An unmanned aerial vehicle image-based water and fertilizer intelligent control method and system

Abstract

The application is suitable for the field of precision agriculture, and provides a water and fertilizer intelligent control method and system based on unmanned aerial vehicle images.The method comprises the following steps: acquiring unmanned aerial vehicle multispectral image data of a target farmland, and pre-processing the image data; extracting vegetation index features of the pre-processed image, and combining a preset farmland grid division to construct a crop water and fertilizer stress distribution map; inputting the crop water and fertilizer stress distribution map into a preset water and fertilizer decision model, combining multi-source environmental parameters for fusion calculation, and obtaining target water and fertilizer application amounts of each grid region of the target farmland; generating a variable operation prescription map according to the target water and fertilizer application amounts, and executing precision application operation.The application introduces a particle swarm optimization algorithm to adaptively optimize vegetation nutrition, soil moisture and meteorological compensation weights, takes the maximum water and fertilizer utilization efficiency of a single grid region as an objective function, and outputs an optimal weight combination of a current growth stage through iterative optimization.

Description

Technical Field

[0001] This invention belongs to the field of precision agriculture, and in particular relates to a method and system for intelligent water and fertilizer control based on drone images. Background Technology

[0002] In precision agriculture, water and fertilizer management plays a decisive role in crop yield and quality. Traditional management methods mainly rely on manual experience or uniform application in fixed plots, failing to fully consider the significant differences in soil properties and crop growth in spatial distribution, resulting in low fertilizer utilization efficiency, serious resource waste, and prominent environmental pollution problems.

[0003] With the development of remote sensing technology and intelligent control theory, crop growth monitoring and variable fertilization technology based on UAV multispectral images has gradually attracted attention. However, existing UAV remote sensing water and fertilizer management methods still have significant shortcomings. The fusion architecture of multi-source parameters is relatively simple, usually using only single spectral indicators such as the normalized vegetation index to characterize crop nutrient status and roughly adjusting fertilizer application based on this, without incorporating key physical parameters such as actual soil moisture and real-time meteorological evapotranspiration into a unified compensation calculation framework. Crop fertilizer requirements are closely related to soil available water content and transpiration intensity. Under water-scarce conditions, root absorption capacity is limited, and simply increasing fertilizer application may exacerbate stress. Existing systems lack a decision model that normalizes and coordinates the spectral response of vegetation, measured soil moisture values, and meteorological driving factors.

[0004] Meanwhile, the weighting coefficients in the decision-making model are mostly fixed empirical values, which cannot be dynamically adjusted according to different growth stages and climatic conditions. This leads to the amount of fertilizer applied deviating from actual needs under specific environments, affecting water and fertilizer utilization efficiency.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] The purpose of this invention is to provide a water and fertilizer intelligent control method based on UAV images, aiming to solve the above-mentioned problems.

[0007] This invention is implemented as follows: it provides a smart water and fertilizer control method based on UAV imagery, comprising the following steps:

[0008] S1. Acquire multispectral image data of the target farmland using a drone, and preprocess the image data;

[0009] S2. Extract vegetation index features from the preprocessed image and construct a crop water and fertilizer stress distribution map by combining it with the preset farmland grid division.

[0010] S3. Input the crop water and fertilizer stress distribution map into the preset water and fertilizer decision model, and perform fusion calculations in combination with multi-source environmental parameters to obtain the target water and fertilizer application amount for each grid area of ​​the target farmland.

[0011] S4. Generate a variable operation prescription map based on the target water and fertilizer application rate, and control the integrated water and fertilizer equipment to perform precise application operations according to the prescription map.

[0012] As a further aspect of the present invention, in step S1, the preprocessing includes: sequentially performing radiometric calibration, atmospheric correction, geometric correction, and image stitching on the UAV multispectral image to obtain an orthophoto map with geographic coordinate information.

[0013] As a further aspect of the present invention, in step S2, the vegetation index features include at least one or more of the following: normalized vegetation index, normalized red edge index, and water stress index.

[0014] The construction of the crop water and fertilizer stress distribution map includes: mapping the acquired vegetation index features to the farmland grid, and calculating the mean value of the vegetation index features in each grid as the water and fertilizer stress characterization value of that grid.

[0015] As a further aspect of the present invention, in step S3, the preset water and fertilizer decision model is constructed based on a normalized compensation architecture of multi-dimensional parameters, and the target water and fertilizer application rate for the i-th grid area of ​​the target farmland is calculated in the following way:

[0016] The target water and fertilizer application rate is equal to the recommended basic water and fertilizer application rate for the current crop growth stage multiplied by the normalized total value. The normalized total value is 1 plus the product of vegetation nutrient weight and the first normalized value, the product of soil moisture weight and the second normalized value, and the product of meteorological compensation weight and the third normalized value. The first normalized value is the difference between the upper limit of the vegetation nutrient index and the normalized vegetation index value of the grid divided by the difference between the upper and lower limits of the vegetation nutrient index. The second normalized value is the difference between field capacity and the measured soil moisture value of the grid divided by the difference between field capacity and the water content at the wilting point. The third normalized value is the difference between the real-time meteorological evapotranspiration of the grid and the minimum evapotranspiration of the historical statistical period divided by the difference between the maximum and minimum evapotranspiration of the historical statistical period.

[0017] Among them, the sum of vegetation nutrient weight, soil moisture weight and meteorological compensation weight is 1. The calculation process uses range standardization to uniformly map heterogeneous parameters to the interval of 0 to 1.

[0018] As a further aspect of the present invention, the upper and lower limits of the vegetation nutrient index are determined by statistically analyzing the normalized vegetation index quantiles of historical crop multispectral data for the same period; the field water holding capacity and wilting point water content are determined by referring to a table based on the soil texture of the target farmland; the maximum and minimum evapotranspiration of the historical statistical period are determined by sorting the evapotranspiration of historical meteorological data; the real-time meteorological evapotranspiration is calculated using the Penman-Montes formula; and the measured soil moisture value is obtained by measuring a multi-layer moisture sensor buried in the soil.

[0019] As a further aspect of the present invention, the vegetation nutrient compensation weight, soil moisture compensation weight, and meteorological compensation weight are determined through an adaptive dynamic adjustment architecture. This architecture employs a particle swarm optimization algorithm, with the objective function being to maximize the water and fertilizer utilization efficiency of a single grid area. The optimal weight combination for the current growth stage is output through iterative optimization. The water and fertilizer utilization efficiency is the expected economic yield predicted based on historical yield data and the current water and fertilizer application rate, divided by the actual water and fertilizer application rate of the grid area.

[0020] As a further aspect of the present invention, in step S4, generating the variable operation prescription map includes: converting the target water and fertilizer application rate of each grid into the electromagnetic valve opening degree and variable frequency pump frequency control command sequence of the integrated water and fertilizer equipment, and binding it with the grid geographic coordinates to generate a GIS-based prescription map.

[0021] The method also includes closed-loop feedback: during the execution process, the flow rate and pressure parameters of the equipment are collected in real time and compared with the theoretical values ​​for feedback adjustment; after the operation is completed, the drone image is acquired again to calculate the gradient of vegetation index change before and after fertilization, and if the gradient is lower than the preset threshold, an abnormal warning is triggered.

[0022] The present invention also provides a water and fertilizer intelligent control system based on UAV imagery, the system being used to implement the aforementioned water and fertilizer intelligent control method based on UAV imagery, the system comprising:

[0023] The data acquisition module is used to acquire UAV multispectral image data of the target farmland and preprocess the image data;

[0024] The stress analysis module is used to extract vegetation index features from the preprocessed image and, in conjunction with the preset farmland grid division, construct a crop water and fertilizer stress distribution map.

[0025] The decision calculation module is used to input the crop water and fertilizer stress distribution map into the preset water and fertilizer decision model, and integrate multi-source environmental parameters for normalization calculation to obtain the target water and fertilizer application amount for each grid area of ​​the target farmland.

[0026] The execution control module is used to generate a variable operation prescription map based on the target water and fertilizer application rate, and control the integrated water and fertilizer equipment to perform precise application operations according to the prescription map.

[0027] As a further aspect of the present invention, the preprocessing in the data acquisition module includes: sequentially performing radiometric calibration, atmospheric correction, geometric correction, and image stitching on the UAV multispectral images to obtain an orthophoto map with geographic coordinate information.

[0028] As a further aspect of the present invention, in the working content of the stress analysis module, the vegetation index features include at least one or more of the following: normalized vegetation index, normalized red edge index, and water stress index.

[0029] The construction of the crop water and fertilizer stress distribution map includes: mapping the acquired vegetation index features to the farmland grid, and calculating the mean value of the vegetation index features in each grid as the water and fertilizer stress characterization value of that grid.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0031] This invention introduces a particle swarm optimization algorithm to adaptively optimize the weights of vegetation nutrition, soil moisture, and meteorological compensation. The objective function is to maximize the water and fertilizer use efficiency of a single grid region. Through iterative optimization, the optimal weight combination for the current growth stage is output. Compared to fixed empirical weights, this dynamically adjusted architecture can automatically adjust the contribution of each factor according to different crop growth stages and climatic conditions, ensuring that the decision-making model maintains high applicability and robustness in various environmental scenarios.

[0032] This invention collects equipment flow rate and pressure parameters in real time during precise application operations for feedback adjustment, ensuring that the actual application amount matches the target instruction. After the operation is completed, it acquires drone images again to calculate the gradient of vegetation index changes before and after fertilization. If the gradient is lower than a preset threshold, an anomaly warning is triggered. This closed-loop architecture can not only promptly detect equipment execution deviations and environmental changes, but also quantitatively evaluate the application effect, providing data support for subsequent decision optimization and significantly improving the reliability and management efficiency of the system.

[0033] This invention divides farmland into grids, using the average vegetation index of each grid as a characterization value for water and fertilizer stress. Based on the target application rate, a GIS-based variable application prescription map with geographic coordinates is generated, controlling the integrated water and fertilizer system to perform differentiated and precise application operations. This spatial differentiation strategy effectively solves the problem of traditional uniform fertilization methods neglecting the spatial heterogeneity within farmland. While ensuring crop growth needs, it significantly reduces excessive input of chemical fertilizers and irrigation water, which helps reduce the risk of agricultural non-point source pollution and promotes cost reduction, efficiency improvement, and sustainable development in agricultural production. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention.

[0035] Figure 1 This is a flowchart of a water and fertilizer intelligent control method based on UAV images.

[0036] Figure 2 This is a block diagram of the composition structure of a water and fertilizer intelligent control system based on UAV images. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0038] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0039] Figure 1 The flowchart illustrates a method for intelligent water and fertilizer control based on UAV imagery, and one embodiment of the present invention provides a method for intelligent water and fertilizer control based on UAV imagery, comprising the following steps:

[0040] S1. Acquire multispectral image data of the target farmland using a drone and preprocess the image data. The multispectral image data refers to farmland surface image information acquired by a drone platform equipped with a multispectral sensor. This data typically contains spectral reflectance information for multiple specific bands, including visible light (such as red, green, and blue) and near-infrared and red-edge bands, reflecting crop growth, nutrient levels, and health status. Preprocessing involves a series of correction and optimization operations on the raw multispectral image data. Its purpose is to eliminate radiation errors, atmospheric effects, geometric distortions, etc., generated during image acquisition, ensuring the accuracy and usability of the image data and providing high-quality input for subsequent feature extraction and analysis.

[0041] S2. Extract vegetation index features from the preprocessed image and combine them with a pre-defined farmland grid to construct a crop water and fertilizer stress distribution map. The vegetation index features are one or more indicators calculated using the reflectance of different bands in multispectral image data. These indicators quantify the biophysical characteristics of vegetation, such as chlorophyll content, biomass, and water content, thus indirectly reflecting the crop's growth potential and stress status. Farmland grid division involves dividing the target farmland area into multiple regular or irregular cells of the same or different sizes according to pre-defined rules. This division method helps to discretize continuous farmland areas, facilitating independent data analysis, decision calculation, and precise application management for each cell. The crop water and fertilizer stress distribution map is a spatial distribution map constructed based on vegetation index features and farmland grid division. This map visually displays the degree of water and nutrient stress suffered by crops in different areas of the farmland, providing a spatial basis for subsequent water and fertilizer decisions.

[0042] S3. Input the crop water and fertilizer stress distribution map into the preset water and fertilizer decision model, and perform fusion calculations in combination with multi-source environmental parameters to obtain the target water and fertilizer application amount for each grid area of ​​the target farmland.

[0043] The water and fertilizer decision model is a mathematical model or algorithm used to calculate the amount of water and fertilizer applied to crops. This model comprehensively considers multiple influencing factors, such as crop growth status, soil conditions, and environmental climate, and outputs the required amounts of water and fertilizer for each region through specific computational logic. Multi-source environmental parameters refer to various data related to the crop growth environment obtained from other sources besides UAV image data. These parameters can include soil moisture content, soil nutrient content, meteorological data (such as temperature, humidity, rainfall, and evapotranspiration), crop type, and growth stage, used to supplement and improve the input information of the water and fertilizer decision model.

[0044] The target water and fertilizer application rate refers to the precise amount of water and fertilizer recommended for a specific grid area or individual crop within a farmland, calculated based on a water and fertilizer decision-making model. This application rate aims to meet the actual needs of the crop in that area, avoiding over- or under-application.

[0045] S4. Generate a variable application prescription map based on the target water and fertilizer application rate, and control the integrated water and fertilizer equipment to perform precise application operations according to the prescription map. The variable application prescription map is a spatial distribution map or instruction sequence generated based on the target water and fertilizer application rate. This prescription map binds the target application rate for different areas with corresponding geographical location information, guiding the integrated water and fertilizer equipment to perform differentiated and precise water and fertilizer application operations in the farmland. Integrated water and fertilizer equipment is an agricultural machine that integrates irrigation and fertilization functions. This equipment can dissolve and deliver fertilizer to the crop root zone while irrigating, according to preset instructions or prescription maps, achieving simultaneous and precise water and fertilizer management. Precise application operation refers to the integrated water and fertilizer equipment applying different amounts of water and fertilizer to different areas of the farmland according to the variable application prescription map. This operation method aims to improve water and fertilizer utilization efficiency, reduce resource waste, and promote healthy crop growth.

[0046] The method provided in this embodiment constructs a complete intelligent control chain from data acquisition to execution by introducing UAV remote sensing data, multi-source parameter fusion decision-making, and precise application of variables. It effectively solves the problems of uniform application and single decision-making basis in traditional water and fertilizer management, and provides technical support for realizing the refinement and sustainable development of agricultural production.

[0047] In a preferred embodiment of the present invention, step S1 includes preprocessing the UAV multispectral image sequentially by performing radiometric calibration, atmospheric correction, geometric correction, and image stitching to obtain an orthophoto map with geographic coordinate information.

[0048] In this embodiment, radiometric calibration converts the digital quantization values ​​recorded by the sensor into physically meaningful radiance or reflectance values. Its purpose is to eliminate differences in sensor response and the influence of lighting conditions, ensuring that the image data reflects the true radiometric characteristics of the ground features. This process can be based on laboratory calibration parameters and field reference board data for radiometric correction, or it can employ relative radiometric correction methods, using stable ground features within the image or statistical relationships between multiple temporal images for correction.

[0049] Atmospheric correction aims to eliminate the influence of the atmosphere on remote sensing images, such as atmospheric scattering, absorption, and reflection, to obtain the true surface reflectance of ground features. This is crucial for comparing images from different times and locations and for the accurate calculation of vegetation indices. This process can employ methods based on radiative transfer models combined with atmospheric parameters, or it can use empirical or semi-empirical methods such as dark target subtraction and histogram matching.

[0050] Geometric correction is used to correct geometric distortions in images caused by factors such as sensor attitude, terrain undulation, and Earth curvature, ensuring that the positions of ground features in the image correspond to their actual geographic coordinates. This is fundamental to ensuring accurate matching between the image and the farmland grid. This process can be performed using orthorectification based on ground control points and a high-precision digital elevation model, or by using external sensor orientation parameters and inertial measurement unit data for direct geolocation and correction.

[0051] Image stitching is the seamless combination of multiple adjacent drone images into a single, large image covering the entire target area, which is essential for generating complete orthophoto maps of farmland. This process can be automated using feature point matching and image fusion techniques, or it can be achieved through image registration and brightness equalization to ensure color and texture consistency in the stitched areas.

[0052] Finally, through the above series of processes, an orthophoto map with geographic coordinate information is obtained. This map is a high-precision image with true geographic location information and eliminated geometric distortion after all correction steps, which is the basis for subsequent accurate analysis and decision-making.

[0053] In a preferred embodiment of the present invention, in step S2, the vegetation index features include at least one or more of the normalized vegetation index, the normalized red edge index, and the water stress index.

[0054] The construction of the crop water and fertilizer stress distribution map includes: mapping the acquired vegetation index features to the farmland grid, and calculating the mean value of the vegetation index features in each grid as the water and fertilizer stress characterization value of that grid.

[0055] In this embodiment, vegetation index features are numerical values ​​derived from the differences in reflectance characteristics of plants across different spectral bands, combined using specific algorithms. These values ​​are used to quantify vegetation growth status and health. The Normalized Difference Vegetation Index (NDVI) is widely used to monitor vegetation cover, biomass, and photosynthetic capacity, reflecting crop nutritional status. The Normalized Red Edge Index (NDI) is more sensitive to changes in crop chlorophyll content, especially in the later stages of crop growth, reflecting nitrogen nutrition status earlier and more accurately. The Water Stress Index is used to assess crop water content and the degree of water stress. These indices can be obtained by performing band calculations on multispectral image data from UAVs. Furthermore, machine learning algorithms can be used to train models to directly extract features related to these indices from multispectral data, representing vegetation status in a more intelligent way.

[0056] Mapping the acquired vegetation index features to the farmland grid aims to associate continuous image data with discrete farmland management units (grids), providing a foundation for refined management. This can be achieved through spatial overlay analysis of a vegetation index layer with geographic coordinate information and a farmland grid layer using geographic information system software. Alternatively, a program can be written using an image processing library to accurately extract the vegetation index values ​​of all pixels falling within each grid based on its geographic boundaries.

[0057] The goal of calculating the mean vegetation index feature within each grid is to aggregate vegetation index information within the grid, eliminate local noise, and obtain a comprehensive index representing the entire grid for easier subsequent decision-making. For each farmland grid, the vegetation index values ​​of all pixels within it can be statistically analyzed, and then the arithmetic mean of these values ​​can be calculated. Furthermore, to reduce the influence of outliers, the median or a weighted average (e.g., assigning different weights based on the pixel's distance from the grid center) can be used as the representative value for that grid.

[0058] The calculated mean value is used as the water and fertilizer stress characterization value for the grid. The aim is to directly use the quantified vegetation index mean as an indicator of the degree of water and fertilizer stress in the region, simplifying the transformation process from raw data to decision-making basis.

[0059] The mean vegetation index can be directly used as a quantitative indicator of water and fertilizer stress. For example, a low mean normalized vegetation index may directly indicate nutrient deficiency. Alternatively, these mean values ​​can be converted into specific stress levels (such as mild stress, moderate stress, and severe stress) based on preset thresholds or grading standards for easier intuitive judgment and decision-making.

[0060] The following is a concrete example. For instance, a drone equipped with a multispectral camera takes aerial photographs of a target farmland, acquiring image data in the red, red-edge, near-infrared, and short-wave infrared bands. After preprocessing including radiometric calibration, atmospheric correction, geometric correction, and image stitching, an orthophoto map with geographic coordinate information is generated. Subsequently, the system calculates the Normalized Difference Vegetation Index (NDVI), Normalized Red Edge Index (NBE), and Water Stress Index (MSI) for each pixel based on these band data. Assume the farmland is divided into 10m x 10m grids. For each 10m x 10m grid, the system extracts the NVI, NBE, and MSI values ​​for all pixels within that grid and calculates their arithmetic mean. For example, the mean NVI for a given grid might be 0.65, the mean NBE might be 0.40, and the mean MSI might be 0.80. These mean values ​​are aggregated as the current water and fertilizer stress characteristics of the grid area and stored in the geographic information system to form a detailed crop water and fertilizer stress distribution map.

[0061] In a preferred embodiment of the present invention, in step S3, the preset water and fertilizer decision model is constructed based on a normalized compensation architecture of multi-dimensional parameters, and the target farmland is calculated using the following formula. Target water and fertilizer application rate for each grid area :

[0062]

[0063] in The recommended application rate of basic water and fertilizer for the current crop growth stage. For the first Normalized vegetation index values ​​for each grid region , These are the upper and lower threshold values ​​for the crop's nutrient index, respectively. For the first The measured values ​​of soil moisture sensors corresponding to each grid area. , These are field water holding capacity and wilting point water content, respectively. This represents real-time meteorological evapotranspiration. , These are the maximum and minimum evapotranspiration rates for the historical statistical period, respectively. , , The weights are respectively for vegetation nutrition, soil moisture, and meteorological compensation. The formula maps heterogeneous parameters to the [0,1] interval through range standardization.

[0064] In this embodiment, the preset water and fertilizer decision-making model is constructed based on a normalized compensation architecture of multi-dimensional parameters. It aims to integrate various parameters affecting crop water and fertilizer requirements, eliminate dimensional differences through normalization, and then finely adjust the basic application rate through the compensation architecture. This architecture can employ an expert system combined with fuzzy logic to perform compensation adjustments based on the normalized values ​​of different parameters and preset rules; alternatively, it can be based on a machine learning model, such as a neural network or support vector machine, to learn the complex nonlinear relationship between multi-dimensional parameters and water and fertilizer requirements through training, and output a compensation factor.

[0065] The formula described is a specific mathematical expression for implementing the normalized compensation architecture, used to quantitatively calculate the precise water and fertilizer application rate for each grid area. It obtains the final application rate by multiplying the base application rate by a compensation factor, which is a weighted combination of parameters such as normalized vegetation nutrition, soil moisture, and meteorological evapotranspiration. This formula can be directly implemented in the software program of the decision-making calculation module, receiving various parameter inputs and performing calculations according to mathematical logic; alternatively, it can be used for high-speed parallel computation through an algorithm module integrated in dedicated hardware to meet the needs of real-time decision-making.

[0066] The The basic recommended water and fertilizer application rate for the current crop growth stage represents the standard amount of water and fertilizer required by a specific crop at a specific growth stage under ideal or average conditions. This basic application rate can be determined based on the experience of agricultural experts, historical planting data, crop growth models, or the recommended standards of local agricultural extension departments; alternatively, it can be determined through field trials by measuring the water and fertilizer response curves of crops at different growth stages, thereby deriving the basic recommended rates for different stages.

[0067] The For the first The normalized vegetation index (NDI) of a grid area is an important indicator reflecting crop growth status, chlorophyll content, and photosynthetic capacity, and is often used to assess crop nutritional status. This value can be calculated from multispectral image data from UAVs using reflectance in the red and near-infrared bands; alternatively, it can be obtained by sampling and measuring within the grid area using a ground-based handheld spectrometer, followed by averaging or interpolation.

[0068] The , These are the upper and lower thresholds for the crop's nutrient index, defining the normal range of crop nutrient status and used for normalization. They can be determined based on the normalized vegetation index quantiles of historical multispectral crop data from the same period; alternatively, they can be set by consulting crop nutrient diagnosis manuals, expert experience, or conducting nutrient stress experiments.

[0069] The For the first The measured values ​​of soil moisture sensors corresponding to each grid area directly reflect the moisture content in the soil and are a key parameter for assessing crop water stress. This value can be obtained in real time by multi-layer moisture sensors (such as TDR and FDR sensors) buried in the soil; alternatively, it can be measured manually at regular intervals using soil moisture probes, or estimated in conjunction with soil moisture models.

[0070] The , These are field capacity and wilting point water content, respectively. These two parameters define the upper and lower limits of available soil moisture and are important references for soil moisture management. They can be determined by referring to tables based on the soil texture of the target farmland; alternatively, they can be accurately measured by determining soil moisture characteristic curves in the laboratory or by field experiments (such as the saturation infiltration method or pressure membrane method).

[0071] The Real-time meteorological evapotranspiration refers to the sum of water evaporation from the crop canopy and soil surface, and plant transpiration, reflecting the crop's water requirements. This value can be calculated using the Penman-Montes formula, combined with data such as temperature, humidity, wind speed, and solar radiation provided by weather stations; alternatively, it can be measured directly using a lysimeter, or estimated using remote sensing data (such as surface temperature and vegetation index) combined with an energy balance model.

[0072] The , These are the maximum and minimum evapotranspiration rates for the historical statistical period, respectively. These thresholds are used to normalize real-time evapotranspiration, reflecting water demand under historical extreme conditions. They can be determined based on evapotranspiration ranking from historical meteorological data; alternatively, they can be extracted from long-term meteorological observation station data or obtained through climate model simulation.

[0073] The , , These are the compensation weights for vegetation nutrition, soil moisture, and weather conditions. These weights reflect the importance or sensitivity of different parameters to the decision on water and fertilizer application rates, and their sum of 1 ensures the rationality of the compensation framework. They can be initially set through expert experience, field trial data analysis, or sensitivity analysis; alternatively, they can be determined through iterative optimization using optimization algorithms (such as particle swarm optimization), genetic algorithms, or reinforcement learning, with objective functions such as crop yield and water and fertilizer use efficiency.

[0074] The formula described above maps heterogeneous parameters to the [0,1] interval using range standardization. Range standardization is a commonly used data preprocessing method that scales parameters with different dimensions and value ranges to the [0,1] interval, thereby eliminating the influence of dimensions and making the parameters comparable within the compensation framework. This prevents certain parameters from dominating the decision-making process due to their large values. This standardization process can be performed in a software program using the formula for each parameter X. Perform calculations. It can be , , , It can be , , Alternatively, this can be achieved during the data preprocessing stage through a dedicated data processing module or function library, ensuring that all parameters input into the decision model are standardized.

[0075] The following is a concrete example. Assume the target farmland is planted with corn and is in the jointing stage. In a certain grid region i, the normalized vegetation index value obtained through UAV image analysis... Slightly below the normal upper limit for this growth stage This indicates a slight nutrient deficiency in the crop. Meanwhile, the measured values ​​from the soil moisture sensor... This indicates that the soil moisture content is below field capacity. However, the moisture content is higher than the wilting point. This indicates that the crop is facing some water stress. In addition, real-time meteorological evapotranspiration... The level is high, exceeding the average for historical statistical periods, indicating an increased water requirement for crops. At this point, the water and fertilizer decision-making model will first determine the recommended basic water and fertilizer application rates for this growth stage. Next, the model will , and Each is compared to its respective upper and lower thresholds, and three compensation terms are calculated through range standardization. For example, due to Low, the first item ( - ) / ( - The result will be a positive value, indicating that fertilizer needs to be added. Similarly, because... Low and If the value is too high, the second and third terms will also be positive, indicating a need to increase the moisture content. These normalized compensation terms will be multiplied by their respective weights. , , Then sum them up to form a total compensation factor. Finally, Multiplying by this compensation factor yields the target water and fertilizer application rate for that grid area. This application rate will be higher than the baseline recommendation to compensate for the nutrient and water stress faced by the crop.

[0076] Through the aforementioned technical solution, the water and fertilizer decision-making model can fully integrate multi-dimensional parameters and employ a normalized compensation architecture for refined calculations, thereby overcoming the shortcomings of traditional water and fertilizer decision-making models in terms of adaptability and accuracy. This allows the calculated target water and fertilizer application rates to more accurately reflect the actual needs of crops in different grid areas, effectively avoiding excessive or insufficient application of water and fertilizer. Ultimately, this solution significantly improves water and fertilizer use efficiency, promotes healthy crop growth, and provides strong technical support for achieving water and fertilizer conservation and increased yield and efficiency in agricultural production.

[0077] As a preferred embodiment of the present invention, the , The vegetation index quantiles were determined by statistically analyzing historical multispectral data of crops from the same period; , The soil texture of the target farmland was determined by referring to a table; , The evapotranspiration ranking is determined based on historical meteorological data; The calculation is performed using the Penman-Montes formula. The moisture was measured by a multi-layer moisture sensor buried in the soil.

[0078] In this embodiment, the Normalized Difference Vegetation Index (NDVI) is an important indicator reflecting crop growth status and nutrient levels. The distribution characteristics of NDVI can be obtained by statistically analyzing multispectral data of crops at different growth stages during the same historical period. The quantile method can objectively determine the upper and lower thresholds of NDVI; for example, the 5th percentile of historical data can be used as the threshold. 95th percentile as This reflects the range of nutrient fluctuations in crops during their normal growth cycle. Furthermore, these thresholds can also be set through expert experience or predicted using crop growth models.

[0079] Soil texture is a key factor affecting soil water-holding capacity. By sampling and analyzing the soil in the target farmland to determine its soil texture type (such as sandy soil, loam, clay, etc.), and then consulting the generally accepted soil hydraulic parameter tables in agricultural science, the field water holding capacity for the corresponding soil texture type can be obtained. and moisture content at the wilting point These parameters can also be measured directly in the field using saturated-drainage experiments, or estimated using soil moisture characteristic curve models.

[0080] Evapotranspiration is an important indicator of crop water consumption. By collecting historical meteorological data of the target farmland area, including temperature, humidity, wind speed, and sunshine duration, a historical evapotranspiration series can be calculated. These historical evapotranspiration data can then be sorted and analyzed; for example, selecting the historical highest value as... The historical lowest value is used as Alternatively, a specific percentile can be selected as a threshold to reflect the extreme evapotranspiration conditions in the region under different meteorological conditions. Furthermore, predictions can be made based on regional climate models or long-term meteorological forecast data.

[0081] The Penman-Montes formula is an internationally recognized and relatively accurate model for calculating crop evapotranspiration. This formula comprehensively considers multiple factors such as radiation, aerodynamics, and crop physiology, and can calculate the actual evapotranspiration of crops under current conditions based on real-time meteorological parameters (such as air temperature, humidity, wind speed, and solar radiation). Besides the Penman-Montes formula, simplified evapotranspiration models, such as the Hargreaves formula or empirical formulas based on temperature and sunshine duration, can also be used for estimation.

[0082] Multilayer moisture sensors embedded in the soil can monitor soil moisture content at different depths in real time and continuously. By placing multiple sensors in the main distribution area of ​​crop roots, the moisture distribution of the soil profile can be obtained, thus more accurately reflecting the soil moisture status available to crops. These sensors can be capacitive sensors, time-domain reflectometer sensors, or frequency-domain reflectometer sensors, which calculate soil moisture content by measuring the soil's dielectric constant or resistivity.

[0083] As a preferred embodiment of the present invention, the weight , , The adaptive dynamic adjustment architecture was determined, which employs a particle swarm optimization algorithm to measure the water and fertilizer use efficiency of a single grid region. Maximize the objective function, where This is the projected economic output based on historical yield data and current water and fertilizer application rates. The actual water and fertilizer application amount for this grid area is used to output the optimal weight combination for the current growth stage through iterative optimization.

[0084] In this embodiment, the adaptive dynamic adjustment architecture refers to a system that can automatically adjust its parameters or behavior according to changes in the external environment or the internal state of the system. Its "adaptive" aspect is reflected in its ability to learn and adjust based on real-time data and feedback information, while "dynamic" indicates that this adjustment is continuous rather than static. Implementation methods can include rule-based expert systems, fuzzy logic control, or various optimization algorithms. Particle Swarm Optimization (PSO) is a swarm intelligence optimization algorithm that simulates the foraging behavior of bird flocks. It finds the optimal solution through cooperation and competition among individuals within the group. During the algorithm's operation, each "particle" represents a potential solution and updates its velocity and position based on its own best position (individual optimal) and the best position found by the entire group (global optimal), thus iteratively searching the solution space. Besides the standard PSO algorithm, variants such as inertial weight PSO, shrinkage factor PSO, or discrete PSO can also be used. Water and fertilizer use efficiency is a key indicator for measuring the input-output ratio of water and fertilizer, defined as the ratio of expected economic output to actual water and fertilizer application. The higher this indicator, the more fully the water and fertilizer resources are absorbed and utilized by the crop, and the less waste of the input resources. Expected economic output refers to the crop yield predicted by a model under given water and fertilizer application rates and other environmental conditions. This prediction can be based on historical crop yield data, meteorological data, soil data, and water and fertilizer application records, achieved by constructing statistical regression models, machine learning models (such as support vector machines, neural networks, random forests, etc.), or crop growth simulation models. Actual water and fertilizer application rate refers to the total amount of water and fertilizer actually applied to crops within a specific grid area, calculated according to a water and fertilizer decision model. Iterative optimization refers to the process of gradually approaching or finding the optimal solution to a problem by repeatedly executing a series of computational steps. In each iteration, the algorithm adjusts its search direction based on the evaluation result of the current solution until a preset convergence condition is met (e.g., reaching the maximum number of iterations, the change in the objective function value being less than a threshold, etc.). Optimal weight combination refers to the combination of weights under specific constraints (e.g., ...). Under the condition that the objective function (i.e., water and fertilizer utilization efficiency) is maximized, the set of weight parameters is defined.

[0085] In a preferred embodiment of the present invention, step S4, generating a variable operation prescription map includes: converting the target water and fertilizer application rate of each grid into the electromagnetic valve opening degree and variable frequency pump frequency control command sequence of the integrated water and fertilizer equipment, and binding it with the grid geographic coordinates to generate a GIS-based prescription map.

[0086] The method also includes closed-loop feedback: during the execution process, the flow rate and pressure parameters of the equipment are collected in real time and compared with the theoretical values ​​for feedback adjustment; after the operation is completed, the drone image is acquired again to calculate the gradient of vegetation index change before and after fertilization, and if the gradient is lower than the preset threshold, an abnormal warning is triggered.

[0087] In this embodiment, the target water and fertilizer application rate for each grid is converted into a sequence of control commands for the solenoid valve opening and variable frequency pump frequency of the integrated water and fertilizer system. This aims to transform the abstract application rate into specific physical control parameters that the equipment can recognize and execute, which is a key step in achieving precise variable application. This conversion can be achieved through pre-established equipment performance curves or lookup tables, which describe the correspondence between different solenoid valve openings or pump frequencies and the actual water and fertilizer flow rates. Alternatively, the required solenoid valve opening and variable frequency pump frequency can be dynamically calculated based on the target application rate through real-time calibration or model-based control algorithms.

[0088] Binding control command sequences to grid geographic coordinates to generate GIS-based prescription maps links control commands with geographic location information, forming a spatialized operational command map. This forms the foundation for achieving precise, differentiated application of control commands across regions. GIS (Geographic Information System) provides a standardized way to store, manage, and display this spatial data. Specifically, the control commands for each grid can be stored as attribute data in a GIS layer and associated with the grid's geometric features (such as polygons); or a raster data containing georeferenced information can be generated, where the value of each pixel or grid cell represents the corresponding control command.

[0089] A closed-loop feedback architecture involves real-time acquisition of equipment flow rate and pressure parameters during execution, comparing them with theoretical values ​​for feedback adjustment. This aims to monitor equipment operating status in real time, ensuring that actual application matches theoretical commands, promptly correcting deviations, and improving application accuracy. This can be achieved by acquiring data in real time through flow and pressure sensors installed on the integrated water and fertilizer equipment, transmitting this data to the controller. The controller adjusts the solenoid valve opening or pump frequency according to a preset PID (proportional-integral-derivative) algorithm or other control strategies. Alternatively, adaptive control algorithms based on fuzzy logic or neural networks can be used to dynamically adjust control commands based on real-time acquired flow rate and pressure data to adapt to equipment wear or environmental changes.

[0090] After the operation is completed, drone images are acquired again to calculate the gradient of vegetation index changes before and after fertilization. If the gradient is lower than a preset threshold, an anomaly warning is triggered. This post-evaluation architecture is used to verify the effectiveness of water and fertilizer application. By comparing changes in crop growth, it determines whether the application is effective and identifies potential problems in a timely manner. Specifically, this can be achieved by registering and calculating vegetation indices on multispectral drone images before and after the operation, then calculating the difference or rate of change of vegetation indices pixel by pixel or grid by grid, and comparing it with a preset threshold. Alternatively, a machine learning model can be used, inputting data such as vegetation indices and application rates before and after fertilization, to predict crop responses, and comparing the predicted results with the actual observed changes in vegetation indices. If the difference is too large, an warning is triggered.

[0091] Figure 2 This is a block diagram of the composition structure of a water and fertilizer intelligent control system based on UAV imagery. In one example of the invention's technical solution, a water and fertilizer intelligent control system 10 based on UAV imagery is provided. The water and fertilizer intelligent control system 10 based on UAV imagery includes:

[0092] The data acquisition module 11 is used to acquire UAV multispectral image data of the target farmland and preprocess the image data;

[0093] The stress analysis module 12 is used to extract vegetation index features from the preprocessed image and, in conjunction with the preset farmland grid division, construct a crop water and fertilizer stress distribution map.

[0094] The decision calculation module 13 is used to input the crop water and fertilizer stress distribution map into the preset water and fertilizer decision model, and integrate multi-source environmental parameters for normalization calculation to obtain the target water and fertilizer application amount for each grid area of ​​the target farmland.

[0095] The execution control module 14 is used to generate a variable operation prescription map based on the target water and fertilizer application rate, and control the integrated water and fertilizer equipment to perform precise application operations according to the prescription map.

[0096] The data acquisition module 11 includes preprocessing of the UAV multispectral images by performing radiometric calibration, atmospheric correction, geometric correction, and image stitching to obtain an orthophoto map with geographic coordinate information. The stress analysis module 12 includes vegetation index features that include at least one or more of the following: normalized vegetation index, normalized red border index, and water stress index.

[0097] The construction of the crop water and fertilizer stress distribution map includes: mapping the acquired vegetation index features to the farmland grid, and calculating the mean value of the vegetation index features in each grid as the water and fertilizer stress characterization value of that grid.

[0098] In this embodiment, by combining the multi-dimensional parameter normalization compensation architecture in the decision calculation module 13 with the closed-loop feedback architecture in the execution control module, the technical problems of simple multi-source parameter fusion methods, fixed weight coefficients that cannot be adaptively adjusted, and lack of post-operation effect verification and closed-loop feedback in the prior art are solved. Specifically, the decision calculation module 13 uses a normalization compensation architecture to uniformly map heterogeneous parameters such as vegetation nutrition, soil moisture, and meteorological evapotranspiration to the [0,1] interval for fusion calculation, avoiding the limitations of single spectral index decision-making; at the same time, the execution control module 14 collects equipment flow rate and pressure parameters in real time during the application process for feedback adjustment, and verifies the effect by re-acquiring UAV images to calculate the gradient of vegetation index changes before and after fertilization after the operation, realizing dynamic correction of water and fertilizer management. Since the crop's demand for fertilizer is closely related to the soil's effective water content and transpiration intensity, the root absorption capacity is limited under water-deficient conditions, and simply increasing fertilizer application will aggravate stress. Therefore, this normalization compensation architecture can dynamically adjust the application amount according to the measured soil moisture value and meteorological driving factors, ensuring the scientific nature of water and fertilizer application under specific environmental conditions.

[0099] Through the aforementioned technical solution, this system integrates vegetation spectral response, measured soil moisture, and meteorological evapotranspiration into a unified compensation calculation framework. It also uses a closed-loop feedback architecture to detect device execution deviations or environmental abrupt changes, significantly improving the accuracy and adaptability of water and fertilizer management decisions. Furthermore, the system dynamically adjusts the contributions of vegetation nutrition, soil moisture, and meteorological factors through an adaptive weight optimization architecture, avoiding deviations in fertilization amounts caused by fixed empirical values. In practical applications, for example, when water stress occurs in a farmland grid area, the system automatically reduces the recommended fertilization amount for that area to prevent a decrease in fertilizer absorption efficiency due to water shortage. Simultaneously, post-application vegetation index change gradient monitoring can promptly trigger anomaly warnings, ensuring that water and fertilizer application effects meet expectations. Overall, this technical solution effectively overcomes the shortcomings of traditional water and fertilizer management, such as resource waste and environmental pollution, providing reliable technical support for precision agriculture.

[0100] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent water and fertilizer control based on UAV imagery, characterized in that, Includes the following steps: S1. Acquire multispectral image data of the target farmland using a drone, and preprocess the image data; S2. Extract vegetation index features from the preprocessed image and construct a crop water and fertilizer stress distribution map by combining it with the preset farmland grid division. S3. Input the crop water and fertilizer stress distribution map into the preset water and fertilizer decision model, and perform fusion calculations in combination with multi-source environmental parameters to obtain the target water and fertilizer application amount for each grid area of ​​the target farmland. S4. Generate a variable operation prescription map based on the target water and fertilizer application rate, and control the integrated water and fertilizer equipment to perform precise application operations according to the prescription map.

2. The intelligent water and fertilizer control method based on UAV imagery according to claim 1, characterized in that, In step S1, the preprocessing includes: performing radiometric calibration, atmospheric correction, geometric correction, and image stitching on the UAV multispectral image in sequence to obtain an orthophoto map with geographic coordinate information.

3. The intelligent water and fertilizer control method based on UAV imagery according to claim 1, characterized in that, In step S2, the vegetation index features include at least one or more of the following: normalized vegetation index, normalized red edge index, and water stress index. The construction of the crop water and fertilizer stress distribution map includes: mapping the acquired vegetation index features to the farmland grid, and calculating the mean value of the vegetation index features in each grid as the water and fertilizer stress characterization value of that grid.

4. The intelligent water and fertilizer control method based on UAV imagery according to claim 1, characterized in that, In step S3, the preset water and fertilizer decision model is constructed based on a normalized compensation architecture with multi-dimensional parameters, and the target water and fertilizer application rate for the i-th grid area of ​​the target farmland is calculated in the following way: The target water and fertilizer application rate is equal to the basic water and fertilizer application rate recommended for the current crop growth stage multiplied by the normalized total value. The normalized total value is 1 plus the product of vegetation nutrient weight and the first normalized value, the product of soil moisture weight and the second normalized value, and the product of meteorological compensation weight and the third normalized value. The first normalized value is the difference between the upper limit of the vegetation nutrient index and the normalized vegetation index value of the grid, divided by the difference between the upper and lower limits of the vegetation nutrient index; the second normalized value is the difference between field capacity and the measured soil moisture value of the grid, divided by the difference between field capacity and the moisture content at the wilting point; the third normalized value is the difference between the real-time meteorological evapotranspiration of the grid and the minimum evapotranspiration of the historical statistical period, divided by the difference between the maximum and minimum evapotranspiration of the historical statistical period. Among them, the sum of vegetation nutrient weight, soil moisture weight and meteorological compensation weight is 1. The calculation process uses range standardization to uniformly map heterogeneous parameters to the interval of 0 to 1.

5. The intelligent water and fertilizer control method based on UAV imagery according to claim 4, characterized in that, The upper and lower limits of the vegetation nutrient index are determined by statistically analyzing the normalized vegetation index quantiles of historical crop multispectral data for the same period; the field water holding capacity and wilting point water content are determined by referring to a table based on the soil texture of the target farmland; the maximum and minimum evapotranspiration of the historical statistical period are determined by sorting the evapotranspiration of historical meteorological data; the real-time meteorological evapotranspiration is calculated using the Penman-Montes formula; and the measured soil moisture value is obtained by measuring the soil moisture through a multi-layer moisture sensor buried in the soil.

6. The intelligent water and fertilizer control method based on UAV imagery according to claim 5, characterized in that, The vegetation nutrient compensation weight, soil moisture compensation weight, and meteorological compensation weight are determined through an adaptive dynamic adjustment architecture. This architecture uses a particle swarm optimization algorithm with the objective function of maximizing the water and fertilizer use efficiency of a single grid area. It iterative optimization is used to output the optimal weight combination for the current growth stage. The water and fertilizer use efficiency is the expected economic yield predicted based on historical yield data and the current water and fertilizer application rate divided by the actual water and fertilizer application rate of the grid area.

7. The intelligent water and fertilizer control method based on UAV imagery according to claim 1, characterized in that, In step S4, generating the variable operation prescription map includes: converting the target water and fertilizer application rate of each grid into the solenoid valve opening degree and variable frequency pump frequency control command sequence of the integrated water and fertilizer equipment, and binding it with the grid geographic coordinates to generate a GIS-based prescription map; The method also includes closed-loop feedback: during the execution process, the flow rate and pressure parameters of the equipment are collected in real time and compared with the theoretical values ​​for feedback adjustment; after the operation is completed, the drone image is acquired again to calculate the gradient of vegetation index change before and after fertilization, and if the gradient is lower than the preset threshold, an abnormal warning is triggered.

8. A water and fertilizer intelligent control system based on UAV imagery, the system being used to implement the water and fertilizer intelligent control method based on UAV imagery as described in any one of claims 1 to 7, characterized in that, The system includes: The data acquisition module is used to acquire UAV multispectral image data of the target farmland and preprocess the image data; The stress analysis module is used to extract vegetation index features from the preprocessed image and, in conjunction with the preset farmland grid division, construct a crop water and fertilizer stress distribution map. The decision calculation module is used to input the crop water and fertilizer stress distribution map into the preset water and fertilizer decision model, and integrate multi-source environmental parameters for normalization calculation to obtain the target water and fertilizer application amount for each grid area of ​​the target farmland. The execution control module is used to generate a variable operation prescription map based on the target water and fertilizer application rate, and control the integrated water and fertilizer equipment to perform precise application operations according to the prescription map.

9. The intelligent water and fertilizer control system based on UAV imagery according to claim 8, characterized in that, The preprocessing in the data acquisition module includes: performing radiometric calibration, atmospheric correction, geometric correction, and image stitching on the UAV multispectral images in sequence to obtain an orthophoto map with geographic coordinate information.

10. The intelligent water and fertilizer control system based on UAV imagery according to claim 8, characterized in that, In the work content of the stress analysis module, the vegetation index features include at least one or more of the following: normalized vegetation index, normalized red edge index, and water stress index; The construction of the crop water and fertilizer stress distribution map includes: mapping the acquired vegetation index features to the farmland grid, and calculating the mean value of the vegetation index features in each grid as the water and fertilizer stress characterization value of that grid.

Images