Wine grape branch pruning weight prediction method and system and storage medium

Abstract

The invention discloses a wine grape branch pruning weight prediction method and system and a storage medium. The method comprises the following steps: acquiring grape canopy spectral data in different growth periods through an unmanned aerial vehicle; respectively pruning and weighing the branches of the experimental plots in the growing period and the dormancy period; removing interference of canopy shadows and weeds on the spectral data by using a U-net network model, extracting a vegetation index and a texture index in the experimental area, and analyzing the correlation between the vegetation index and the branch pruning weight; based on a correlation analysis result, respectively constructing machine learning models of the wine grapes in a growth period and a dormancy period to predict branch pruning weight; and performing spatial distribution prediction on the pruning weight of the branches of the whole vineyard by using a prediction result. By adopting the technical scheme of the invention, the problems of high labor intensity, low efficiency, precision limited by sample representativeness, experience dependence and strong subjectivity in the traditional wine grape branch pruning weight measurement are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of pruning techniques for wine grape branches, and particularly relates to a method and system for predicting the weight of pruned wine grape branches, and a storage medium. Background Art

[0002] The cultivation of wine grapes is an important cash crop in global agriculture. Reasonable pruning of branches can not only promote the healthy growth of plants, improve yield and fruit quality, but also help to shape an ideal tree form. As a key link in grape cultivation management, the weight of pruned branches is an important indicator for measuring growth potential and yield. Therefore, accurately predicting the weight of pruned wine grape branches is crucial for yield assessment, growth monitoring and agronomic management.

[0003] The traditional method for predicting the weight of pruned wine grape branches mainly involves manually weighing and measuring the pruned branches in different areas, including the following common methods:

[0004] 1. Direct weighing method: After pruning, the pruned branches of each grapevine are collected and weighed together, and the total weight of the pruned branches of each grapevine or each row of grapevines is recorded.

[0005] 2. Sampling method for sample plants: Select representative sample plants in the vineyard, prune and weigh the branches of each plant. Then, the pruning weight of the sample plants is extended to the entire orchard, and the total pruning weight is estimated using a formula.

[0006] 3. Visual estimation method: Based on years of experience, estimate the weight of the pruned branches by observing the number, thickness and growth density of the branches on the tree.

[0007] Currently, the traditional method for predicting the weight of pruned wine grape branches mainly relies on manual weighing and measurement. The weighing method is time-consuming and laborious, the accuracy of the sampling method is limited by the representativeness of the samples, and the visual estimation method relies on experience and is highly subjective. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a method and system for predicting the weight of pruned wine grape branches, and a storage medium, so as to solve the problems existing in the traditional method for estimating the weight of pruned branches, such as high labor intensity, low efficiency, limited accuracy, low information utilization rate and strong subjectivity, which are difficult to provide comprehensive growth information and data support and cannot meet the requirements of modern, refined and large-scale management. Therefore, it is necessary to predict the weight of pruned wine grape branches during the growth period and dormancy period.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] A method for predicting the weight of pruned wine grape branches, comprising:

[0011] Step S1: Collect spectral data of grape canopies at different growth stages using drones;

[0012] Step S2: Prune and weigh the branches of the experimental plots during the growth period and dormancy period respectively;

[0013] Step S3: After using the U-net network model to eliminate the interference of canopy shadows and weeds on the spectral data, extract the vegetation index and texture index within the experimental area, and analyze their correlation with the branch pruning weight;

[0014] Step S4: Based on the results of the correlation analysis, construct machine learning models for pruning grape branches during the growth period and dormancy period respectively to predict the branch pruning weight;

[0015] Step S5: Use the prediction results to predict the spatial distribution of the branch pruning weight of the entire vineyard.

[0016] Preferably, in Step S2, the branches are pruned during the new shoot growth period, flowering period, berry swelling period, coloring and maturity period, and dormancy period of the grape. The pruned branches of each experimental plot are weighed respectively according to the calculation formulas of the summer branch pruning weight SPW and the winter branch pruning weight WPW. The calculation formulas of the summer and winter branch pruning weights are as follows:

[0017]

[0018] where W n is the new shoot pruning weight, W s is the lateral shoot pruning weight, W nf is the pruning weight of non-fruiting branches, W d is the pruning weight of diseased and pest branches, W o is the pruning weight of old branches, W f is the pruning weight of fruiting mother branches, W t is the pruning weight of weak branches, and S is the sampling area of the research plot.

[0019] Preferably, in Step S5, through the co-Kriging interpolation spatial statistics method, based on the spherical model in the semi-variogram model, the branch pruning weights predicted by the artificial neural network model during the growth period and dormancy period are used as the target variables, and the spectral data is used as the auxiliary variable to predict the spatial distribution of the branch pruning weight of the entire vineyard; where

[0020] The co-Kriging interpolation predicts the spatial distribution of the branch pruning weight based on the following formula:

[0021]

[0022] where is the estimated value of the target variable Z 1 at the position s to be predicted; Z1 (s i ) is the target variable Z 1 At the known sampling point s i value; Z 2 (s j ), Z p (s k ) are the auxiliary variables Z 2 , …, Z p values at their sampling points; λ 1i , λ 2j , λ pk are the interpolation weights, corresponding to the target variable and the auxiliary variables respectively; n 1 , n 2 , n p are the number of samples of the target variable and the auxiliary variables.

[0023] The present invention also provides a prediction system for the pruning weight of wine grape branches, including:

[0024] A collection device for collecting spectral data of grape canopies at different growth stages through an unmanned aerial vehicle;

[0025] A calculation device for pruning and weighing the branches in the experimental plots during the growth period and the dormancy period respectively;

[0026] A processing device for using a U-net network model to eliminate the interference of canopy shadows and weeds on the spectral data, extracting the vegetation index and texture index in the experimental area, and analyzing their correlation with the pruning weight of the branches;

[0027] A first prediction device for respectively constructing machine learning models for pruning weight prediction of wine grapes during the growth period and the dormancy period based on the correlation analysis results;

[0028] A first prediction device for using the prediction results to predict the spatial distribution of the pruning weight of the branches in the entire vineyard.

[0029] Preferably, the calculation device prunes the branches during the new shoot growth period, flowering period, berry swelling period, coloring and maturity period, and dormancy period of the grapes, and weighs the pruned branches in each experimental plot respectively according to the summer branch pruning weight SPW and winter branch pruning weight WPW calculation formulas. The summer and winter branch pruning weight calculation formulas are as follows:

[0030]

[0031] Among them, W n is the new shoot pruning weight, W s is the lateral shoot pruning weight, W nf is the pruning weight of non-fruiting branches, W d is the pruning weight of pest and disease branches, Wo is the pruning weight of old branches, W f is the pruning weight of fruiting mother branches, W t is the pruning weight of weak branches, and S is the sampling area of the research plot.

[0032] Preferably, the second prediction device uses the co-Kriging interpolation spatial statistics method, based on the spherical model in the semi-variogram model, takes the pruning weights of branches during the growth period and dormancy period predicted by the artificial neural network model as target variables, and spectral data as auxiliary variables to predict the spatial distribution of the pruning weights of branches in the entire vineyard; wherein,

[0033] The co-Kriging interpolation predicts the spatial distribution of the pruning weight of branches based on the following formula:

[0034]

[0035] wherein, is the estimated value of the target variable Z 1 at the position s to be predicted; Z 1 (s i ) is the target variable Z 1 at the known sampling point s i value; Z 2 (s j )、Z p (s k ) are the values of the auxiliary variable Z 2 ,…,Z p at its sampling points; λ 1i ,λ 2j ,λ pk are the interpolation weights, corresponding to the target variable and the auxiliary variable respectively; n 1 ,n 2 ,n p are the sample numbers of the target variable and the auxiliary variable.

[0036] The present invention also provides a storage medium, on which a computer program is stored, and the computer program executes the method for predicting the pruning weight of wine grape branches when running.

[0037] The present invention effectively solves the problems in the traditional pruning weight measurement of wine grape branches, such as high labor intensity, low efficiency, limited accuracy due to sample representativeness, dependence on experience and strong subjectivity. By using drones to collect the spectral data of grape canopies at different growth stages, the branches in the experimental plots are pruned and weighed respectively based on the calculation formulas of summer pruning weight (SPW) and winter pruning weight (WPW) during the growth period and dormancy period. After using the U-net network model to eliminate the interference of canopy shadows and weeds on the spectral data, the vegetation index and texture index in the experimental area are extracted, and their correlations with the pruning weight of branches are analyzed. Based on the results of the correlation analysis, machine learning models for pruning weight prediction are constructed for wine grapes during the growth period and dormancy period respectively. Combining with the model-driven geostatistical interpolation technology, using the pruning weight and spectral data of known prediction points, the spatial distribution of the pruning weight of branches in the entire vineyard is predicted, providing a scientific guiding basis for the growth monitoring of wine grapes, the utilization of pruning branch waste and the growth management of canopy biomass. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0039] Figure 1 Schematic flow chart of the method for predicting the pruning weight of wine grape branches in the embodiment of the present invention;

[0040] Figure 2 Design of the research area and layout of ground control points in the embodiment of the present invention;

[0041] Figure 3 Information map of the annotation categories of the U-net dataset in the embodiment of the invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0043] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0044] Embodiment 1:

[0045] As Figure 1 shown, an embodiment of the present invention provides a method for predicting the pruning weight of wine grape branches, comprising the following steps:

[0046] Step 1: Layout of the research area. As Figure 2 shown, the experimental area is divided into 10 columns, each column is divided into 7 experimental plots, for a total of 70 experimental plots. Based on the topographic features, 12 ground control points (GCPs) are arranged in the experimental area to correct the errors of the UAV positioning system and improve the accuracy of geographical information.

[0047] The experimental area is 90 m long and 36 m wide. Each experimental plot is 6 m long and 1 m wide, with an area of 6 m 2 . The geographical coordinate information of the GCPs and each plot is obtained by a real-time kinematic (RTK) device (Chixun Xingju SRmini).

[0048] Step 2: Acquisition and processing of UAV multispectral images and branch pruning weight data. A DJI Matrice 600 Pro equipped with a RedEdge-MX (Micasense, USA) multispectral camera is used to acquire images of 5 spectral bands (R, G, B, REG, NIR) of the grape canopy during the new shoot growth period, flowering period, berry swelling period, and coloring and maturity period. The camera automatically captures one photo every 5 seconds and stores the images in TIFF format. Each UAV flight is scheduled between 10:00 am and 2:00 pm on sunny and cloudless days, and the measurement of the branch pruning weight during the growth period is carried out synchronously.

[0049] The flight altitude of the UAV is set at 30 m relative to the takeoff point, and the flight speed is 1 m / s. The flight line and side overlap rates are 80% respectively. In addition, two Lambertian rubber sheets with a size of 1 m × 1 m are arranged on the ground as standard reflectance sheets for radiometric calibration of the spectral images, with reflectivities of 10% and 30% respectively.

[0050] The Pix4Dmapper software is used to process the UAV images. The processing process includes image stitching, geometric correction, band combination, control point optimization, and region of interest extraction, generating orthophotos, digital elevation models (DEMs), digital surface models (DSMs), and three-dimensional point cloud data.

[0051] After the UAV image processing is completed, the obtained GCP and geographical coordinate information of each plot are imported into QGIS-3.34.3, and then each 6x1-meter plot is visualized through the new polygon Shapefile function.

[0052] Pruning is carried out on the branches during the new shoot growth period, flowering period, berry swelling period, coloring and ripening period, and dormancy period of the grape. According to the calculation formulas of the summer branch pruning weight (SPW) and winter branch pruning weight (WPW) respectively, the pruned branches in each experimental plot are weighed, and the weight is uniformly converted to the unit kg / m 2 .

[0053] The new shoot growth period, flowering period, berry swelling period, and coloring and ripening period are for summer pruning (growth period pruning) and are synchronized with the drone flight time. The dormancy period is for winter pruning. The branch pruning is carried out by professional agronomic pruners in the vineyard. The pruning weight of the branches in each plot is weighed using an electronic scale. The calculation formulas for the summer and winter branch pruning weights are as follows:

[0054]

[0055] Among them, W n is the new shoot pruning weight, W s is the lateral shoot pruning weight, W nf is the non-fruiting branch pruning weight, W d is the pest and disease branch pruning weight, W o is the old branch pruning weight, W f is the fruiting mother branch pruning weight, W t is the thin and weak branch pruning weight, and S is the sampling area of the research plot.

[0056] Step 3: Soil background removal. The grape canopy image is segmented through the U-net network model to remove the interference of canopy shadows, soil, and ground weeds on the spectral data.

[0057] As Figure 3 shown, the training of the U-Net network model annotates the dataset through LabelMe 5.5.0. The annotation categories are soil, canopy, and shadow respectively. The orthophoto image is segmented into non-repeating sub-images of 1024x1024 pixels for the dataset used for model training. The number of image channels is 3 (RGB). Finally, the accuracy of the segmentation is evaluated using the intersection over union (IoU) and overall accuracy (OA) of each category. The model training is carried out on a desktop computer workstation equipped with a Windows 11 system. The source code is based on TensorFlow 2.12.0. The U-Net hyperparameters are shown in Table 1.

[0058] Table 1

[0059]

[0060]

[0061] Step 4: Extraction of vegetation indices and texture indices. For the UAV multispectral images after removing canopy shadows and ground weeds, that is, the images that only retain the grape canopy after image segmentation, vegetation indices and texture indices are extracted. The vegetation indices include Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Normalized Difference Red Edge (NDRE), Red Edge Chlorophyll Index (CIRE), Ratio Vegetation Index (RVI), and Difference Vegetation Index (DVI). The calculation formulas are as follows:

[0062] NDVI = (NIR - Red) / (NIR + Red) (1)

[0063] GNDVI=(NIR - Green) / (NIR + Green) (2)

[0064] NDRE=(NIR - Red Edge) / (NIR + Red Edge) (3)

[0065] CIRE=(NIR / Red Edge)-1 (4)

[0066] RVI=NIR / Red (5)

[0067] DVI=NIR - Red (6)

[0068] where Red, Green, Red Edge, and NIR are the reflectance values of each band

[0069] The texture indices include Normalized Difference Texture Index (NDTI), Ratio Texture Index (RTI), and Difference Texture Index (DTI). The calculation formulas are as follows:

[0070] NDTI=(T1 - T2) / (T1 + T2) (7)

[0071] RTI=(T1 / T2) (8)

[0072] DTI=(T1 - T2) (9)

[0073] where T1 and T2 are the texture feature values of the selected bands. The texture features are calculated by the gray-level co-occurrence matrix (GLCM) to obtain 8 texture feature values: mean (Mea), variance (Var), homogeneity (Hom), contrast (Con), dissimilarity (Dis), entropy (Ent), second moment (Sec), and correlation (Cor). The index calculations are performed through the raster calculator and zonal statistics functions in QGIS - 3.34.3.

[0074] Step 5: Feature variable screening. Use the Pearson correlation coefficient to analyze the relationship between vegetation indices, texture indices, and the pruning weights of branches during the growth period and dormancy period to determine the sensitive indices.

[0075] Step 6: Construct a prediction model for grapevine branch pruning weight. As shown in Tables 2 and 3, during the new shoot growth period, flowering period, berry swelling period, and coloring and maturity period, for each growth period, select the 8 groups of vegetation indices and texture indices with the highest correlation with the branch pruning weight as independent variables, and use the corresponding branch pruning weight at each growth period as the dependent variable to establish a machine learning model. At the same time, for the entire growth period and dormancy period, select the 8 groups of vegetation indices and texture indices with the highest correlation with the branch pruning weight as independent variables, and use the branch pruning weight during the dormancy period as the dependent variable to construct another machine learning model.

[0076] Table 2

[0077]

[0078] Table 3

[0079]

[0080] The machine learning models are partial least squares regression (PLSR), random forest (RF), and artificial neural network (ANN). The accuracy of the models is evaluated by the coefficient of determination (R 2 ) and the root mean square error (RMSE). Finally, the artificial neural network (ANN) is determined to be the optimal prediction model for grapevine branch pruning weight.

[0081] Step 7: According to the prediction results of the grapevine branch pruning weight model, use co-kriging interpolation to predict the spatial distribution of the branch pruning weight of the entire vineyard.

[0082] Through the co-kriging interpolation spatial statistics method, based on the spherical model in the semi-variogram model, use the branch pruning weights during the growth period and dormancy period predicted by the artificial neural network model as the target variables and the spectral data as the auxiliary variables to predict the spatial distribution of the branch pruning weight of the entire vineyard.

[0083] The co-kriging interpolation is used to predict the spatial distribution of the branch pruning weight based on the following formula:

[0084]

[0085] where is the estimated value of the target variable Z 1 at the position s to be predicted; Z 1 (s i ) is the target variable Z 1 at the known sampling point si Value; Z 2 (s j ), Z p (s k ) is the auxiliary variable Z 2 , …, Z p at its sampling points; λ 1i , λ 2j , λ pk is the interpolation weight, corresponding to the target variable and the auxiliary variable respectively; n 1 , n 2 , n p is the number of sample points of the target variable and the auxiliary variable.

[0086] In the embodiment of the present invention, the spectral data of the grape canopy at different growth stages is collected by an unmanned aerial vehicle (UAV). During the growth period and the dormant period, the branches of the experimental plots are pruned and weighed respectively based on the calculation formulas of summer pruning weight (SPW) and winter pruning weight (WPW). After using the U-net network model to eliminate the interference of canopy shadows and weeds on the spectral data, the vegetation index and texture index in the experimental area are extracted, and their correlations with the pruning weight of the branches are analyzed. Based on the results of the correlation analysis, machine learning models for predicting the pruning weight of wine grapes are constructed respectively during the growth period and the dormant period, and combined with the model-driven geostatistical interpolation technology, using the pruning weight and spectral data of the known prediction points, the spatial distribution of the pruning weight of the entire vineyard is predicted. The present invention effectively solves the problems in the traditional measurement of pruning weight of branches, such as high labor intensity, low efficiency, limited accuracy by sample representativeness, dependence on experience and strong subjectivity, etc., and provides a scientific basis for the yield evaluation, growth monitoring and agronomic management of wine grapes.

[0087] Embodiment 2:

[0088] The embodiment of the present invention also provides a prediction system for the pruning weight of wine grape branches, including:

[0089] A collection device for collecting the spectral data of the grape canopy at different growth stages by an unmanned aerial vehicle;

[0090] A calculation device for pruning and weighing the branches of the experimental plots respectively during the growth period and the dormant period;

[0091] A processing device for using the U-net network model to eliminate the interference of canopy shadows and weeds on the spectral data, extracting the vegetation index and texture index in the experimental area, and analyzing their correlations with the pruning weight of the branches;

[0092] A first prediction device for constructing machine learning models for predicting the pruning weight of wine grapes respectively during the growth period and the dormant period based on the results of the correlation analysis;

[0093] The first prediction device is used to predict the spatial distribution of the pruning weight of branches in the entire vineyard by using the prediction results.

[0094] As an implementation manner of an embodiment of the present invention, the calculation device prunes the branches during the new shoot growth period, flowering period, berry swelling period, coloring and maturity period, and dormancy period of the grapes, and weighs the pruned branches in each experimental plot according to the summer pruning weight (SPW) and winter pruning weight (WPW) calculation formulas respectively. The summer and winter pruning weight calculation formulas are as follows:

[0095]

[0096] Among them, W n is the new shoot pruning weight, W s is the lateral shoot pruning weight, W nf is the non-fruiting branch pruning weight, W d is the pest and disease branch pruning weight, W o is the old branch pruning weight, W f is the fruiting mother branch pruning weight, W t is the thin and weak branch pruning weight, and S is the sampling area of the research plot.

[0097] As an implementation manner of an embodiment of the present invention, the second prediction device uses the co-Kriging interpolation spatial statistics method, based on the spherical model in the semi-variogram model, takes the pruning weights of branches during the growth period and dormancy period predicted by the artificial neural network model as the target variable, and the spectral data as the auxiliary variable to predict the spatial distribution of the pruning weights of branches in the entire vineyard; among them,

[0098] The co-Kriging interpolation is used to predict the spatial distribution of the pruning weight of branches based on the following formula:

[0099]

[0100] Among them, is the estimated value of the target variable Z 1 at the position s to be predicted; Z 1 (s i ) is the value of the target variable Z 1 at the known sampling point s i ; Z 2 (s j ), Z p (s k ) are the values of the auxiliary variables Z 2 ,…,Z p at their sampling points; λ 1i ,λ 2j ,λ pk are the interpolation weights, corresponding to the target variable and the auxiliary variable respectively; n 1,n 2 ,n p The number of sample points for the target variable and the auxiliary variable.

[0101] Example 3:

[0102] The embodiment of the present invention further provides a storage medium, on which a computer program is stored, and the computer program executes the method for predicting the pruning weight of wine grape branches when running.

[0103] The above-described embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A wine grape branch pruning weight prediction method, characterized in that: include: Step S1, collecting grape canopy spectral data at different growth stages by using a drone; Step S2, pruning and weighing the branches in the experimental plot during the growth period and the dormancy period respectively; Step S3, after removing the interference of canopy shadows and weeds on the spectral data using the U-net network model, extract the vegetation index and texture index in the experimental area, and analyze their correlation with the branch pruning weight; Step S4: Based on the correlation analysis results, construct machine learning models for wine grapes in the growth period and dormancy period to predict branch pruning weight; Step S5: using the prediction results, predict the spatial distribution of the branch pruning weight of the entire vineyard.

2. The wine grape branch pruning weight prediction method according to claim 1, characterized in that: In step S2, the branches are pruned during the growth period of new grape shoots, flowering period, berry expansion period, coloring and ripening period and dormancy period. The pruned branches in each experimental plot are weighed according to the calculation formulas of summer branch pruning weight SPW and winter branch pruning weight WPW. The calculation formulas of summer and winter branch pruning weight are as follows: Among them, W n is the new shoot pruning weight, W s is the pruning weight of the secondary shoots, W nf Pruning weight for fruitless branches, W d W is the pruning weight of diseased and insect-damaged branches. o Pruning weight for old branches, W f is the pruning weight of the fruiting mother branch, W t is the pruning weight of thin and weak branches, and S is the sampling area of ​​the research plot.

3. The wine grape branch pruning weight prediction method according to claim 2, characterized in that: In step S5, the spatial distribution of the branch pruning weight in the entire vineyard is predicted by using the cokriging interpolation spatial statistics method, based on the spherical model in the semivariogram model, taking the branch pruning weight in the growing period and the dormant period predicted by the artificial neural network model as the target variable and the spectral data as the auxiliary variable; wherein, Cokriging interpolation predicts the spatial distribution of branch pruning weight based on the following formula: in, is the estimated value of the target variable Z1 at the position to be predicted s; Z1(s i ) is the target variable Z1 at the known sampling point s i The value of Z2(s j ), Z p (s k ) are auxiliary variables Z2,…,Z p The value at its sampling point; 1i ,λ 2j ,λ pk are interpolation weights, corresponding to the target variable and auxiliary variable respectively; n1, n2, n p is the number of sample points for the target variable and auxiliary variable.

4. A wine grape branch pruning weight prediction system, characterized in that: include: A collection device, used to collect spectral data of grape canopies at different growth stages through drones; The computing device is used to prune and weigh the branches in the experimental plot during the growth period and the dormancy period; A processing device, used to extract vegetation index and texture index in the experimental area after eliminating interference of canopy shadow and weeds on spectral data by using a U-net network model, and analyze the correlation between the vegetation index and texture index and the branch pruning weight; The first prediction device is used to construct machine learning models of wine grapes in the growth period and dormancy period to predict the branch pruning weight based on the correlation analysis results; The first prediction device is used to use the prediction results to predict the spatial distribution of the branch pruning weight of the entire vineyard.

5. The wine grape branch pruning weight prediction system according to claim 4, characterized in that: The calculation device prunes the branches during the growth period of new shoots, flowering period, berry expansion period, coloring and ripening period and dormancy period of grapes. The pruned branches of each experimental plot are weighed according to the calculation formulas of summer branch pruning weight SPW and winter branch pruning weight WPW. The calculation formulas of summer and winter branch pruning weight are as follows: Among them, W n is the new shoot pruning weight, W s is the pruning weight of the secondary shoots, W nf Pruning weight for fruitless branches, W d W is the pruning weight of diseased and insect-damaged branches. o Pruning weight for old branches, W f is the pruning weight of the fruiting mother branch, W t is the pruning weight of thin and weak branches, and S is the sampling area of ​​the research plot.

6. The wine grape branch pruning weight prediction system according to claim 5, characterized in that: The second prediction device uses the co-kriging interpolation spatial statistics method, based on the spherical model in the semivariogram model, and uses the branch pruning weight predicted by the artificial neural network model during the growth period and the dormancy period as the target variable, and the spectral data as the auxiliary variable to predict the spatial distribution of the branch pruning weight of the entire vineyard; Cokriging interpolation predicts the spatial distribution of branch pruning weight based on the following formula: in, is the estimated value of the target variable Z1 at the position to be predicted s; Z1(s i ) is the target variable Z1 at the known sampling point s i The value of Z2(s j ), Z p (s k ) are auxiliary variables Z2,…,Z p The value at its sampling point; 1i ,λ 2j ,λ pk are interpolation weights, corresponding to the target variable and auxiliary variable respectively; n1, n2, n p is the number of sample points for the target variable and auxiliary variable.

7. A storage medium, characterized in that: The storage medium stores a computer program, which executes the wine grape branch pruning weight prediction method according to claim 1 when running.

Images