Method, device and medium for estimating density of new shoots of wine grapes based on drone images

Abstract

The present invention discloses a method, device and medium for estimating the density of new shoots of wine grapes based on drone images, wherein the method comprises: determining a first orthophoto and a second orthophoto respectively according to drone image data; calculating a dual-band vegetation index based on the first orthophoto; calculating visible light texture features based on the second orthophoto; splicing the dual-band vegetation index and the visible light texture features to form a feature vector; taking the density of new shoots of grapes in the corresponding image area collected on the ground as the target feature, constructing an integrated model based on a stacking strategy, and using the feature vector to train the integrated model to obtain a model for estimating the density of new shoots of grapes. The present invention can solve the problems that the existing traditional estimation method requires a lot of manpower and time, has subjectivity and inconsistency, lacks real-time and dynamic performance in estimating the density of new shoots of grapes, and has a very limited spatial coverage.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle remote sensing technology, and more specifically, to a method, device and medium for estimating the density of new shoots of wine grapes based on unmanned aerial vehicle images. Background Art

[0002] In recent years, the wine grape industry has grown rapidly and has become one of the most important industries in the global agricultural field. The growth status and yield of grape plants are closely related to the growth density of their new shoots. Therefore, accurately estimating the density of grape shoots is crucial for vineyard management and yield prediction.

[0003] Traditional methods for estimating the density of new grape shoots are mainly based on manual measurement and statistics, including the following two common methods: The first method is to manually select a certain number of sample plots in the vineyard and manually count the new grape shoots in each sample plot, and then obtain the new shoot density based on the ratio of the sample area to the total area. This method requires a high labor intensity and is prone to human subjective errors. The second method uses aerial images or satellite images to interpret the vineyard, and identifies and counts the new grape shoots in the image through manual or computer-assisted means. This method can improve work efficiency, but it still requires manual interpretation and counting, and for large-scale vineyards, the workload and time are still large.

[0004] At present, the traditional estimation method requires a lot of manpower and time, and there are problems of subjectivity and inconsistency. In addition, the traditional method for estimating the density of grape shoots lacks real-time and dynamics, and its spatial coverage is very limited. Therefore, it is necessary to develop a wide-ranging and accurate method for estimating the density of wine grape shoots. Summary of the invention

[0005] The present invention is provided to solve the above problems existing in the prior art. Therefore, a method, device and medium for estimating the density of new shoots of wine grapes based on drone images are needed to solve the problems that the existing traditional estimation method requires a lot of manpower and time, has subjectivity and inconsistency, lacks real-time and dynamic performance in estimating the density of new shoots of grapes, and has a very limited spatial coverage, thereby providing a wide-range and accurate method for estimating the density of new shoots of wine grapes.

[0006] According to a first aspect of the present invention, a method for estimating the density of new shoots of wine grapes based on drone images is provided, the method comprising:

[0007] Acquire drone image data, wherein the drone image data includes vineyard drone hyperspectral image and vineyard drone visible light image;

[0008] Obtaining a first orthophoto based on the vineyard drone hyperspectral image;

[0009] Obtain a second orthophoto based on the vineyard drone visible light image;

[0010] Based on the first orthophoto, a sensitive spectral wavelength is determined by correlation analysis, and a dual-band vegetation index is constructed using the sensitive spectral bands, wherein the dual-band vegetation index includes a normalized vegetation index, a photochemical reflectance index, a structure-insensitive pigment index, and a green normalized difference vegetation index;

[0011] Based on the second orthophoto, calculating visible light texture features, the visible light texture features including contrast, entropy and correlation;

[0012] splicing the dual-band vegetation index and the visible light texture feature to form a feature vector;

[0013] Taking the density of new grape shoots in the corresponding image area collected on the ground as the target feature, an integrated model based on the stacking strategy is constructed, and the integrated model is trained using the feature vector to obtain a new grape shoot density estimation model.

[0014] Further, obtaining a first orthophoto according to the vineyard drone hyperspectral image specifically includes:

[0015] Using background data collected by the drone with the lens cap on and the calibration file, the lens of the vineyard drone hyperspectral image is calibrated;

[0016] Based on the lens calibrated data, reflectivity calibration is performed under a dark background and a white frame background to obtain reflectivity calibrated data;

[0017] Using the gray cloth curve, batch processing is performed on the reflectivity calibrated data to obtain final data;

[0018] Perform image stitching on the final data to obtain a first orthophoto.

[0019] Furthermore, the normalized difference vegetation index, photochemical reflectance index, structure-insensitive pigment index, and green normalized difference vegetation index were calculated by the following formulas:

[0020]

[0021]

[0022] Among them, NDVI, PRI, SIPI, and GNDVI represent the normalized difference vegetation index, photochemical reflectance index, structural insensitive pigment index, and green normalized difference vegetation index, respectively; 800 、r 670 、r531 、r 570 、r 445 、r 680 、r 750 and r 550 They represent the spectra at wavelengths of 800nm, 670nm, 531nm, 570nm, 445nm, 680nm, 750nm, and 550nm respectively.

[0023] Further, based on the second orthophoto, the visible light texture feature is calculated, specifically including:

[0024] Based on the second orthophoto, the numerical values ​​of the RGB channels are weighted averaged to obtain a grayscale image. For each pixel in the grayscale image, the relationship between each pixel and the adjacent pixels is calculated. By sliding a window on the image, the frequency of occurrence of pixel pairs in the window is calculated to obtain a grayscale co-occurrence matrix, and the grayscale co-occurrence matrix is ​​normalized. According to the normalized grayscale co-occurrence matrix, three texture features of contrast, entropy, and correlation are calculated and extracted.

[0025] Furthermore, contrast, entropy and correlation are calculated by the following formulas:

[0026]

[0027] Where Con represents contrast, Ent represents entropy, corr represents correlation, i represents the grayscale of any point (x, y), j represents the grayscale of another point deviating from this point, P(i, j) represents the frequency of occurrence of a pixel starting from (x, y) with a distance d from (x, y) and a grayscale j, and μ x Expressed as the variance of the texture parameters in the row direction, μ y Expressed as the variance of the texture parameters in the column direction, σ x Expressed as the mean value of the texture parameter in the row direction, σ y Expressed as the mean of the texture parameters in the column direction.

[0028] Furthermore, the integrated model based on the stacking strategy includes a support vector machine, a random forest module, a K-nearest neighbor module and a linear regressor.

[0029] Furthermore, when the integrated model is trained using the feature vector, the feature vector is input into the support vector machine, random forest module, and K-nearest neighbor module included in the integrated model based on the stacking strategy for training and prediction, and multiple prediction results are combined through a linear regressor to obtain a grape shoot density estimation model.

[0030] According to a second aspect of the present invention, a device for estimating the density of new shoots of wine grapes based on drone images is provided, the device comprising:

[0031] An image data acquisition module is configured to acquire drone image data, wherein the drone image data includes vineyard drone hyperspectral images and vineyard drone visible light images;

[0032] A first image processing module is configured to obtain a first orthophoto according to the vineyard UAV hyperspectral image;

[0033] A second image processing module is configured to obtain a second orthophoto according to the vineyard drone visible light image;

[0034] A vegetation index calculation module is configured to determine the sensitive spectral wavelength through correlation analysis based on the first orthophoto, and construct a dual-band vegetation index using the sensitive spectral band, wherein the dual-band vegetation index includes a normalized vegetation index, a photochemical reflectance index, a structure-insensitive pigment index, and a green normalized difference vegetation index;

[0035] A texture feature calculation module is configured to calculate visible light texture features based on the second orthophoto, wherein the visible light texture features include contrast, entropy and correlation;

[0036] A feature vector determination module is configured to concatenate the dual-band vegetation index with the visible light texture feature to form a feature vector;

[0037] The estimation model building module is configured to take the density of new grape shoots in the corresponding image area collected on the ground as the target feature, construct an integrated model based on the stacking strategy, and use the feature vector to train the integrated model to obtain a new grape shoot density estimation model.

[0038] Furthermore, the first image processing module is further configured to:

[0039] Using background data collected by the drone with the lens cap on and the calibration file, the lens of the vineyard drone hyperspectral image is calibrated;

[0040] Based on the lens calibrated data, reflectivity calibration is performed under a dark background and a white frame background to obtain reflectivity calibrated data;

[0041] Using the gray cloth curve, batch processing is performed on the reflectivity calibrated data to obtain final data;

[0042] Perform image stitching on the final data to obtain a first orthophoto.

[0043] According to a third aspect of the present invention, there is provided a readable storage medium storing one or more programs, wherein the one or more programs can be executed by one or more processors to implement the method as described above.

[0044] The present invention has at least the following beneficial effects:

[0045] The present invention can solve the problems that the existing traditional estimation methods require a lot of manpower and time, and are subjective and inconsistent, and the grape shoot density estimation lacks real-time and dynamics and its spatial coverage is very limited. Based on the visible light and hyperspectral image data of unmanned aerial vehicles, an integrated model based on a stacking strategy is constructed by fusing visible light texture features and vegetation indexes constructed based on hyperspectral images, thereby providing a wide-ranging and accurate wine grape shoot density estimation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A flow chart of a method for estimating the density of new shoots of wine grapes based on drone images according to an embodiment of the present invention is shown.

[0047] Figure 2 A flowchart of processing vineyard drone hyperspectral images according to an embodiment of the present invention is shown.

[0048] Figure 3 A schematic diagram of feature vector concatenation according to an embodiment of the present invention is shown.

[0049] Figure 4 A structural diagram of a wine grape new shoot density estimation device based on drone images according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and specific embodiments, but are not intended to limit the present invention. For the various steps described herein, if there is no necessity for a causal relationship between each other, the order in which they are described as examples herein should not be regarded as a limitation, and those skilled in the art should know that they can be adjusted in order, as long as the logic between them is not destroyed, resulting in the inability to implement the entire process.

[0051] The embodiment of the present invention provides a method for estimating the density of new shoots of wine grapes based on drone images. Figure 1 As shown, it is a flow chart of the method, which includes the following steps S100 to S700, which are described in detail as follows:

[0052] Step S100, obtaining drone image data, wherein the drone image data includes vineyard drone hyperspectral images and vineyard drone visible light images.

[0053] Step S200, obtaining a first orthophoto according to the vineyard UAV hyperspectral image.

[0054] In some embodiments, Figure 2 As shown in the figure, it is a flow chart for obtaining the first orthophoto according to the vineyard UAV hyperspectral image. The first orthophoto is obtained through the following steps:

[0055] Step S201, using background data collected by the drone with the lens cover on and a calibration file to perform lens calibration on the vineyard drone hyperspectral image.

[0056] Exemplarily, lens calibration was performed in the SpecView software, importing the original data (i.e., the vineyard drone hyperspectral image), background data collected with the lens cap, and accurate calibration files, and exporting the lens-calibrated data.

[0057] Step S202 , based on the lens calibrated data, reflectivity calibration is performed under a dark background and a white frame background to obtain reflectivity calibrated data.

[0058] Exemplarily, when performing reflectivity calibration, the lens calibrated data, dark background, and white frame background are imported into the SpecView software, and the reflectivity calibrated data is exported.

[0059] Step S203, using the gray cloth curve to batch process the reflectivity calibrated data to obtain final data.

[0060] Exemplarily, during atmospheric correction, the data after reflectivity calibration is loaded, gray cloth is selected, the gray cloth curve is imported, batch processing is performed, and the final data is output.

[0061] Step S204: stitching the final data to obtain a first orthophoto.

[0062] Exemplarily, the final data obtained in step S203 is imported into Photoscan, image stitching is performed, and an orthophoto is output to obtain a first orthophoto.

[0063] Step S300, obtaining a second orthophoto according to the vineyard drone visible light image.

[0064] For example, after obtaining the visible light image of the vineyard from the drone, it is imported into Photoscan, image stitching is performed, and the second orthophoto is output.

[0065] Step S400, based on the first orthophoto, determine the sensitive spectral wavelength through correlation analysis, and use the sensitive spectral band to construct a dual-band vegetation index, wherein the dual-band vegetation index includes a normalized vegetation index, a photochemical reflectance index, a structure-insensitive pigment index, and a green normalized difference vegetation index.

[0066] In some embodiments, the normalized vegetation index, the photochemical reflectance index, the structure-insensitive pigment index, and the green normalized difference vegetation index are calculated by the following formulas:

[0067]

[0068]

[0069] Among them, NDVI, PRI, SIPI, and GNDVI represent the normalized difference vegetation index, photochemical reflectance index, structural insensitive pigment index, and green normalized difference vegetation index, respectively; 800 、r 670 、r 531 、r 570 、r 445 、r 680 、r 750 and r 550 They represent the spectra at wavelengths of 800nm, 670nm, 531nm, 570nm, 445nm, 680nm, 750nm, and 550nm respectively.

[0070] Step S500: Calculate visible light texture features based on the second orthophoto, where the visible light texture features include contrast, entropy and correlation.

[0071] In some embodiments, based on the second orthophoto, the visible light texture feature is calculated by the following method:

[0072] The spliced ​​visible light image (i.e., the second orthophoto) is processed, and the grayscale image is obtained by weighted averaging the values ​​of the RGB channels. For each pixel, its relationship with the adjacent pixels is calculated. A window is slid across the image, the frequency of occurrence of pixel pairs within the window is calculated, and the calculated gray-level co-occurrence matrix (GLCM) is normalized to eliminate the influence of image brightness and contrast. Based on the normalized GLCM, three texture features, contrast, entropy, and correlation, are calculated and extracted.

[0073] In some embodiments, contrast, entropy, and correlation are calculated using the following formulas:

[0074]

[0075] Where Con represents contrast, Ent represents entropy, corr represents correlation, i represents the grayscale of any point (x, y), j represents the grayscale of another point deviating from this point, P(i, j) represents the frequency of occurrence of a pixel starting from (x, y) with a distance d from (x, y) and a grayscale j, and μ x Expressed as the variance of the texture parameters in the row direction, μ y Expressed as the variance of the texture parameters in the column direction, σx Expressed as the mean value of the texture parameter in the row direction, σ y Expressed as the mean of the texture parameters in the column direction.

[0076] Contrast reflects the clarity of the image and the depth of the texture grooves. The deeper the texture grooves, the greater the contrast and the clearer the effect; conversely, the shallower the grooves, the smaller the contrast and the more blurred the effect.

[0077] The entropy value shows the complexity of the grayscale distribution of the image. The larger the entropy value, the more complex the image. When the entropy value is the largest, it indicates that all values ​​in the co-occurrence matrix are equal or the pixel values ​​show the greatest randomness.

[0078] The correlation reflects the similarity of the gray levels of the image in the row or column direction, and its size shows the local gray level correlation. The larger the value, the greater the correlation.

[0079] Step S600: concatenate the dual-band vegetation index and the visible light texture feature to form a feature vector.

[0080] See also Figure 3 , which shows a schematic diagram of splicing dual-band vegetation index and visible light texture features, where NDVI, PRI, SIPI, and GNDVI represent normalized vegetation index, photochemical reflectance index, structure-insensitive pigment index, and green normalized difference vegetation index, respectively, Con represents contrast, Ent represents entropy, and corr represents correlation. As an example only, you can use Figure 3 The splicing method shown is to splice NDVI, PRI, SIPI, GNDVI, Con, Ent, and corr from left to right to obtain a characteristic vector. It should be noted that the splicing order of each index and each texture feature described above is only an example and is not a limitation of the present invention.

[0081] Step S700, taking the density of new grape shoots in the corresponding image area collected on the ground as the target feature, constructing an integrated model based on the stacking strategy, and using the feature vector to train the integrated model to obtain a new grape shoot density estimation model.

[0082] It should be noted that the density of new grape shoots in the corresponding image area collected on the ground refers to the measured data overlapped with the drone image area measured on the ground using cameras and other shooting equipment, that is, this embodiment trains the constructed integrated model in a supervised manner, and then obtains a new grape shoot density estimation model to achieve accurate estimation of the new grape shoot density.

[0083] In some embodiments, the density of new grape shoots in the corresponding image area collected on the ground is used as the target feature, and an integrated model based on a stacking strategy is constructed. The obtained feature vector is input into a support vector machine, a random forest, and a K-nearest neighbor for training and prediction. Multiple prediction results are combined through a linear regressor to obtain a grape shoot density estimation model based on drone images.

[0084] The embodiment of the present invention also provides a device for estimating the density of new shoots of wine grapes based on drone images, such as Figure 4 As shown, the device 400 includes:

[0085] An image data acquisition module 401 is configured to acquire drone image data, wherein the drone image data includes vineyard drone hyperspectral images and vineyard drone visible light images;

[0086] The first image processing module 402 is configured to obtain a first orthophoto according to the vineyard UAV hyperspectral image;

[0087] The second image processing module 403 is configured to obtain a second orthophoto according to the vineyard drone visible light image;

[0088] The vegetation index calculation module 404 is configured to determine the sensitive spectral wavelength through correlation analysis based on the first orthophoto, and construct a dual-band vegetation index using the sensitive spectral bands, wherein the dual-band vegetation index includes a normalized vegetation index, a photochemical reflectance index, a structure-insensitive pigment index, and a green normalized difference vegetation index;

[0089] The texture feature calculation module 405 is configured to calculate visible light texture features based on the second orthophoto, wherein the visible light texture features include contrast, entropy and correlation;

[0090] A feature vector determination module 406 is configured to concatenate the dual-band vegetation index with the visible light texture feature to form a feature vector;

[0091] The estimation model building module 407 is configured to take the density of new grape shoots in the corresponding image area collected on the ground as the target feature, construct an integrated model based on the stacking strategy, and use the feature vector to train the integrated model to obtain a new grape shoot density estimation model.

[0092] In some embodiments, the first image processing module is further configured to:

[0093] Using background data collected by the drone with the lens cap on and the calibration file, the lens of the vineyard drone hyperspectral image is calibrated;

[0094] Based on the lens calibrated data, reflectivity calibration is performed under a dark background and a white frame background to obtain reflectivity calibrated data;

[0095] Using the gray cloth curve, batch processing is performed on the reflectivity calibrated data to obtain final data;

[0096] Perform image stitching on the final data to obtain a first orthophoto.

[0097] In some embodiments, the vegetation index calculation module is further configured to calculate the normalized vegetation index, the photochemical reflectance index, the structure-insensitive pigment index, and the green normalized difference vegetation index by the following formula:

[0098]

[0099]

[0100] Among them, NDVI, PRI, SIPI, and GNDVI represent the normalized difference vegetation index, photochemical reflectance index, structural insensitive pigment index, and green normalized difference vegetation index, respectively; 800 、r 670 、r 531 、r 570 、r 445 、r 680 、r 750 and r 550 They represent the spectra at wavelengths of 800nm, 670nm, 531nm, 570nm, 445nm, 680nm, 750nm, and 550nm respectively.

[0101] In some embodiments, the texture feature calculation module is further configured to:

[0102] Based on the second orthophoto, the numerical values ​​of the RGB channels are weighted averaged to obtain a grayscale image. For each pixel in the grayscale image, the relationship between each pixel and the adjacent pixels is calculated. By sliding a window on the image, the frequency of occurrence of pixel pairs in the window is calculated to obtain a grayscale co-occurrence matrix, and the grayscale co-occurrence matrix is ​​normalized. According to the normalized grayscale co-occurrence matrix, three texture features of contrast, entropy, and correlation are calculated and extracted.

[0103] In some embodiments, the texture feature calculation module is further configured to:

[0104]

[0105] Where Con represents contrast, Ent represents entropy, corr represents correlation, i represents the grayscale of any point (x, y), j represents the grayscale of another point deviating from this point, P(i, j) represents the frequency of occurrence of a pixel starting from (x, y) with a distance d from (x, y) and a grayscale j, and μ x Expressed as the variance of the texture parameters in the row direction, μ y Expressed as the variance of the texture parameters in the column direction, σ x Expressed as the mean value of the texture parameter in the row direction, σ y Expressed as the mean of the texture parameters in the column direction.

[0106] In some embodiments, the stacking strategy-based integrated model includes a support vector machine, a random forest module, a K-nearest neighbor module, and a linear regressor.

[0107] In some embodiments, the estimation model building module is further configured to input the feature vector into an integrated model based on a stacking strategy including a support vector machine, a random forest module, and a K-nearest neighbor module for training and prediction when the integrated model is trained using the feature vector, and combine multiple prediction results through a linear regressor to obtain a grape shoot density estimation model.

[0108] It should be noted that the various device structures described in this embodiment belong to the same technical concept as the previously described method, and achieve the same technical effect through the same principle, which will not be repeated here.

[0109] An embodiment of the present invention further provides a readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the method described in any of the above embodiments.

[0110] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present invention with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. The elements in the claims will be interpreted broadly based on the language adopted in the claims, and are not limited to the examples described in this specification or during the implementation of this application, and the examples will be interpreted as non-exclusive. Therefore, this specification and examples are intended to be considered as examples only, and the true scope and spirit are indicated by the following claims and the full scope of their equivalents.

[0111] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art can use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features can be grouped together to simplify the present invention. This should not be interpreted as a feature of an invention that is not claimed for protection being necessary for any claim. On the contrary, the subject matter of the present invention may be less than all the features of the embodiments of a specific invention. Thus, the following claims are incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently used as a separate embodiment, and it is considered that these embodiments can be combined with each other in various combinations or arrangements. The scope of the present invention should be determined with reference to the full scope of the equivalent forms of the attached claims and these claims.

Claims

1. A method for estimating the density of new shoots of wine grapes based on drone images, characterized in that: The method comprises: Acquire drone image data, wherein the drone image data includes vineyard drone hyperspectral image and vineyard drone visible light image; Obtaining a first orthophoto based on the vineyard drone hyperspectral image; Obtain a second orthophoto based on the vineyard drone visible light image; Based on the first orthophoto, a sensitive spectral wavelength is determined by correlation analysis, and a dual-band vegetation index is constructed using the sensitive spectral bands, wherein the dual-band vegetation index includes a normalized vegetation index, a photochemical reflectance index, a structure-insensitive pigment index, and a green normalized difference vegetation index; Based on the second orthophoto, calculating visible light texture features, the visible light texture features including contrast, entropy and correlation; splicing the dual-band vegetation index and the visible light texture feature to form a feature vector; Taking the density of new grape shoots in the corresponding image area collected on the ground as the target feature, an integrated model based on the stacking strategy is constructed, and the integrated model is trained using the feature vector to obtain a new grape shoot density estimation model; The normalized difference vegetation index, photochemical reflectance index, structure-insensitive pigment index, and green normalized difference vegetation index were calculated using the following formulas: Among them, NDVI, PRI, SIPI, and GNDVI represent the normalized difference vegetation index, photochemical reflectance index, structural insensitive pigment index, and green normalized difference vegetation index, respectively; 800 、r 670 、r 531 、r 570 、r 445 、r 680 、r 750 and r 550 Respectively represent the spectra at wavelengths of 800nm, 670nm, 531nm, 570nm, 445nm, 680nm, 750nm, and 550nm; The integrated model based on the stacking strategy includes a support vector machine, a random forest module, a K-nearest neighbor module and a linear regressor; When the integrated model is trained using the feature vector, the feature vector is input into the support vector machine, random forest module and K-nearest neighbor module included in the integrated model based on the stacking strategy for training and prediction, and multiple prediction results are combined through a linear regressor to obtain a grape shoot density estimation model.

2. The method according to claim 1, characterized in that The first orthophoto is obtained according to the vineyard UAV hyperspectral image, specifically including: Using background data collected by the drone with the lens cap on and the calibration file, the lens of the vineyard drone hyperspectral image is calibrated; Based on the lens calibrated data, reflectivity calibration is performed under a dark background and a white frame background to obtain reflectivity calibrated data; Using the gray cloth curve, batch processing is performed on the reflectivity calibrated data to obtain final data; Perform image stitching on the final data to obtain a first orthophoto.

3. The method according to claim 1, characterized in that Based on the second orthophoto, the visible light texture feature is calculated, specifically including: Based on the second orthophoto, the numerical values ​​of the RGB channels are weighted averaged to obtain a grayscale image. For each pixel in the grayscale image, the relationship between each pixel and the adjacent pixels is calculated. By sliding a window on the image, the frequency of occurrence of pixel pairs in the window is calculated to obtain a grayscale co-occurrence matrix, and the grayscale co-occurrence matrix is ​​normalized. According to the normalized grayscale co-occurrence matrix, three texture features of contrast, entropy, and correlation are calculated and extracted.

4. The method according to claim 1, characterized in that: Contrast, entropy and correlation are calculated by the following formulas: Where Con represents contrast, Ent represents entropy, corr represents correlation, i represents the grayscale of any point (x, y), j represents the grayscale of another point deviating from this point, P(i, j) represents the frequency of occurrence of a pixel starting from (x, y) with a distance d from (x, y) and a grayscale j, and μ x Expressed as the variance of the texture parameters in the row direction, μ y Expressed as the variance of the texture parameters in the column direction, σ x Expressed as the mean value of the texture parameter in the row direction, σ y Expressed as the mean of the texture parameters in the column direction.

5. A wine grape new shoot density estimation device based on drone images, characterized in that: The device comprises: An image data acquisition module is configured to acquire drone image data, wherein the drone image data includes vineyard drone hyperspectral images and vineyard drone visible light images; A first image processing module is configured to obtain a first orthophoto according to the vineyard UAV hyperspectral image; A second image processing module is configured to obtain a second orthophoto according to the vineyard drone visible light image; A vegetation index calculation module is configured to determine the sensitive spectral wavelength through correlation analysis based on the first orthophoto, and construct a dual-band vegetation index using the sensitive spectral band, wherein the dual-band vegetation index includes a normalized vegetation index, a photochemical reflectance index, a structure-insensitive pigment index, and a green normalized difference vegetation index; A texture feature calculation module is configured to calculate visible light texture features based on the second orthophoto, wherein the visible light texture features include contrast, entropy and correlation; A feature vector determination module is configured to concatenate the dual-band vegetation index with the visible light texture feature to form a feature vector; The estimation model building module is configured to take the density of new grape shoots in the corresponding image area collected on the ground as the target feature, build an integrated model based on the stacking strategy, and train the integrated model using the feature vector to obtain a new grape shoot density estimation model; The vegetation index calculation module is further configured to calculate the normalized vegetation index, the photochemical reflectance index, the structure-insensitive pigment index, and the green normalized difference vegetation index by the following formula: Among them, NDVI, PRI, SIPI, and GNDVI represent the normalized difference vegetation index, photochemical reflectance index, structural insensitive pigment index, and green normalized difference vegetation index, respectively; 800 、r 670 、r 531 、r 570 、r 445 、r 680 、r 750 and r 550 Respectively represent the spectra at wavelengths of 800nm, 670nm, 531nm, 570nm, 445nm, 680nm, 750nm, and 550nm; The integrated model based on the stacking strategy includes a support vector machine, a random forest module, a K-nearest neighbor module and a linear regressor; The estimation model building module is further configured to input the feature vector into the integrated model based on the stacking strategy including a support vector machine, a random forest module, and a K-nearest neighbor module for training and prediction when the integrated model is trained using the feature vector, and combine multiple prediction results through a linear regressor to obtain a grape shoot density estimation model.

6. The device according to claim 5, characterized in that The first image processing module is further configured to: Using background data collected by the drone with the lens cap on and the calibration file, the lens of the vineyard drone hyperspectral image is calibrated; Based on the lens calibrated data, reflectivity calibration is performed under a dark background and a white frame background to obtain reflectivity calibrated data; Using the gray cloth curve, batch processing is performed on the reflectivity calibrated data to obtain final data; Perform image stitching on the final data to obtain a first orthophoto. 7 . A non-transitory computer-readable storage medium storing instructions, which, when executed by a processor, executes the method according to claim 1 .

Images