Variable fertilization method and device based on multi-source data fusion

Abstract

The invention provides a multi-source data fusion variable fertilization method and device, and belongs to the field of agricultural production technology and agricultural remote sensing application, and the method comprises the following steps: data acquisition and preprocessing: collecting historical data, ground sampling data, ground sensor data and remote sensing images; spatial data processing and consistency correction; variable fertilization model construction: constructing a fertilization amount and yield effect curve, constructing a yield and growth vigor index VI distribution diagram, and further constructing a variable fertilization model representing the relationship between the fertilization amount and the growth vigor index VI; the spatial heterogeneity quantification and zoning comprises the following steps: quantifying the spatial heterogeneity of environmental factors on fertilization requirements by using a geographic detector, and dividing a region into a plurality of sub-regions; and differential fertilization strategy optimization: generating a dynamic adjustment coefficient, and determining a final fertilization amount in combination with the fertilization amount recommended by the variable fertilization model. The crop growth condition can be described more comprehensively and accurately, and accurate agricultural management can be guided.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of agricultural production technology and remote sensing application, and particularly relates to a variable fertilization method and device based on multi-source data fusion. BACKGROUND

[0002] In recent years, precision agriculture technology has made significant progress in crop fertilization management, providing various innovative methods for optimizing fertilization strategies. On the one hand, sensor technology is widely used for real-time monitoring of soil nutrients, crop growth conditions, and environmental variables, providing real-time data with high resolution for fertilization decision-making. On the other hand, remote sensing technology uses satellite or unmanned aerial vehicle images to assess crop nitrogen requirements and calculate the optimal fertilization amount based on vegetation indices such as normalized difference vegetation index (NDVI) and leaf area index (LAI). In addition, geographic information system (GIS) technology is also used to analyze spatial heterogeneity of farmland, including soil types and terrain features, providing scientific support for variable rate fertilization. However, current research still faces some challenges. First, sensors have limitations in data collection accuracy and coverage, especially in large-scale farmland, where soil and crop heterogeneity cannot be fully captured. Second, remote sensing technology is limited by weather and light conditions, which may result in inaccurate data acquisition. Overall, existing technologies have problems such as single data source, lack of dynamicity in models, and insufficient analysis of spatial heterogeneity, which limit the precision and relevance of fertilization programs.

[0003] In recent years, methods combining remote sensing data, environmental factors, and ground sensor data have been proven to effectively improve fertilization utilization and have been widely used in crop precision fertilization research. Among them, the existing technology proposes a winter wheat fertilization method, which includes: collecting soil data, meteorological data and management measure data of the target area; inputting the collected data into the applicability process model for simulation, and using the measured yield to calibrate and verify the model; using the calibrated model to determine the maximum potential target yield of winter wheat in the target area; then determining the target yield of recommended fertilization in the target area; calculating the nitrogen agronomic efficiency of winter wheat crops in the target area according to the target yield; then inferring the nitrogen fertilizer use amount of the target area according to the nitrogen agronomic efficiency; determining the amount and type of organic fertilizer and chemical fertilizer in the target area according to the nitrogen fertilizer use amount of the target area; determining the proportion of base nitrogen fertilizer and topdressing nitrogen fertilizer and the type of nitrogen fertilizer according to the winter wheat nutrient absorption curve simulated by the applicability process model and the fertilizer distribution result.

[0004] However, the following main disadvantages exist:

[0005] (1) Only through fixed formula calculation of nitrogen agronomic efficiency, nitrogen fertilizer yield response and final nitrogen fertilizer use, such empirical formula may not have complete adaptability in different soil and crop growth environment, and the complexity of local conditions is easily ignored;

[0006] (2) Over-reliance on historical data, unable to adjust in real time, under the influence of uncontrollable factors such as extreme weather, pests and diseases, the model prediction and actual crop response may have large deviation, thereby affecting the effect of fertilization management;

[0007] (3) The precision is low to ensure precision fertilization by interpolating the soil sampling data to obtain the conditions in different regions.

[0008] (4) The average value of the whole field is used for fertilization design, ignoring the spatial variability of soil fertility, terrain and microclimate and other factors in the field. SUMMARY

[0009] The present application proposes a variable fertilization method and device based on multi-source data fusion, which constructs a dynamic adaptive model by fusing remote sensing data and environmental factors, optimizes fertilization zoning by combining geographic detectors, and significantly improves fertilization accuracy and resource utilization rate. The specific technical scheme is as follows:

[0010] A variable fertilization method based on multi-source data fusion, comprising the following steps:

[0011] Step 1, data acquisition and preprocessing, including: collecting historical data of yield corresponding to fertilization amount, ground sampling data, ground sensor data and remote sensing image, and preprocessing the remote sensing image;

[0012] Step 2, spatial data processing and consistency correction, including: unifying data scale, time and space alignment, processing ground sensor data to generate soil nutrient and weather data;

[0013] Step 3, variable fertilization model construction, including: constructing fertilization amount and yield effect curve, constructing yield and vigor index VI distribution map, and constructing variable fertilization model representing the relationship between fertilization amount and vigor index VI according to the fertilization amount and yield effect curve and the yield and vigor index VI distribution map;

[0014] Step 4, spatial heterogeneity quantification and zoning, including: quantifying the spatial heterogeneity of fertilization demand of environmental factors by using geographic detector, and dividing the region into multiple sub-regions; wherein, the environmental factors include soil nutrient and weather data;

[0015] Step 5, differential fertilization strategy optimization, including: generating dynamic adjustment coefficient for each sub-region, combining the recommended fertilization amount of the variable fertilization model, and determining the final fertilization amount.

[0016] A variable fertilization device based on multi-source data fusion comprises the following modules:

[0017] A data acquisition and preprocessing module is configured to collect historical data of fertilization amount corresponding to yield, ground sampling data, data obtained by ground sensors, and remote sensing images, and to preprocess the remote sensing images.

[0018] A spatial data processing and consistency correction module is configured to unify data scales, perform spatio-temporal alignment, and process ground sensor data to generate soil nutrient and meteorological data.

[0019] A variable fertilization model construction module is configured to construct a fertilization amount-yield effect curve, construct a yield-vegetation index (VI) distribution map, and construct a variable fertilization model representing the relationship between fertilization amount and VI based on the fertilization amount-yield effect curve and the yield-vegetation index (VI) distribution map.

[0020] A spatial heterogeneity quantification and partitioning module is configured to quantify the spatial heterogeneity of fertilization demand of environmental factors, including soil nutrients and meteorological data, by using a geographic detector, and divide the area into multiple sub-regions.

[0021] A differentiated fertilization strategy optimization module comprises generating dynamic adjustment coefficients for each sub-region and determining the final fertilization amount based on the fertilization amount recommended by the variable fertilization model.

[0022] An electronic device comprises one or more processors and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above method.

[0023] A computer-readable storage medium stores executable instructions, which, when executed by a processor, cause the processor to implement the above method.

[0024] The present application has the following advantages:

[0025] The present application integrates multi-source data, including soil data, meteorological data, crop growth conditions, and remote sensing image data, to construct a dynamic and adaptive fertilization model. Compared with traditional methods, this model not only considers soil and meteorological conditions, but also inverses crop vigor parameters through remote sensing technology, thereby more comprehensively reflecting the actual growth needs of crops. This comprehensive method can effectively capture small-scale growth differences in the field, providing a more scientific basis for precision fertilization.

[0026] The present application quantifies the spatial heterogeneity of environmental factors on fertilizer demand using geographic probes, and divides the study area into multiple sub-regions through clustering algorithms. This method can accurately determine the fertilizer demand of each sub-region according to the differences in soil fertility, meteorological conditions and other factors, avoiding the uneven fertilization problem caused by ignoring spatial variability in traditional methods. Through differentiated fertilization strategies, fertilizer resources can be more effectively utilized, and fertilizer utilization rate can be improved.

[0027] The fertilization model of the present application can dynamically adjust the amount of fertilizer according to real-time data. Unlike traditional methods that rely on historical data and fixed formulas, the present application dynamically updates fertilization decisions by monitoring crop growth conditions and environmental changes in real time. This dynamic adjustment mechanism can better cope with the impact of uncontrollable factors such as extreme weather, pests and diseases, ensuring the timeliness and accuracy of the fertilization plan.

[0028] The present application realizes high-precision inversion of various physicochemical parameters through remote sensing inversion technology, and generates physicochemical composite vigor parameters of crops by combining machine learning modeling. This method not only improves monitoring accuracy, but also quickly obtains crop growth information in large areas of farmland, providing technical support for precision fertilization in large areas of farmland.

[0029] The present application has high flexibility in data acquisition. When it is difficult to obtain data by unmanned aerial vehicle, satellite remote sensing images can be used as a substitute; in the absence of ground sensor data, relevant information can be obtained through field sampling and public weather data. In addition, the clustering algorithm can select different methods according to specific needs, further enhancing the applicability and flexibility of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 The technical roadmap of the present application is shown in Figure

[0031] Figure 2 The fertilizer effect curve is shown in Figure DETAILED DESCRIPTION

[0032] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other. In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme.

[0033] The present application provides a variable fertilization method based on multi-source data fusion, as shown in Figure Figure 1 The method comprises the following steps:

[0034] Step 1, data acquisition and preprocessing, including: collecting historical data of fertilizer amount corresponding to yield, ground sampling data, ground sensor data acquisition data and remote sensing image, and preprocessing the remote sensing image;

[0035] Step 2, spatial data processing and consistency correction, including: unifying data scale, time and space alignment, processing ground sensor data to generate soil nutrient and meteorological data;

[0036] Step 3, variable fertilization model construction, including: constructing the fertilizer amount and yield effect curve, constructing the yield and vigor index VI distribution graph, and constructing the variable fertilization model representing the relationship between the fertilizer amount and the vigor index VI according to the fertilizer amount and yield effect curve and the yield and vigor index VI distribution graph;

[0037] Step 4, spatial heterogeneity quantification and partitioning, including: quantifying the spatial heterogeneity of fertilizer demand of environmental factors by using geographic detector, and dividing the region into multiple sub-regions; wherein the environmental factors include soil nutrient and meteorological data;

[0038] Step 5, differential fertilization strategy optimization, including: generating dynamic adjustment coefficient for each sub-region, combining the recommended fertilizer amount of the variable fertilization model, and determining the final fertilizer amount.

[0039] Step 1 is specifically:

[0040] In the present application, four types of data are used, including historical data of fertilizer amount corresponding to yield, ground sampling data (aboveground fresh biomass, leaf area index, leaf chlorophyll relative content and leaf nitrogen content), environmental and meteorological data obtained by ground sensors and high-resolution multispectral remote sensing images obtained by unmanned aerial vehicles, wherein the historical data includes fertilizer-yield effect curve.

[0041] The remote sensing image is preprocessed before use. The preprocessing work is specifically: for the unmanned aerial vehicle multispectral remote sensing image, first, geometric correction is performed, the mapping relationship between the image and the geographic coordinates is established through the layout of ground control points (GCPs) combined with the second order polynomial model, and the orthographic correction is performed based on the digital terrain model DEM to eliminate the terrain distortion; secondly, through radiation correction, the original DN (Digital Number) value is converted into radiation brightness by using the sensor gain or bias parameter, and the atmospheric interference is eliminated by using the radiation transfer model (such as FLAASH) to obtain the ground reflectivity; then the image position matching is carried out, the sub-pixel level registration of multi-temporal images is realized through the scale invariant feature transformation SIFT feature point extraction and the random sampling consensus RANSAC algorithm, and the bilinear interpolation resampling to the unified coordinate system is adopted; finally, the denoising processing is completed through the median filter or the frequency domain band stop filter to generate the high-quality image without cloud and noise.

[0042] Step 2 is specifically:

[0043] Data standardization: Convert data from different sources, scales, and formats into a unified data format that can be directly compared and integrated. Align remote sensing images and ground sensor data in time and space to the same spatio-temporal reference system and unified numerical scale. Unified numerical scale includes: For ground sensor data outliers (including soil and weather sensors), define physically unreasonable values (such as soil moisture > 100%) and statistical outliers, directly remove through threshold filtering, linear interpolation to fill missing segments, or sliding window mean to smooth high-frequency noise, to ensure data reliability. Use Kriging interpolation method to predict soil nutrients and weather data at target locations based on observation point data and their spatial correlation. The interpolation formula is:

[0044] (1)

[0045] where, represents the value of the point to be interpolated, is the weight coefficient of each observation point, is the actual measured value of each observation point.

[0046] Temporal and spatial alignment includes: High-resolution multispectral remote sensing images collected by unmanned aerial vehicles and ground sensor data are matched through coordinates and aligned to the same crop growth period through time window.

[0047] Step 3 is specifically:

[0048] According to the historical data of fertilization and yield, construct the fertilizer yield effect curve to obtain the relationship between yield and fertilizer amount:

[0049] (2)

[0050] where Y represents annual yield and N is annual fertilizer amount.

[0051] Using the fertilizer yield effect curve, the best fertilizer amount N best for the season is calculated.

[0052] Then, using the entropy method to assign weights to four growth-stage physicochemical parameters collected during field trials—aboveground fresh biomass, leaf area index, relative chlorophyll content, and leaf nitrogen content—a physicochemical parameter was constructed to quantify crop growth. First, each physicochemical parameter was normalized to its range, mapping the original values ​​to the [0, 1] interval to eliminate dimensional differences. A weighting matrix was then constructed based on the normalized data to calculate the relative contribution of each sample to each physicochemical parameter. The information entropy formula was then used to quantitatively evaluate the information entropy value of each physicochemical parameter, and the "redundancy" was calculated and normalized to obtain the entropy weight of each physicochemical parameter. A higher entropy weight indicates a richer information component and a higher contribution to overall growth. Finally, the normalized values ​​of each physicochemical parameter were weighted and summed according to the entropy weight to obtain the overall growth index (VI) for each sample. Here, the sample refers to the ground sampling data, corresponding to the ground sampling values ​​of aboveground fresh biomass, leaf area index, relative chlorophyll content, and leaf nitrogen content.

[0053] Using multispectral UAV imagery data to invert the physical and chemical parameters of each growth period, a crop growth distribution map was constructed. A random forest regression model was constructed using remote sensing spectra, vegetation indices, and texture features. The spatiotemporal distribution of key growth physical and chemical parameters, such as aboveground fresh biomass, leaf area index, leaf chlorophyll relative content, and leaf nitrogen content, was obtained for each key growth period. A comprehensive growth index (VI) distribution map was constructed using the entropy method, and a yield-growth index (VI) relationship model was established.

[0054] 3)

[0055] By correlating equations (2) and (3) through yield, a variable fertilization model is established to characterize the relationship between growth index and fertilizer amount:

[0056] (4)

[0057] On this basis, by substituting the growth parameters of the study area into formula (4), the optimal fertilizer application distribution map in a specific area is calculated, and the fertilizer application amount of the corresponding category is calculated based on the agronomic efficiency and fertilizer yield response.

[0058] Fertilizer-yield effect curve The fertilization effect curve is a functional model of the relationship between the yield in previous years (kg / mu) and the total annual fertilizer application (kg / mu), such as Figure 2 As shown: The horizontal axis is the winter wheat yield Y, and the vertical axis is the fertilizer amount N. When the winter wheat yield Y is between 0 and 400 kg / mu, the nitrogen application amount N is constant at 12 kg / mu. When the yield exceeds 400 kg / mu, the fertilizer amount N increases with the yield Y, and the function is N=-0.84(100 / Y) 2+11.29 (100 / Y) -19.66. When the yield reaches 600 kg / acre and above, the curve again tends to flatten, and the amount of fertilizer stabilizes at about 17.6 kg / acre.

[0059] Step 4 is specifically:

[0060] The ground sensor data includes meteorological and soil nutrient data, and the fertilizer demand is affected by soil nutrients, meteorological and environmental factors, and these factors often show significant spatial heterogeneity. By quantifying the spatial distribution characteristics of the influence of each factor on fertilizer demand, the differences in fertilizer demand in different regions can be determined, thereby providing a basis for subsequent fine management. Geographical detector is a method for detecting spatial heterogeneity between variables. By calculating the q value, the explanatory power of an environmental factor on the spatial heterogeneity of fertilizer demand can be quantified. The formula is as follows:

[0061] (5)

[0062] In the formula, is the number of sub-regions or partitions of each factor, that is, the entire study area is divided into how many non-overlapping sub-regions according to the value range of the factor. is the total number of global samples in the geographical detector, that is, the sum of all samples in the study area, is the number of samples in the sub-region, is the variance of the sub-region, is the global variance.

[0063] According to the influence of each environmental factor (q value) screening, by normalizing and weighted sum of selected factors, a comprehensive index is formed to reflect the comprehensive influence of the environment. By minimizing the intra-class distance, K-means clustering can divide the entire region into multiple sub-regions that are homogeneous internally and heterogeneous externally according to the distribution of the comprehensive index. This partitioning method can more accurately reflect the differences in fertilizer demand in different regions, thereby achieving more targeted fertilizer management. The formula is as follows:

[0064] (6)

[0065] In the formula, represents the clustering objective function, which is the sum of the squared distances of all samples to their respective cluster centers. represents the pre-set number of clusters, that is, how many sub-regions the study area is divided into. represents the sample vector, represents the sample set within the i-th cluster, represents the center (mean vector) of the i-th cluster, represents the Euclidean distance squared of the sample to the cluster center.

[0066] Euclidean distance, also known as Euclidean distance, in n-dimensional space, given two points and , the formula for calculating the Euclidean distance is:

[0067] (7)

[0068] Step 5 is as follows:

[0069] Quantify the spatial heterogeneity of environmental factors through the geographic detector to obtain multiple sub-regions. For each sub-region, generate a dynamic adjustment coefficient. Combine the recommended fertilizer amount of the variable fertilizer model with the spatial heterogeneity to achieve precise regulation and control in different regions. The entire region is divided into five levels through the K-means clustering algorithm, which are mapped to 0.8, 0.9, 1.0, 1.1, and 1.2 adjustment coefficients. The final fertilizer amount formula is as follows:

[0070] (8)

[0071] In the formula, is the final fertilizer amount, is the base fertilizer amount (kg / ha). is the adjustment coefficient, with a value range of 0.8-1.2 to avoid extreme adjustments.

[0072] The basis for clustering is the comprehensive score or eigenvalue of environmental factors. For example, areas with high soil fertility and adequate water may be classified into one level, while areas with poor soil and drought may be classified into another level.

[0073] Level 1 (worst conditions): Adjustment coefficient is 1.2. These areas have poor environmental conditions (such as poor soil and drought), and crops have high demand for fertilizer, so the amount of fertilizer needs to be increased.

[0074] Level 2 (poor conditions): Adjustment coefficient is 1.1. The environmental conditions in these areas are slightly worse, but the amount of fertilizer still needs to be increased appropriately.

[0075] Level 3 (moderate conditions): Adjustment coefficient is 1.0. The environmental conditions in these areas are moderate, and the base fertilizer amount (F_base) can be used directly.

[0076] Level 4 (better conditions): Adjustment coefficient is 0.9. The environmental conditions in these areas are good (such as fertile soil and adequate water), and the crop has high efficiency in using fertilizer, so the amount of fertilizer can be appropriately reduced.

[0077] Level 5 (best conditions): Adjustment coefficient is 0.8. The environmental conditions in these areas are very good, and the crop has great growth potential, so the amount of fertilizer can be further reduced to avoid waste and pollution.

[0078] In the present application, when the unmanned aerial vehicle data acquisition is difficult, satellite remote sensing image can be used for substitution.

[0079] In the present application, when there is no ground sensor data, the soil nutrient condition can be obtained through on-site soil sampling and testing, and the meteorological data can be obtained through the national meteorological center data website.

[0080] In the present application, K-means clustering can be replaced by DBSCAN (density-based clustering), hierarchical clustering (Hierarchical Clustering) or self-organizing map (Self-Organizing Maps, SOM), and the most suitable partition method is selected according to the specific distribution of the experimental field.

[0081] The present application also provides a variable fertilization device for multi-source data fusion, comprising the following modules:

[0082] The data acquisition and preprocessing module is used for collecting historical data of yield corresponding to fertilization amount, ground sampling data, ground sensor data and remote sensing image, and preprocessing the remote sensing image.

[0083] The spatial data processing and consistency correction module is used for unifying data scale, time and space alignment, processing ground sensor data to generate soil nutrient and meteorological data.

[0084] The variable fertilization model construction module is used for constructing fertilization amount and yield effect curve, constructing yield and vigor index VI distribution map, and constructing variable fertilization model representing the relationship between fertilization amount and vigor index VI according to the fertilization amount and yield effect curve and the yield and vigor index VI distribution map.

[0085] The spatial heterogeneity quantification and partition module is used for quantifying the spatial heterogeneity of fertilization demand of environmental factors by using geographic detector, and dividing the region into multiple sub-regions; wherein, the environmental factors include soil nutrient and meteorological data.

[0086] The differential fertilization strategy optimization module comprises generating dynamic adjustment coefficient for each sub-region, and determining the final fertilization amount combined with the fertilization amount recommended by the variable fertilization model.

[0087] The present application also provides an electronic device, comprising one or more processors; a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method.

[0088] The present application also provides a computer readable storage medium having executable instructions stored thereon, which are executed by a processor to make the processor implement the above-mentioned method.

Claims

1. A variable-rate fertilization method based on multi-source data fusion, characterized in that: The following steps are involved: Step 1: Data collection and preprocessing, including: collecting historical data on fertilizer application and yield, ground sampling data, data obtained by ground sensors, and remote sensing images, and preprocessing the remote sensing images; Step 2: Spatial data processing and consistency correction, including: unifying data scale, spatial and temporal alignment, and processing ground sensor data to generate soil nutrient and meteorological data; Step 3, constructing a variable fertilization model, including: constructing a fertilizer amount and yield effect curve, constructing a yield and growth index VI distribution map, and constructing a variable fertilization model that characterizes the relationship between fertilizer amount and growth index VI based on the fertilizer amount and yield effect curve and the yield and growth index VI distribution map; Step 4: Quantifying and zoning spatial heterogeneity, including: using geographic detectors to quantify the spatial heterogeneity of environmental factors on fertilization demand and dividing the region into multiple sub-regions; where environmental factors include soil nutrients and meteorological data; Step 5: Optimize the differentiated fertilization strategy, including: generating a dynamic adjustment coefficient for each sub-region, combining the fertilization amount recommended by the variable fertilization model, and determining the final fertilization amount.

2. The variable rate fertilization method based on multi-source data fusion according to claim 1, characterized in that: In step 1, the preprocessing work is as follows: first, geometric correction is performed to establish the mapping relationship between the remote sensing image and the geographic coordinates, and orthorectification is performed to eliminate terrain distortion; second, through radiation correction, the original DN value is converted into radiation brightness, and then atmospheric interference is eliminated to obtain surface reflectivity; then image position matching is performed to achieve sub-pixel level registration of multi-temporal images and resample them to a unified coordinate system; finally, denoising is performed to generate a cloud-free and noise-free remote sensing image.

3. The variable rate fertilization method based on multi-source data fusion according to claim 1, characterized in that: In step 2, the unified numerical scale includes: for outliers in ground sensor data, directly eliminating them through threshold filtering, filling missing segments through linear interpolation, or smoothing high-frequency noise through sliding window averaging; Processing ground sensor data to generate soil nutrient and meteorological data includes: using the Kriging interpolation method to predict the soil nutrient and meteorological data at the target location. The interpolation formula is: （1） in, represents the value of the point to be interpolated, is the weight coefficient of each observation point, is the actual measurement value of each observation point.

4. The variable rate fertilization method based on multi-source data fusion according to claim 1, characterized in that: In step 2, spatiotemporal alignment includes: remote sensing images and ground sensor data are matched by coordinates, and the time windows are aligned to the same crop growth period.

5. The variable rate fertilization method based on multi-source data fusion according to claim 1, characterized in that: Step 3 includes: According to the fertilization and yield data in the historical data, the relationship between yield and fertilization amount is obtained: （2） Among them, Y represents the annual yield and N represents the annual fertilizer application amount; Four growth period physicochemical parameters including aboveground fresh biomass, leaf area index, leaf chlorophyll relative content and leaf nitrogen content were collected, and weights were assigned to each parameter to construct physicochemical parameters for quantifying crop growth VI; Using remote sensing images to invert the physical and chemical parameters of each growth period, we constructed a crop growth distribution map. We obtained the spatiotemporal distribution of the physical and chemical parameters of aboveground fresh biomass, leaf area index, leaf chlorophyll relative content, and leaf nitrogen content in each growth period, constructed a comprehensive growth index (VI) distribution map, and established a yield-growth index (VI) relationship model: （3） By correlating equations (2) and (3) through yield, a variable fertilization model is established to characterize the relationship between growth index and fertilizer amount: （4）， The fertilizer application rate was calculated using formula (4).

6. The variable rate fertilization method based on multi-source data fusion according to claim 1, characterized in that: In step 4, ground sensor data includes meteorological and soil nutrient data. These data are used as environmental factors to construct a geographic detector. The q value is calculated to quantify the explanatory power of a certain environmental factor on the spatial heterogeneity of fertilization demand. The formula is as follows: （5） Where, is the number of strata or partitions for each factor; is the total number of global samples in the geographic detector, that is, the sum of all samples in the study area, is the number of sub-region samples, is the sub-region variance, is the global variance; According to the q value screening of each environmental factor, it is divided into multiple sub-areas. The formula is as follows: （6） Where, Represents the clustering objective function, which is the sum of the squares of the distances from all samples to the cluster center to which they belong; Indicates the pre-set number of clusters, that is, how many sub-regions are divided into. represents the sample vector, represents the sample set in the i-th cluster, represents the center of the i-th cluster, Representation sample The squared Euclidean distance to the cluster center.

7. The variable rate fertilization method based on multi-source data fusion according to claim 1, characterized in that: In step 5, the sub-areas are divided into five levels and mapped to five adjustment coefficients of 0.8, 0.9, 1.0, 1.1, and 1.2 respectively; the final fertilizer application formula is as follows: （8） Where, is the final fertilizer amount, As the basic fertilizer amount, is the adjustment factor.

8. A variable-rate fertilization device with multi-source data fusion, characterized in that: Includes the following modules: The data acquisition and preprocessing module is used to collect historical data on yield corresponding to fertilizer application amount, ground sampling data, data obtained by ground sensors, and remote sensing images, and preprocess the remote sensing images; Spatial data processing and consistency correction module, used to unify data scale, align time and space, and process ground sensor data to generate soil nutrient and meteorological data; The variable fertilization model construction module is used to: construct a fertilizer amount and yield effect curve, construct a yield and growth index VI distribution map, and construct a variable fertilization model that characterizes the relationship between fertilizer amount and growth index VI based on the fertilizer amount and yield effect curve and the yield and growth index VI distribution map; The spatial heterogeneity quantification and partitioning module is used to quantify the spatial heterogeneity of fertilization demand due to environmental factors using geographic detectors and divide the region into multiple sub-regions. Environmental factors include soil nutrients and meteorological data. The differentiated fertilization strategy optimization module includes: generating a dynamic adjustment coefficient for each sub-area, combining the fertilization amount recommended by the variable fertilization model, and determining the final fertilization amount.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Executable instructions are stored thereon, and when the instructions are executed by a processor, the processor implements the method according to any one of claims 1 to 7.

Images