Method for obtaining crude protein yield of forage grass and related equipment

Abstract

The invention discloses a forage crude protein yield obtaining method and related equipment, and relates to the technical field of smart livestock and agricultural remote sensing, and the method comprises the steps: collecting and preprocessing multispectral data of a preset area through a multispectral sensor carried by an unmanned aerial vehicle, and then obtaining the surface reflectance of the preset area; the vegetation nitrogen content index and the vegetation index of the preset area and the grass yield of the preset area are calculated; extracting terrain factors of the preset area based on the digital elevation model of the preset area; inputting the vegetation nitrogen content index, the vegetation index, the grass yield and the topographic factor into the trained learning model to obtain the crude protein content of the preset area; and calculating the forage grass crude protein yield of the preset area according to the grass yield and the crude protein content. According to the method, a high-precision and high-robustness quantitative inversion method from unmanned aerial vehicle multispectral data to forage grass crude protein yield is established.

Description

Technical Field

[0001] This invention relates to the field of smart animal husbandry and agricultural remote sensing technology, and in particular to a method and related equipment for obtaining crude protein yield from forage. Background Technology

[0002] In traditional alpine grassland grazing management and carrying capacity assessment, relying solely on grassland yield to evaluate carrying capacity fails to accurately reflect the actual nutritional needs of herbivores, thus hindering the scientific revelation of the grassland's true carrying capacity. This leads to management measures often remaining at the level of extensive control over forage quantity, failing to meet the demands of modern animal husbandry for refined nutritional management. Particularly in ecologically fragile areas like the Qinghai-Tibet Plateau, achieving high-quality development of grassland animal husbandry requires improving breeding efficiency and profitability while meeting the basic nutritional needs of livestock. This necessitates a method that can simultaneously quantify the grassland's nutrient quality supply capacity.

[0003] To enhance the scientific rigor of assessments, existing technologies have expanded beyond the carrying capacity of edible forage to incorporate livestock maintenance needs and forage nutritional quality. Technically, satellite remote sensing technologies such as MODIS and Sentinel-2 are used for large-scale vegetation monitoring. Furthermore, a vegetation nitrogen content index developed based on Sentinel-2 data provides a potential tool for regional-scale assessment of forage nutritional quality. Simultaneously, UAV remote sensing platforms, due to their flexibility and high resolution, have been introduced into grassland monitoring, primarily for biomass estimation, attempting to overcome the limitations of satellite remote sensing in terms of spatiotemporal resolution.

[0004] However, existing technologies still have significant limitations. Traditional carrying capacity assessments are mostly based on dry matter yield, failing to fully consider the spatial heterogeneity of forage nutritional quality and thus unable to support refined management decisions based on quality-based grazing. While satellite remote sensing has a wide coverage area, its spatial resolution is low, and revisit cycles are constrained by weather, making it difficult to depict the detailed characteristics of small-scale patchy distributions in alpine grasslands. Although the vegetation nitrogen content index based on Sentinel-2 has been developed, its spatial resolution of 10-20 meters is still insufficient and it is easily affected by frequent cloud and rain in high-altitude areas. While UAV platforms have solved some data acquisition bottlenecks, their current applications are mostly isolated and focused on biomass estimation, and a complete, operationally feasible technical system integrating accurate inversion of forage nutritional quality, simultaneous calculation of forage yield, and integrated calculation of nutrient carrying capacity has not yet been formed.

[0005] Therefore, given the shortcomings of the existing technologies in terms of assessment accuracy, technology integration, and ranch-level applicability, there is an urgent need to develop a new technical solution to achieve rapid, accurate, and integrated assessment of the nutrient production capacity of forage grasses in alpine grasslands, providing direct and reliable technical support for refined grazing management centered on nutrient requirements. Summary of the Invention

[0006] The technical problem to be solved by this invention is to address the shortcomings of existing technologies, specifically by providing a method and related equipment for obtaining crude protein yield from forage, as detailed below: 1) In a first aspect, the present invention provides a method for obtaining crude protein yield from forage, the specific technical solution of which is as follows: Multispectral data, including red-edge bands, of a predetermined area is collected using a multispectral sensor mounted on a drone. The multispectral data is preprocessed. The surface reflectance of the predetermined area is obtained based on the preprocessed multispectral data. The vegetation nitrogen content index of the predetermined area is calculated based on the red-edge bands in the preprocessed multispectral data. The vegetation index and grass yield of the predetermined area are obtained based on the preprocessed multispectral data. Topographic factors of the predetermined area are extracted based on a digital elevation model. The vegetation nitrogen content index, vegetation index, grass yield, and topographic factors are input into a trained learning model to obtain the crude protein content of the predetermined area. The crude protein yield of the forage in the predetermined area is calculated based on the grass yield and crude protein content.

[0007] The beneficial effects of the method for obtaining crude protein yield from forage provided by this invention are as follows: By utilizing a drone platform to collect multispectral data including the red-edge band, sub-meter spatial resolution spectral information for a pre-defined area was acquired, overcoming the limitations of insufficient satellite remote sensing resolution and enabling precise characterization of the spatial heterogeneity of forage nutritional quality in alpine grasslands. By fusing multi-source features such as vegetation nitrogen content index, vegetation index, forage yield, and topographic factors, and inputting them into a trained learning model for inversion, a high-precision and robust quantitative inversion method for crude protein content from drone multispectral data was established, breaking through the bottleneck of nutritional quality assessment in existing technologies. The technical solution integrates a unified process from data acquisition, preprocessing, feature extraction, crude protein content inversion to forage crude protein yield calculation, achieving for the first time the simultaneous acquisition of both the "quality" and "quantity" information of forage within the same technical framework, forming a complete nutritional output capacity assessment technology system. Fourth, the final spatial distribution map of forage crude protein yield provides a direct and accurate data foundation for refined grassland management and scientific assessment of carrying capacity based on nutritional requirements, effectively supporting the decision-making needs of "quality-based grazing."

[0008] Based on the above scheme, the method for obtaining crude protein yield from forage grass according to the present invention can be further improved as follows.

[0009] Furthermore, based on the red-edge bands in the preprocessed multispectral data, the vegetation nitrogen content index for the preset area is calculated, including: The reflectance of bands including different wavelengths is obtained from the red-edge bands in the preprocessed multispectral data, and the vegetation nitrogen content index of the preset area is calculated based on the reflectance of bands including different wavelengths.

[0010] The beneficial effects of adopting the above-mentioned further scheme are: directly extracting surface reflectance at multiple specific wavelengths within the red-edge band from the preprocessed multispectral data, and calculating the vegetation nitrogen content index based on these reflectance values. This process fully utilizes the sensitive response characteristics of the red-edge band to vegetation nitrogen content, achieving rapid and non-destructive quantitative estimation of vegetation nitrogen content in a preset area. The spatial distribution results of the calculated vegetation nitrogen content index provide key and direct input features for the subsequent crude protein content inversion model, effectively enhancing the physical basis and accuracy of the model inversion.

[0011] Furthermore, the process of obtaining the grass yield of the preset area includes: calculating the grass yield of the preset area based on the preprocessed multispectral data and using the light energy utilization rate model.

[0012] The beneficial effects of adopting the above-mentioned further scheme are as follows: By applying a light energy utilization model to calculate the preprocessed multispectral data, a quantitative and spatial estimation of forage yield in a preset area is achieved. This method, based on the principle of vegetation photosynthesis, transforms spectral information into biophysical parameters, providing a clear ecological mechanism to support the acquisition of forage yield. The calculated spatial distribution data of forage yield, combined with the crude protein content inversion results, provides a reliable quantitative basis for the accurate calculation of forage crude protein yield and grassland nutrient carrying capacity, realizing an effective transformation from spectral information to biomass products.

[0013] Furthermore, the multispectral data is preprocessed, including spatiotemporal alignment, radiometric calibration, and atmospheric correction.

[0014] The beneficial effects of adopting the above-mentioned further scheme are: it effectively eliminates errors caused by sensor differences, imaging geometry, and atmospheric scattering and absorption. This process transforms the original UAV multispectral data into surface reflectance data with consistent spatial reference, accurate radiometric measurements, and true surface physical properties, providing a reliable and comparable data foundation for subsequent accurate calculations of vegetation nitrogen content index, vegetation index, and grass yield, and ensuring the quality and consistency of input data for all derived inversion models.

[0015] 2) In a second aspect, the present invention also provides a system for obtaining crude protein yield from forage, the specific technical solution of which is as follows: The system includes modules for data acquisition, data preprocessing, surface reflectance acquisition, vegetation nitrogen content index acquisition, vegetation index forage yield acquisition, topographic factors acquisition, crude protein content acquisition, and forage crude protein yield acquisition. The data acquisition module collects multispectral data of a preset area using a multispectral sensor mounted on a drone. The multispectral data includes the red-edge band. The data preprocessing module preprocesses the multispectral data. The surface reflectance acquisition module obtains the surface reflectance of the preset area based on the preprocessed multispectral data. The vegetation nitrogen content index acquisition module obtains the surface reflectance of the preset area based on the preprocessed multispectral data. The system uses the red-edge bands of the multispectral data to calculate the vegetation nitrogen content index of a preset area; the vegetation index and grass yield acquisition module is used to: acquire the vegetation index and grass yield of a preset area based on the preprocessed multispectral data; the terrain factor acquisition module is used to: extract the terrain factors of a preset area based on the digital elevation model of the preset area; the crude protein content acquisition module is used to: input the vegetation nitrogen content index, vegetation index, grass yield, and terrain factors into the trained learning model to obtain the crude protein content of the preset area; and the forage crude protein yield acquisition module is used to: calculate the forage crude protein yield of the preset area based on the grass yield and crude protein content.

[0016] Based on the above scheme, the system for obtaining crude protein yield from forage according to the present invention can be further improved as follows.

[0017] Furthermore, the vegetation nitrogen content index acquisition module is specifically used to: obtain the reflectance of bands including different wavelengths from the red-edge bands in the preprocessed multispectral data, and calculate the vegetation nitrogen content index of the preset area based on the reflectance of bands including different wavelengths.

[0018] Furthermore, the vegetation index grass yield acquisition module is specifically used to: calculate the grass yield of a preset area based on preprocessed multispectral data and using a light energy utilization model.

[0019] Furthermore, the data preprocessing module is specifically used for: spatiotemporal alignment, radiometric calibration, and atmospheric correction of multispectral data.

[0020] 3) In a third aspect, the present invention also provides an electronic device, the electronic device including a processor coupled to a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor, so as to enable the electronic device to implement any of the above-described methods for obtaining crude protein yield from forage.

[0021] 4) In a fourth aspect, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-described methods for obtaining crude protein yield from forage.

[0022] It should be noted that the beneficial effects of the technical solutions of the second to fourth aspects of the present invention and their corresponding possible implementations can be found in the above description of the technical effects of the first aspect and its corresponding possible implementations, and will not be repeated here. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments of the present invention will be briefly introduced below: Figure 1 This is a flowchart illustrating a method for obtaining crude protein yield from forage according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a system for obtaining crude protein yield from forage according to an embodiment of the present invention. Detailed Implementation

[0024] The principles and features of the present invention are described below. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0025] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0026] like Figure 1 As shown, a method for obtaining crude protein yield from forage according to an embodiment of the present invention includes the following steps: S1. Collect multispectral data of a preset area using a multispectral sensor mounted on a drone. The multispectral data includes the red-edge band. The specific implementation process includes: S10. Select a drone platform equipped with a multispectral sensor, such as the DJI Phantom 4 Multispectral Edition or a model with equivalent performance. The multispectral sensor must have multiple spectral channels, including the red-edge band. Before the flight operation, survey the predetermined area to determine the geographical boundaries of the flight. Using professional flight mission planning software, set the drone's flight altitude according to the required ground resolution. Simultaneously, set the forward overlap to be no less than 80% and the lateral overlap to be no less than 75% to ensure the generation of complete and accurate orthophotos. Flight mission planning also needs to consider environmental conditions, selecting a time period with clear skies, low winds, and a suitable solar altitude angle, typically between 10:00 AM and 2:00 PM local time.

[0027] S11. Upload the planned flight mission to the UAV flight control system. After takeoff, the UAV will automatically fly along the preset route. During flight, the multispectral sensor onboard the UAV will simultaneously image the ground surface, collecting spectral information of each pixel within the preset area in multiple specific bands. This information will collectively constitute the raw multispectral data. The multispectral data must include band data with the center wavelength located near the red edge region. Simultaneously, the UAV's integrated illumination sensor will record solar irradiance data in real time. Before and after the flight, operators must place and photograph a calibration board with known reflectivity on the ground for subsequent radiometric calibration processing.

[0028] S12. Raw multispectral data, geographic location information, solar irradiance data, and calibration board images collected by the UAV during flight are stored in real time on the UAV's onboard storage device or transmitted to the ground station via data link. After the flight mission is completed, all these raw data files are exported completely and archived. This raw multispectral data, including red-edge band information, serves as the foundational input data for all subsequent processing steps.

[0029] Among them, a multispectral sensor is an imaging device capable of simultaneously acquiring the reflected or radiated energy of a target across multiple discrete, narrow spectral bands. Unlike ordinary cameras that only record red, green, and blue, it can capture information from wavelengths beyond the visible light spectrum, such as near-infrared and red-edge wavelengths, which are sensitive to vegetation physiological parameters, by setting optical filters at specific wavelengths. This provides rich spectral dimension data for quantitative remote sensing analysis.

[0030] Multispectral data refers to datasets acquired by multispectral sensors that reflect the response intensity of surface targets in different spectral bands. It typically manifests as a set of spatially registered images, where each image represents the reflectance or radiance distribution of the surface within a specific wavelength range. Through analysis and computation of this data across different bands, feature information required for various vegetation indices and physicochemical parameter inversion models can be extracted.

[0031] The red-edge band refers to a specific range of electromagnetic wavelengths between the end of the visible red band and the beginning of the near-infrared band, with its center wavelength typically between 700 and 750 nanometers. The spectral reflectance of vegetation leaves increases sharply within this wavelength range, creating the so-called "red-edge" effect. The spectral response of the red-edge band is closely related to the chlorophyll content, nitrogen status, and biomass of vegetation, making it an indispensable key piece of information for remote sensing monitoring of vegetation nutritional quality, especially for retrieving vegetation nitrogen content.

[0032] S2. Preprocess the multispectral data, specifically performing spatiotemporal alignment, radiometric calibration, and atmospheric correction. The specific implementation process is as follows: S20. Spatiotemporal alignment of multispectral data: This step aims to integrate multiple single, overlapping multispectral raw images acquired by the UAV during a single flight mission into a complete, spatially accurate, two-dimensional orthorectified image. The process begins by importing the raw images for all bands and their corresponding POS data. The POS data records the latitude, longitude, altitude, and aircraft attitude angle at the moment each image was captured. Using this information, an aerial triangulation algorithm is used to calculate the precise external orientation elements of all images. Next, based on the calculated orientation elements and the internal geometric model of the multispectral sensors on the UAV, each tilted raw image is corrected to a vertically downward viewing angle, and all pixels are projected onto the same plane coordinate system—a process called orthorectification. Afterward, images of the same ground feature captured by different band sensors need to be precisely registered to ensure complete pixel-level overlap in each band, including red, green, blue, red-edge, and near-infrared. Finally, based on the set ground resolution, all the geometrically corrected and registered band images are stitched and cropped to generate a multi-band, spatially aligned composite image covering the entire preset area.

[0033] S21. The goal of radiometric calibration of multispectral data is to convert the raw digital quantization value of each pixel in the image into a physically meaningful value, namely the apparent radiance at the sensor's entrance pupil. The processing relies on the absolute radiometric calibration coefficients provided by the multispectral sensor at the factory and ground calibration board data collected before and after flight. The absolute radiometric calibration coefficients establish a linear relationship between the digital quantization value and the radiance. For a specific band, the radiometric calibration process can be expressed by the formula: in, Indicates the sensor at wavelength The apparent radiance received in the band is measured in watts per square meter per steradian per micrometer. This represents the original digital quantization value of a pixel in the image of that band, and is a dimensionless integer. and These represent the calibration gain coefficient and bias coefficient for that band, respectively, and are provided by the sensor manufacturer. Using this formula, calculations are performed on each pixel and each band in the spatiotemporally aligned multiband composite image, converting the digital quantization value into an apparent radiance value, thus obtaining a radiometrically calibrated multispectral radiance image.

[0034] S22. The goal of atmospheric correction for multispectral data is to eliminate the influence of atmospheric scattering and absorption on sensor signals, converting the apparent radiance received by the sensor into the true reflectance of the land surface. This is the basis for subsequent quantitative inversion of vegetation parameters. The processing employs an atmospheric correction method based on a radiative transfer model. Input data includes multispectral radiance images obtained after radiometric calibration, imaging time, and atmospheric environmental parameters at the imaging location. Atmospheric environmental parameters can be selected from standard atmospheric models or estimated using real-time meteorological data; key parameters include aerosol optical thickness, water vapor content, and ozone content. Radiance values, imaging geometry, and atmospheric parameters are input into a radiative transfer model, such as the MODTRAN or 6S model, for calculation. The radiative transfer model simulates the interaction of light with the atmosphere throughout its transmission from the sun to the land surface and then to the sensor. Through model inversion, the influence of atmospheric path radiation is ultimately eliminated, and the apparent radiance of each pixel in each band is calculated. Converted to true reflectance of ground surface This transformation process can be expressed as a conceptual formula for solving: Among them, the function This represents the atmospheric radiative transfer process described by the radiative transfer model. By numerically resolving this model, the surface reflectance can be obtained. After atmospheric correction, the output is the surface reflectance value of each pixel in each band. These data form a reliable basis for subsequent calculations of surface feature information such as vegetation nitrogen content index and vegetation index.

[0035] S3. Based on the preprocessed multispectral data, the surface reflectance of the preset area is obtained. The specific implementation process is as follows: S30. Call and load the data file that has been fully processed in the previous steps. Preprocessed multispectral data refers to the final output data that has undergone a series of operations such as spatiotemporal alignment, radiometric calibration, and atmospheric correction. Open these data files in the software environment. This data is usually stored in a multi-band image file format, such as GeoTIFF, which contains multiple layers, each corresponding to numerical information for a specific spectral band. It is necessary to check and confirm the integrity of the data, ensuring that data for all necessary bands, such as the red-edge band and near-infrared band, exists, and that all band images are spatially perfectly aligned, with accurate geographic coordinates for each pixel.

[0036] S31. After successfully loading the preprocessed multispectral data file, extract the numerical information representing surface reflectance from the dataset. The atmospheric correction step in the preprocessing workflow directly outputs the conversion of the apparent radiance value of each pixel into the true reflectance value of the ground surface. Therefore, the values ​​stored in each band layer of the data file at this point directly represent the surface reflectance of the corresponding pixel in a specific band. Operationally, the image data of each band is read into the computing memory as a two-dimensional numerical matrix. For a given band, its surface reflectance matrix can be conceptually represented as: in, Indicates the wavelength The surface reflectance matrix for the band. Each element in the matrix... Indicates the first [unit] within the preset area line, number The data is calculated based on the surface reflectance value of a pixel in this band, which is a dimensionless physical quantity ranging from 0 to 1 or from 0% to 100%. Several pixel points of known land cover types need to be selected to check whether their reflectance values ​​are within a reasonable physical range to verify the correctness of the data processing.

[0037] S32. After extracting and validating the surface reflectance matrix for each independent band, these matrices need to be integrated into a unified multidimensional dataset. The surface reflectance matrices for all bands, including the red-edge band, near-infrared band, and red band, are overlaid in the same spatial row and column order. This overlaid multidimensional dataset completely covers the entire preset area spatially and contains surface reflectance information for all processed bands spectrally. This dataset is the final "surface reflectance of the preset area." Subsequently, this dataset can be saved as a new image file containing multi-band surface reflectance information for subsequent feature extraction and model calculation. Additionally, multi-band surface reflectance curves can be extracted from it for specific geographic coordinates or areas for further analysis, depending on requirements.

[0038] The surface reflectance of the preset area refers to the quantitative data of surface target reflectance characteristics covering the entire user-specified survey area, obtained through a UAV multispectral remote sensing system and after rigorous radiometric calibration and atmospheric correction. It characterizes the ratio of solar radiation energy reflected by the surface object to the incident solar radiation energy at each spatial pixel point within the area across multiple specific electromagnetic wave bands. This data is a surface attribute parameter with clear physical meaning, eliminating the influence of the atmosphere and the sensor itself. It serves as the most direct and reliable data foundation for connecting the raw remote sensing signal with surface biophysical and chemical parameters, enabling quantitative inversion work such as vegetation nitrogen content index calculation and vegetation index calculation.

[0039] S4. Based on the red-edge bands in the preprocessed multispectral data, calculate the vegetation nitrogen content index of the preset area. Specifically, obtain the reflectance of bands including different wavelengths from the red-edge bands in the preprocessed multispectral data, and calculate the vegetation nitrogen content index of the preset area based on the reflectance of bands including different wavelengths. The specific implementation process is as follows: S40. Load the multi-band dataset containing the surface reflectance of the preset area, generated in step S3. Open the dataset file in remote sensing image processing software or a programming environment. This dataset is a data cube combining spatial and spectral dimensions, storing the surface reflectance values ​​of each pixel in multiple discrete bands. Clearly identify and locate the band channels associated with the red-edge area. Calculating the vegetation nitrogen content index requires reflectance data at three specific wavelengths, centered around approximately 665 nm, 705 nm, and 740 nm. Therefore, locate and confirm the band layers corresponding to these three center wavelengths in the dataset.

[0040] S41. From the loaded multi-band surface reflectance data cube, extract reflectance data corresponding to three bands near wavelengths of 665nm, 705nm, and 740nm. The data for each band is extracted into a two-dimensional numerical matrix, where each element represents the surface reflectance of a pixel within a preset region in that specific band. Represent these three matrices using mathematical notation: Let... To record the numerical matrix of surface reflectance in the band with a center wavelength of approximately 665 nm. Let... To record the numerical matrix of surface reflectivity in the band with a center wavelength of approximately 705 nm. Let... This is a numerical matrix recording the surface reflectance in the band with a center wavelength of approximately 740 nm. These three matrices are spatially perfectly registered, meaning that for any given pixel location (row index i, column index j), the matrix... elements in ,matrix elements in sum matrix elements in It describes the spectral information of the same point on the ground.

[0041] S42. For each pixel within the preset area, use the reflectance values ​​of that pixel in the three bands and apply the given calculation formula to perform the calculation. The formula for calculating the vegetation nitrogen content index is: in, This represents the calculated vegetation nitrogen content index value. This indicates the surface reflectance value of the pixel in the center wavelength band of approximately 665nm. This indicates the surface reflectance value of the pixel in the center wavelength band of approximately 705nm. This represents the surface reflectance value of the pixel in the band with a center wavelength of approximately 740nm. The calculation process iterates through every pixel in the entire preset area. Specifically, the matrix... With matrix By subtracting element by element, we obtain a new difference matrix. At the same time, the matrix With matrix By subtracting element by element, we obtain another difference matrix. Finally, the first difference matrix and the second difference matrix are divided element-wise. To avoid the error of dividing by zero, a logical check needs to be added to the calculation program; when a certain pixel... When the value is very close to zero, a specific invalid value is assigned to the vegetation nitrogen content index result of that pixel.

[0042] S43. After completing the traversal calculation of all pixels, a new two-dimensional numerical matrix with the same number of rows and columns as the input reflectance matrix is ​​obtained. This matrix is ​​the spatial distribution result of the vegetation nitrogen content index of the preset area, denoted as matrix S43. .matrix Each element in This represents the vegetation nitrogen content index value at pixel (i,j). Subsequently, this calculated matrix is ​​correlated with the geographic coordinates and projection information of the original surface reflectance data to generate a georeferenced raster image file. This image visually displays the spatial distribution of the vegetation nitrogen content index within the preset area. Areas with higher values ​​typically indicate areas with relatively high vegetation nitrogen content. This vegetation nitrogen content index distribution map will serve as a key input feature for subsequently constructing a crude protein content inversion model.

[0043] S5. Based on the preprocessed multispectral data, obtain the vegetation index and grass yield of the preset area; The grass yield of the preset area is calculated based on the preprocessed multispectral data and using a light energy utilization model. The specific implementation process is as follows: S50. The light energy use efficiency (LEE) model is a remote sensing estimation model based on ecological principles, used to convert photosynthetically active radiation absorbed by vegetation into dry matter yield. The core idea of ​​this model is that vegetation converts absorbed solar radiation energy into biochemical energy through photosynthesis, and its conversion efficiency, i.e., LEE, is affected by environmental conditions. A completed LEE model that can be used to calculate grass yield typically consists of the following key parts: the first part is the photosynthetically active radiation absorbed by the vegetation, which depends on the total solar radiation and the proportion of photosynthetically active radiation absorbed by the vegetation; the second part is the LEE, i.e., the efficiency with which the vegetation converts absorbed radiation energy into organic dry matter. This efficiency is not constant but varies with environmental stress factors such as temperature and moisture. A typical LEE model expression can be summarized as follows: in, Gross primary productivity (GLP) is the total amount of carbon fixed by vegetation through photosynthesis per unit time and unit area, expressed in grams of carbon per square meter. This represents the total amount of solar radiation reaching the Earth's surface. This indicates the proportion of photosynthetically active radiation absorbed by the vegetation layer. This represents the maximum light energy utilization rate of vegetation under ideal conditions. and These are temperature and water stress factors, with values ​​between 0 and 1, used to simulate the attenuation effect of non-ideal environmental conditions on light energy utilization efficiency. The light energy utilization efficiency model used to calculate grass yield is a specialized model obtained by converting carbon content into dry matter through a conversion factor and calibrating the model parameters for alpine grassland vegetation types.

[0044] S51. Calculating the grass yield of a preset area requires preparing multiple spatial data as inputs to the light energy utilization efficiency model. The primary input data is preprocessed multispectral data, i.e., surface reflectance data that has undergone spatiotemporal alignment, radiometric calibration, and atmospheric correction. From this data, the Normalized Difference Vegetation Index (NDVI) needs to be calculated. The calculation of the NDVI requires surface reflectance data in the near-infrared and red bands. The calculation formula is: In the formula, This represents the normalized difference vegetation index. This represents the surface reflectance of a pixel in the near-infrared band. This represents the surface reflectance of a pixel in the red band. The calculated normalized difference vegetation index will be used to estimate the proportion of photosynthetically active radiation absorbed by vegetation. Another key input is the total solar radiation data for the calculation period, which can be obtained from meteorological station observation records or meteorological reanalysis data products and interpolated to the preset area. In addition, atmospheric temperature data and soil moisture data for the same period are also required.

[0045] S52. Using the input data prepared in S51, calculate each component of the light energy utilization model pixel by pixel. First, using the empirical relationship between the normalized difference vegetation index and the proportion of photosynthetically active radiation absorbed by vegetation, estimate the proportion of photosynthetically active radiation absorbed by vegetation in each pixel. The commonly used linear relationship is: Where a and b are empirical coefficients. Next, this is combined with the total solar radiation. The proportion of photosynthetically active radiation absorbed by vegetation The photosynthetically active radiation absorbed by the vegetation is calculated. Then, based on the geographical location and imaging time of the preset area, the maximum light energy utilization parameter of the alpine grassland in that area is determined. The values ​​of [values ​​to be determined] are then used. Finally, using temperature and moisture data, the temperature stress factor for each pixel is calculated. and water stress factors These factors are typically converted from environmental data into efficiency coefficients between 0 and 1 using piecewise linear or exponential functions.

[0046] S53. Substitute the components calculated in S52 into the light energy utilization model to calculate the total primary productivity of vegetation pixel by pixel. The calculation formula is: The calculation process iterates through every pixel in the preset region. Due to the model output... The unit is grams of carbon per square meter, which needs to be converted to the unit of grass yield, i.e., grams of dry matter per square meter. This conversion is achieved using a conversion factor that reflects the average carbon content in the plant's dry matter. The formula for calculating grass yield is: in, It indicates the yield of grass, that is, the dry matter mass of the aboveground part per unit area, and the unit is grams of dry matter per square meter. This represents the total primary productivity calculated for that pixel, expressed in grams of carbon per square meter. The carbon content ratio in plant dry matter is a dimensionless empirical constant, typically ranging from 0.45 to 0.5. This conversion yields the current yield of grass per pixel. .

[0047] S54. After completing the calculations for all pixels, a forage yield numerical matrix is ​​obtained that matches the spatial range of the input data. This matrix represents the spatial distribution data of forage yield in the preset area. This resulting matrix is ​​then associated with geographic coordinate information to generate a spatial distribution map of forage yield. To verify the reliability of the calculation results, the forage yield estimated by the model needs to be compared and analyzed with the actual forage yield data obtained through on-site sampling, harvesting, drying, and weighing within the preset area. Statistical indicators such as the coefficient of determination and root mean square error are calculated to evaluate the accuracy of the light energy utilization model in this area. Based on the verification results, the empirical parameters in the model can be fine-tuned if necessary to further improve the accuracy of forage yield estimation. The final forage yield distribution map will serve as key input data for subsequent calculations of forage crude protein yield and grassland carrying capacity.

[0048] The process of obtaining the vegetation index of the preset area is as follows: S55. It is necessary to clarify the specific type of vegetation index to be calculated, which can be the Normalized Difference Vegetation Index (NDV). Taking the acquisition of the NIV for a preset area as an example, we first load the multispectral dataset containing the surface reflectance of the preset area generated in step S3. In this dataset, locate and identify the two spectral bands necessary for calculating the NIV: the red band and the near-infrared band. Confirm that the center wavelength corresponding to the red band is typically around 650 nm, and the center wavelength corresponding to the near-infrared band is typically around 850 nm. Extract the complete surface reflectance data for these two bands from the dataset. The data for each band will be retrieved in the form of a two-dimensional numerical matrix, where each element represents the surface reflectance value of a pixel in that band.

[0049] S56. During the calculation process, the software program or algorithm will process the entire preset area pixel by pixel. For each pixel covering the preset area, the program simultaneously reads its surface reflectance value in the red band and its surface reflectance value in the near-infrared band. Since the data has undergone rigorous preprocessing and spatial alignment, for any given spatial location, its red band reflectance value and near-infrared band reflectance value are spatially perfectly matched, representing the spectral information of the same ground feature. The reflectance value of the pixel in the near-infrared band is denoted as... The reflectance value of a pixel in the red light band is denoted as .here, Specifically refers to the surface reflectance of a pixel in the near-infrared band, which is a dimensionless value. Specifically refers to the surface reflectance of a pixel in the red light band, and is also a dimensionless value.

[0050] S57. For the currently being processed cell, use the acquired... and The value is then substituted into the standard formula for calculating the Normalized Difference Vegetation Index (NDVI). The formula is: in, This represents the calculated normalized difference vegetation index value. The entire calculation process is performed pixel-by-pixel. The software first calculates the molecular component, namely the difference between near-infrared reflectance and red reflectance. Then calculate the denominator, which is the sum of near-infrared reflectance and red reflectance. Finally, the numerator is divided by the denominator to obtain the normalized difference vegetation index value for that pixel. The program will perform this operation repeatedly, traversing every pixel in the preset area image until all pixels have been calculated.

[0051] S58. After all pixels have been calculated, a new two-dimensional numerical matrix with the same number of rows and columns as the input reflectance data is obtained. This matrix represents the spatial distribution result of the normalized difference vegetation index for the preset area. Each value in this matrix corresponds to the normalized difference vegetation index of a geographic pixel, ranging from -1 to 1, with typical values ​​for vegetation-covered areas between 0.2 and 0.8. Subsequently, this numerical matrix is ​​bound to the geographic coordinate system and projection information of the original multispectral data to generate a georeferenced, single-band raster image file. This image is one of the final results of the "vegetation index for the preset area," visually reflecting the vegetation growth status and spatial distribution differences within the area. This vegetation index layer will serve as an important input feature variable for the subsequent construction of a crude protein content inversion model.

[0052] The vegetation index is a quantitative indicator obtained by combining multiple spectral band data acquired by remote sensing sensors through mathematical operations. It aims to highlight specific information about vegetation and reduce interference from non-vegetation factors. It is based on an understanding of the unique spectral response characteristics of green vegetation: vegetation exhibits strong absorption in the red visible light band and strong reflection in the near-infrared band. By designing specific calculation formulas, such as the ratio of the difference between the near-infrared and red light bands to their sum, vegetation signals can be effectively separated from background signals such as soil and water bodies. The vegetation index value is usually closely correlated with parameters such as vegetation cover, leaf area index, biomass, and photosynthetic activity. Therefore, it is widely used in large-scale vegetation monitoring, growth assessment, and ecological environment research, serving as an important bridge connecting raw remote sensing spectral data with surface vegetation biophysical parameters.

[0053] S6. Extract terrain factors of the preset area based on the digital elevation model of the preset area. The specific implementation process is as follows: S60. Obtain digital elevation model (DEM) data covering the entire preset area. DEM data can be downloaded from publicly available geographic information databases or generated using lidar on a drone or photogrammetry. The acquired data file should contain elevation values ​​for every pixel within the preset area. Load the DEM data file into GIS software or a remote sensing processing platform. Check the spatial resolution, coordinate system, and projection information of the DEM data. To perform fusion analysis with the aforementioned multispectral data and surface reflectance data, it is necessary to ensure that the DEM data is completely consistent with the spatial reference system of these data. If inconsistent, the DEM data must be reprojected to convert it to the same coordinate system and projection as the multispectral data. Simultaneously, based on the precise boundaries of the preset area, the DEM data is cropped, removing redundant portions outside the boundaries to obtain a subset of the DEM data that strictly corresponds to the preset area.

[0054] S61. Slope, a topographic factor, is an important parameter describing the degree of surface tilt. Using pre-processed digital elevation model (DEM) data for a pre-defined area, a spatial analysis algorithm is used to calculate the slope value for each pixel. The calculation process is based on the elevation difference between each pixel and its surrounding pixels in the DEM. For a given pixel, its slope value reflects the angle between the tangent plane of the ground surface at that point and the horizontal plane. Specific mathematical calculations typically employ a finite difference method based on a moving window. For example, in a 3x3 pixel window, the slope calculation for the central pixel e uses the elevation values ​​of its eight neighboring pixels. The formula for calculating slope S can be expressed as: In the formula, This represents the calculated slope value, usually expressed in degrees. It represents the rate of change of elevation in the east-west direction (x-direction). This represents the rate of change of elevation along the north-south direction (y-direction). Rate of change and The specific value is calculated by dividing the elevation difference between the center pixel and its adjacent pixels by the pixel spacing. The software or algorithm moves the calculation window pixel by pixel, traversing the entire digital elevation model, and finally generates a slope raster layer of the same size as the digital elevation model, where the value of each pixel is its slope.

[0055] S61. Elevation in topographic factors refers to the height of a ground point along a plumb line from the geoid. In the pre-obtained digital elevation model data of the preset area, the value stored in each cell is itself the elevation of the ground point represented by that cell. Therefore, the process of extracting elevation factors is relatively straightforward. The prepared digital elevation model data of the preset area, after coordinate system matching and clipping, is directly output as an independent data layer. Each cell value in this data layer is the required elevation factor, denoted as... The unit is usually meters. The altitude factor data layer does not require complex calculations or transformations, but it is necessary to ensure the accuracy and validity of its values.

[0056] S62. After obtaining the slope factor raster layer and the elevation factor raster layer respectively, they need to be integrated into a unified topographic factor dataset. Check and ensure that the two layers have exactly the same spatial extent, number of cell rows and columns, and geographic coordinate information. Subsequently, the slope layer and elevation layer can be spatially overlaid to form a multi-band topographic factor data cube, or saved as two independent but spatially aligned data files. This dataset contains the slope and elevation values ​​corresponding to each spatial location within the preset area, collectively constituting the so-called "topographic factors" in subsequent steps. These topographic factors will be used as feature variables, along with vegetation nitrogen content index, vegetation index, and grass yield, and input into the random forest model for inverting crude protein content.

[0057] A Digital Elevation Model (DEM) is a data model used to digitally represent and simulate the morphology of the Earth's surface. It consists of a series of spatially arranged pixels, each assigned a specific numerical value representing the elevation of its corresponding geographic location. DEMs contain only topographic elevation information and do not include the height of any surface cover such as buildings or trees. It is one of the most fundamental geospatial data sources for describing topography. Its data format is typically raster image format, which facilitates computer storage, processing, and analysis. It is a core data source for extracting various topographic factors and conducting geoscientific analyses, such as slope, aspect, and watershed analysis.

[0058] S7. Input the vegetation nitrogen content index, vegetation index, grass yield, and topographic factors into the trained learning model to obtain the crude protein content of the preset area; the specific implementation process is as follows: S70. Random forest regression algorithm is adopted as the learning model for this implementation scheme. Random forest regression is an ensemble machine learning algorithm that improves prediction accuracy and stability by constructing multiple decision trees and averaging them. It is suitable for handling nonlinear regression problems involving multi-feature fusion. Model training requires two main datasets: feature variable dataset and label variable dataset. The feature variable dataset comes from the results of previous processing. For each ground sample point with laboratory-measured crude protein content, its corresponding multi-source feature values ​​need to be collected. These features include: the vegetation nitrogen content index value at that point, the normalized difference vegetation index value, the grass yield value calculated by the light energy utilization rate model, and the altitude and slope values ​​extracted from the digital elevation model. These feature values ​​of all sample points are arranged in the same order to construct a feature matrix. The label variable dataset consists of the percentage crude protein content of these corresponding sample points, obtained through ground sampling and laboratory measurement using the Kjeldahl method, which is used to construct label vectors. .

[0059] S71. Prepare the feature matrix and label vector The data is input into the random forest regression algorithm framework. Before training begins, a series of hyperparameters need to be set for the model, such as the number of decision trees, the maximum depth of the trees, and the minimum number of samples required for leaf nodes. After initial training, to obtain optimal model performance, a grid search and cross-validation strategy is used for hyperparameter optimization. The grid search systematically traverses a preset range of hyperparameter combinations. For each set of hyperparameters, K-fold cross-validation is used for evaluation: the training dataset is randomly divided into K parts, one part is used as the validation set, and the remaining K-1 parts are used as the training set. This process is repeated K times, and the average evaluation metric (such as root mean square error or coefficient of determination) of the K validation results is used as the measure of the performance of that set of hyperparameters. The hyperparameter combination with the optimal evaluation metric is selected using all training data. and Retraining yields the final, fully trained random forest regression model. This model internally encapsulates a complex mapping relationship between input features and output crude protein content.

[0060] S72. Spatially align and integrate the generated feature layers covering the entire preset area. These layers include: vegetation nitrogen content index distribution map, normalized difference vegetation index distribution map, grass yield distribution map, elevation distribution map, and slope distribution map. Ensure that all these layers have the exact same spatial extent, cell size, and number of rows and columns. For each cell within the preset area, extract all five feature values ​​at that cell location to form a feature vector. Organize the feature vectors of all cells within the entire area sequentially to form a large feature matrix. .matrix Each row represents a cell, and each column represents a feature variable.

[0061] S73, Construct the full feature matrix of the preset region The input is fed into a trained random forest regression model. The model's internal mechanism operates on the feature vector of each input (i.e., the data for each pixel). Each decision tree, based on the feature value of that pixel, starts from the root node and follows a specific path to a leaf node according to the decision rules learned during training. This leaf node contains a predicted crude protein content value determined by the training data. The random forest model aggregates the predictions from all decision trees and generates the final predicted crude protein content value for that pixel by calculating the average of all tree predictions. This calculation process can be represented by the following conceptual formula: In the formula, The model represents the first 1 pixel (corresponding feature vector) The predicted percentage of crude protein content. This represents the total number of decision trees in the random forest. Indicates the first Decision trees are based on the input feature vector The output is the predicted crude protein content. The model automatically iterates through the model. For each row of the matrix, this prediction process is repeated for each cell.

[0062] S74. After the model completes the prediction of all pixels within the preset area, it will obtain a new numerical matrix with the same size as the input feature layer. Each element in this matrix is ​​the predicted crude protein content of the corresponding spatial pixel. The physical meaning of this value is the mass percentage of crude protein in the dry matter of forage. Subsequently, this predicted matrix is ​​correlated with the spatial geographic coordinates of the original data to generate a georeferenced raster image file. This image is the spatial distribution map of the "crude protein content of the preset area," visually demonstrating the spatial differentiation of forage nutritional quality (crude protein) within the area, completing the inversion process from multi-source remote sensing features to target biochemical parameters.

[0063] Crude protein content is one of the core indicators for evaluating the nutritional quality of forage, referring to the percentage of crude protein in the total dry matter of a forage sample. Crude protein is a general term for nitrogenous substances in feed, including true protein and non-protein nitrogenous substances. Its value is usually obtained by determining the total nitrogen content using the Kjeldahl method in the laboratory and then multiplying it by a specific conversion factor. Crude protein is an essential nutrient for ruminants to maintain life, growth, reproduction, and production. The level of crude protein content in forage directly determines the nutritional level that grassland can provide and is a key parameter for calculating grassland nutrient carrying capacity. In remote sensing inversion, crude protein content serves as the target variable for model prediction, and its spatial distribution information is crucial for achieving precision livestock management.

[0064] S8. Calculate the crude protein yield of forage in the preset area based on the forage yield and crude protein content. The specific implementation process is as follows: S80. Calculating forage crude protein yield requires two key input data layers: a spatial distribution map of forage yield and a spatial distribution map of crude protein content. The forage yield spatial distribution map is calculated using a light energy utilization model, where the value of each pixel represents the dry matter mass of the forage per unit area at that location. The crude protein content spatial distribution map is inverted using a random forest regression model, where the value of each pixel represents the percentage of crude protein in the forage dry matter at that location. Before starting the calculation, it is essential to ensure that these two data layers are spatially perfectly matched. This means they need to have identical geographic coordinate systems, projection methods, spatial extents, and cell sizes and row / column numbers. Load both raster layers simultaneously in the software and perform a spatial consistency check. If a slight spatial offset or cell misalignment is found between the two layers, a resampling or registration tool should be used to realign the other layer using one layer as a reference, ensuring that for any point on the ground, the forage yield value and crude protein content value originate from the same spatial cell.

[0065] S81. After completing spatial alignment, the software program will traverse the entire preset area pixel by pixel. For each pixel within the area, the program synchronously reads the value stored in the forage yield spatial distribution map and the value stored in the crude protein content spatial distribution map for that pixel. The forage yield value of pixel i is recorded as... The crude protein content value of pixel i is denoted as .here, This represents the grass yield of the i-th pixel. Its physical meaning is the dry matter mass of the above-ground portion of the pasture per unit area, usually expressed in grams of dry matter per square meter. This represents the crude protein content of the i-th pixel. Its physical meaning is the percentage of crude protein in the dry matter of the forage. It is a dimensionless value, but it is usually used in calculations as a percentage, for example, 15.5 represents 15.5%.

[0066] S82. For the currently processed pixel i, set its grass production value. and crude protein content value Substitute the values ​​into the formula for calculating crude protein yield in forage grass. The formula is: in, This represents the calculated crude protein yield of forage in the i-th pixel. This represents the grass yield of the i-th pixel, expressed in grams of dry matter per square meter. This represents the percentage of crude protein content in the i-th pixel. Because... Since it's expressed as a percentage, you need to divide it by 100 to convert it to a decimal before dividing it by the grass yield. Multiplication. The result of multiplication. The unit is grams of crude protein per square meter, which physically represents the absolute mass of crude protein that existing pasture can provide within a unit area of ​​ground represented by that pixel. The program will perform this calculation cyclically, traversing every valid pixel within the preset area.

[0067] S83. After all pixels have been calculated, the program will obtain a result consisting of all... A new two-dimensional numerical matrix is ​​constructed from the values. This matrix represents the spatial distribution data of crude protein yield in forage within the preset region. Next, this resulting matrix is ​​bound to the geographic coordinates and projection parameters of the input data layer to generate a new, georeferenced single-band raster image, i.e., a spatial distribution map of crude protein yield in forage. To verify the reasonableness of the calculation results, statistical analysis and threshold checks can be performed on the output data. For example, checking... Whether the values ​​are negative or abnormally high, and whether their spatial distribution pattern is logically consistent with the distribution of forage yield and crude protein content. Ultimately, this map of forage crude protein yield distribution is the most direct basis for calculating grassland nutrient carrying capacity.

[0068] Crude protein yield is a composite evaluation indicator that combines the "quantity" and "quality" of forage. It specifically refers to the absolute mass of crude protein that currently growing forage can provide per unit area of ​​land. Its value is obtained by multiplying the forage yield per unit area by the percentage of crude protein content in that yield, typically expressed as grams of crude protein per square meter or kilograms of crude protein per hectare. Unlike forage yield, which focuses solely on dry matter weight, or crude protein content, which focuses solely on nutrient concentration, crude protein yield comprehensively reflects the actual nutrient supply capacity of grassland at a specific moment. Crude protein yield is a core input parameter for calculating the nutrient-based carrying capacity of grassland. By comparing the total crude protein yield of a certain area of ​​grassland with the daily crude protein requirement of a single standard sheep unit, the theoretically supported livestock number and grazing time of this grassland can be scientifically calculated. Furthermore, it can be used to assess the balance between grassland nutrient supply and livestock nutrient requirements, providing a quantitative basis for precise supplemental feeding and avoiding nutrient overfeeding or underfeeding. Furthermore, by comparing the crude protein yield of pasture in different regions or at different times, we can guide rotational grazing by region, prioritize the use of high-yield and high-quality grasslands, and achieve efficient and sustainable use of grassland resources.

[0069] Further elaboration of the technical solution of the present invention will be provided through another embodiment. The present invention provides a novel method, the core of which lies in the organic integration of four links: UAV multispectral data acquisition, vegetation nitrogen / crude protein co-inversion, grass yield fusion calculation, and standardized carrying capacity measurement. The method includes: UAV acquisition of multispectral data; vegetation index and environmental factors are extracted after preprocessing of the multispectral data; crude protein content is inverted by fusing vegetation nitrogen index and grass yield and other features using a random forest algorithm; crude protein yield is obtained by combining grass yield calculated by a light energy utilization model; and finally, theoretical carrying capacity potential and actual carrying capacity pressure are calculated according to industry standards to achieve accurate assessment of grassland nutrition. Specifically, a multispectral sensor mounted on a drone is used to conduct flight operations over a predetermined area, simultaneously collecting high-resolution spectral data, including the red-edge band. Data preprocessing and feature extraction are then performed, including spatiotemporal alignment, radiometric calibration, and atmospheric correction of the raw data to obtain surface reflectance. Based on the reflectance data, the vegetation nitrogen content index (MTNI) is calculated, and other environmental factors such as the normalized difference vegetation index (NDVI) and existing grass yield calculated using a light energy utilization model are extracted. Next, the crude protein content inversion model is constructed. In this stage, the extracted MTNI, NDVI, grass yield, and topographic factors are used as input features, combined with ground-measured crude protein sample data as training labels. A random forest regression algorithm is used for model training and validation. The trained model is then applied to the entire process. A pre-defined region is used to generate a distribution map of crude protein content in forage. Simultaneously, a light energy utilization model runs independently, calculating forage yield based on pre-processed multispectral data and generating a distribution map of existing forage yield. Subsequent processes integrate quality and quantity information by multiplying the crude protein content value of each pixel in the forage crude protein content distribution map with the forage yield value of the corresponding pixel in the existing forage yield distribution map to calculate the crude protein yield of each pixel, thus forming spatial information on the crude protein yield of forage in the region. Finally, based on the national industry standard input of the standard daily crude protein requirement (Sp) per sheep unit, and combined with the calculated crude protein yield of forage, grassland utilization rate, edible forage rate, and other parameters, the theoretical carrying capacity of grassland based on the theoretical maximum forage yield and the actual carrying capacity pressure based on the existing forage yield measured by UAVs are calculated respectively, completing an accurate assessment and early warning of the grassland's nutrient carrying capacity.

[0070] The process of UAV flight operation and data acquisition is as follows: In the mission preparation phase, a UAV platform equipped with a multispectral sensor is selected and a flight path is planned. The overlap between the heading and lateral directions is set to be greater than 80% to ensure image integrity. At the same time, the status of UAV equipment, such as camera function, RTK positioning module accuracy, and battery power, is checked. During the operation, a ground calibration board with known reflectivity needs to be laid in the flight area for subsequent radiometric calibration. Then, the automatic flight acquisition program is started. The UAV flies according to the preset flight path. The RTK positioning system acquires high-precision POS data in real time and records the spatial position and attitude of each image. The multispectral sensor simultaneously acquires spectral reflectivity data of five bands, including blue light, green light, red light, red edge, and near-infrared bands. The sensor exposure parameters are dynamically adjusted through intelligent real-time dimming calculation to ensure the consistency of data quality under different lighting conditions. After the flight, the original data is backed up immediately to prevent data loss. The end-to-cloud collaborative data upload process is started. The original multispectral data, POS data, and calibration board images are uploaded to the cloud server through encrypted transmission via 4G or 5G network. After receiving and storing this data, the cloud system automatically triggers the subsequent data preprocessing and feature extraction process.

[0071] For data acquisition, the hardware configuration is as follows: a DJI Phantom 4 Multispectral Edition or a DJI Mavic 3M or a drone with equivalent performance is used. This drone must have RTK positioning capabilities to ensure high-precision spatial position and attitude data. The multispectral sensor onboard the drone is the core of the data acquisition, and it must include a red-edge band with a center wavelength of approximately 730 nm and a near-infrared band with a center wavelength of approximately 840 nm. Priority is given to sensors with center wavelengths close to the B4, B5, and B6 bands of the Sentinel-2 satellite, specifically bands with center wavelengths close to 665 nm, 705 nm, and 740 nm, respectively. This design ensures that the vegetation nitrogen content index algorithm developed for satellite-scale applications can be successfully transferred and applied to the higher spatial resolution multispectral data from the drone.

[0072] During flight operations, missions are conducted in clear, windless weather conditions, with the optimal time window being between 10:00 AM and 2:00 PM local time. The flight mission planning software sets a forward overlap of ≥80% and a lateral overlap of ≥75%. The specific flight altitude setting depends on the desired ground resolution; for example, a flight altitude is set to achieve a ground resolution of 3-5 cm. The UAV automatically flies along the planned route. During flight, multispectral sensors simultaneously collect spectral reflectance data for the preset area in various wavelengths, while the UAV's integrated illumination sensor records solar irradiance data. For accurate radiometric calibration, images of a pre-laid standard calibration board with known reflectance are taken before takeoff and after landing.

[0073] In the data processing and feature extraction process, a series of processing steps are performed on the raw digitally quantized images acquired by the UAV to obtain physically meaningful and geometrically accurate data. The processing steps include: radiometric calibration based on calibration board data and sensor calibration coefficients, converting the digitally quantized values ​​into radiance at the sensor's entrance pupil; atmospheric correction using a radiative transfer model to eliminate the effects of atmospheric scattering and absorption, converting the radiance into surface reflectance; and finally, geometric stitching and orthorectification using the image's POS data and aerial triangulation techniques to generate an accurate surface reflectance orthophoto image. The vegetation nitrogen content index is calculated using the red-edge band information in the preprocessed surface reflectance orthophoto image. The calculation formula is as follows: in, This represents the surface reflectance value in the band with a center wavelength of approximately 665nm; This represents the surface reflectance value in the band with a center wavelength of approximately 705 nm. This represents the surface reflectance value in the band with a center wavelength of approximately 740 nm. The vegetation nitrogen content index is calculated using the red-edge band information from the preprocessed surface reflectance orthophoto image. The calculation formula is: in, This represents the surface reflectance value in the band with a center wavelength of approximately 665nm; This represents the surface reflectance value in the band with a center wavelength of approximately 705 nm. This represents the surface reflectance value in the band with a center wavelength of approximately 740 nm.

[0074] The random forest regression algorithm was used to construct the crude protein content inversion model. The UAV inversion value corresponding to each ground sampling quadrat point constituted the input feature vector, denoted as... These characteristics include: the vegetation nitrogen content index at that location. Value, Normalized Difference Vegetation Index Value, current stock of grass production The values, as well as the elevation and slope values ​​extracted from the digital elevation model, are used. The percentage of crude protein content measured in the corresponding quadrat using the laboratory Kjeldahl method is used as the training label for the model, denoted as... 70% to 80% of all ground samples were randomly allocated as the training set to train the random forest regression model. Key parameters of the model, such as the number of decision trees and the maximum tree depth, were optimized using grid search and cross-validation. The remaining 20% ​​to 30% of the samples were used as the validation set to evaluate the model's prediction accuracy. The trained and validated random forest regression model was then applied to every pixel in the entire study area. For each pixel, its corresponding feature vector was input. The model outputs the predicted crude protein content of the pixel, expressed as a percentage. After traversing all pixels, a spatial distribution map of crude protein content in the study area can be generated. Specifically, the multispectral raw images collected by the UAV first undergo a data preprocessing process, including radiometric calibration, atmospheric correction, and geometric correction, to generate accurate multiband reflectance orthophoto images. Based on the generated multiband reflectance orthophoto images, several feature extraction tasks are carried out in parallel: first, calculating multiple vegetation indices, including the vegetation nitrogen content index (MTNI); second, obtaining the existing grass yield through light energy utilization model inversion and recording it as grass yield RY; and third, extracting topographic factors such as altitude and slope from digital elevation model data. Subsequently, these parallel extraction results—the vegetation nitrogen content index (MTNI), grass yield RY, and other features—are used to extract the predicted crude protein content of the pixel. RY (Rapid Protein Content), altitude, and slope are integrated into an input feature vector. Meanwhile, crude protein content data from ground samples collected in the field and measured using the Kjeldahl method in the laboratory are used as training labels for supervised training. The prepared input features and training labels are fed into a random forest machine learning algorithm for supervised training and optimization of the model. The trained random forest model is applied to the entire study area. By inputting the feature vector corresponding to each pixel, the model outputs the predicted crude protein content value for that pixel. Finally, the prediction results for all pixels are spatially integrated to generate a high-precision spatial distribution map of crude protein content.

[0075] Crude protein yield in forage is calculated by combining forage yield with crude protein content, using the following formula: in, This indicates the crude protein yield of pasture per unit area, expressed in grams of dry matter per square meter. This represents the current amount of grass produced, calculated using the drone light energy utilization model, in grams of dry matter per square meter. This represents the percentage of crude protein content in the corresponding pixel obtained from the crude protein content inversion model. According to the People's Republic of China agricultural industry standard "NY / T816-2021 Nutritional Requirements for Meat Sheep," a standard sheep unit weighing 50 kg, aiming for a daily weight gain of 100 grams while maintaining its body weight, requires 158 grams of crude protein daily, denoted as... The formula for calculating the crude protein carrying capacity per unit area of ​​grassland is: in, This indicates the number of standard sheep units that a unit area of ​​grassland can support, expressed in sheep units per hectare per year. This represents grassland utilization rate, and the value is determined according to relevant standards based on different types such as alpine meadows or alpine steppes. This indicates the availability of grassland, typically taken as 92%. This indicates the percentage of edible forage grass in a grassland, typically taken as 80%. This represents the annual grazing time of the grassland, taken as 365 days. The theoretical maximum grass yield is used; this data is derived from a model estimate within fenced areas unaffected by grazing, denoted as [value missing]. .Will Substituting the values ​​into the above carrying capacity calculation formula, the calculated result is called the theoretical carrying capacity, denoted as . It represents the maximum nutrient supply potential of grassland under ideal conditions. It is the current yield of existing grass, as currently monitored using drones. Substituting the values ​​into the bearing capacity calculation formula, the result obtained is called the actual bearing pressure, denoted as . This represents the actual carrying capacity of the grassland that it can support for livestock. It is calculated by comparing the actual number of grazing livestock with the calculated actual carrying capacity. This can be used to assess whether an area is overgrazing. The specific assessment uses a stress index, calculated as follows: Stress Index = Actual Carrying Capacity divided by Actual Carrying Pressure. .

[0076] The core calculation parameters include the daily requirement of crude protein per unit of sheep, as determined according to national industry standards. The grassland utilization rate is set at 158 ​​grams per 50 kilograms of sheep per day, based on the grassland type. The commonly set grassland availability coefficient The edible forage rate of grassland is 92%. 80%, and the grazing time throughout the year. The calculation period is 365 days; the calculation process is divided into two parallel branches, one of which calculates the theoretical carrying capacity potential, with the input parameter being the theoretical maximum grass yield. Its crude protein content is similar to that of the same region. Substitute the above core parameters into the formula Perform calculations, where This indicates the maximum potential grass yield under conditions unaffected by grazing. The formula outputs the result representing the percentage of crude protein content. This represents the maximum theoretical carrying capacity of grassland per unit area; another method calculates the actual carrying capacity pressure, with the input parameter being the existing grass yield obtained from UAV inversion. Its relationship with crude protein content Substitute the core parameters into the formula The formula outputs the result of the calculation. This indicates the current actual carrying capacity of grassland per unit area; ultimately, a carrying capacity status assessment and early warning are conducted. By obtaining the actual livestock carrying capacity data of the region, a pressure index is calculated, the formula of which is: pressure index equals actual livestock carrying capacity divided by actual carrying capacity pressure. The system determines the status based on the pressure index value: when the pressure index is less than or equal to 1, it is determined to be in a green equilibrium state; when the pressure index is greater than 1 but less than 1.2, it is determined to be in a yellow critical state; and when the pressure index is greater than or equal to 1.2, it is determined to be in a red overload state, thereby achieving scientific early warning for grassland grazing management.

[0077] The technical solution proposed in this invention innovatively transplants the vegetation nitrogen content index at the satellite scale to a higher-resolution UAV multispectral platform, utilizing sensors with precise red-edge bands to achieve high-resolution, non-destructive detection of grassland vegetation nitrogen content at the pasture scale. In the inversion model, it overcomes the limitations of relying on a single spectral index, constructing a multi-source feature fusion crude protein content inversion model. This model innovatively uses the vegetation nitrogen content index, the existing grass yield calculated by the light energy utilization rate model, and topographic factors such as altitude and slope as input features, and uses a random forest regression algorithm for collaborative training and prediction. This fusion strategy greatly improves the accuracy of spatial prediction of crude protein content and the robustness of the model under different environments. Furthermore, this invention creates a "quality-quantity combined" grassland carrying capacity measurement system, by coupling the inverted nutrient quality, i.e., crude protein content, with the simultaneously obtained existing grass yield, according to the formula... Calculate the crude protein yield of forage, where, Indicates crude protein yield in forage. This indicates the current amount of grass produced. This invention represents the percentage of crude protein content, thus constructing a measurement model that moves from "dry matter carrying capacity" to "nutrient carrying capacity." This model strictly adheres to national industry standards, and the calculation results are more consistent with the actual nutritional needs of livestock, providing scientific guidance for precise supplementary feeding and rotational grazing. The beneficial effects of this invention are reflected in several aspects: First, it is highly refined; the sub-meter spatial resolution provided by the drone clearly depicts the spatial heterogeneity of grassland nutrient patches, providing a direct basis for implementing precision grazing. Second, it is highly efficient and real-time; the drone operation is fast and flexible, and combined with cloud data processing capabilities, it can complete the assessment of large areas of grassland in a short time, achieving dynamic monitoring of grassland nutrient status. Third, it is scientifically sound; the entire measurement process, from data inversion to carrying capacity calculation, follows rigorous models and standards, ensuring the scientific validity and practicality of the results. Finally, it is forward-looking; by calculating the actual carrying capacity pressure and comparing it with the theoretical carrying capacity potential, the system can provide early warning when grasslands are on the verge of overloading, prompting managers to adjust herd size and grazing strategies in a timely manner, thereby effectively preventing grassland degradation and supporting the sustainable development of grassland animal husbandry.

[0078] In the above embodiments, although the steps are numbered S1, S2, etc., they are only specific embodiments given by the present invention. Those skilled in the art can adjust the execution order of S1, S2, etc. according to the actual situation. The scheme after adjusting the order is also within the protection scope of the present invention. It can be understood that in some embodiments, some or all of the above embodiments may be included.

[0079] like Figure 2 As shown, an embodiment of the present invention provides a system for obtaining crude protein yield from forage, comprising a data acquisition module, a data preprocessing module, a surface reflectance acquisition module, a vegetation nitrogen content index acquisition module, a vegetation index forage yield acquisition module, a topographic factor acquisition module, a crude protein content acquisition module, and a forage crude protein yield acquisition module. The data acquisition module is used to: acquire multispectral data of a preset area using a multispectral sensor mounted on a drone; the multispectral data includes red-edge bands. The data preprocessing module is used to: preprocess the multispectral data. The surface reflectance acquisition module is used to: obtain the surface reflectance of the preset area based on the preprocessed multispectral data. The vegetation nitrogen content index acquisition module... The following modules are used: The vegetation nitrogen content module is used to calculate the vegetation nitrogen content index of a preset area based on the red-edge bands in the preprocessed multispectral data; the vegetation index and grass yield acquisition module is used to acquire the vegetation index and grass yield of a preset area based on the preprocessed multispectral data; the terrain factor acquisition module is used to extract the terrain factor of a preset area based on the digital elevation model of the preset area; the crude protein content acquisition module is used to input the vegetation nitrogen content index, vegetation index, grass yield, and terrain factor into the trained learning model to obtain the crude protein content of the preset area; and the forage crude protein yield acquisition module is used to calculate the forage crude protein yield of the preset area based on the grass yield and crude protein content.

[0080] Optionally, in the above technical solution, the vegetation nitrogen content index acquisition module is specifically used to: obtain the reflectance of bands including different wavelengths from the red-edge bands in the preprocessed multispectral data, and calculate the vegetation nitrogen content index of the preset area based on the reflectance of bands including different wavelengths.

[0081] Optionally, in the above technical solution, the vegetation index grass yield acquisition module is specifically used to: calculate the grass yield of a preset area based on preprocessed multispectral data and using a light energy utilization model.

[0082] Optionally, in the above technical solution, the data preprocessing module is specifically used for: spatiotemporal alignment, radiometric calibration, and atmospheric correction of multispectral data.

[0083] It should be noted that the beneficial effects of the forage crude protein yield acquisition system provided in the above embodiments are the same as those of the forage crude protein yield acquisition method described above, and will not be repeated here. Furthermore, the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, and will not be repeated here.

[0084] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-described methods for obtaining crude protein yield from forage.

[0085] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-described methods for obtaining crude protein yield from forage.

[0086] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0087] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for obtaining crude protein yield from forage, characterized in that, include: Multispectral data of a preset area is collected using a multispectral sensor mounted on a drone, and the multispectral data includes the red-edge band. The multispectral data is preprocessed; The surface reflectance of the preset area is obtained based on the preprocessed multispectral data; Based on the red-edge bands in the preprocessed multispectral data, the vegetation nitrogen content index of the preset area is calculated. Based on the preprocessed multispectral data, the vegetation index and grass yield of the preset area are obtained. Extract terrain factors of the preset area based on the digital elevation model of the preset area; The vegetation nitrogen content index, the vegetation index, the grass yield, and the topographic factor are input into the trained learning model to obtain the crude protein content of the preset area. The crude protein yield of the forage in the preset area is calculated based on the grass yield and the crude protein content.

2. The method for obtaining crude protein yield from forage according to claim 1, characterized in that, Based on the red-edge bands in the preprocessed multispectral data, the vegetation nitrogen content index of the preset region is calculated, including: The reflectance of bands including different wavelengths is obtained from the red-edge bands in the preprocessed multispectral data, and the vegetation nitrogen content index of the preset area is calculated based on the reflectance of bands including different wavelengths.

3. The method for obtaining crude protein yield from forage according to claim 1, characterized in that, The process of obtaining the grass yield of the preset area includes: The grass yield of the preset area is calculated based on the preprocessed multispectral data and using a light energy utilization model.

4. A method for obtaining crude protein yield from forage according to any one of claims 1 to 3, characterized in that, The multispectral data is preprocessed, including: spatiotemporal alignment, radiometric calibration, and atmospheric correction.

5. A system for obtaining crude protein yield from forage, characterized in that, It includes a data acquisition module, a data preprocessing module, a surface reflectance acquisition module, a vegetation nitrogen content index acquisition module, a vegetation index grass yield acquisition module, a topographic factor acquisition module, a crude protein content acquisition module, and a forage crude protein yield acquisition module. The data acquisition module is used to: acquire multispectral data of a preset area using a multispectral sensor mounted on a drone, wherein the multispectral data includes the red-edge band; The data preprocessing module is used to: preprocess the multispectral data; The surface reflectance acquisition module is used to: obtain the surface reflectance of the preset area based on the preprocessed multispectral data; The vegetation nitrogen content index acquisition module is used to: calculate the vegetation nitrogen content index of the preset area based on the red edge band in the preprocessed multispectral data; The vegetation index and grass yield acquisition module is used to: acquire the vegetation index and grass yield of the preset area based on the preprocessed multispectral data; The terrain factor acquisition module is used to: extract the terrain factors of the preset area based on the digital elevation model of the preset area; The crude protein content acquisition module is used to: input the vegetation nitrogen content index, the vegetation index, the grass yield and the terrain factor into the trained learning model to obtain the crude protein content of the preset area; The forage crude protein yield acquisition module is used to calculate the forage crude protein yield of the preset area based on the forage yield and the crude protein content.

6. The system for obtaining crude protein yield from forage according to claim 5, characterized in that, The vegetation nitrogen content index acquisition module is specifically used to: obtain the reflectance of bands including different wavelengths from the red-edge bands in the preprocessed multispectral data, and calculate the vegetation nitrogen content index of the preset area based on the reflectance of bands including different wavelengths.

7. The system for obtaining crude protein yield from forage according to claim 5, characterized in that, The vegetation index grass yield acquisition module is specifically used to: calculate the grass yield of the preset area based on the preprocessed multispectral data and using a light energy utilization model.

8. A system for obtaining crude protein yield from forage according to any one of claims 5 to 7, characterized in that, The data preprocessing module is specifically used for: spatiotemporal alignment, radiometric calibration, and atmospheric correction of multispectral data.

9. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for obtaining crude protein yield of forage as described in any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements a method for obtaining crude protein yield from forage as described in any one of claims 1 to 4.

Images