Fertilization control method and system based on corn conservation farming

Abstract

The invention is suitable for the field of crop fertilization control, and provides a fertilization control method and system based on corn protective farming, and the method comprises the steps: synchronously collecting time-space multi-dimensional data of a target planting region, and constructing a multi-modal data cube; generating a seed fertilizer ratio decision vector; according to the seed-fertilizer ratio decision vector, controlling the multi-channel ratio of mechanical precision seed-fertilizer simultaneous sowing equipment, so as to realize the dynamic ratio and variable fertilization of the seed fertilizer; a multi-spectral image and soil moisture content data are collected through an unmanned aerial vehicle group, and a space-time continuous crop growth monitoring network is constructed; dynamically adjusting the topdressing amount and the water and fertilizer concentration of the unmanned aerial vehicle group based on the growth monitoring network data; according to the method, by constructing a multi-modal data cube, the dynamic proportioning and variable fertilization of seeds and fertilizers are realized, the proportioning precision is improved by utilizing fuzzy PID and machine vision, a space-time monitoring network is constructed by virtue of an unmanned aerial vehicle group, the dynamic adjustment of topdressing is realized, and a data-driven closed-loop fertilization system is formed.

Description

Technical Field

[0001] The present invention belongs to the field of crop fertilization control, and in particular relates to a fertilization control method and system based on corn conservation tillage. Background Art

[0002] In traditional agricultural production, seed and fertilizer co-seeding and topdressing operations rely primarily on empirical decision-making, resulting in problems such as mismatches between fertilizer application and crop needs and low fertilizer utilization. While existing mechanical precision seed and fertilizer co-seeding technologies can achieve quantitative fertilization, they lack the ability to respond in real time to spatial variations in soil fertility and the dynamic growth needs of crops. This results in fixed fertilizer ratios being applied to the same plot of land, making it difficult to meet the differentiated needs of crops in different regions. Furthermore, conventional topdressing methods are typically based on empirical estimates or simple soil sampling and analysis, which cannot accurately capture changes in nutrient requirements during the crop growth cycle, easily leading to fertilizer waste and environmental pollution.

[0003] In recent years, the application of artificial intelligence and drone technology in agriculture has gradually increased, but significant shortcomings still exist. Some fertilization decision-making methods based on remote sensing imagery only consider vegetation indices for a single period and fail to establish a spatiotemporal continuous crop growth monitoring model, resulting in decision-making results lagging behind the actual needs of crops. At the same time, existing drone fertilization systems are mostly open-loop control and lack real-time feedback mechanisms, making it difficult to dynamically adjust fertilization parameters based on crop growth conditions. Therefore, there is an urgent need for an intelligent fertilization technology solution that can integrate multi-source data, realize dynamic ratio decision-making and precise execution, and address the problems of decision lag, low execution accuracy, and poor environmental friendliness existing in existing technologies. Summary of the Invention

[0004] The object of the present invention is to provide a fertilization control method and system based on corn conservation tillage, aiming to solve the problems raised in the above background technology.

[0005] The present invention is achieved by, on the one hand, a fertilization control method based on corn conservation tillage, the method comprising:

[0006] Synchronously collect spatiotemporal multidimensional data of the target planting area and construct a multimodal data cube;

[0007] The spatiotemporal multidimensional data includes historical soil fertility spatiotemporal distribution data, current remote sensing image data, meteorological data and crop growth model data;

[0008] Generate a seed-fertilizer ratio decision vector based on the multimodal data cube, wherein the decision vector includes a base fertilizer ratio parameter and a topdressing strategy parameter;

[0009] According to the seed-fertilizer ratio decision vector, the multi-channel ratio of the mechanical precision seed-fertilizer simultaneous sowing equipment is controlled to achieve dynamic seed-fertilizer ratio and variable fertilization;

[0010] During the crop growth cycle, drone swarms collect multispectral images and soil moisture data to build a spatiotemporal continuous crop growth monitoring network.

[0011] Based on the growth monitoring network data, dynamically adjust the amount of topdressing and water and fertilizer concentration of the drone group;

[0012] The construction of the multimodal data cube specifically includes:

[0013] Constructing a spatiotemporal indexed soil fertility database S = {S i，j，k};

[0014] Where i is the spatial location, j is the time dimension, and k is the nutrient type;

[0015] Based on the convolutional neural network, the vegetation characteristic index vector VI including the normalized vegetation index, enhanced vegetation index, greenness normalized vegetation index, red edge normalized vegetation index and leaf area index is extracted from the remote sensing image data;

[0016] The soil fertility database S={S i，j，k} performs spatiotemporal registration with the vegetation feature index vector VI to generate a fused feature tensor T.

[0017] As a further solution of the present invention, the seed-fertilizer ratio decision vector is generated based on the multimodal data cube, and the decision vector includes base fertilizer ratio parameters and topdressing strategy parameters, specifically including:

[0018] Constructing a multi-task deep neural network model M, wherein the multi-task deep neural network model M includes a shared feature extraction layer and multiple task-specific output layers;

[0019] Obtain historical soil sampling data and the amount of organic matter returned to the field, calculate the nutrient values released after the decomposition of organic matter per unit area in the target planting area, and generate a set of nutrient values to be deducted per unit area;

[0020] Generate a unit nutrient requirement matrix based on the set of nutrient values to be deducted per unit area;

[0021] The unit nutrient requirement matrix uses 1m×1m as the smallest unit and is used to quantify the target nutrient gap of each unit;

[0022] Based on the unit nutrient requirement matrix, combined with the soil fertility database S = {S i，j，k}, calculate the application amount of organic fertilizer and inorganic fertilizer for each unit;

[0023] Input the fusion feature tensor T and the application amount of organic fertilizer and inorganic fertilizer for each unit, and output the base fertilizer ratio decision vector F based on the multi-task deep neural network model M. base =(fb N , f b P ,,f b K , ···) and topdressing strategy matrix S t =[s ij ];

[0024] Where, f b N Indicates the ratio coefficient of nitrogen fertilizer in base fertilizer, f b P Indicates the ratio coefficient of phosphorus fertilizer in base fertilizer, f b K Indicates the ratio coefficient of phosphorus fertilizer in base fertilizer, s ij represents the demand coefficient for the jth nutrient in the i-th growth stage;

[0025] The multi-task deep neural network model M is trained and optimized using the following loss function:

[0026] L=αL r +βL s +γL y ;

[0027] Where, L r is the matching loss function, L s is the policy loss function, L y is the yield prediction loss function, and α, β, and γ are the corresponding weight coefficients.

[0028] As a further embodiment of the present invention, the multi-channel ratio control of the mechanical precision seed and fertilizer simultaneous sowing equipment according to the seed and fertilizer ratio decision vector is used to achieve dynamic seed and fertilizer ratio and variable fertilization, which specifically includes:

[0029] Based on the base fertilizer ratio decision vector F base =(f b N , f b P ,,f b K ,···), the flow control valve of each fertilizer channel is adjusted by fuzzy PID controller to achieve a dynamic ratio accuracy error of less than ±2%;

[0030] Use machine vision technology to monitor the status of fertilizer discharge in real time, and use the attention mechanism neural network to identify abnormal conditions such as blockage and fertilizer leakage;

[0031] When an abnormal situation is detected, the adaptive compensation algorithm is triggered to adjust the fertilizer amount of adjacent fertilizer outlets to maintain the overall fertilization uniformity coefficient of variation CV ≤ 8%.

[0032] As a further embodiment of the present invention, the method of collecting multispectral images and soil moisture data by a swarm of drones during the crop growth cycle to construct a spatiotemporally continuous crop growth monitoring network specifically includes:

[0033] Deploy drone swarms to form a grid monitoring network, with the overlap rate of adjacent drone monitoring areas ≥30%;

[0034] Build a spatiotemporal series prediction model to predict crop growth trends in the next 3-7 days based on historical monitoring data;

[0035] When the predicted growth indicators deviate from the preset threshold range, the early warning mechanism is triggered and intervention recommendations are generated;

[0036] Recommendations for intervention include:

[0037] When nitrogen is insufficient, it generates nitrogen, phosphorus and potassium topdressing ratio and dosage adjustment information for adding 5-8 kg of urea per mu;

[0038] During the drought period, the drone generates water and fertilizer concentration adjustment information to adjust the nitrogen fertilizer concentration from 1:10 to 1:8;

[0039] When crop density is uneven, it generates information to adjust the drone's flight parameters, reducing flight speed and narrowing spray width;

[0040] When the soil pH is unbalanced, the soil pH value and trace element adjustment information of applying 20-30 kg of quicklime per mu is generated.

[0041] As a further solution of the present invention, the dynamic adjustment of the amount of topdressing and the concentration of water and fertilizer of the drone group based on the growth monitoring network data specifically includes:

[0042] A Markov decision process model was constructed, where the state space U contained crop growth indicators, soil moisture conditions, and meteorological data, and the action space A contained fertilizer application amount and water and fertilizer concentration.

[0043] Solve the optimal strategy π through deep reinforcement learning algorithm * : U→A, maximize the long-term cumulative reward function R;

[0044] According to the optimal strategy π * , dynamically adjust the operating parameters of the drone swarm, including flight speed, spray width, nozzle flow rate and fertilizer concentration;

[0045] The calculation process of the reward function R is:

[0046] R=ω1Y p -ω2C f -ω3C e -ω4E e ;

[0047] Where Y p To predict production, C f is the fertilizer cost, Ce is the energy consumption cost, Ee is the environmental impact index, ω1, ω2, ω3, ω4 are the corresponding normalized weight coefficients.

[0048] As a further embodiment of the present invention, on the other hand, a fertilization control system based on corn conservation tillage is provided, the system comprising:

[0049] An acquisition module, used to synchronously collect spatiotemporal multidimensional data of the target planting area;

[0050] The spatiotemporal multidimensional data include historical soil fertility spatiotemporal distribution data, current remote sensing image data, meteorological data and crop growth model data.

[0051] Multimodal data cube module, used to build multimodal data cube;

[0052] A generation module, configured to generate a seed-fertilizer ratio decision vector based on the multimodal data cube;

[0053] The decision vector includes base fertilizer ratio parameters and topdressing strategy parameters;

[0054] A control module, configured to control the multi-channel ratio of the mechanical precision seed and fertilizer simultaneous sowing equipment according to the seed and fertilizer ratio decision vector;

[0055] The crop growth monitoring network module is used to collect multispectral images and soil moisture data through drone swarms during the crop growth cycle to build a spatiotemporally continuous crop growth monitoring network;

[0056] A dynamic adjustment module is used to dynamically adjust the amount of topdressing and water and fertilizer concentration of the drone group based on the growth monitoring network data.

[0057] As a further solution of the present invention, the multimodal data cube module specifically includes:

[0058] Soil fertility database unit, used to construct a spatiotemporal indexed soil fertility database S = {S i，j，k};

[0059] An extraction unit, configured to extract a vegetation characteristic index vector VI including a normalized vegetation index, an enhanced vegetation index, a greenness normalized vegetation index, a red edge normalized vegetation index, and a leaf area index from remote sensing image data based on a convolutional neural network;

[0060] Generating unit, used to convert soil fertility database S={S i，j，k} performs spatiotemporal registration with the vegetation feature index vector VI to generate a fused feature tensor T.

[0061] As a further solution of the present invention, the generation module specifically includes:

[0062] A multi-task deep neural network model unit, used to build a multi-task deep neural network model M;

[0063] The nutrient deduction value set unit is used to obtain historical soil sampling data and the amount of organic matter returned to the field, calculate the nutrient values released after the decomposition of organic matter per unit area in the target planting area, and generate a set of nutrient values to be deducted per unit area;

[0064] Calculation unit, used based on unit nutrient requirement matrix, combined with soil fertility database S = {S i，j，k}, calculate the application amount of organic fertilizer and inorganic fertilizer for each unit;

[0065] Input and output unit, used to input the fusion feature tensor T, based on the multi-task deep neural network model M, output the base fertilizer ratio decision vector F base =(f b N , f b P ,,f b K , ···) and topdressing strategy matrix S t =[s ij ].

[0066] As a further solution of the present invention, the control module specifically includes:

[0067] Adjustment unit, used for making decision based on base fertilizer ratio vector F base =(f b N , f b P ,,f b K ,···), the flow control valve of each fertilizer channel is adjusted by fuzzy PID controller to achieve a dynamic ratio accuracy error of less than ±2%;

[0068] Real-time monitoring unit, used to monitor the status of seed fertilizer discharge in real time using machine vision technology;

[0069] The recognition unit is used to identify abnormal conditions such as blockage and fertilizer leakage through the attention mechanism neural network;

[0070] The adjustment unit is used to detect abnormal conditions, trigger the adaptive compensation algorithm, adjust the fertilizer amount of adjacent fertilizer discharge ports, and maintain the overall fertilization uniformity coefficient of variation CV ≤ 8%.

[0071] The present invention provides a fertilization control method and system based on corn conservation tillage. By constructing a multimodal data cube, dynamic seed-fertilizer ratio and variable fertilization are achieved. Fuzzy PID and machine vision are used to improve the ratio accuracy. A spatiotemporal monitoring network is built based on a swarm of drones to achieve dynamic adjustment of topdressing, forming a data-driven closed-loop fertilization system. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a main flow chart of a fertilization control method based on corn conservation tillage.

[0073] Figure 2 This is a flowchart for constructing a multimodal data cube in a fertilization control method based on corn conservation tillage.

[0074] Figure 3 The invention relates to a flow chart for generating a seed-fertilizer ratio decision vector based on the multimodal data cube in a fertilization control method based on corn conservation tillage.

[0075] Figure 4 The invention relates to a flow chart for controlling the multi-channel ratio of mechanical precision seed and fertilizer simultaneous sowing equipment according to the seed-fertilizer ratio decision vector in a fertilization control method based on corn conservation tillage.

[0076] Figure 5 This is a flowchart of a fertilization control method based on corn conservation tillage, which uses a swarm of drones to collect multispectral images and soil moisture data during the crop growth cycle to build a spatiotemporal continuous crop growth monitoring network.

[0077] Figure 6 The invention relates to a flow chart for dynamically adjusting the amount of topdressing and water-fertilizer concentration of a drone group based on the growth monitoring network data in a fertilization control method based on corn conservation tillage.

[0078] Figure 7 It is the main structure diagram of a fertilization control system based on corn conservation tillage.

[0079] Figure 8 This is a structural block diagram of a multimodal data cube module in a fertilization control system based on corn conservation tillage.

[0080] Figure 9 It is a structural block diagram of the generation module in a fertilization control system based on corn conservation tillage.

[0081] Figure 10 It is a structural block diagram of the control module in a fertilization control system based on corn conservation tillage. DETAILED DESCRIPTION

[0082] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present 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 only used to explain the present invention and are not intended to limit the present invention.

[0083] The specific implementation of the present invention is described in detail below with reference to specific embodiments.

[0084] The present invention provides a fertilization control method and system based on corn conservation tillage, which solves the technical problems in the background technology.

[0085] like Figure 1 FIG. 1 is a main flow chart of a fertilization control method based on corn conservation tillage according to an embodiment of the present invention. The fertilization control method based on corn conservation tillage includes:

[0086] Step S100: synchronously collecting spatiotemporal multidimensional data of the target planting area and constructing a multimodal data cube;

[0087] The spatiotemporal multidimensional data includes historical soil fertility spatiotemporal distribution data, current remote sensing image data, meteorological data and crop growth model data;

[0088] Step S200: generating a seed-fertilizer ratio decision vector based on the multimodal data cube, wherein the decision vector includes a base fertilizer ratio parameter and a topdressing strategy parameter;

[0089] Step S300: Controlling the multi-channel ratio of the mechanical precision seed and fertilizer simultaneous sowing equipment according to the seed-fertilizer ratio decision vector to achieve dynamic seed-fertilizer ratio and variable fertilization;

[0090] Step S400: During the crop growth cycle, a swarm of drones is used to collect multispectral images and soil moisture data to build a spatiotemporally continuous crop growth monitoring network.

[0091] Step S500: Dynamically adjusting the amount of topdressing and water and fertilizer concentration of the drone group based on the growth monitoring network data;

[0092] When this embodiment is applied, during the construction of a multimodal data cube, the spatiotemporal indexed soil fertility database integrates nutrient data such as nitrogen, phosphorus, and potassium from multiple years and multiple sampling points to form a spatiotemporal fertility evolution map. The five vegetation indices extracted from current remote sensing image data through a convolutional neural network can reflect crop photosynthetic capacity and growth status in real time. The generation of the seed-fertilizer ratio decision vector relies on the feature fusion of multimodal data to achieve the coordinated optimization of the base fertilizer ratio and topdressing strategy. The multi-channel ratio control of mechanical precision seed and fertilizer simultaneous sowing equipment addresses the drawbacks of traditional fixed ratios through variable fertilization technology. Simultaneously, the drone swarm monitoring network constructs a dynamic monitoring grid with a resolution of 30 meters x 30 meters through the spatiotemporal continuous acquisition of multispectral images and soil moisture conditions. Dynamic adjustment of topdressing amount and water and fertilizer concentration, combined with growth model predictions and real-time monitoring data, forms an adaptive nutrient supply mechanism.

[0093] Furthermore, the mechanical precision seed and fertilizer sowing equipment is an integrated equipment with sowing and fertilizing functions. It can independently store different fertilizers such as nitrogen, phosphorus, and potassium, support precise supply according to the ratio decision vector, and is equipped with a fuzzy PID controller. The fuzzy PID controller adjusts the flow control valves of each channel to achieve dynamic ratio fertilization. Through the binocular camera and attention neural network, it monitors the fertilizer discharge status in real time and identifies anomalies. It is also equipped with an adaptive compensation mechanism. When blockage is detected, the fertilizer amount of adjacent fertilizer discharge ports can be automatically adjusted to maintain uniformity.

[0094] like Figure 2 As shown, as a preferred embodiment of the present invention, the construction of a multimodal data cube specifically includes:

[0095] Step S101: Constructing a spatiotemporal indexed soil fertility database S = {S i，j，k};

[0096] Where i is the spatial location, j is the time dimension, and k is the nutrient type;

[0097] Step S102: extracting a vegetation characteristic index vector VI including the normalized vegetation index, enhanced vegetation index, greenness normalized vegetation index, red edge normalized vegetation index, and leaf area index from the remote sensing image data based on a convolutional neural network;

[0098] Step S103: Soil fertility database S={S i，j，k} performs spatiotemporal registration with the vegetation feature index vector VI to generate a fused feature tensor T.

[0099] like Figure 3 As shown, as a preferred embodiment of the present invention, generating a seed-fertilizer ratio decision vector based on the multimodal data cube specifically includes:

[0100] Step S201: constructing a multi-task deep neural network model M, wherein the multi-task deep neural network model M includes a shared feature extraction layer and multiple task-specific output layers;

[0101] Step S202: Acquire historical soil sampling data and the amount of organic matter returned to the field, calculate the nutrient values released after decomposition of organic matter per unit area in the target planting area, and generate a set of nutrient values to be deducted per unit area;

[0102] Step S203: generating a unit nutrient requirement matrix based on the set of nutrient values to be deducted per unit area;

[0103] The unit nutrient requirement matrix uses 1m×1m as the smallest unit and is used to quantify the target nutrient gap of each unit;

[0104] Step S204: Based on the unit nutrient requirement matrix, combined with the soil fertility database S = {S i，j，k}, calculate the application amount of organic fertilizer and inorganic fertilizer for each unit;

[0105] Step S205: Input the fusion feature tensor T and the application amount of organic fertilizer and inorganic fertilizer for each unit, and output the base fertilizer ratio decision vector F based on the multi-task deep neural network model M. base =(f b N , f b P ,,f b K , ···) and topdressing strategy matrix S t =[s ij ];

[0106] Where, f b N Indicates the ratio coefficient of nitrogen fertilizer in base fertilizer, f b P Indicates the ratio coefficient of phosphorus fertilizer in base fertilizer, f b K Indicates the ratio coefficient of phosphorus fertilizer in potassium fertilizer, s ij represents the demand coefficient for the jth nutrient in the i-th growth stage;

[0107] The multi-task deep neural network model M is trained and optimized using the following loss function:

[0108] L=αL r +βL s +γL y ;

[0109] Where, L r is the matching loss function, L s is the policy loss function, L yis the yield prediction loss function, α, β, and γ are the corresponding weight coefficients;

[0110] In application, this embodiment addresses the problem of neglecting the return of organic matter to the field and the extensive application of fertilizers in conservation tillage. First, based on historical soil sampling data and the amount of organic matter returned to the field, the nitrogen, phosphorus, potassium and other nutrients released by the degradation of organic matter are calculated to form a set of nutrient values to be deducted per unit area, and a 1m×1m unit nutrient demand matrix is generated, thereby accurately quantifying the target nutrient gap of each unit and filling the nutrient accounting loopholes in the conservation tillage scenario; then, combined with the soil fertility database S={S i，j，k}, converting the nutrient gap into the application amount of organic fertilizer and inorganic fertilizer to achieve fertilizer efficiency complementarity; finally, input the fusion feature tensor T and fertilizer application amount into the multi-task deep neural network model M, and output the base fertilizer ratio decision vector F base =(f b N , f b P ,,f b K ,···) (clear nitrogen, phosphorus, potassium ratio) and topdressing strategy matrix S t =[s ij ](Planning nutrient supplementation plan during the growth period), providing precise parameters for the actuators that control the fertilizer amount and the electric drive ratio under the power wheel drive, effectively realizing the spatial precision of organic and inorganic fertilizers in conservation tillage scenarios.

[0111] The multi-task deep neural network model M employs an encoder-decoder architecture. Its shared feature extraction layer consists of three convolutional neural network layers and two fully connected layers, extracting deep correlation features from soil, vegetation, and meteorological data. Task-specific output layers include a subnetwork for outputting base fertilizer ratios and a subnetwork for outputting topdressing strategies, each employing different activation functions to handle linear and nonlinear requirements.

[0112] The multi-task deep neural network model M is trained using a transfer learning strategy. After pre-training on an existing publicly available agricultural dataset, it is fine-tuned using historical fertilization data from the target region. The ratio loss in the loss function uses mean squared error to measure the deviation between the predicted ratio and actual fertilization. The strategy loss uses cross entropy to assess the accuracy of nutrient demand forecasts for each growth stage. The yield prediction loss uses mean absolute error to optimize the model's ability to estimate final yield. The weight coefficients α, β, and γ are determined using a Bayesian optimization algorithm.

[0113] like Figure 4 As shown, as a preferred embodiment of the present invention, the multi-channel ratio control of the mechanical precision seed and fertilizer simultaneous sowing equipment according to the seed and fertilizer ratio decision vector specifically includes:

[0114] Step S301: Based on the base fertilizer ratio decision vector Fbase =(f b N , f b P ,,f b K ,···), the flow control valve of each fertilizer channel is adjusted by fuzzy PID controller to achieve a dynamic ratio accuracy error of less than ±2;

[0115] Step S302: Using machine vision technology to monitor the status of fertilizer discharge in real time, and identifying abnormal conditions such as blockage and fertilizer leakage through an attention mechanism neural network;

[0116] Step S303: When an abnormality is detected, the adaptive compensation algorithm is triggered to adjust the fertilizer amount of adjacent fertilizer outlets to maintain the coefficient of variation CV of the overall fertilizer uniformity ≤ 8%;

[0117] In this embodiment, the fuzzy PID controller uses a two-dimensional fuzzy controller structure, with a sampling period of 50ms. The machine vision monitoring system uses a binocular vision sensor to capture images of the fertilizer outlets at a frame rate of 30fps. The YOLOv model is used to identify abnormal fertilizer particle distribution and flow rates. When a fertilizer outlet is blocked, the fertilizer application rate to adjacent outlets is increased based on distance weighting (closer distances result in higher weighting). The compensation coefficient is determined by fitting a blockage-yield loss curve based on historical data.

[0118] like Figure 5 As shown in FIG. , as a preferred embodiment of the present invention, the method of collecting multispectral images and soil moisture data by a swarm of drones during the crop growth cycle to build a spatiotemporally continuous crop growth monitoring network specifically includes:

[0119] Step S401: deploying a swarm of drones to form a grid monitoring network, with the overlap rate of adjacent drone monitoring areas being ≥30%;

[0120] Step S402: constructing a spatiotemporal sequence prediction model to predict crop growth trends in the next 3-7 days based on historical monitoring data;

[0121] Step S403: When the predicted growth indicator deviates from the preset threshold range, triggering the early warning mechanism and generating intervention suggestions;

[0122] Recommendations for intervention include:

[0123] When nitrogen is insufficient, it generates nitrogen, phosphorus and potassium topdressing ratio and dosage adjustment information for adding 5-8 kg of urea per mu;

[0124] During the drought period, the drone generates water and fertilizer concentration adjustment information to adjust the nitrogen fertilizer concentration from 1:10 to 1:8;

[0125] When crop density is uneven, it generates information to adjust the drone's flight parameters, reducing flight speed and narrowing spray width;

[0126] When the soil pH is unbalanced, it generates soil pH value and trace element adjustment information for applying 20-30 kg of quicklime per mu;

[0127] It should be understood that the drone swarm grid monitoring system adopts a layered deployment strategy. High-altitude drones acquire regional-scale vegetation indices, while low-altitude drones collect field-scale multispectral data. Adjacent drone monitoring areas overlap by 35%, ensuring seamless data integration. The spatiotemporal series prediction model, based on an LSTM-Attention architecture, takes as input NDVI, soil moisture, and temperature data from the past 15 days and outputs leaf area index and plant height predictions for the next five days. This model uses a bidirectional LSTM to capture the dependencies between time series, while the attention mechanism focuses on the impact of environmental factors during key growth periods (such as the jointing stage).

[0128] The early warning mechanism sets three levels of thresholds (mild, moderate, and severe deviation). When the predicted growth indicator deviates from the historical mean by ±15%, a mild warning is triggered; when the predicted growth indicator deviates from the historical mean by ±35%, a moderate warning is triggered; and when the predicted growth indicator deviates from the historical mean by ±55%, a severe deviation warning is triggered.

[0129] like Figure 6 As shown, as a preferred embodiment of the present invention, the dynamic adjustment of the drone group's topdressing amount and water and fertilizer concentration based on the growth monitoring network data specifically includes:

[0130] Step S501: Constructing a Markov decision process model, where the state space U includes crop growth indicators, soil moisture conditions, and meteorological data, and the action space A includes the amount of topdressing and water and fertilizer concentrations;

[0131] Step S502: Solve the optimal strategy π through deep reinforcement learning algorithm * : U→A, maximize the long-term cumulative reward function R;

[0132] Step S503: According to the optimal strategy π * , dynamically adjust the operating parameters of the drone swarm, including flight speed, spray width, nozzle flow rate and fertilizer concentration;

[0133] The calculation process of the reward function R is:

[0134] R=ω1Y p -ω2C f -ω3C e -ω4E e ;

[0135] Where Yp To predict production, C f is the fertilizer cost, Ce is the energy consumption cost, Ee is the environmental impact index, ω1, ω2, ω3, ω4 are the corresponding normalized weight coefficients;

[0136] In this embodiment, the deep reinforcement learning algorithm adopts the proximal policy optimization framework. The policy network consists of 2 fully connected layers, and the value network uses a 3-layer convolutional network to process state features. The predicted output Y in the reward function R p Based on the random forest model, the fertilizer cost C f Calculated based on the real-time prices of urea, potassium dihydrogen phosphate, etc., the energy consumption cost Ce includes the drone flight energy consumption and water pump power consumption, and the environmental impact indicator Ee is evaluated through the soil nitrogen leaching model.

[0137] like Figure 7 As shown, as another preferred embodiment of the present invention, on the other hand, a fertilization control system based on corn conservation tillage, the system includes:

[0138] The acquisition module 100 is used to synchronously acquire spatiotemporal multi-dimensional data of the target planting area;

[0139] The spatiotemporal multidimensional data include historical soil fertility spatiotemporal distribution data, current remote sensing image data, meteorological data and crop growth model data.

[0140] Multimodal data cube module 200, used to construct a multimodal data cube;

[0141] A generating module 300 is configured to generate a seed-fertilizer ratio decision vector based on the multimodal data cube;

[0142] The decision vector includes base fertilizer ratio parameters and topdressing strategy parameters;

[0143] The control module 400 is used to control the multi-channel ratio of the mechanical precision seed and fertilizer simultaneous sowing equipment according to the seed and fertilizer ratio decision vector;

[0144] The crop growth monitoring network module 500 is used to collect multispectral images and soil moisture data through a swarm of drones during the crop growth cycle to build a spatiotemporally continuous crop growth monitoring network;

[0145] The dynamic adjustment module 600 is used to dynamically adjust the amount of topdressing and the concentration of water and fertilizer applied by the drone group based on the growth monitoring network data.

[0146] When this embodiment is applied, the acquisition module 100 synchronously collects spatiotemporal multidimensional data of the target planting area, the spatiotemporal multidimensional data including historical soil fertility spatiotemporal distribution data, current remote sensing image data, meteorological data and crop growth model data; the multimodal data cube module 200 constructs a multimodal data cube; based on the multimodal data cube, the generation module 300 generates a seed-fertilizer ratio decision vector, the decision vector including base fertilizer ratio parameters and topdressing strategy parameters; according to the seed-fertilizer ratio decision vector, the control module 400 controls the multi-channel ratio of the mechanical precision seed-fertilizer simultaneous sowing equipment; during the crop growth cycle, multispectral images and soil moisture data are collected by a swarm of drones; the crop growth monitoring network module 500 constructs a spatiotemporal continuous crop growth monitoring network; at the same time, based on the growth monitoring network data, the dynamic adjustment module 600 dynamically adjusts the topdressing amount and water and fertilizer concentration of the drone swarm.

[0147] like Figure 8 As shown in FIG. 1 , as another preferred embodiment of the present invention, the multimodal data cube module 200 specifically includes:

[0148] Soil fertility database unit 201, used to construct a soil fertility database S with spatiotemporal indexing = {S i，j，k};

[0149] The extraction unit 202 is configured to extract a vegetation characteristic index vector VI including the normalized vegetation index, enhanced vegetation index, greenness normalized vegetation index, red edge normalized vegetation index, and leaf area index from the remote sensing image data based on a convolutional neural network.

[0150] Generating unit 203, used for converting soil fertility database S={S i，j，k} performs spatiotemporal registration with the vegetation feature index vector VI to generate a fused feature tensor T.

[0151] When this embodiment is used, the soil fertility database unit 201 constructs a soil fertility database S with a spatiotemporal index. i，j，k Based on the convolutional neural network, the extraction unit 202 extracts the vegetation characteristic index vector VI including the normalized vegetation index, enhanced vegetation index, greenness normalized vegetation index, red edge normalized vegetation index, and leaf area index from the remote sensing image data, and the generation unit 203 converts the soil fertility database S = {S i，j，k} performs spatiotemporal registration with the vegetation feature index vector VI to generate a fused feature tensor T.

[0152] like Figure 9 As shown, as another preferred embodiment of the present invention, the generating module 300 specifically includes:

[0153] A multi-task deep neural network model unit 301 is used to construct a multi-task deep neural network model M;

[0154] The nutrient deduction value set unit 302 is used to obtain historical soil sampling data and the amount of organic matter returned to the field, calculate the nutrient values released after the decomposition of organic matter per unit area in the target planting area, and generate a set of nutrient values per unit area to be deducted;

[0155] The calculation unit 303 is used to calculate the unit nutrient requirement matrix based on the soil fertility database S = {S i，j，k}, calculate the amount of organic fertilizer and inorganic fertilizer applied for each unit

[0156] The input and output unit 304 is used to input the fusion feature tensor T and output the base fertilizer ratio decision vector F based on the multi-task deep neural network model M. base =(f b N , f b P ,,f b K , ···) and topdressing strategy matrix S t =[s ij ].

[0157] When this embodiment is applied, the multi-task deep neural network model unit 301 constructs a multi-task deep neural network model M, the input and output unit 302 inputs the fusion feature tensor T, and outputs the base fertilizer ratio decision vector F based on the multi-task deep neural network model M. base =(f b N , f b P ,,f b K , ···) and topdressing strategy matrix S t =[s ij ]

[0158] like Figure 10 As shown, as another preferred embodiment of the present invention, the control module 400 specifically includes:

[0159] Adjustment unit 401, for adjusting the ratio of the base fertilizer based on the decision vector F base =(f b N , f b P ,,f b K ,···), the flow control valve of each fertilizer channel is adjusted by fuzzy PID controller to achieve a dynamic ratio accuracy error of less than ±2%;

[0160] A real-time monitoring unit 402 is used to monitor the discharge status of the seed fertilizer in real time using machine vision technology;

[0161] Identification unit 403, used to identify abnormal conditions such as blockage and fertilizer leakage through an attention mechanism neural network;

[0162] The adjustment unit 404 is used to detect an abnormal situation, trigger an adaptive compensation algorithm, adjust the fertilizer amount of adjacent fertilizer outlets, and maintain the overall fertilization uniformity coefficient of variation CV ≤ 8%.

[0163] In this embodiment, when applied, based on the base fertilizer ratio decision vector F base =(f b N , f b P ,,f b K , ···), the regulating unit 401 adjusts the flow regulating valve of each fertilizer channel through the fuzzy PID controller to achieve a dynamic ratio accuracy error of less than ±2%. The real-time monitoring unit 402 uses machine vision technology to monitor the seed fertilizer discharge status in real time. The recognition unit 403 uses the attention mechanism neural network to identify abnormal conditions such as blockage and fertilizer leakage. The adjustment unit 404 will detect the abnormal condition and trigger the adaptive compensation algorithm to adjust the fertilizer amount of adjacent fertilizer discharge ports to maintain the overall fertilization uniformity coefficient of variation CV ≤ 8%.

[0164] The above-mentioned embodiments of the present invention provide a fertilization control method based on corn conservation tillage, as well as a fertilization control system based on corn conservation tillage. In the construction of a multimodal data cube, a spatiotemporal-indexed soil fertility database integrates nutrient data such as nitrogen, phosphorus, and potassium from multiple years and multiple sampling points to form a spatiotemporal fertility evolution map. Five vegetation indices extracted from current remote sensing image data via a convolutional neural network can reflect crop photosynthetic capacity and growth status in real time. The generation of a seed-fertilizer ratio decision vector relies on the feature fusion of multimodal data to achieve coordinated optimization of base fertilizer ratios and topdressing strategies. Multi-channel ratio control for mechanical precision seed-fertilizer simultaneous sowing equipment addresses the drawbacks of traditional fixed ratios through variable-rate fertilization technology. Simultaneously, a swarm of drones monitors a dynamic monitoring grid with a resolution of 30 meters by 30 meters through the spatiotemporal continuous acquisition of multispectral images and soil moisture conditions. Dynamic adjustment of topdressing amount and water-fertilizer concentration, combined with growth model prediction and real-time monitoring data, forms an adaptive nutrient supply mechanism; by constructing a multimodal data cube, dynamic seed-fertilizer ratio and variable fertilization are achieved, fuzzy PID and machine vision are used to improve the ratio accuracy, and relying on drone swarms to build a spatiotemporal monitoring network to achieve dynamic adjustment of topdressing, forming a data-driven closed-loop fertilization system.

[0165] In order to enable the above-mentioned method and system to be loaded and run smoothly, in addition to the various modules mentioned above, the system may also include more or fewer components than described above, or a combination of certain components, or different components, for example, it may include input and output devices, network access devices, buses, processors and memories, etc.

[0166] The processor may be a central processing unit, other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the system, connecting various components using various interfaces and lines.

[0167] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0168] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

[0169] 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 in the scope of protection of the present invention.

Claims

1. A fertilization control method based on corn conservation tillage, characterized in that: The method comprises: Synchronously collect spatiotemporal multidimensional data of the target planting area and construct a multimodal data cube; The spatiotemporal multidimensional data includes historical soil fertility spatiotemporal distribution data, current remote sensing image data, meteorological data and crop growth model data; Generate a seed-fertilizer ratio decision vector based on the multimodal data cube, wherein the decision vector includes a base fertilizer ratio parameter and a topdressing strategy parameter; According to the seed-fertilizer ratio decision vector, the multi-channel ratio of the mechanical precision seed-fertilizer simultaneous sowing equipment is controlled to achieve dynamic seed-fertilizer ratio and variable fertilization; During the crop growth cycle, drone swarms collect multispectral images and soil moisture data to build a spatiotemporal continuous crop growth monitoring network. Based on the growth monitoring network data, dynamically adjust the amount of topdressing and water and fertilizer concentration of the drone group; The construction of the multimodal data cube specifically includes: Constructing a spatiotemporal indexed soil fertility database S = {S i，j，k }; Where i is the spatial location, j is the time dimension, and k is the nutrient type; Based on the convolutional neural network, the vegetation characteristic index vector VI including the normalized vegetation index, enhanced vegetation index, greenness normalized vegetation index, red edge normalized vegetation index and leaf area index is extracted from the remote sensing image data; The soil fertility database S={S i，j，k } performs spatiotemporal registration with the vegetation feature index vector VI to generate a fused feature tensor T.

2. The fertilization control method based on corn conservation tillage according to claim 1, characterized in that: The seed-fertilizer ratio decision vector is generated based on the multimodal data cube, and the decision vector includes base fertilizer ratio parameters and topdressing strategy parameters, specifically including: Constructing a multi-task deep neural network model M, wherein the multi-task deep neural network model M includes a shared feature extraction layer and multiple task-specific output layers; Obtain historical soil sampling data and the amount of organic matter returned to the field, calculate the nutrient values released after the decomposition of organic matter per unit area in the target planting area, and generate a set of nutrient values to be deducted per unit area; Generate a unit nutrient requirement matrix based on the set of nutrient values to be deducted per unit area; The unit nutrient requirement matrix uses 1m×1m as the smallest unit and is used to quantify the target nutrient gap of each unit; Based on the unit nutrient requirement matrix, combined with the soil fertility database S = {S i，j，k }, calculate the application amount of organic fertilizer and inorganic fertilizer for each unit; Input the fusion feature tensor T and the application amount of organic fertilizer and inorganic fertilizer for each unit, and output the base fertilizer ratio decision vector F based on the multi-task deep neural network model M. base =(f b N， f b P，， f b K， ···) and topdressing strategy matrix S t =[s ij ]; Where, f b N Indicates the ratio coefficient of nitrogen fertilizer in base fertilizer, f b P Indicates the ratio coefficient of phosphorus fertilizer in base fertilizer, f b K Indicates the ratio coefficient of phosphorus fertilizer in base fertilizer, s ij represents the demand coefficient for the jth nutrient in the i-th growth stage; The multi-task deep neural network model M is trained and optimized using the following loss function: L=αLr+βLs+γLy; Where, L r is the matching loss function, L s is the policy loss function, L y is the yield prediction loss function, and α, β, and γ are the corresponding weight coefficients.

3. The fertilization control method based on corn conservation tillage according to claim 2, characterized in that: The method of controlling the multi-channel ratio of the mechanical precision seed and fertilizer simultaneous sowing equipment according to the seed and fertilizer ratio decision vector to realize dynamic seed and fertilizer ratio and variable fertilization specifically includes: Based on the base fertilizer ratio decision vector F base =(f b N， f b P，， f b K， ···), the flow control valve of each fertilizer channel is adjusted by fuzzy PID controller to achieve a dynamic ratio accuracy error of less than ±2%; Use machine vision technology to monitor the status of fertilizer discharge in real time, and use the attention mechanism neural network to identify abnormal conditions such as blockage and fertilizer leakage; When an abnormal situation is detected, the adaptive compensation algorithm is triggered to adjust the fertilizer amount of adjacent fertilizer outlets to maintain the overall fertilization uniformity coefficient of variation CV ≤ 8%.

4. The fertilization control method based on corn conservation tillage according to claim 1, characterized in that: The method of collecting multispectral images and soil moisture data by drone swarms during the crop growth cycle to build a spatiotemporal continuous crop growth monitoring network specifically includes: Deploy drone swarms to form a grid monitoring network, with the overlap rate of adjacent drone monitoring areas ≥30%; Build a spatiotemporal series prediction model to predict crop growth trends in the next 3-7 days based on historical monitoring data; When the predicted growth indicators deviate from the preset threshold range, the early warning mechanism is triggered and intervention recommendations are generated; Recommendations for intervention include: When nitrogen is insufficient, it generates nitrogen, phosphorus and potassium topdressing ratio and dosage adjustment information for adding 5-8 kg of urea per mu; During the drought period, the drone generates water and fertilizer concentration adjustment information to adjust the nitrogen fertilizer concentration from 1:10 to 1:8; When crop density is uneven, it generates information to adjust the drone's flight parameters, reducing flight speed and narrowing spray width; When the soil pH is unbalanced, the soil pH value and trace element adjustment information of applying 20-30 kg of quicklime per mu is generated.

5. The fertilization control system based on corn conservation tillage according to claim 1, characterized in that: The dynamically adjusting the amount of topdressing and water-fertilizer concentration of the drone group based on the growth monitoring network data specifically includes: A Markov decision process model was constructed, where the state space U contained crop growth indicators, soil moisture conditions, and meteorological data, and the action space A contained fertilizer application amount and water and fertilizer concentration. Solve the optimal strategy π through deep reinforcement learning algorithm * : U→A, maximize the long-term cumulative reward function R; According to the optimal strategy π * , dynamically adjust the operating parameters of the drone swarm, including flight speed, spray width, nozzle flow rate and fertilizer concentration; The calculation process of the reward function R is: R=ω1Y p -ω2C f -ω3Ce-ω4Ee; Where Y p To predict production, C f is the fertilizer cost, Ce is the energy consumption cost, Ee is the environmental impact index, ω1, ω2, ω3, ω4 are the corresponding normalized weight coefficients.

6. A fertilization control system based on corn conservation tillage, characterized in that: The fertilization control method based on corn conservation tillage according to any one of claims 1 to 5 is applied, wherein the system comprises: An acquisition module, used to synchronously collect spatiotemporal multidimensional data of the target planting area; The spatiotemporal multidimensional data include historical soil fertility spatiotemporal distribution data, current remote sensing image data, meteorological data and crop growth model data. Multimodal data cube module, used to build multimodal data cube; A generation module, configured to generate a seed-fertilizer ratio decision vector based on the multimodal data cube; The decision vector includes base fertilizer ratio parameters and topdressing strategy parameters; A control module, configured to control the multi-channel ratio of the mechanical precision seed and fertilizer simultaneous sowing equipment according to the seed and fertilizer ratio decision vector; The crop growth monitoring network module is used to collect multispectral images and soil moisture data through drone swarms during the crop growth cycle to build a spatiotemporally continuous crop growth monitoring network; A dynamic adjustment module is used to dynamically adjust the amount of topdressing and water and fertilizer concentration of the drone group based on the growth monitoring network data.

7. The fertilization control system based on corn conservation tillage according to claim 6, characterized in that: The multimodal data cube module specifically includes: Soil fertility database unit, used to construct a spatiotemporal indexed soil fertility database S = {S i，j，k }; An extraction unit, configured to extract a vegetation characteristic index vector VI including a normalized vegetation index, an enhanced vegetation index, a greenness normalized vegetation index, a red edge normalized vegetation index, and a leaf area index from remote sensing image data based on a convolutional neural network; Generating unit, used to convert soil fertility database S={S i，j，k } performs spatiotemporal registration with the vegetation feature index vector VI to generate a fused feature tensor T.

8. The fertilization control system based on corn conservation tillage according to claim 6, characterized in that: The generation module specifically includes: A multi-task deep neural network model unit, used to build a multi-task deep neural network model M; The nutrient deduction value set unit is used to obtain historical soil sampling data and the amount of organic matter returned to the field, calculate the nutrient values released after the decomposition of organic matter per unit area in the target planting area, and generate a set of nutrient values to be deducted per unit area; Calculation unit, used based on unit nutrient requirement matrix, combined with soil fertility database S = {S i，j，k }, calculate the application amount of organic fertilizer and inorganic fertilizer for each unit; Input and output unit, used to input the fusion feature tensor T, based on the multi-task deep neural network model M, output the base fertilizer ratio decision vector F base =(f b N， f b P，， f b K， ···) and topdressing strategy matrix S t =[s ij ].

9. The fertilization control system based on corn conservation tillage according to claim 6, characterized in that: The control module specifically includes: Adjustment unit, used for making decision based on base fertilizer ratio vector F base =(f b N， f b P，， f b K， ···), the flow control valve of each fertilizer channel is adjusted by fuzzy PID controller to achieve a dynamic ratio accuracy error of less than ±2%; Real-time monitoring unit, used to monitor the status of seed fertilizer discharge in real time using machine vision technology; The recognition unit is used to identify abnormal conditions such as blockage and fertilizer leakage through the attention mechanism neural network; The adjustment unit is used to detect abnormal conditions, trigger the adaptive compensation algorithm, adjust the fertilizer amount of adjacent fertilizer discharge ports, and maintain the overall fertilization uniformity coefficient of variation CV ≤ 8%.

Images