of the F-th fruit, ΔBC(0) represents the preset average wavelength offset, and M 3 represents the accurate factor for wavelength offset extraction. Among them, by monitoring the absolute value of the difference between the preset wavelength and the preset wavelength (pre-set by a preset person) of the equatorial center image of the fruit with a preset hyperspectral imaging device, and taking its average value as the average wavelength offset.

[0038] AA4. Coupling processing is performed on the hyperspectral extraction data to obtain the extraction accuracy quantization value, which is used to reflect the combined effect of the hyperspectral extraction accurate parameters and the preset hyperspectral extraction accurate parameters on the accuracy of hyperspectral data extraction of the equatorial center image of the fruit. Among them, the hyperspectral extraction data includes the accurate value for resolution deviation extraction, the accurate value for signal-to-noise ratio extraction, and the accurate value for wavelength offset extraction, and all the hyperspectral extraction data are greater than 0.

[0039] Among them, the extraction accuracy quantization value is obtained through the following method:

[0040] GGP(F) = GGP1(F) + GGP2(F) + GGP3(F);

[0041] In the formula, GGP(F) represents the extraction accuracy quantization value of the equatorial center image of the F-th fruit.

[0042] Among them, before designing a grape landmark recognition method based on hyperspectral integrated learning provided in this application, a database for storing various set data is established. The database includes, but is not limited to, a preset average spectral resolution deviation, a preset average spectral signal-to-noise ratio, and a preset average wavelength offset, etc. All kinds of values therein are directly set by technicians; the accurate hyperspectral extraction parameters include an average spectral resolution deviation, an average spectral signal-to-noise ratio, and an average wavelength offset; the preset accurate hyperspectral extraction parameters include a preset average spectral resolution deviation, a preset average spectral signal-to-noise ratio, and a preset average wavelength offset; among them, the preset accurate hyperspectral extraction parameters are represented by the average value of the accurate hyperspectral extraction parameters in a historical time period; the units of both the average spectral resolution deviation and the preset average spectral resolution deviation are nanometers; the units of both the average spectral signal-to-noise ratio and the preset average spectral signal-to-noise ratio are decibels; the units of both the average wavelength offset and the preset average wavelength offset are nanometers.

[0043] In this embodiment, an extraction accuracy quantization value is further obtained by analyzing the hyperspectral extraction data. The larger the resolution deviation - extraction accurate value, the stronger the effect of the spectral resolution deviation on the extraction accuracy of the hyperspectral data of the fruit equator center image, resulting in a larger extraction accuracy quantization value; the larger the signal-to-noise ratio - extraction accurate value, the stronger the effect of the average spectral signal-to-noise ratio on the extraction accuracy of the hyperspectral data of the fruit equator center image, resulting in a larger extraction accuracy quantization value; the larger the wavelength offset - extraction accurate value, the stronger the effect of the average wavelength offset on the extraction accuracy of the hyperspectral data of the fruit equator center image, resulting in a larger extraction accuracy quantization value; in summary, in this embodiment, there is a positive correlation between the hyperspectral extraction data and the extraction accuracy quantization value.

[0044] The accurate hyperspectral extraction parameters monitored in this embodiment do not exist in isolation, but have the characteristic of being interrelated, and correlation analysis is required to describe their combined action. When the average spectral resolution deviation increases, it means that there is greater non-uniformity in the wavelength interval of the spectral data, which may lead to the distortion of the spectral signal, and then lead to a smaller average spectral signal-to-noise ratio; the average spectral resolution deviation may affect the accuracy of wavelength measurement. The larger the average spectral resolution deviation, the greater the error in wavelength measurement by the spectrometer, which may lead to inaccurate wavelength measurement, and then lead to a larger average wavelength offset; the smaller the average spectral signal-to-noise ratio, the more noise in the spectral signal, which may interfere with the accurate measurement of the wavelength, and then lead to a larger average wavelength offset. By analyzing the comprehensive influence between parameters, the accurate evaluation of the extraction accuracy of the hyperspectral data of the fruit equator center image is realized, and then the accuracy of the prediction of the SSC of Keketerek grapes and the accuracy of landmark accurate recognition are improved.

[0045] Further, the specific process of obtaining hyperspectral data is as follows: Determine whether the extraction accuracy quantization value meets the extraction accuracy condition; when the extraction accuracy quantization value meets the extraction accuracy condition, send a qualified extraction prompt and obtain the corresponding hyperspectral data; when the extraction accuracy quantization value does not meet the extraction accuracy condition, send an unqualified extraction prompt and perform extraction accuracy optimization; the extraction accuracy optimization is used to improve the extraction accuracy of the hyperspectral data of the fruit equatorial center image; the extraction accuracy condition means that the extraction accuracy quantization value is not lower than the preset extraction accuracy quantization value obtained from the database.

[0046] The specific steps involved in the extraction accuracy optimization are as follows:

[0047] BB1. Obtain optimized qualified parameters during the data extraction process of the fruit equatorial center image; after analyzing the proportion degree between the light source intensity uniformity ratio and the preset light source intensity uniformity ratio, perform a weighted operation in combination with the first light source uniformity factor to obtain the first light source uniformity value, which is used to reflect the effect of the light source intensity uniformity ratio on the light source uniformity during the hyperspectral data extraction process of the fruit equatorial center image. Specifically, the expression of the first light source uniformity value is F = 1, 2,..., J, where F represents the number of the fruit equatorial center image, J represents the total number of fruit equatorial center images, JYX1(F) represents the first light source uniformity value of the F-th fruit equatorial center image, JYB(F) represents the light source intensity uniformity ratio of the F-th fruit equatorial center image, JYB(0) represents the preset light source intensity uniformity ratio, and E 1 represents the first light source uniformity factor. Among them, the ratio of the maximum value to the minimum value of the light source intensity in the preset irradiation area of the fruit equatorial center image is monitored by a preset hyperspectral imaging device and a light intensity sensor, and its average value is used as the light source intensity uniformity ratio.

[0048] BB2. After analyzing the proportion degree between the average light source intensity and the preset average light source intensity, perform a weighted operation in combination with the second light source uniformity factor to obtain the second light source uniformity value, which is used to reflect the effect of the average light source intensity on the light source uniformity during the hyperspectral data extraction process of the fruit equatorial center image. Specifically, the expression of the second light source uniformity value is JYX2(F) represents the second light source uniformity value of the F-th fruit equatorial center image, GQP(F) represents the average light source intensity of the F-th fruit equatorial center image, GQP(0) represents the preset average light source intensity, and E 2 represents the second light source uniformity factor. Among them, the light source intensity at the points in the preset irradiation area of the fruit equatorial center image is monitored by a preset hyperspectral imaging device and a light intensity sensor, and its average value is used as the average light source intensity.

[0049] BB3. After analyzing the proportion of the preset average image reflectance and the average image reflectance, a weighted operation is performed in combination with the third light source uniformity factor to obtain the third light source uniformity value, which is used to reflect the role of the average image reflectance in the light source uniformity during the hyperspectral data extraction process of the fruit equatorial center image. Specifically, the expression of the third light source uniformity value is JYX3(F) represents the third light source uniformity value of the F-th fruit equatorial center image, FS(F) represents the average image reflectance of the F-th fruit equatorial center image, FS(0) represents the preset average image reflectance, and E 3 represents the third light source uniformity factor. Among them, the reflectance of the pixel points corresponding to the preset irradiation area points of the fruit equatorial center image is monitored by a preset hyperspectral imaging device, and its average value is used as the average image reflectance.

[0050] BB4. After performing coupling processing on the extraction accuracy optimization data, an extraction accuracy optimization value is obtained. The extraction accuracy optimization value is used to reflect the combined effect of the optimized qualified parameters and the preset optimized qualified parameters on the light source uniformity during the hyperspectral data extraction process of the fruit equatorial center image. The extraction accuracy optimization data includes the first light source uniformity value, the second light source uniformity value, and the third light source uniformity value, and all the extraction accuracy optimization data are greater than 0.

[0051] Among them, the extraction accuracy optimization value is obtained through the following method:

[0052] JYX(F) = JYX1(F) + JYX2(F) + JYX3(F);

[0053] In the formula, JYX(F) represents the extraction accuracy optimization value of the F-th fruit equatorial center image.

[0054] In summary, the optimized qualified parameters include the light source intensity uniformity ratio, the average light source intensity, and the average image reflectance. The preset optimized qualified parameters include the preset light source intensity uniformity ratio, the preset average light source intensity, and the preset average image reflectance. Among them, the preset optimized qualified parameters are represented by the average value of the optimized qualified parameters in the historical time period. Both the light source intensity uniformity ratio and the preset light source intensity uniformity ratio have no unit. The units of both the preset average light source intensity and the average light source intensity are watts per square meter. Both the preset average image reflectance and the average image reflectance have no unit.

[0055] Perform a ratio analysis on the optimized extraction accuracy value and the preset optimized extraction accuracy value obtained from the database to obtain an extraction adjustment factor. Among them, the preset optimized extraction accuracy value is represented by the average value of the optimized extraction accuracy values in the historical time period; the extraction adjustment factor is represented by the ratio of the optimized extraction accuracy value to the preset optimized extraction accuracy value, and is used to adjust the extraction accuracy of the hyperspectral data of the fruit equatorial center image; set the light source intensity, and setting the light source intensity means sending a prompt to the preset personnel to gradually increase the light source intensity in the corresponding amplitude of the extraction adjustment factor; set the working distance, and setting the working distance means sending a prompt to the preset personnel to gradually decrease the working distance of the preset hyperspectral imaging device in the corresponding amplitude of the extraction adjustment factor; when the working distance of the preset hyperspectral imaging device is reduced to the preset minimum working distance or the light source intensity is increased to the preset maximum light source intensity, if the extraction accuracy quantization value does not meet the extraction accuracy condition, send an alarm prompt, where the working distance of the preset hyperspectral imaging device and the preset maximum light source intensity are preset by the preset personnel in advance.

[0056] In this embodiment, when the light source intensity setting prompt is monitored, the light source intensity is gradually increased in the corresponding amplitude of the extraction adjustment factor to compensate for the light attenuation in the equatorial region. When the working distance setting prompt is monitored, the working distance of the preset hyperspectral imaging device is gradually decreased in the corresponding amplitude of the extraction adjustment factor to enhance the spectral detail capture ability; by dynamically adjusting the light source intensity and the working distance, it helps to improve the performance and reliability of the hyperspectral imaging device in grape SSC detection.

[0057] In this embodiment, the optimized extraction accuracy value is further obtained by analyzing the optimized extraction accuracy data. The larger the first light source uniformity value, the stronger the role of the light source intensity uniformity ratio in the light source uniformity during the hyperspectral data extraction process of the fruit equatorial center image, resulting in a larger optimized extraction accuracy value; the larger the second light source uniformity value, the stronger the role of the average light source intensity in the light source uniformity during the hyperspectral data extraction process of the fruit equatorial center image, resulting in a larger optimized extraction accuracy value; the larger the third light source uniformity value, the stronger the role of the average image reflectance in the light source uniformity during the hyperspectral data extraction process of the fruit equatorial center image, resulting in a larger optimized extraction accuracy value; in summary, in this embodiment, there is a positive correlation between the optimized extraction accuracy data and the optimized extraction accuracy value.

[0058] In this embodiment, the optimized qualified parameters monitored do not exist in isolation, but have interrelated characteristics, and correlation analysis is required to describe their combined effects. The larger the light source intensity uniformity ratio, the more uniform the intensity distribution of the light source in the irradiation area, which helps to maintain the stability and effectiveness of the average light source intensity, and thus leads to a larger average light source intensity; the smaller the light source intensity uniformity ratio, the more uneven the intensity distribution of the light source in the preset irradiation area, and the relatively uneven the reflected light distribution, which in turn leads to a smaller average image reflectance; the larger the average image reflectance, it may cause overexposure of the fruit equator center image, resulting in the loss of details in the preset area, reducing the light source uniformity in the preset area, and thus leading to a smaller light source intensity uniformity ratio. By analyzing the comprehensive effects between parameters, the accurate evaluation of the light source uniformity in the process of hyperspectral data extraction of the fruit equator center image is realized, and further the accuracy of the SSC prediction and landmark accurate identification of Koriketerek grapes is improved.

[0059] Further, the specific process of obtaining the spectral set by splitting the spectral dataset is as follows: Step 1, set the splitting ratio, and setting the splitting ratio means sending a prompt to a preset person to set the splitting ratio of the spectral dataset; Step 2, stratified random sampling, and stratified random sampling means splitting the spectral dataset by the stratified random sampling method; Stratified random sampling is used to ensure that the landmark categories and SSCs in the spectral set are normally distributed.

[0060] The specific process of obtaining the shared features by feature recognition is as follows: SS1, generate BA, and BA is obtained through the random forest ensemble; SS2, perform code reproduction, and code reproduction means reproducing through R code; SS3, perform feature-level fusion, and feature-level fusion means identifying the features corresponding to the SSC and landmark categories through BA and combining them to obtain the fusion features; SS4, perform average importance evaluation, and average importance evaluation means calculating the average importance score based on the reduction of the random forest node impurity; SS5, perform importance threshold setting, and importance threshold setting means sending a prompt to a preset person to set the importance threshold (generally p = 0.6); SS6, perform feature screening, and feature screening means screening the fusion features based on the importance threshold to obtain the shared features.

[0061] In this embodiment, a stratified random sampling method is adopted to split the spectral dataset. For example, through a preset person, the splitting ratio of the spectral training set and the spectral test set is set to be greater than 4:1 to ensure the normal distribution of the landmark categories and SSC corresponding to the spectral set; the BA of the spectral dataset is obtained through the random forest ensemble algorithm, which helps to evaluate the improvement effect of feature selection on the prediction performance; the random forest algorithm is reproduced through R language, which helps to ensure the reproducibility of the results and optimize the feature recognition effect; based on the features identified by the random forest, the features corresponding to the SSC and landmark categories are identified through BA, which helps to remove redundant features and enhance the prediction ability of the SSC and landmarks of the Keketie Rek grape; based on the reduction of the node impurity of the random forest, the average importance score of each fused feature is calculated, providing a quantitative basis for feature screening; and then the accuracy of the results of the SSC prediction and landmark accurate identification of the Keketie Rek grape is improved.

[0062] Further, the specific process of obtaining the coefficient of determination is as follows: CC1, the grape SSC difference value is obtained by processing the difference between the actual value of the grape SSC and the predicted value of the grape SSC; the grape SSC difference value is used to reflect the relative deviation degree between the actual value of the grape SSC and the predicted value of the grape SSC, and jointly the effect on the fitting accuracy of the regression model. The grape SSC difference value is represented by the sum of the squared errors of the actual value of the grape SSC and the predicted value of the grape SSC. Specifically, the expression of the grape SSC difference value is y obs represents the actual value of the grape SSC, y pre represents the predicted value of the grape SSC, i represents the sample number, and n represents the number of samples.

[0063] CC2, the average grape SSC difference value is obtained by processing the difference between the actual value of the grape SSC and the mean value of the actual value of the grape SSC; the average grape SSC difference value is used to reflect the relative deviation degree between the actual value of the grape SSC and the mean value of the actual value of the grape SSC, and jointly the effect on the fitting accuracy of the regression model. The average grape SSC difference value is represented by the sum of the squared errors of the actual value of the grape SSC and the mean value of the actual value of the grape SSC. Specifically, the expression of the average grape SSC difference value is y ave represents the mean value of the actual value of the grape SSC.

[0064] CC3, after analyzing the degree of approximation between the grape SSC difference value and the average grape SSC difference value, the fitting degree deviation processing is performed to obtain the coefficient of determination; the coefficient of determination is used to reflect the comprehensive effect of the grape evaluation parameters and the predicted value of the grape SSC on the fitting accuracy of the regression model; the grape evaluation parameters include the actual value of the grape SSC and the mean value of the actual value of the grape SSC.

[0065] Among them, the coefficient of determination is obtained by the following method:

[0066]

[0067] In the formula, R 2 represents the coefficient of determination, S1 represents the difference value of grape SSC, and S2 represents the average difference value of grape SSC.

[0068] The specific process of obtaining the root mean square error is as follows:

[0069] DD1. By performing a ratio analysis on the difference value of grape SSC and the number of samples, a root mean square error reflection value is obtained. The root mean square error reflection value is used to evaluate the degree of deviation of the actual value of grape SSC from the accuracy of grape SSC. Specifically, the expression of the root mean square error reflection value is y obs represents the actual value of grape SSC, and y pre represents the predicted value of grape SSC.

[0070] DD2. By taking the square root of the root mean square error reflection value, the root mean square error is obtained; the root mean square error is used to reflect the combined effect of the actual value of grape SSC and the predicted value of grape SSC on the degree of deviation from the accuracy of grape SSC.

[0071] Among them, the root mean square error is obtained by the following method:

[0072]

[0073] In the formula, RMSE represents the root mean square error, and H represents the root mean square error reflection value.

[0074] It should be added that for constructing the SSC prediction model based on the feature training set and globally tuning the hyperparameters by Bayesian to improve the prediction accuracy of SSC and landmark categories, it also includes evaluating the accuracy of the classification model; evaluating the accuracy of the classification model means obtaining the accuracy rate and recall rate based on the confusion matrix, which are used to evaluate the classification accuracy of the XGBoost ensemble learning model.

[0075] Specifically, accuracy rate = (TP + TN) / (TP + TN + FP + FN); recall rate = TP / (TP + FN); where TP is the positive correct sample; TN is the negative correct sample; FP is the positive wrong sample; FN is the negative wrong sample.

[0076] In this embodiment, the larger the difference value of grape SSC, the stronger the combined effect of the actual value of grape SSC and the relative deviation degree of the predicted value of grape SSC on the fitting accuracy of the regression model, resulting in a smaller coefficient of determination; the larger the average difference value of grape SSC, the stronger the combined effect of the relative deviation degree between the actual value of grape SSC and the mean value of the actual values of grape SSC on the fitting accuracy of the regression model (the regression model corresponding to the XGBoost ensemble learning model), resulting in a larger coefficient of determination; the larger the reflected value of the root mean square error, the stronger the effect of the actual value of grape SSC on the deviation degree of the accuracy of grape SSC, resulting in a larger root mean square error.

[0077] By monitoring the coefficient of determination and the root mean square error, it helps to improve the prediction ability of the model, and thus achieves the effect of improving the accuracy of the prediction of grape SSC and the precise identification of landmarks for Koriketerek grapes.

[0078] This embodiment provides a specific comparative example 1. NIR (Near-Infrared Spectroscopy) spectra are used to construct traditional PLS (Partial Least Squares) models to predict SSC and origin, and the specific method is as follows:

[0079] 1) 131 landmark and 231 non-landmark grape fruits are collected from Koriketerek grapes in Xinjiang respectively. Each fruit is marked, and a preset hyperspectral imaging device is used to perform imaging scans on each sample. ENVI software is used to select 3 regions at the equatorial part of the surface of each sample to extract hyperspectra, and then the SSC corresponding to each sample is measured by a refractometer; the wavelength range is 400 to 1000 nm, and the soluble solid content is measured 2 times.

[0080] 2) A full-wavelength dataset is obtained with full wavelength, SSC, and landmark categories, and it is split into a training set of 290 samples and a test set of 72 samples.

[0081] 3) The traditional PLS model is trained using the training set and predictions are made using the prediction set.

[0082] This embodiment provides a specific comparative example 2. NIR spectra are used to construct a PLS model with BA characteristic wavelengths to predict SSC and origin, and the specific method is as follows:

[0083] 1) 131 landmark and 231 non-landmark grape fruits are collected from Koriketerek grapes in Xinjiang respectively. Each fruit is marked, and a preset hyperspectral imaging device is used to perform imaging scans on each sample. ENVI software is used to select 3 regions at the equatorial part of the surface of each sample to extract hyperspectra, and then the SSC corresponding to each sample is measured by a refractometer; the wavelength range is 400 to 1000 nm, and the soluble solid content is measured 2 times.

[0084] 2) Obtain a full-wavelength dataset with full wavelength, SSC, and landmark categories, and split it into a training set of 290 samples and a test set of 72 samples;

[0085] 3) Use the BA algorithm to screen the characteristic wavelengths of SSC and landmarks, and reconstruct the training set and prediction set.

[0086] 4) Use the reconstructed training set to build a traditional PLS model.

[0087] In summary, through the extraction and monitoring of spectral information using grape landmark samples, and then based on the spectral dataset to construct data to obtain a spectral dataset and split the dataset to obtain a spectral set, then perform feature recognition to obtain shared features and construct a feature set to obtain a feature set, and finally build an SSC prediction model based on the feature training set and perform Bayesian global tuning of hyperparameters, thereby improving the effectiveness of the SSC and landmark category prediction results, and further improving the SSC prediction of Keketerek grapes and the accuracy of landmark precise identification, effectively solving the problem of high deviation degree of the SSC prediction of Keketerek grapes and the results of landmark precise identification caused by planting adjustment differences in the prior art.

[0088] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0089] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0090] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means embodying the function specified in the flowchart Figure 1 a flow or flows and / or block Figure 1 or blocks or blocks specified in the plurality of blocks.

[0091] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the function specified in the flowchart Figure 1 a flow or flows and / or block Figure 1 or blocks or blocks specified in the plurality of blocks.

[0092] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made by those skilled in the art once they learn of the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0093] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for grape landmark recognition based on hyperspectral integrated learning, characterized in that, It includes the following steps: S1. Extract and monitor spectral information based on the obtained grape landmark samples to obtain hyperspectral images and determine the SSC. The grape landmark samples include the Keketerek landmark in Xinjiang and non-landmark grape samples; S2. Construct a spectral dataset based on the obtained spectral dataset construction data to obtain a spectral dataset. Split the spectral dataset to obtain spectral sets, which include a spectral training set and a spectral test set. The spectral dataset construction data includes hyperspectral data, SSC, and landmark categories; S3. Perform feature recognition to obtain shared features, and construct a feature set based on the shared features to obtain a feature set, which includes a feature training set and a feature test set. The feature recognition means identifying the features corresponding to the SSC and landmark categories through BA. The feature training set is used to improve the accuracy of constructing the SSC prediction model, and the feature test set is used to predict and evaluate the XGBoost ensemble learning model; S4. Construct an SSC prediction model based on the feature training set and perform Bayesian global tuning of hyperparameters to improve the prediction accuracy of the SSC and landmark categories. The construction of the SSC prediction model is used to obtain the XGBoost ensemble learning model, the coefficient of determination, and the root mean square error. The Bayesian global tuning of hyperparameters is used to optimize the XGBoost ensemble learning model to improve the model performance.

2. The method for identifying grape landmarks by hyperspectral integrated learning according to claim 1, wherein The hyperspectral image is obtained by setting the spectral image of the marked grape landmark sample information through a preset hyperspectral imaging device; The marked grape landmark sample information represents grape landmark sample information not less than the preset sample size obtained from the database; The spectral image setting includes band range setting and resolution setting; The band range setting means sending a prompt to a preset person to set the band corresponding to the grape landmark sample information within a preset band range; The resolution setting means sending a prompt to a preset person to set the spectral resolution corresponding to the grape landmark sample information within a preset resolution range.

3. The method for identifying grape landmarks by hyperspectral integrated learning according to claim 1, characterized in that, The process of constructing a spectral dataset based on the obtained spectral dataset construction data to obtain a spectral dataset is as follows: Obtain the fruit equatorial center image, which is obtained by imaging and scanning through a preset hyperspectral imaging device; Extract hyperspectral data by extracting data from the fruit equatorial center image. The data extraction means extracting from a preset image area at the fruit equator based on ENVI software; Determine the SSC of the marked grape landmark samples by a refractometer; Divide the landmark categories according to the SSC to obtain the landmark categories; Construct a spectral dataset by constructing the spectral dataset construction data. The dataset construction means summarizing the spectral dataset construction data into a set; Perform smoothing processing on the obtained spectral dataset; The smoothing processing means eliminating the noise of the spectral dataset based on Savitzky-Golay smoothing.

4. The method for identifying grape landmarks by hyperspectral integrated learning according to claim 3, wherein The process of constructing a spectral dataset based on the obtained spectral dataset construction data to obtain a spectral dataset also includes quantifying the extraction accuracy; The specific process of quantifying the extraction accuracy is as follows: Obtain accurate hyperspectral extraction parameters during the data extraction process of the fruit equatorial center image; Quantify the accuracy of hyperspectral data extraction for the fruit equatorial center image based on the obtained accurate hyperspectral extraction parameters and the preset accurate hyperspectral extraction parameters to obtain an extraction accuracy quantification value; The specific process of obtaining the extraction accuracy quantification value is as follows: After analyzing the proportion degree of the preset spectral resolution deviation and the spectral resolution deviation, perform a weighted operation in combination with the resolution deviation - extraction accuracy factor to obtain a resolution deviation - extraction accuracy value, which is used to reflect the effect of the spectral resolution deviation on the accuracy of hyperspectral data extraction for the fruit equatorial center image; After analyzing the proportion degree of the average spectral signal signal - to - noise ratio and the preset average spectral signal signal - to - noise ratio, perform a weighted operation in combination with the signal - to - noise ratio - extraction accuracy factor to obtain a signal - to - noise ratio - extraction accuracy value, which is used to reflect the effect of the average spectral signal signal - to - noise ratio on the accuracy of hyperspectral data extraction for the fruit equatorial center image; After analyzing the proportion degree of the preset average wavelength offset and the average wavelength offset, perform a weighted operation in combination with the wavelength offset - extraction accuracy factor to obtain a wavelength offset - extraction accuracy value, which is used to reflect the effect of the average wavelength offset on the accuracy of hyperspectral data extraction for the fruit equatorial center image; Perform coupling processing on the hyperspectral extraction data to obtain an extraction accuracy quantification value; The extraction accuracy quantification value is used to reflect the combined effect of the accurate hyperspectral extraction parameters and the preset accurate hyperspectral extraction parameters on the accuracy of hyperspectral data extraction for the fruit equatorial center image; The hyperspectral extraction data includes a resolution deviation - extraction accuracy value, a signal - to - noise ratio - extraction accuracy value, and a wavelength offset - extraction accuracy value, and all the hyperspectral extraction data are greater than 0; The accurate hyperspectral extraction parameters include the average spectral resolution deviation, the average spectral signal signal - to - noise ratio, and the average wavelength offset; 5. The method for identifying grape landmarks by hyperspectral integrated learning according to claim 4, wherein, The specific process of obtaining the hyperspectral data is as follows: Judge whether the extraction accuracy quantification value meets the extraction accuracy condition; When the extraction accuracy quantification value meets the extraction accuracy condition, send an extraction qualified prompt and obtain the corresponding hyperspectral data; When the extraction accuracy quantification value does not meet the extraction accuracy condition, send an extraction unqualified prompt and perform extraction accuracy optimization; The extraction accuracy optimization is used to improve the accuracy of hyperspectral data extraction for the fruit equatorial center image; The extraction accuracy condition means that the extraction accuracy quantification value is not lower than the preset extraction accuracy quantification value obtained from the database; 6. The method for identifying grape landmarks by hyperspectral integrated learning according to claim 5, characterized in that, The specific steps of the extraction accuracy optimization are as follows: Obtain optimized qualified parameters during the data extraction process of the fruit equatorial center image; After analyzing the proportion degree of the light source intensity uniformity ratio and the preset light source intensity uniformity ratio, perform a weighted operation in combination with the first light source uniformity factor to obtain a first light source uniformity value, which is used to reflect the effect of the light source intensity uniformity ratio on the light source uniformity during the hyperspectral data extraction process of the fruit equatorial center image; After analyzing the proportion of the average light source intensity and the preset average light source intensity, a weighted operation is performed in combination with the second light source uniformity factor to obtain the second light source uniformity value, which is used to reflect the role of the average light source intensity in the light source uniformity during the extraction process of the hyperspectral data of the fruit equatorial center image; After analyzing the proportion of the preset average image reflectance and the average image reflectance, a weighted operation is performed in combination with the third light source uniformity factor to obtain the third light source uniformity value, which is used to reflect the role of the average image reflectance in the light source uniformity during the extraction process of the hyperspectral data of the fruit equatorial center image; After coupling the extraction accuracy optimization data, an extraction accuracy optimization value is obtained; The extraction accuracy optimization value is used to reflect the combined effect of the optimized qualified parameters and the preset optimized qualified parameters on the light source uniformity during the extraction process of the hyperspectral data of the fruit equatorial center image; The extraction accuracy optimization data includes the first light source uniformity value, the second light source uniformity value, and the third light source uniformity value, and all the extraction accuracy optimization data are greater than 0; The optimized qualified parameters include the light source intensity homogenization ratio, the average light source intensity, and the average image reflectance; A proportion analysis is performed on the extraction accuracy optimization value and the preset extraction accuracy optimization value obtained from the database to obtain an extraction adjustment factor; Set the light source intensity, and the set light source intensity means sending a prompt to a preset person to gradually increase the light source intensity in the amplitude corresponding to the extraction adjustment factor; Set the working distance, and the set working distance means sending a prompt to a preset person to gradually decrease the working distance of the preset hyperspectral imaging device in the amplitude corresponding to the extraction adjustment factor; When the working distance of the preset hyperspectral imaging device is reduced to the preset minimum working distance or the light source intensity is increased to the preset maximum light source intensity, if the extraction accuracy quantization value does not meet the extraction accuracy condition, an alarm prompt is sent.

7. The method for identifying grape landmarks by hyperspectral integrated learning according to claim 1, wherein The specific process of splitting the spectral dataset to obtain the spectral set based on the spectral dataset is as follows: Step 1, set the splitting ratio, and the set splitting ratio means sending a prompt to a preset person to set the splitting ratio of the spectral dataset; Step 2, stratified random sampling, and the stratified random sampling means splitting the spectral dataset by the stratified random sampling method; The stratified random sampling is used to ensure that the landmark categories and SSC in the spectral set are normally distributed.

8. The method for identifying grape landmarks by hyperspectral integrated learning according to claim 1, characterized in that, The specific process of performing feature recognition to obtain shared features is as follows: SS1, generate BA, and the BA is obtained through a random forest ensemble; SS2, perform code reproduction, and the code reproduction means reproducing through R code; SS3, perform feature-level fusion, and the feature-level fusion means identifying the features corresponding to SSC and landmark categories through BA and combining them to obtain fusion features; SS4, perform average importance evaluation, and the average importance evaluation means calculating the average importance score based on the reduction of the random forest node impurity; SS5, perform importance threshold setting, and the importance threshold setting means sending a prompt to a preset person to set the importance threshold; SS6, perform feature screening, and the feature screening means screening the fusion features based on the importance threshold to obtain shared features.

9. The method for identifying grape landmarks by hyperspectral integrated learning according to claim 1, wherein The specific process of obtaining the coefficient of determination is as follows: The grape SSC difference value is obtained by performing difference processing on the actual value of grape SSC and the predicted value of grape SSC; The grape SSC difference value is used to reflect the relative deviation degree between the actual value of grape SSC and the predicted value of grape SSC, and jointly the effect on the fitting accuracy of the regression model; The average grape SSC difference value is obtained by performing difference processing on the actual value of grape SSC and the mean value of the actual values of grape SSC; The average grape SSC difference value is used to reflect the relative deviation degree between the actual value of grape SSC and the mean value of the actual values of grape SSC, and jointly the effect on the fitting accuracy of the regression model; After performing an approach degree analysis on the grape SSC difference value and the average grape SSC difference value, a fitting degree deviation processing is performed to obtain the determination coefficient; The determination coefficient is used to reflect the comprehensive effect of the grape evaluation parameters and the predicted value of grape SSC on the fitting accuracy of the regression model; The grape evaluation parameters include the actual value of grape SSC and the mean value of the actual values of grape SSC; The construction of the SSC prediction model based on the feature training set and the Bayesian global tuning of hyperparameters to improve the prediction accuracy of SSC and landmark categories further includes performing classification model accuracy evaluation; The performing of classification model accuracy evaluation means obtaining the accuracy rate and recall rate based on the confusion matrix, and is used to evaluate the classification accuracy of the XGBoost ensemble learning model.

10. A grape landmark recognition system based on hyperspectral integrated learning, characterized in that, It includes a spectral information extraction and monitoring module, a data set construction and splitting module, a feature set construction module, and a prediction accuracy evaluation module: Among them, the spectral information extraction and monitoring module is used to perform spectral information extraction and monitoring according to the obtained grape landmark samples to obtain hyperspectral images and measure SSC. The grape landmark samples include Xinjiang Keketerek landmarks and non-landmark grape samples; The data set construction and splitting module is used to construct data based on the obtained spectral data set to obtain a spectral data set, and perform data set splitting on the spectral data set to obtain a spectral set. The spectral set includes a spectral training set and a spectral test set. The data for constructing the spectral data set includes hyperspectral data, SSC, and landmark categories; The feature set construction module is used to perform feature recognition to obtain shared features, and construct a feature set based on the shared features to obtain a feature set. The feature set includes a feature training set and a feature test set. The feature recognition means identifying the features corresponding to SSC and landmark categories through BA. The feature training set is used to improve the accuracy of constructing the SSC prediction model, and the feature test set is used to predict and evaluate the XGBoost ensemble learning model; The prediction accuracy evaluation module is used to construct the SSC prediction model based on the feature training set and perform Bayesian global tuning of hyperparameters to improve the prediction accuracy of SSC and landmark categories. The construction of the SSC prediction model is used to obtain the XGBoost ensemble learning model, the determination coefficient, and the root mean square error. The Bayesian global tuning of hyperparameters is used to optimize the XGBoost ensemble learning model to improve the model performance.

Images