Fertilizer proportioning management method and system combined with remote sensing data analysis

Abstract

The application provides a fertilizer matching management method and system combined with remote sensing data analysis, relates to the technical field of remote sensing data, and comprises the following steps: using a UAV to shoot original data sets of crops in a target field during a key growth period, pre-processing the original data sets, and obtaining a remote sensing image set; calculating a vegetation index according to the remote sensing image set, combining with GIS software, and generating a vegetation index map; obtaining a soil nutrient content data set and a crop nutrient content data set; applying a random forest regression algorithm to correlate and model the crop nutrient content data set and the vegetation index map, using the model to calculate nutrient inversion of the target field, and obtaining a nutrient spatial distribution map; generating a fertilizer matching decision; and using GIS software to spatially express the decision result and obtaining a digital fertilization prescription map. The technical problems of low efficiency, large error, incomplete coverage and slow response in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing data, and particularly relates to a fertilizer ratio management method and system combined with remote sensing data analysis. BACKGROUND

[0002] In modern agricultural production, scientific and reasonable fertilizer ratio not only directly affects the yield and quality of crops, but also relates to soil health and sustainable development of agriculture. Excessive fertilization is easy to cause resource waste and environmental pollution, while insufficient fertilization will lead to crop nutrient deficiency and reduce yield potential. Therefore, how to dynamically adjust the fertilizer ratio according to the actual growth status of crops and the soil nutrient supply capacity has become a key problem to improve the efficiency and ecological benefits of agricultural production.

[0003] The traditional fertilizer ratio method is based on manual sampling and laboratory analysis, and its limitations are increasingly prominent. Manual sampling is limited by time and cost, and usually only a few points of farmland can be covered, which is difficult to fully reflect the spatial heterogeneity of soil nutrients in the field, resulting in significant deviation of the fertilization scheme. The limitations of the traditional method make it difficult for the existing technology to meet the needs of precision agriculture for efficient, real-time and full-coverage fertilization management.

[0004] In summary, the traditional fertilizer ratio method has problems such as low efficiency, large error, incomplete coverage, and slow response, and it is difficult to achieve precise, efficient and dynamic fertilization management, so there is an urgent need for a method to solve the above problems. SUMMARY

[0005] The present disclosure provides a fertilizer ratio management method and system combined with remote sensing data analysis to solve the technical problems of low efficiency, large error, incomplete coverage and slow response in the prior art.

[0006] According to a first aspect of the present disclosure, a fertilizer ratio management method combined with remote sensing data analysis is provided, comprising:

[0007] An unmanned aerial vehicle is used to shoot an original data set of crops in a target field during a key growth period, the original data set is preprocessed to obtain a remote sensing image set, and the preprocessing includes radiation calibration, atmospheric correction, geometric correction and cloud mask;

[0008] A vegetation index is calculated according to the remote sensing image set, and a vegetation index map is generated in combination with GIS software;

[0009] Based on the soil type of the target field, historical yield data and the vegetation index map, representative points of the target field are selected, soil and crop samples are collected, and key nutrient content analysis is performed on the samples to obtain a soil nutrient content data set and a crop nutrient content data set;

[0010] The random forest regression algorithm is applied to correlate the crop nutrient content dataset and the vegetation index map, and the model is used to calculate the nutrient inversion of the target field plot to obtain a nutrient spatial distribution map.

[0011] The vegetation index map, the soil nutrient content dataset, the nutrient spatial distribution map, and the historical yield data are comprehensively analyzed, the clustering analysis algorithm is applied to spatial partitioning, the target field plot is divided into M management units, the required fertilizer application amount and element abundance index of each management unit are determined, and a fertilizer ratio decision is generated.

[0012] The decision result is spatialized by using the GIS software to obtain a digital fertilization prescription map.

[0013] According to a second aspect of the present disclosure, a fertilizer ratio management system combined with remote sensing data analysis is provided, comprising:

[0014] A remote sensing image acquisition module is configured to capture the original dataset of crops in the target field plot during the key growth period by using a UAV, and to obtain a remote sensing image set by preprocessing the original dataset, wherein the preprocessing includes radiation calibration, atmospheric correction, geometric correction, and cloud mask.

[0015] A vegetation index calculation module is configured to calculate the vegetation index according to the remote sensing image set, and to generate a vegetation index map in combination with the GIS software.

[0016] A soil and crop nutrient analysis module is configured to select representative points of the target field plot based on the soil type, historical yield data, and vegetation index map of the target field plot, to collect soil and crop samples, and to obtain a soil nutrient content dataset and a crop nutrient content dataset by analyzing the key nutrient content of the samples.

[0017] A nutrient inversion modeling module is configured to correlate the crop nutrient content dataset and the vegetation index map by using the random forest regression algorithm, and to obtain a nutrient spatial distribution map by using the model to calculate the nutrient inversion of the target field plot.

[0018] A precise partitioning and fertilization decision module is configured to comprehensively analyze the vegetation index map, the soil nutrient content dataset, the nutrient spatial distribution map, and the historical yield data, to divide the target field plot into M management units by using the clustering analysis algorithm for spatial partitioning, to determine the required fertilizer application amount and element abundance index of each management unit, and to generate a fertilizer ratio decision.

[0019] A digital fertilization prescription generation module is configured to spatialize the decision result by using the GIS software to obtain a digital fertilization prescription map.

[0020] The one or more technical solutions provided in the disclosure have at least the following technical effects or advantages: original data sets of crops in a key growth period of a target field are photographed by a drone, the original data sets are preprocessed to obtain a remote sensing image set, the preprocessing includes radiation calibration, atmospheric correction, geometric correction and a cloud mask; a vegetation index is calculated according to the remote sensing image set, and a vegetation index map is generated in combination with GIS software; representative points of the target field are selected based on a soil type of the target field, historical yield data and the vegetation index map, soil and crop samples are collected, key nutrient content analysis is performed on the samples, and a soil nutrient content data set and a crop nutrient content data set are obtained; a random forest regression algorithm is applied to associate modeling of the crop nutrient content data set and the vegetation index map, and nutrient inversion calculation is performed on the target field by using the model to obtain a nutrient spatial distribution map; the vegetation index map, the soil nutrient content data set, the nutrient spatial distribution map and the historical yield data are comprehensively applied to spatial partitioning by using a clustering analysis algorithm, the target field is divided into M management units, the required fertilizer application amount and element abundance and deficiency index of each management unit are determined, and a fertilizer ratio decision is generated; the decision result is spatialized and expressed by using GIS software to obtain a digital fertilization prescription map. The technical problems of low efficiency, large error, incomplete coverage and slow response in the prior art are solved. The technical effects of improving the calculation precision, coverage range and decision efficiency of the fertilizer ratio are achieved.

[0021] The above description is only a summary of the technical solutions of the present application. In order to enable the technical means of the present application to be more clearly understood, and to be implemented in accordance with the content of the description, and in order to enable the above and other purposes, characteristics and advantages of the present application to be more apparent and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the present disclosure or the prior art, the following will briefly introduce the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are only exemplary, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.

[0023] Figure 1 The flowchart of the fertilizer ratio management method combined with remote sensing data analysis provided by the embodiments of the present application;

[0024] Figure 2 The structural schematic diagram of the fertilizer ratio management system combined with remote sensing data analysis provided by the embodiments of the present application.

[0025] The reference signs are explained as follows: remote sensing image acquisition module 11, vegetation index calculation module 12, soil and crop nutrient analysis module 13, nutrient inversion modeling module 14, precision zoning and fertilization decision module 15, and digital fertilization prescription generation module 16. DETAILED DESCRIPTION

[0026] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to assist in understanding, which should be considered in their context only. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Also, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0027] In an embodiment, the present disclosure provides a fertilizer ratio management method combined with remote sensing data analysis, which is described with reference to Figure 1 The method comprises the following steps:

[0028] S1: Using a UAV to shoot an original data set of crops in a target field during a key growth period, pre-processing the original data set to obtain a remote sensing image set, the pre-processing including radiation calibration, atmospheric correction, geometric correction, and cloud mask;

[0029] Further, the step S1 further comprises:

[0030] Selecting N ground control points uniformly in the target field, spraying L-shaped markers and recording the coordinate positions synchronously;

[0031] Setting flight parameters of the UAV based on the geographical conditions of the target field, selecting an original data set of the target field collected during the key growth period of the crops, the UAV being equipped with a multi-spectral sensor, and the original data set including an original DN image set and a flight parameter log;

[0032] Obtaining the original data set, performing radiation calibration, converting the original DN value into apparent reflectance by combining the sensor calibration file and the flight parameter log, and obtaining an apparent reflectance image set;

[0033] Based on the apparent reflectance image set, converting the apparent reflectance into ground reflectance by using an empirical linear method assisted by a ground calibration plate, and obtaining a ground reflectance image set;

[0034] Using POS data and SIFT feature points to perform coarse stitching on the ground reflectance image set to generate an initial orthographic image set, performing second-order polynomial transformation through ground control point coordinates, using bilinear interpolation resampling, and outputting an orthographic reflectance image set in a projection coordinate system;

[0035] The cloud mask processing is performed based on the spectral characteristics of the ortho-reflectance image set, and invalid areas are removed, and finally the remote sensing image set is generated.

[0036] Specifically, N ground control points are uniformly selected in the target field block, and no less than 5 points are arranged per 10 mu. Each point is sprayed with a permanent L-shaped mark, and the high-precision coordinates of the mark center point are recorded by an RTK measuring instrument. The target field block boundary is determined based on GIS software and field surveying and mapping, and the unmanned aerial vehicle flight parameters are planned accordingly: the flight height is set to 80 to 100 meters, the ground resolution is guaranteed to be 3 to 5 cm, the heading overlap rate is greater than or equal to 80%, the lateral overlap rate is greater than or equal to 70%, and the flight direction is parallel to the long side of the field block to optimize the coverage efficiency. The key growth period of crops is selected, and the flight task is executed in sunny and cloudless weather conditions with a wind speed of less than 5 m / s at 10:00 to 14:00 every day when the sun height is greater than 45 degrees. The unmanned aerial vehicle carries a multi-spectral sensor to automatically cruise and synchronously collect the original data set of the target field block, specifically including the original DN image set and the flight parameter log, wherein the original DN image set includes the DN value data of the blue, green, red, red edge and near-infrared bands, and the flight parameter log includes the GPS timestamp, sensor attitude angle, sun zenith angle and real-time irradiance.

[0037] The DN value is converted to apparent reflectance through radiometric calibration. The gain coefficient K band and the offset value Offset band of each band are obtained based on the calibration file provided by the sensor manufacturer, and the real-time irradiance in the flight parameter log is combined to perform radiometric calibration. The original DN value is converted to radiance L λ , and the specific formula is: , wherein L λ is the radiance at the entrance pupil of the sensor, is the DN value data in the original data set, Offset band is the offset value, and K band is the gain coefficient of each band. The day-earth distance correction factor d is calculated according to the GPS timestamp, and the atmospheric layer outside solar irradiance Esun corresponding to the center wavelength of the band is queried, such as about 1530 Wm -2 μm -1 for the red light band. The apparent reflectance is calculated combined with the sun zenith angle θ in the flight parameter log, and the specific formula is: , wherein represents the apparent reflectance, is a constant, is the radiance, d is the day-earth distance correction factor, Esun is the atmospheric layer outside solar irradiance, is the cosine function of the sun zenith angle. All apparent reflectance data are integrated, and the output result is a floating-point apparent reflectance image set, and the histogram is used to verify whether the values of each band meet the expected range.

[0038] The apparent reflectance image set is acquired and is subjected to atmospheric correction. An experience linear method assisted by ground calibration board is adopted to eliminate the influence of atmospheric scattering. Three groups of reflectance calibration boards with standard reflectance of 10%, 30% and 50% are laid in the field before flight, and the real reflectance of the calibration boards is measured by using a handheld spectrometer The average value of the image element of the calibration board region is extracted from the apparent reflectance image set Linear equations are independently fitted for each waveband: Wherein a represents the slope, b represents the intercept, and the coefficients of specific wavebands are different, for example, the typical coefficient of the near-infrared waveband is a = 1.12, b = -0.03. The regression model is applied to the panoramic image to generate the ground surface reflectance data with clear physical meaning, all the ground surface reflectance data are integrated to obtain the ground surface reflectance image set.

[0039] Combined with the previously set N ground control points and the POS data, the spatial accuracy is improved, the ground surface reflectance influence set is subjected to geometric correction and image splicing. The POS data of the unmanned aerial vehicle and the SIFT feature matching algorithm are used to perform initial splicing on the single-scene ground surface reflectance image to generate an initial orthographic image set with residual error of 5-10 pixels. The measured ground control point coordinates are imported, and a second-order polynomial geometric transformation is performed: , Wherein is the corrected map coordinate, x and y represent the original image element coordinates, a0 to a5 represent the X-direction transformation coefficients, and b0 to b5 represent the Y-direction transformation coefficients. Bilinear interpolation resampling is adopted to ensure that the RMS of all GCP residual errors is less than or equal to 0.5 pixels, and high-resolution GIS base maps such as ridge vector lines are superimposed, the alignment accuracy of ground objects is visually checked, and the edge registration error is required to be less than 1 pixel element. The final output result is the orthographic reflectance image set in the projection coordinate system, and the edge registration error is less than 1 pixel element.

[0040] The orthographic reflectance image set is subjected to cloud mask processing, and invalid areas are identified and removed based on spectral characteristics. Specifically, the cloud detection standard is that the near-infrared waveband and the ratio of the blue light waveband to the red light waveband The shadow detection standard is that the near-infrared waveband and the red light waveband An expansion operation of a 3*3 pixel rectangular structure element is performed on the identified area to eliminate small spots, and finally manual review is performed to correct misjudgment areas by referring to the RGB true color preview image, such as misjudging a white plastic film as a cloud, a binary mask file is generated, effective farmland = 1, cloud or shadow = 0. The mask result is applied to the ortho reflectance image set, and the invalid area is assigned as NaN, the pure farmland observation data is retained, and a remote sensing image set is generated. The remote sensing image set stores the reflectance data in GeoTIFF format of each band, and the metadata file is attached, and all images match the same coordinate system and resolution.

[0041] S2: Calculate the vegetation index according to the remote sensing image set, and generate a vegetation index map in combination with GIS software.

[0042] Further, the step S2 further comprises:

[0043] Based on the remote sensing image set and the crop growth stage, the vegetation index is calculated, if the crop is in the early to middle growth stage, the normalized difference vegetation index is selected, and if the crop is in the middle to late growth stage, the red edge normalized index is selected, wherein the normalized difference vegetation index is calculated according to the following formula:

[0044] ;

[0045] Wherein, represents the normalized difference vegetation index, represents the near-infrared band surface reflectance, represents the red light band surface reflectance;

[0046] The red edge normalized index is calculated according to the following formula:

[0047] ;

[0048] Wherein, NDRE represents the red edge normalized index, represents the near-infrared band surface reflectance, represents the red edge band surface reflectance;

[0049] The spatial optimization of the calculated vegetation index includes spatial relationship matching and clipping operation;

[0050] Based on the spatial optimized vegetation index, the statistical characteristics are obtained, the hierarchical threshold is set according to the statistical characteristics, the different threshold areas are mapped into gradual color bands by using the HSV color space model, and the vegetation index map is generated, and the statistical characteristics include minimum value, maximum value and quartile.

[0051] Specifically, a remote sensing image set is acquired and data standardization is performed to ensure computational reliability. First, spatial consistency verification is conducted, checking the geographic coordinates, projection system, resolution, and pixel alignment of each band image. GIS software is used to confirm that the spatial offset of the near-infrared, red, and red-edge bands does not exceed 0.5 pixels. If deviations exist, resampling is performed to force alignment. Next, value range validity screening is implemented, and the near-infrared reflectance ρ of vegetation-covered areas is considered. NIR The red light reflectance ρ should be in the range of 0.1-0.9. Red Within the range of 0.02-0.25. Cells outside this range are marked as invalid values.

[0052] Vegetation indices are calculated based on standardized reflectance data and crop growth stages. If the crop is in the early to mid-growth stage, the Normalized Difference Vegetation Index (NDVI) is used; if the crop is in the mid-to-late growth stage, the Red Edge Normalized Vegetation Index (RBVI) ​​is used. The specific formula for calculating the NDVI is as follows:

[0053] ;

[0054] in, Represents the normalized difference vegetation index. Representing near-infrared surface reflectance, it originates from multiple scattering by plant cell walls. Represents the surface reflectance in the red light band. Due to strong absorption by chlorophyll, this value is usually <0.1 in vegetated areas. The molecule characterizes the photosynthetic capacity of vegetation. Healthy plants have strong absorption of red light by chlorophyll and high reflectance of near-infrared by cell structure. The higher the value, the stronger the chlorophyll activity. The denominator represents the standardization process, which normalizes the light conditions, eliminates the influence of solar altitude angle, and limits the output value to [-1,1].

[0055] The specific formula for calculating the red-edge normalization index is as follows:

[0056] ;

[0057] NDRE stands for Red Edge Normalized Index, which is mainly used to describe the crop status in the mid-to-late stages of growth. Represents the surface reflectance in the near-infrared band. Represents the surface reflectance of the red edge band. The red edge band is sensitive to chlorophyll concentration and can penetrate high-density canopies. Its calculation process is the same as NDVI, but the band input needs to be replaced.

[0058] The vegetation index obtained by calculation is spatially optimized. The boundary points are collected in the field along the target field plot by RTK measuring instrument, and a coordinate point is recorded every 10 meters for straight boundary. The point interval is encrypted to 3 meters at corners. Avoid crop shading when collecting. After importing the data into GIS software, generate farmland boundary vector file, and the attribute table contains field number and area field. Use ArcGIS topology verification tool to check the topology of farmland boundary vector, ensure that the polygon is completely closed, there is no hanging line, and there is no overlap or gap between adjacent fields. At the same time, the coordinate of farmland boundary vector is corrected. If the vector coordinate system is inconsistent with the remote sensing image, use seven-parameter affine transformation to unify to the target coordinate system. Load the vegetation index data and farmland boundary vector into the GIS platform at the same time, check the consistency of the coordinate system, and match the spatial relationship. If deviation is found, use the projection tool ProjectRaster to convert the raster coordinate system, and apply spatial correction Spatial Adjustment to translate the vector boundary. Use the vector polygon as a mold to extract the raster data within its coverage range, and perform clipping operation. The clipping parameters are set as follows: the clipping mode is selected as accurate clipping, all pixels within the boundary are retained; the area outside the boundary is filled with invalid values; the resampling method is set to bilinear interpolation to avoid jagged edges. Finally, the spatially optimized vegetation index data is output, and the error between the data raster area and the vector area is less than or equal to 0.5%, and the visual image boundary of the target field plot can be aligned without error and loss.

[0059] The vegetation index after spatial optimization is obtained, at this time the vegetation index data is a floating point GeoTIFF file, each pixel stores a vegetation index value between -1.0 and 1.0, contains accurate geographic coordinate information, and has completed spatial optimization steps such as farmland boundary clipping. In order to realize the intuitive expression of crop information, the continuous vegetation index value is converted into a graded color with clear agronomic significance. The specific steps include: statistical characteristic analysis, calculation of the histogram distribution of the vegetation index, determination of the key demarcation points, including the minimum value, the maximum value and the quartiles, specifically the minimum value Min is the actual minimum value after removing bare soil and water, the maximum value Max is the crop saturation point after removing 5% of the highest value, the quartiles are 25% Q1, 50% Median and 75% Q3; analyze the statistical characteristics of the vegetation index of the target field block to determine the grading threshold, set Min to Q1 as the low-fertilizer area, representing the 25% area with the worst growth, Q1 to Median as the medium-fertilizer area, representing the 25% area with medium growth, Median to Q3 as the balance area, representing the 25% area with medium growth, Q3 to Max as the high-fertilizer area, representing the 25% area with the best growth, and simultaneously adjust the grading threshold according to the actual situation of crop growth to ensure that the threshold interval is flexible and variable, for example, if it is a drought year, the overall threshold interval can be lowered by 0.1 to adapt to the changes in crop physiological state; design a color mapping scheme, use the HSV color space model to realize the scientific conversion from numerical value to color, specifically, the hue is gradually changed from red to green, corresponding to the growth from poor to good, among which the red color corresponds to the low-fertilizer area, indicating the stress state of chlorophyll deficiency, the yellow color corresponds to the medium-fertilizer area, indicating normal growth but not up to standard, the green color corresponds to the balance area, indicating normal crop growth, and the dark green color corresponds to the high-fertilizer area, indicating saturated biomass and potential lodging risk, and the color saturation is set to remain above 85% to ensure bright colors, and the high-value area reduces brightness to enhance visual hierarchy; finally, generate a vegetation index map according to the above settings, the index map contains spatial information and agronomic interpretation, the vegetation index rendering map occupies 70% of the map area, and the scale length corresponds to 100 meters in the field.

[0060] S3: Based on the target field block soil type, historical yield data and vegetation index map, select representative points of the target field block, collect soil and crop samples, and analyze the key nutrient content of the samples to obtain soil nutrient content dataset and crop nutrient content dataset;

[0061] Specifically, to achieve precision agriculture nutrient management, first of all, based on the target field soil type, historical yield data and vegetation index map for multi-source spatial overlay analysis. In the GIS platform, the three types of data are unified to the same coordinate system, and the variation weight value of each position is calculated by weighted fusion algorithm: the difference of soil type is given the basic weight, the historical yield fluctuation reflects the long-term fertility trend, and the vegetation index indicates the real-time crop growth. The three are fused according to the weight ratio of 3:4:3 to generate a field comprehensive variation grid map with a value range of 0-1. According to the grid value, the field is divided into high, medium and low variation levels, and the weight greater than 0.7 is determined as a high variation area, 8 sampling points per hectare, the weight between 0.4 and 0.7 is determined as a medium variation area, 5 sampling points per hectare, and the weight less than 0.4 is determined as a low variation area, 3 sampling points per hectare. When distributing points, strictly follow the principle of spatial uniformity, each point represents a 15-meter radius circular area, avoid 10 meters away from the ridge and far away from the ditch road, and finally generate a sampling point distribution map containing accurate GPS coordinates.

[0062] The professional sampling team carries stainless steel soil drill and liquid nitrogen insulation box for field sampling operation. Soil sampling adopts five-point method: taking 5 drills of 0-20 cm plough layer soil samples within a radius of 5 meters with point as center, mixing 1 sample after removing stones and roots on site, and recording sampling depth error within ±2 cm. Crop sampling strictly follows the organ standard, such as taking the latest fully expanded leaves of corn and the third leaf of the main stem of wheat, and collecting 10 leaves of the same part per point. After natural drying for 7 days, the soil samples are ground through 2 mm nylon screen, sealed and packed according to the standard of 200 grams per bag, and the leaf samples are immediately frozen in liquid nitrogen tank and transferred to-80℃ ultra-low temperature refrigerator for storage, to ensure that the nutrients are not degraded. In the laboratory analysis stage, samples are detected according to industry standard, each batch of samples is inserted into standard material calibration, and 15% parallel sample error control is set to ensure that the relative deviation is less than or equal to 5%.

[0063] After obtaining the original analysis data, two structured data sets are constructed. The crop nutrient content data set retains the sampling point scale characteristics, and the spatial coordinates and nitrogen, phosphorus and potassium contents of each point are stored in CSV format. This data set is directly associated with the vegetation index modeling and does not perform spatial extrapolation. The fields include point, coordinate, total nitrogen, total phosphorus, total potassium content and sampling date. The soil nutrient content data set needs to cover the entire field. The discrete sampling point data is spatialized using the Kriging interpolation method. Based on the semi-variation function model, the autocorrelation weight is calculated using the digital elevation model as the covariate to generate a 1-meter resolution raster layer. The alkali-hydrolyzable nitrogen, available phosphorus and available potassium are stored as independent bands, respectively, and the fourth band records the interpolation standard deviation. After 20% of the reserved points are verified, the interpolation error of alkali-hydrolyzable nitrogen is less than or equal to 8 mg / kg, and the available phosphorus is less than or equal to 2 mg / kg, forming a GeoTIFF file that can fully reflect the spatial heterogeneity of soil fertility. The final output results include the crop nutrient content data set and the soil nutrient content data set, which are spatially related through a unified coordinate system. The crop data set provides true value calibration for vegetation index inversion, and the soil data set directly drives zoning fertilization decision-making, together forming the basis of the data chain for precision agriculture nutrient management. All data are accompanied by metadata documents, which detail sampling specifications, detection methods and accuracy verification reports to ensure that the results are traceable and reusable.

[0064] S4: Apply random forest regression algorithm to associate and model the crop nutrient content data set and the vegetation index map, and use the model to calculate the nutrient content of the target field to obtain the nutrient spatial distribution map;

[0065] Further, the step S4 further comprises:

[0066] Obtain the crop nutrient content data set and the vegetation index map, and perform spatial matching through GIS software to extract the sampling point number, coordinates, average value of vegetation index pixels, band ratio, texture features and measured nutrient data to construct the modeling data set;

[0067] Select the random forest regression algorithm for associated modeling, and the model expression is:

[0068] ;

[0069] wherein, represents the predicted nutrient content, T represents the number of decision trees, and t represents the index value, represents the single decision tree prediction function, and x represents the data column in the modeling data set;

[0070] Extract 70% of the data in the modeling data set as the training set to train the model, and determine the optimal hyperparameters through grid search with the minimum mean square error as the criterion;

[0071] Based on the trained model, the target field nutrient content is inverted, and an initial nutrient spatial distribution map is generated.

[0072] 30% of the independent samples in the modeling data set are used as the validation set. The model accuracy is verified by statistical verification indicators and residual space diagnosis. The statistical verification indicators include the determination coefficient and the root mean square error. The residual space diagnosis uses the Moran index. The specific formulas are as follows:

[0073] Determination coefficient: ;

[0074] Root mean square error: ;

[0075] wherein, represents the determination coefficient, the root mean square error, and the remaining parameters in the two formulas have the same physical meaning. y i is the measured nutrient content of the i-th validation point, represents the nutrient content predicted by the model for the corresponding point, which is obtained from the initial nutrient spatial distribution map, represents the mean value of all nutrient content data in the validation set, and n represents the number of samples in the validation set;

[0076] Moran index: ;

[0077] wherein, I is the Moran index, n represents the total number of residual points, i and j are two adjacent points, represents the residual value of the i-th point, and the specific value is , represents the residual value of the j-th point, and the calculation formula is the same as , represents the global mean value of the residual, and the specific value is , is a spatial weight matrix generated by a distance decay function, and the specific value is , wherein dij is the Euclidean distance between points i and j;

[0078] The predicted data calculated by the model is judged based on the determination coefficient, the root mean square error and the Moran index. If the judgment result has an aggregation deviation, the high estimation area is corrected. The specific correction formula is as follows:

[0079] ;

[0080] wherein, represents the corrected predicted nutrient content, represents the original predicted nutrient content, is the local residual mean value, is the maximum measured nutrient content in the region;

[0081] Finally, the initial nutrient spatial distribution map is rendered into a color picture via the corrected HSV color space model to obtain the nutrient spatial distribution map.

[0082] Specifically, a crop nutrient content dataset and a vegetation index map are obtained, where the crop nutrient content dataset contains the accurate coordinate positions of each sampling point and the corresponding nutrient data, and the vegetation index map contains the multi-band GeoTIFF file of NDVI or NDRE index. The crop nutrient content dataset and the vegetation index map are imported into GIS software for spatial matching, and a circular buffer area with a radius of 5 meters is created at the center of the sampling point coordinate by using the point sampling tool. The average value of all vegetation index pixels in the buffer area is extracted, and the model sensitivity is improved by increasing the calculation of derived features, including band ratio and texture features. The specific formulas are as follows:

[0083] Vegetation index pixel average value: ;

[0084] wherein, represents the vegetation index pixel average value, n represents the number of vegetation index pixels, and i is the index value, represents the specific value of the i th vegetation index pixel;

[0085] Band ratio: ;

[0086] wherein, BD represents the band ratio, represents the near-infrared band surface reflectivity, represents the red band surface reflectivity.

[0087] Texture feature: ;

[0088] wherein, represents the texture feature, represents the contrast of the near-infrared band gray level co-occurrence matrix, i represents the gray value of the reference pixel, j represents the gray value of the adjacent pixel, p(i,j) is the joint probability of the pixel gray level, represents the frequency of the gray level combination (i,j) appearing, and represents the canopy structure roughness.

[0089] The sampling point number, coordinate, vegetation index pixel average value, band ratio, texture feature, and measured nutrient data are integrated to form a modeling dataset.

[0090] The random forest regression algorithm is selected for correlation modeling, and the model expression is as follows:

[0091] ;

[0092] wherein, T represents the predicted nutrient content, T represents the number of decision trees, t represents the index value, and the specific value is 500, T represents the predicted nutrient content, T represents the number of decision trees, t represents the index value, and the specific value is 500,

[0093] According to the set model, 70% of the samples in the modeling data set were extracted as the training set according to the soil type, and the optimal hyperparameters were determined by grid search with the least mean square error as the criterion, and finally the tree depth was optimized to 3-15 and the minimum sample number of the leaf node was 1-10. The trained model is applied to the target field pixel-level prediction, the vegetation index is reorganized into a three-dimensional feature array, the dimensions include row number, column number and feature number, the model is input pixel by pixel to obtain the predicted value, and a floating-point initial nutrient spatial distribution map is generated. At the same time, in order to eliminate the random fluctuations of the model, 5*5 pixel mean filtering is performed, and the farmland boundary vector mask is used to remove non-cultivated areas.

[0094] 30% of the independent samples reserved in the modeling data set are used as the validation set to verify the accuracy of the model, and the determination coefficient and the root mean square error are calculated, and the specific formulas are:

[0095] Determination coefficient: ;

[0096] Root mean square error: ;

[0097] Among them, represents the determination coefficient, the root mean square error, the physical meanings of the remaining parameters in the two formulas are the same, i is the measured nutrient content of the i-th validation point, represents the nutrient content predicted by the model for the corresponding point, which is obtained from the initial nutrient spatial distribution map, represents the average value of all nutrient content data in the validation set, n represents the number of samples in the validation set, when R² ≥ 0.75, RMSE ≤ 0.35%, it indicates that the predicted data obtained by the model is accurate.

[0098] Since the determination coefficient and the root mean square error can only reflect the overall accuracy and cannot capture the spatial heterogeneity of the error, the residual spatial diagnosis is introduced to further verify the model. For spatial residual distribution, the Moran index is used to test spatial autocorrelation, and the specific formula is:

[0099] ;

[0100] Among them, I is the Moran index, n represents the total number of point positions, i and j are two adjacent point positions, represents the residual value of the i-th point, the specific value is , represents the residual value of the j-th point, the calculation formula is the same as same, representing the residual global mean, reflecting the system bias direction, the specific value is , is a spatial weight matrix generated by a distance decay function, and the specific value is , where d ij is the Euclidean distance between points i and j.

[0101] If I>0.3 is detected, it is judged as an aggregation bias, and correction needs to be implemented in the overestimation area. The determination standard of the overestimation area is the area with a residual value less than -2σ, where σ is the global residual standard deviation. The specific correction formula is:

[0102] ;

[0103] where, represents the corrected predicted nutrient content, represents the original predicted nutrient content, is the local residual mean, is the maximum measured nutrient content in the area.

[0104] The final initial nutrient spatial distribution map after verification and correction is in the floating-point GeoTIFF format. Finally, the mapping of agronomic significance is realized through the HSV color space model. The area with a nutrient content less than 2.5% is rendered as red, indicating the nitrogen stress area. The area with a nutrient content between 2.5% and 3.2% is rendered as yellow, indicating the potential deficiency area. The area with a nutrient content greater than 3.2% is rendered as dark green, indicating the sufficient area. Finally, the complete nutrient spatial distribution map is obtained.

[0105] S5: Integrating the vegetation index map, soil nutrient content dataset, nutrient spatial distribution map, and historical yield data, applying clustering analysis algorithm for spatial partitioning, dividing the target field block into M management units, determining the required fertilizer application amount and element abundance and deficiency index of each management unit, and generating fertilizer ratio decision;

[0106] Further, the step S5 further includes:

[0107] Integrating the vegetation index map, soil nutrient content dataset, nutrient spatial distribution map, and historical yield data, performing spatial registration, unifying the coordinate system and data standardization;

[0108] Using K-means++ clustering algorithm to realize management unit division, obtaining M homogeneous management units;

[0109] Obtaining the target yield of each management unit based on the vegetation index map and historical yield data, and the specific formula is:

[0110] ;

[0111] wherein, represents the target yield of the kth unit, represents the average yield of the kth unit in the past three years, and S represents the yield-increasing coefficient, represents the vegetation index of the unit, represents the health threshold of the vegetation index;

[0112] The available soil nutrient supply is obtained based on the soil nutrient content dataset, and the specific formula is:

[0113] ;

[0114] wherein, represents the available soil nutrient supply of the e nutrient element of the kth unit, represents the soil nutrient content of the e nutrient element of the kth unit, represents the available coefficient of the e nutrient element, and G represents the conversion factor indicating the conversion from mg / kg unit to kg / acre unit;

[0115] The soil available coefficient correction factor is obtained based on the soil nutrient content dataset and the nutrient spatial distribution map, and the specific formula is:

[0116] ;

[0117] wherein represents the soil crop nutrient transfer efficiency ratio of the e nutrient element, and the specific value is wherein, represents the crop nutrient content of the e nutrient element, represents the soil nutrient content of the e nutrient element;

[0118] The target yield, the available soil nutrient supply, and the soil available coefficient correction factor are obtained, and the nutrient balance method is used to calculate the fertilizer application amount, and the specific formula is:

[0119] ;

[0120] wherein, represents the fertilizer application amount of the e element of the kth unit, k represents the unit number, and there are M in total, e represents the nutrient element type, and U e × C k is the target fertilizer requirement, wherein U e represents the nutrient demand per unit yield of crops, and C k represents the target yield of the kth unit, represents the available soil nutrient supply of the e nutrient element of the kth unit, is the soil available coefficient correction factor, represents the fertilizer utilization rate of the e nutrient element of the kth unit, representing fertilizer efficiency adjustment coefficient, based on soil leaf nutrient transfer efficiency ratio determination, representing unit area;

[0121] obtaining fertilizer application amount, calculating the abundance and deficiency index of each element in each management unit, the specific calculation formula is:

[0122]

[0123] wherein representing fertilizer application amount, Line e representing the critical value of each element;

[0124] Finally, according to the fertilizer application amount and the element abundance and deficiency index, the fertilizer ratio decision is generated based on the minimum nutrient law.

[0125] Specifically, the vegetation index map, soil nutrient content dataset, nutrient spatial distribution map and historical yield data are obtained, wherein the vegetation index map represents the real-time growth of crops, which is calculated and generated by a set of unmanned aerial vehicle remote sensing images, the soil nutrient content dataset is generated by laboratory analysis combined with spatial interpolation, which contains nitrogen, phosphorus, potassium and other nutrient content data, the nutrient spatial distribution map is generated by random forest model inversion, which contains crop nutrient content data, and the historical yield data comes from the yield monitoring system of combine harvester. Four types of key datasets with spatial explicit expression are integrated, and the four types of datasets are uniformly input into GIS software to perform strict spatial registration. Through ground control point constraint, the coordinate system is moved to EPSG:32650 projection system, and the pixel size is resampled to 1*1 meter to ensure accurate matching of spatial position. In view of the dimension difference of data, Z-score standardization processing is adopted, and finally a multi-source spatial data cube is generated, which contains 8 feature layers in the depth direction: vegetation index, soil nitrogen, soil phosphorus, soil potassium, leaf nitrogen, leaf phosphorus, leaf potassium, and three-year average yield.

[0126] Based on the standardized multi-source spatial data cube, the K-means++ clustering algorithm is applied to realize fine partitioning of the field. This method first avoids the randomness deviation of traditional K-means through intelligent initialization of the center of mass, ensuring that the partitioning result is stable and reliable. In the feature space, the algorithm measures the similarity of each pixel to the center of mass according to the Euclidean distance, and performs the assignment and update cycle through multiple iterations: dynamically classifying the spatial pixels to the nearest center of mass cluster, and then recalculating the geometric center position within the cluster. The iteration process continues until the center of mass moves less than the preset threshold, which is set to 0.001 units in this case. At this time, the spatial pattern tends to be stable, and the partitioning raster map is output. This map divides the target field into M homogeneous management units, and the spectral, soil, nutrient and yield characteristics within each unit are highly uniform, with a variation coefficient strictly controlled within 15%, which is significantly lower than the spatial fluctuation level of more than 35% of the original field.​

[0127] For each management unit, the fertilizer application amount is calculated by using the nutrient balance formula, and a personalized fertilization scheme is formulated:

[0128]

[0129] wherein, represents the fertilizer application amount of the kth unit e nutrient element, k represents the unit number, a total of M, e represents the nutrient element type, determined by agronomic diagnosis as nitrogen, phosphorus, and potassium, U e ×C k is the target fertilizer requirement, wherein U e represents the nutrient requirement per unit yield of crops, derived from variety test data, C k represents the target yield of the kth unit, represents the soil available nutrient supply of the kth unit e nutrient element, is a soil available coefficient correction factor, represents the fertilizer utilization rate of the kth unit e nutrient element, with specific values of 40% for urea, 20% for phosphorus fertilizer, and 50% for potassium fertilizer, represents a fertilizer efficiency adjustment coefficient, when is less than 0.8, the value is 1.2, and otherwise the value is 1, represents the unit area;

[0130] wherein the specific obtaining formula of the target yield C k is wherein represents the average yield of the kth unit in the past three years, and S represents the yield increase coefficient, with a specific value of 1.1~1.3, represents the vegetation index of the unit, represents the health threshold of the vegetation index;

[0131] wherein the specific obtaining formula of the soil available nutrient supply U is wherein represents the soil nutrient content of the kth unit e nutrient element, represents the available coefficient of e nutrient element, with specific values of 0.3 for nitrogen, 0.4 for phosphorus, and 0.5 for potassium, and G represents the conversion factor, indicating the conversion from mg / kg unit to kg / mu unit;

[0132] wherein the specific obtaining formula of the soil available coefficient correction factor C is wherein represents the soil crop nutrient transfer efficiency ratio of e nutrient element, with a specific value of wherein, ​crop nutrient content representing e nutrient element, soil nutrient content representing e nutrient element.

[0133] After the fertilizer application amount is obtained, a fertilizer ratio decision is generated based on a minimum nutrient law, and an element abundance index of each element in each management unit is calculated, and the specific calculation formula is , wherein represent the fertilizer application amount, Line e represent the critical value of each element, nitrogen is 90 mg / kg, phosphorus is 15 mg / kg, potassium is 100 mg / kg, and the element abundance index of the element Ie is less than 80%. For example, the element abundance index of nitrogen in unit k is 65%, and the element abundance index of phosphorus is 110%. At this time, only nitrogen fertilizer needs to be supplemented.

[0134] S6: using GIS software to spatialize the decision results to obtain a digital fertilization prescription map.

[0135] Further, the step S6 further includes:

[0136] Based on the vector boundary of the M management units generated by spatial clustering, a fertilization prescription thematic map layer is created using GIS software, the fertilizer decision data is connected through the attribute table association function, and the universal geodetic coordinate system is used to ensure the uniformity of the spatial reference;

[0137] Implement multi-level visual rendering to generate a digital fertilization prescription map, and the multi-level visual rendering includes fertilizer type expression, application amount intensity mapping, and dynamic agronomic annotation.

[0138] Specifically, the fertilizer ratio decision is obtained, and the decision result is converted into a spatialized and executable digital fertilization prescription map through a GIS platform. First, based on the M management unit vector boundaries obtained through spatial clustering analysis, a fertilization prescription thematic layer is created in QGIS or ArcGIS Pro. The layer inherits all the geometric properties of the management unit and dynamically connects the fertilizer ratio decision through the attribute table. To ensure accurate identification of the agricultural machinery system, the universal geodetic coordinate system is used, and the unit boundary is verified by the topological checking tool to ensure that there is no overlap or gap, and the boundary positioning error is strictly controlled within 0.3 meters. Then, multi-level visualization rendering is implemented, including: fertilizer type expression, a composite symbol system is used, a 45° blue diagonal line is filled to represent the urea application area, an orange dot matrix is filled to represent the phosphorus fertilizer area, and a red grid is filled to represent the potassium fertilizer area. The three symbols are displayed in superposition according to the actual ratio, for example, urea and phosphorus fertilizer are needed in a unit, and then a blue and orange interlaced diagonal dot matrix mixed texture is displayed; application amount intensity mapping, according to the pure nitrogen per mu amount, five color gradients are divided, light green marks the ecological protection area, representing 0-5 kg / mu, yellow indicates the optimized control area, representing 5-10 kg / mu, orange represents the conventional management area, representing 10-15 kg / mu, red warns the soil fertility improvement area, representing 15-20 kg / mu, and dark red highlights the capacity breakthrough area, representing more than 20 kg / mu, and the layer superposition does not cover the background map through semi-transparent rendering; dynamic agricultural annotation, a text box is automatically generated at the center of each unit, and a customized fertilization formula is displayed, such as N:8.5 kg / mu, an operation prompt label is hung at the boundary, for example, unit 7 applies 60% urea before September 15, the font size is dynamically adjusted according to the map scaling, and the text is clear and identifiable when viewed on the mobile terminal.

[0139] The finally generated prescription map converts the agronomic decision into a spatially locatable, measurable, and traceable geographical entity, realizing the precision management mode.

[0140] In the foregoing embodiments, the fertilizer ratio management method based on remote sensing data analysis is also invented. Please refer to the accompanying drawings Figure 2 , the system comprises:

[0141] The remote sensing image acquisition module 11 is used to shoot the original data set of the crops in the target field during the key growth period by using the unmanned aerial vehicle, to pre-process the original data set, and to obtain a remote sensing image set. The pre-processing includes radiation calibration, atmospheric correction, geometric correction, and cloud mask;

[0142] The vegetation index calculation module 12 is used to calculate the vegetation index according to the remote sensing image set, and to generate a vegetation index map in combination with the GIS software;

[0143] The soil and crop nutrient analysis module 13 is configured to select representative points of the target field based on the target field soil type, historical yield data and vegetation index map, collect soil and crop samples, analyze the key nutrient content of the samples, and obtain a soil nutrient content dataset and a crop nutrient content dataset;

[0144] The nutrient inversion modeling module 14 is configured to apply a random forest regression algorithm to model the correlation between the crop nutrient content dataset and the vegetation index map, and use the model to calculate the nutrient inversion of the target field to obtain a nutrient spatial distribution map.

[0145] The precise zoning and fertilization decision module 15 is configured to comprehensively use the vegetation index map, the soil nutrient content dataset, the nutrient spatial distribution map and the historical yield data, apply a clustering analysis algorithm to spatial zoning, divide the target field into M management units, determine the required fertilizer application amount and element abundance index of each management unit, and generate a fertilizer ratio decision.

[0146] The digital fertilization prescription generation module 16 is configured to use GIS software to spatially express the decision results to obtain a digital fertilization prescription map.

[0147] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, however, as long as the combinations of the technical features do not contradict, they should be considered as within the scope of the present disclosure.

[0148] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fertilizer ratio management method combining remote sensing data analysis, characterized in that, The method includes: The raw dataset of crops during the key growth period of the target field is captured by drone. The raw dataset is then preprocessed to obtain a set of remote sensing images. The preprocessing includes radiometric calibration, atmospheric correction, geometric correction, and cloud masking. Vegetation indices are calculated based on remote sensing image sets, and vegetation index maps are generated using GIS software. Based on the soil type, historical yield data and vegetation index map of the target field, representative locations of the target field were selected, soil and crop samples were collected, and key nutrient content analysis was performed on the samples to obtain soil nutrient content datasets and crop nutrient content datasets. The random forest regression algorithm was applied to model the correlation between the crop nutrient content dataset and the vegetation index map. The model was then used to perform nutrient inversion calculations on the target field to obtain a nutrient spatial distribution map. Based on the integrated vegetation index map, soil nutrient content dataset, nutrient spatial distribution map, and historical yield data, a clustering analysis algorithm is applied to spatially partition the target field into M management units. The required fertilizer application rate and element abundance / deficiency index for each management unit are determined, generating fertilizer ratio decisions. The generation of fertilizer ratio decisions includes: Integrate vegetation index maps, soil nutrient content datasets, nutrient spatial distribution maps, and historical yield data; perform spatial registration; and unify coordinate systems and data standardization. The K-means++ clustering algorithm is used to divide the management units, resulting in M ​​homogeneous management units; The target yield for each management unit is obtained based on the vegetation index map and historical yield data, using the following formula: ； in, Represents the target output of the k-th unit. S represents the average output of unit k over the past three years, and S represents the production growth coefficient. The vegetation index represents the unit. The health threshold representing the vegetation index; The available nutrient supply in soil is obtained based on a soil nutrient content dataset, using the following formula: ； in, This represents the available soil nutrient supply of nutrient element e in the k-th unit. This represents the soil nutrient content of nutrient element e in unit k. The effective coefficient of nutrient element e is represented by G, which represents the conversion factor, indicating the conversion of mg / kg units to kg / acre units. The soil availability coefficient correction factor is obtained based on the soil nutrient content dataset and nutrient spatial distribution map. The specific formula is as follows: ； in The soil crop nutrient transfer efficiency ratio representing nutrient element e is shown in the following figure. ,in, The crop nutrient content representing nutrient element e. This represents the soil nutrient content of nutrient element e; To obtain the target yield, soil available nutrient supply, and soil availability coefficient correction factor, the fertilizer application rate is calculated using the nutrient balance method. The specific formula is as follows: ； in, This represents the fertilizer application amount for element e in unit k, where k represents the unit number (M units in total), and e represents the nutrient element type. e ×C k The target fertilizer requirement, of which U e C represents the nutrient requirement per unit yield of crops. k Represents the target output of the k-th unit. This represents the available soil nutrient supply of nutrient element e in unit k. This is a correction factor for the soil availability coefficient. The fertilizer utilization rate of nutrient element e in unit k is represented. This represents the fertilizer efficiency adjustment coefficient, determined based on the ratio of soil to leaf nutrient transfer efficiency. Represents the area of ​​a unit; Obtain the fertilizer application rate and calculate the abundance / deficiency index of each element in each management unit. The specific calculation formula is as follows: ； in Represents fertilizer application rate, Line e The critical values ​​representing each element; Finally, based on the fertilizer application rate and element abundance / deficiency index, a fertilizer ratio decision is generated according to the law of minimum nutrient requirements. GIS software is used to spatialize the decision-making results and obtain digital fertilizer prescription maps.

2. The fertilizer ratio management method combining remote sensing data analysis as described in claim 1, characterized in that, Obtain a set of remote sensing images, including: N ground control points were evenly selected in the target field, L-shaped markers were sprayed and their coordinate positions were recorded simultaneously. The flight parameters of the UAV are set based on the geographical conditions of the target field. The raw dataset of the target field is collected during the key growth period of the crop. The UAV is equipped with a multispectral sensor. The raw dataset includes the raw DN image set and flight parameter log. Obtain the raw dataset, perform radiometric calibration, and combine the sensor calibration file and flight parameter log to convert the raw DN value into apparent reflectance to obtain an apparent reflectance image set; Based on the apparent reflectance image set, the apparent reflectance is converted to surface reflectance using the empirical linear method assisted by the ground calibration plate, and the surface reflectance image set is obtained. The surface reflectance image set is coarsely stitched together using POS data and SIFT feature points to generate an initial orthophoto image set. A second-order polynomial transformation is performed using ground control point coordinates, and bilinear interpolation is used for resampling to output the orthophoto image set in the projection coordinate system. Cloud masking is performed based on the spectral characteristics of the orthoreflectivity image set to remove invalid regions, ultimately generating a remote sensing image set.

3. The fertilizer ratio management method combining remote sensing data analysis as described in claim 1, characterized in that, Generate a vegetation index map, including: Vegetation indices are calculated based on remote sensing image sets and crop growth stages. If the crop is in the early to mid-growth stage, the normalized difference vegetation index (NDI) is used; if the crop is in the mid-to-late growth stage, the red-edge normalized vegetation index (RBRI) is used. The specific formula for calculating the NDI is as follows: ； in, Represents the normalized difference vegetation index, ρ NIR ρ represents the surface reflectance in the near-infrared band. Red Represents the surface reflectance in the red light band; The specific formula for calculating the red-edge normalization index is as follows: ； Where NDRE represents the red-edge normalization exponent, ρ NIR Represents the surface reflectance in the near-infrared band. Represents the surface reflectance in the red-edge band; Spatial optimization is performed on the calculated vegetation indices, and the spatial optimization includes spatial relationship matching and pruning operations. Statistical features are obtained based on the spatially optimized vegetation index. Grading thresholds are set according to the statistical features. The HSV color space model is used to map different threshold areas into asymptotic color bands to generate a vegetation index map. The statistical features include minimum value, maximum value and quartiles.

4. The fertilizer ratio management method combining remote sensing data analysis as described in claim 1, characterized in that, Obtain a spatial distribution map of nutrients, including: We acquired crop nutrient content datasets and vegetation index maps, and used GIS software for spatial matching to extract sampling point numbers, coordinates, average values ​​of vegetation index pixels, band ratios, texture features, and measured nutrient data to construct a modeling dataset. The random forest regression algorithm was selected for association modeling, and the model expression is as follows: ； in, The value represents the predicted nutrient content, T represents the number of decision trees, and t represents the index value. represents the prediction function of a single decision tree, and x represents a data column in the modeling dataset; 70% of the data in the modeling dataset is extracted as the training set to train the model. The optimal hyperparameters are determined by grid search with the minimum mean squared error as the criterion. Based on the trained model, the nutrient content of the target field is inverted, and an initial nutrient spatial distribution map is generated; Using the remaining 30% of independent samples in the modeling dataset as the validation set, model accuracy was validated through statistical validation metrics and residual space diagnostics. The statistical validation metrics included the coefficient of determination and root mean square error, while the residual space diagnostics used the Moran's index. The specific formulas are as follows: Coefficient of determination: ; Root mean square error: ; in, Represents the coefficient of determination. Root mean square error; the remaining parameters in both formulas have the same physical meaning; y i The measured nutrient content at the i-th verification point. This represents the nutrient content predicted by the model at the corresponding point; this data is extracted from the initial nutrient spatial distribution map. represents the mean of all nutrient content data in the validation set, and n represents the number of samples in the validation set; Moran Index: ; Where I is the Moran index, n represents the total number of residual points, and i and j are two adjacent points. This represents the residual value at point i, specifically the value of... , The residual value at point j is obtained using the formula and same, This represents the global mean of the residuals, with a specific value of [value missing]. , This is the spatial weight matrix, generated by the distance decay function, with specific values ​​as follows: , where d ij Let be the Euclidean distance between point i and point j; The predicted data calculated by the model are evaluated based on the coefficient of determination, root mean square error, and Moran's index. If clustering bias is found in the evaluation results, corrections are applied to the overestimated areas. The specific correction formula is as follows: ； in, This represents the corrected predicted nutrient content. Represents the original predicted nutrient content. The mean of the local residuals. This represents the highest measured nutrient content in the region. Finally, the initial nutrient space distribution map after correction is rendered into a color image using the HSV color space model, thus obtaining the nutrient space distribution map.

5. The fertilizer ratio management method combining remote sensing data analysis as described in claim 1, characterized in that, Obtain digital fertilizer prescription maps, including: Based on the vector boundaries of M management units generated by spatial clustering, a fertilizer prescription thematic layer is created using GIS software. Fertilizer decision data is linked through the attribute table association function, and a universal geodetic coordinate system is used to ensure the uniformity of spatial benchmarks. Implement multi-level visualization rendering to generate digital fertilizer prescription maps. The multi-level visualization rendering includes fertilizer type expression, application rate intensity mapping and dynamic agronomic annotation.

6. A fertilizer ratio management system combining remote sensing data analysis, characterized in that, The system is used to implement the fertilizer ratio management method combining remote sensing data analysis as described in any one of claims 1 to 5, and the system comprises: The remote sensing image acquisition module is used to capture raw datasets of crops during the key growth period of the target field using drones, and to preprocess the raw datasets to obtain a set of remote sensing images. The preprocessing includes radiometric calibration, atmospheric correction, geometric correction, and cloud masking. The vegetation index calculation module is used to calculate the vegetation index based on the remote sensing image set and generate a vegetation index map in conjunction with GIS software. The soil and crop nutrient analysis module is used to select representative points of the target field based on the soil type, historical yield data and vegetation index map of the target field, collect soil and crop samples, analyze the key nutrient content of the samples, and obtain soil nutrient content datasets and crop nutrient content datasets. The nutrient inversion modeling module is used to apply the random forest regression algorithm to model the correlation between the crop nutrient content dataset and the vegetation index map, and to use the model to perform nutrient inversion calculations on the target field to obtain a nutrient spatial distribution map. The precise zoning and fertilization decision module is used to integrate vegetation index map, soil nutrient content dataset, nutrient spatial distribution map and historical yield data, apply cluster analysis algorithm to perform spatial zoning, divide the target field into M management units, determine the fertilizer application amount and element abundance / deficiency index required for each management unit, and generate fertilizer ratio decision. A digital fertilizer prescription generation module is used to spatially express decision results using GIS software to obtain a digital fertilizer prescription map.

Images