Agricultural environment monitoring method and system based on digital agriculture

Abstract

The invention discloses an agricultural environment monitoring method and system based on digital agriculture. The method comprises the steps that S1, original environment data signals are collected through a heterogeneous sensor network deployed in a farmland area; s2, performing space-time fusion processing on the original environment data signal to generate a standardized environment state signal; s3, inputting the standardized environment state signal into a pre-trained multi-mode diagnosis model to generate an environment-crop coupling diagnosis signal; s4, according to the environment-crop coupling diagnosis signal, an agricultural operation decision signal is generated, and the agricultural operation decision signal comprises quantitative parameters of the irrigation amount, the fertilization formula or the disaster prevention and control instruction; and S5, converting the agricultural operation decision signal into an equipment control instruction signal, and driving an execution mechanism to complete precise agricultural operation. According to the agricultural environment monitoring method and system based on digital agriculture, the problems of data splitting, extensive decision making and execution lagging in agricultural environment monitoring can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent environmental monitoring in digital agriculture, and in particular to an agricultural environment monitoring method and system based on digital agriculture. Background Art

[0002] With the development of digital agriculture, agricultural environmental monitoring systems are increasingly adopting IoT technology to deploy sensor networks to collect soil, meteorological, and crop physiological data. Existing technologies primarily aggregate this data to cloud platforms via wireless transmission protocols, incorporating threshold rules to trigger basic control commands such as irrigation and ventilation. However, these systems have significant drawbacks: First, the disparate deployment locations of multiple sensors result in inconsistent temporal and spatial scales, making it difficult to correlate and analyze soil moisture data with canopy light data, hindering the ability to construct a holistic environmental picture. Second, decisions rely on simple threshold rules (e.g., irrigation if soil moisture falls below 30%), failing to consider the dynamic needs of crop growth stages and the influence of multiple parameters, resulting in a high rate of misjudgment. Furthermore, existing patents (such as CN1129508A) propose irrigation systems based on meteorological forecasts, while incorporating external data, fail to establish crop response models, making it impossible to discern the differences between similar symptoms of stresses such as drought and salinization.

[0003] At the same time, traditional system control instructions need to be manually confirmed and executed after they are generated. Delayed responses to emergency scenarios such as rainstorm warnings often lead to disaster losses. Although some studies have attempted to introduce machine learning models (such as CNN processing crop images) in recent years, the models only serve a single task (pest and disease identification), are not linked to environmental data, and lack closed-loop control capabilities. Although the environmental control device proposed in patent CN1145678B is partially automated, its decision-making module still relies on preset fixed parameters and cannot adapt to different crop varieties and regional climate characteristics. In summary, the current agricultural environmental monitoring field urgently needs a systematic solution that can integrate multi-dimensional data, intelligently diagnose root causes, dynamically generate decisions and automatically execute them, so as to solve the industry pain points of insufficient data value mining and excessive reliance on manual labor. Summary of the Invention

[0004] In view of the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide an agricultural environmental monitoring method and system based on digital agriculture, which is used to solve the problems of data fragmentation, extensive decision-making, and delayed execution in agricultural environmental monitoring. The present invention collects multi-dimensional environmental data through a heterogeneous sensor network, eliminates data fragmentation through spatiotemporal fusion processing, and generates standardized signals. Utilizing a multimodal diagnostic model that integrates a knowledge graph and a growth rule library, the environmental signals are converted into diagnostic signals for the root causes of crop stress. Based on the diagnostic results, quantitative agronomic decision signals are dynamically generated and automatically converted into equipment control instructions to drive the actuator, thus realizing a closed loop of "perception-diagnosis-decision-execution", breaking through the bottlenecks of traditional monitoring systems that are characterized by isolated data, reliance on manual experience, and slow response.

[0005] The present invention provides an agricultural environment monitoring method based on digital agriculture, comprising:

[0006] S1: A heterogeneous sensor network deployed in farmland collects raw environmental data signals, including soil parameter signals, meteorological parameter signals, and crop physiological parameter signals;

[0007] S2: Performing spatiotemporal fusion processing on the original environmental data signal to generate a standardized environmental status signal, wherein the spatiotemporal fusion processing includes data alignment based on sensor geographic location information and outlier correction based on timestamps;

[0008] S3: Input standardized environmental status signals into a pre-trained multimodal diagnostic model to generate an environment-crop coupled diagnostic signal. The multimodal diagnostic model is constructed by integrating the environmental knowledge graph with the crop growth rule library to identify the root causes of environmental stress.

[0009] S4: generating an agronomic operation decision signal based on the environment-crop coupling diagnostic signal, wherein the agronomic operation decision signal includes quantitative parameters of irrigation amount, fertilization formula or disaster prevention and control instructions;

[0010] S5: Convert agronomic operation decision signals into equipment control command signals and drive the actuators to complete precision agriculture operations.

[0011] In one embodiment of the present invention, in step S1, the heterogeneous sensor network includes a profile moisture sensor buried deep in the soil, a photosynthetically active radiation sensor installed in the crop canopy, and a stem flow meter attached to the crop stem; the soil parameter signal forms a multi-layer soil moisture gradient signal through the profile moisture sensor, the crop physiological parameter signal forms a crop transpiration rate continuous monitoring signal through the stem flow meter, and the meteorological parameter signal forms a photosynthesis potential assessment signal through the canopy sensor; wherein each sensor node is deployed according to the preset farmland grid coordinates to form a spatial topologically associated raw data acquisition matrix.

[0012] In one embodiment of the present invention, in step S2, data alignment based on sensor geographic location information includes: mapping discretely collected raw environmental data signals to a unified geographic grid coordinate system according to the latitude and longitude coordinates and elevation data of each sensor node to generate a spatially continuously distributed interpolated data field; timestamp-based outlier correction includes: for sensor data with time synchronization deviations within the same geographic grid, using a dynamic threshold filtering algorithm within a sliding time window to eliminate transient interference signals, thereby forming a standardized environmental status signal with temporal and spatial consistency.

[0013] In one embodiment of the present invention, in step S3, the environmental knowledge graph includes a rule chain of pest and disease occurrence conditions with crop species as entities, a nutrient deficiency symptom feature library, and a climate stress response pattern library; the crop growth rule library includes water demand curves, accumulated temperature accumulation thresholds, and photoperiod sensitivity coefficients for different growth periods; the multimodal diagnostic model uses a graph neural network to match the standardized environmental state signals with the knowledge graph nodes for similarity, while triggering the constraint verification in the crop growth rule library, and finally outputting a coupled diagnostic signal that characterizes the type of environmental stress and the probability of occurrence.

[0014] In one embodiment of the present invention, when generating an environment-crop coupling diagnostic signal, the multimodal diagnostic model simultaneously performs a traceability analysis of the root cause of the stress: according to the activation path of the rule chain in the knowledge graph, the core parameter combination that causes the environmental stress is located, and a root cause tracing signal containing the stress contribution weight is generated; the root cause tracing signal and the coupling diagnostic signal together constitute the diagnostic result output.

[0015] In one embodiment of the present invention, in step S4, the generation of the agronomic operation decision signal includes: calling a preset agronomic response strategy library according to the stress type identifier in the environment-crop coupling diagnostic signal; the agronomic response strategy library quantifies the irrigation amount as a combination of the soil target moisture content range and the irrigation duration, quantifies the fertilization formula as the nitrogen, phosphorus and potassium element ratio and the water-fertilizer mixing concentration parameters, and quantifies the disaster prevention and control instructions as the equipment start and stop time window and the operation intensity level; wherein the quantitative parameters are dynamically generated by inversion calculation of the crop growth model.

[0016] In one embodiment of the present invention, the agronomic response strategy library performs multi-objective optimization decision-making before outputting quantitative parameters: using water resource utilization efficiency, fertilizer absorption rate and energy consumption as constraints, and expected crop yield and quality indicators as objective functions, a multi-objective evolutionary algorithm is used to solve the optimal operating parameter combination, and the combination is encoded into an agronomic operation decision signal.

[0017] In one embodiment of the present invention, in step S5, the conversion of the equipment control command signal includes: generating a water pump start-stop status signal and a solenoid valve opening adjustment signal based on the irrigation amount parameter in the agronomic operation decision signal; generating a raw material feed valve control sequence and a mixed liquid concentration feedback calibration signal of the fertilizer liquid proportioning tank based on the fertilization formula parameters; and generating a fan speed step control signal or a sunshade net extension and retraction stroke position signal based on the disaster prevention and control instruction.

[0018] In one embodiment of the present invention, before driving the actuator, a control instruction safety verification step is added: the device control instruction signal and the actuator's status feedback signal are logically interlocked and verified. When a conflict is detected between the instruction signal and the actual working condition of the equipment, the preset degradation control strategy is triggered and an equipment safety operation instruction signal is generated to replace the original control instruction to drive the actuator.

[0019] The present invention also includes an agricultural environment monitoring system based on digital agriculture, comprising:

[0020] The acquisition module collects raw environmental data signals through a heterogeneous sensor network deployed in the farmland area;

[0021] The coordination module performs spatiotemporal fusion processing on the original environmental data signals to generate standardized environmental status signals, wherein the spatiotemporal fusion processing includes data alignment based on sensor geographic location information and outlier correction based on timestamps;

[0022] The comparison module inputs standardized environmental status signals into a pre-trained multimodal diagnostic model to generate an environment-crop coupled diagnostic signal. The multimodal diagnostic model is constructed by integrating the environmental knowledge graph with the crop growth rule library to identify the root causes of environmental stress;

[0023] A preview module generates an agronomic operation decision signal based on the environment-crop coupling diagnostic signal. The agronomic operation decision signal includes quantitative parameters of irrigation amount, fertilizer formula or disaster prevention and control instructions;

[0024] Analysis module, the analysis module converts the agronomic operation decision signal into the equipment control command signal, and drives the execution mechanism to complete the precision agriculture operation.

[0025] The agricultural environment monitoring method and system based on digital agriculture provided by the present invention collect multi-dimensional environmental data through a heterogeneous sensor network, eliminate data fragmentation through spatiotemporal fusion processing to generate standardized signals; utilize a multimodal diagnostic model that integrates a knowledge graph and a growth rule library to convert environmental signals into diagnostic signals for the root causes of crop stress; dynamically generate quantitative agronomic decision signals based on the diagnostic results, and automatically convert them into equipment control instructions to drive the actuator, realizing a "perception-diagnosis-decision-execution" closed loop, breaking through the bottlenecks of traditional monitoring systems such as data isolation, reliance on manual experience, and slow response. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 A method flow chart for agricultural environment monitoring method based on digital agriculture;

[0028] Figure 2 This is the system architecture diagram of the agricultural environment monitoring system based on digital agriculture. DETAILED DESCRIPTION

[0029] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0030] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0031] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0032] See Figure 1-Figure 2, showing the agricultural environment monitoring method and system based on digital agriculture of the present invention. The agricultural environment monitoring method based on digital agriculture of the present invention includes: S1: collecting raw environmental data signals through a heterogeneous sensor network deployed in the farmland area, the raw environmental data signals including soil parameter signals, meteorological parameter signals, and crop physiological parameter signals; S2: performing spatiotemporal fusion processing on the raw environmental data signals to generate standardized environmental status signals, wherein the spatiotemporal fusion processing includes data alignment based on sensor geographic location information and outlier correction based on timestamps; S3: inputting the standardized environmental status signals into a pre-trained multimodal diagnostic model to generate an environment-crop coupled diagnostic signal, the multimodal diagnostic model being constructed by fusing an environmental knowledge graph with a crop growth rule library and used to identify the root cause of environmental stress; S4: generating an agronomic operation decision signal based on the environment-crop coupled diagnostic signal, the agronomic operation decision signal including quantitative parameters of irrigation amount, fertilization formula, or disaster prevention and control instructions; S5: converting the agronomic operation decision signal into an equipment control command signal and driving the actuator to complete the precision agriculture operation.

[0033] like Figure 1As shown in the figure, a method for agricultural environmental monitoring based on digital agriculture is provided. Its core lies in establishing a closed-loop control chain of "data acquisition → fusion processing → intelligent diagnosis → decision generation → automatic execution." In specific implementation, raw environmental data signals are first collected through a heterogeneous sensor network deployed in the farmland area. This network consists of sensor nodes with various physical forms and communication protocols, covering three major parameter monitoring categories: soil, meteorological, and crop physiological. Soil parameter signals include at least soil temperature, volumetric moisture content, and electrical conductivity; meteorological parameter signals include at least light intensity, air temperature and humidity, and wind speed and direction; and crop physiological parameter signals include at least leaf temperature, stem micro-variation, and canopy reflectance spectrum. These raw signals, due to differences in sensor type, have inconsistent sampling frequencies, dimensions, and temporal and spatial references. To address this issue, the system performs spatiotemporal fusion processing: It uses the sensor's built-in positioning module to obtain geographic location information (latitude, longitude, and elevation), maps the discretely collected data to a unified geographic grid coordinate system, and generates a continuously distributed data field using the Kriging spatial interpolation algorithm. Furthermore, based on high-precision timestamps, a sliding window outlier detection technique is used to eliminate outliers caused by transient interference (such as birds blocking the light sensor). Ultimately, the system outputs a standardized environmental status signal consistent across spatiotemporal scales. This signal is fed into a pre-trained multimodal diagnostic model. By integrating an environmental knowledge graph (containing rule chains such as pest and disease occurrence conditions and nutrient deficiency thresholds) with a crop growth rule base (containing water requirement curves and accumulated temperature thresholds for different growth stages), the model uses a graph neural network to calculate the match between environmental states and knowledge nodes, identifying the root causes of environmental stress (for example, distinguishing between similar wilting symptoms caused by water stress and salt stress) and generating an environmental-crop coupled diagnostic signal that includes stress type, probability of occurrence, and impact range. Based on this diagnostic signal, the system calls the agronomic response strategy library to generate quantitative operational instructions: It calculates the target soil moisture range and continuous irrigation duration for irrigation needs, generates nitrogen, phosphorus, and potassium ratios and mixing concentration parameters for fertilization needs, and generates fan start / stop time windows or shade net extension / retraction position instructions for disaster prevention and control, forming agronomic operation decision signals. Finally, through a protocol conversion module (for example, converting irrigation volume parameters into water pump start / stop instructions under the Modbus RTU protocol), the decision signals are converted into device control command signals that can be interpreted by the actuator, driving irrigation valves, fertilization pumps, or environmental control equipment to complete precise operations, achieving a fully automated closed-loop process from environmental perception to agronomic operations.

[0034] Furthermore, the hardware deployment architecture and data generation mechanism of the heterogeneous sensor network are defined. The network adopts a layered, three-dimensional deployment strategy: vertically, profile moisture sensors embedded deep in the soil are distributed at depths of 20 cm, 40 cm, and 60 cm, generating a soil moisture gradient monitoring signal. This signal reveals the movement of water within the soil layer and avoids misjudgments caused by single surface measurements. In the canopy space, photosynthetically active radiation sensors mounted atop the crop canopy are fixed with tilt-adjustable brackets to generate a real-time photosynthetic potential assessment signal. This data, combined with canopy temperature and CO2 concentration, can be used to infer light energy utilization efficiency. Within the crop itself, stem flow meters attached to the base of the stem monitor transpiration rate using a heat dissipation method, generating a continuous assessment signal for crop water use efficiency. All sensor nodes are deployed according to the field's grid coordinates, divided into monitoring grids of 100-meter units. Each grid contains at least one soil-canopy-crop physiological sensor set, with the grid center node acting as a data aggregation gateway. The soil profile sensor transmits data using the narrowband IoT protocol, the canopy sensor utilizes a LoRa ad hoc network, and the stemflow meter uses Bluetooth 5.0 to connect directly to the gateway, forming a multi-protocol heterogeneous communication topology. This deployment approach imbues the raw environmental data signals with spatial topological correlation. For example, if a grid detects a sudden drop in deep soil moisture and increased canopy transpiration, it can provide an early warning of the risk of insufficient root water absorption. Furthermore, a temporal comparison of the stemflow meter signal with the canopy light signal reveals the impact of the crop's "midday nap" (decreased transpiration due to midday stomatal closure) on photosynthetic efficiency, providing a multi-dimensional cross-validation basis for the diagnostic model.

[0035] like Figure 1As shown in the figure, the spatiotemporal fusion processing step refines the technical implementation of data alignment and outlier correction. During the data alignment phase, the system reads Beidou / GPS dual-mode positioning data from each sensor node and, combined with elevation barometer readings, constructs three-dimensional geographic coordinates. Discrete coordinates are converted to a plane rectangular coordinate system using a Gauss-Krüger projection. Using a 10m×10m grid as the basic unit, the inverse distance weighted interpolation algorithm is used to spatially aggregate data from different sensors within the same grid. For n soil temperature sensors within a grid, the grid center point temperature, T_c, is calculated using the formula T_c = ∑(T_i / d_i^2) / ∑(1 / d_i^2), where d_i is the distance from the sensor to the center point. This ultimately generates a spatially continuous temperature field covering the entire field. For factors such as meteorological parameters that require gradient variation, gradient-adaptive interpolation is used. For example, in areas with slopes greater than 5°, a vertical lapse rate correction (0.6°C per 100m of elevation) is applied to the temperature data based on the elevation difference. In the time dimension, the system adds a microsecond timestamp to each data point. For sensors with acquisition delays within the same grid (such as the sampling intervals of canopy light and soil temperature are not synchronized), a cubic spline function is used for time resampling and alignment. The outlier correction link uses dynamic threshold filtering: a sliding time window is set (the default is 30 minutes), the mean μ and standard deviation σ of the data in the window are calculated, and data points exceeding μ±3σ are marked as suspected anomalies; combined with device status verification (such as checking whether the sensor voltage at that moment is lower than the critical value) and environmental correlation analysis (such as a sudden increase in soil moisture during a sudden rainstorm is a reasonable phenomenon), if it is verified to be interference, the median within the window is used instead. Especially for canopy light data that is susceptible to transient interference, a spectrum analysis filter is added: after collecting the original light signal, a fast Fourier transform is performed to filter out impulse noise with a frequency higher than 10Hz (such as the shadow of a flying bird), and then reconstruct it into a valid signal. After this processing, the originally fragmented raw data is converted into a standardized environmental status signal that is synchronized in time and space. Its data field can accurately reflect the spatial distribution gradient and temporal evolution continuity of environmental parameters, providing high-quality input for subsequent diagnostic models.

[0036] Furthermore, based on the multimodal diagnostic model, its knowledge base architecture and diagnostic mechanism are further defined. The environmental knowledge graph uses crop species as the main entity node and connects three key sub-graphs through attribute edges: the rule chain for pest and disease occurrence conditions includes the temperature and humidity thresholds for pathogen infection, the accumulated temperature thresholds for pest outbreaks, and the parameters for the symptom development cycle. For example, the anthrax infection rule is defined as "air humidity > 85% for 48 hours and the leaves are wet for > 6 hours"; the nutrient deficiency symptom feature library establishes a mapping relationship between element deficiency and phenotypic characteristics. For example, the potassium deficiency symptom is described as "scorched leaf margins of old leaves accompanied by yellowing between veins" and is associated with the critical value of soil potassium ion concentration; the climate stress response model library defines the physiological impact model of extreme weather on crops. For example, the high temperature heat damage model includes a quantitative curve of "nonlinear increase in pollen abortion rate when the daily maximum temperature is > 35°C." The crop growth rule library stores physiological requirement parameters in segments according to the growth period: the critical soil moisture content for seed water absorption and expansion is set during the germination period, the minimum effective accumulated temperature required for stem elongation is defined during the jointing period, and the mathematical function of the dry matter accumulation of grains and the temperature difference between day and night are associated during the filling period. The multimodal diagnostic model uses a dual-channel processing architecture of a graph neural network when running: Channel 1 performs graph attention matching on the standardized environmental state signal and the knowledge graph node, and calculates the degree of compliance of the current environmental state with each rule condition (for example, if the temperature and humidity conditions for anthrax infection are met at the same time, the node is activated); Channel 2 calls the growth period parameters in the crop growth rule library to verify whether the environmental data exceeds the stage tolerance threshold (for example, the soil moisture during the jointing period is lower than 1.5 times the wilting coefficient). The outputs of the two channels are weightedly integrated by the gated fusion layer. When the activation intensity of the pest and disease rule is higher than the preset threshold and the crop physiological tolerance verification alarm is issued, a high-confidence environment-crop coupling diagnostic signal is output, and its signal vector contains the stress type code, the probability value of occurrence and the affected organ identifier. This design breaks through the limitations of traditional single models that only deal with phenotypes or environments, and achieves a deep integration of agricultural knowledge and data-driven.

[0037] like Figure 1As shown in the figure, the diagnostic model adds a traceability analysis function for the root causes of stress. When the model generates an environment-crop coupling diagnostic signal, the system automatically traces the node activation path in the graph neural network. For example, when diagnosing "fruit sunburn," the traceability reveals that the "canopy high temperature node" (temperature > 42°C) and the "fruit surface transpiration inhibition node" (air humidity < 30%) in the knowledge graph are simultaneously activated, and the "inadequate leaf shading rule" (leaf area index < 2.8) contributes 70%. Based on the connection weights of each node in the activation path and the rule confidence, the system calculates the contribution rate of the core parameter combination leading to the stress, such as "canopy temperature contribution rate 45%, air humidity contribution rate 35%, leaf area index contribution rate 20%" in this case. This analysis result is encoded as a root cause traceability signal, whose data structure consists of three elements: a list of traceability factors (e.g., [temperature, humidity, leaf area index]), the contribution weights of each factor (e.g., [0.45, 0.35, 0.20]), and the critical threshold deviation (e.g., temperature deviation value +8.2°C). This signal, along with the original diagnostic signal, forms the complete output, visualized through the human-computer interface. On the farm management platform, a Sankey diagram displays the contribution flow of each factor, while a heat map identifies the spatial distribution of parameters with high contribution rates. Especially in complex stress scenarios (such as overlapping drought and potassium deficiency symptoms), the root cause tracing signal can clearly distinguish the dominant factor: when the soil moisture contribution reaches 65%, drought is considered the primary cause, while conversely, potassium deficiency is considered the primary cause. This mechanism significantly improves the interpretability of diagnostic results and provides a theoretical basis for the precise implementation of agronomic measures.

[0038] Specifically, the system generates agronomic operation decision signals using a built-in agronomic response strategy library, whose strategy entries are indexed and stored by environmental stress type. For water stress, the irrigation amount is quantified into a target soil moisture range (e.g., sandy loam is set at 22%-25% volume moisture during the flowering period) and an irrigation duration (calculated based on the soil infiltration rate to last 120 minutes). For nutrient stress, the fertilization formula is converted into a specific element ratio (e.g., a nitrogen, phosphorus, and potassium ratio of 18:5:22) and a water-fertilizer mixture concentration parameter (e.g., a mother liquor dilution factor of 1:200). For disaster prevention and control, cold wave response instructions are quantified into a fan start-stop time window (starting 2 hours before frost occurs) and a speed level (set to 80% of rated speed at -3°C). Quantified parameter generation utilizes a crop growth model inversion mechanism: Taking water stress as an example, the stress level (0-100%) from the environmental-crop coupling diagnostic signal is input into the crop water production function to infer the optimal water supply for the target yield. Specifically, the crop coefficient Kc is first determined based on the current growth period, and the actual evapotranspiration (ETa) is calculated using canopy transpiration signals. The ideal evapotranspiration (ETi) is then inverted using the expected yield reduction rate from the diagnostic signal using the Jensen model formula: Yield / Ymax = ∏(ETi / ETm_i)^λ_i. Finally, the irrigation amount is determined from the difference between ETi and ETa, and then converted to irrigation duration based on soil hydraulic conductivity. This process transitions agronomic decision-making from qualitative to quantitative, avoiding the ambiguity inherent in traditional empirical decision-making.

[0039] like Figure 2 As shown, the present invention relates to an agricultural environment monitoring system based on digital agriculture, including an acquisition module, which collects original environmental data signals through a heterogeneous sensor network deployed in a farmland area; a coordination module, which performs spatiotemporal fusion processing on the original environmental data signals to generate standardized environmental status signals, wherein the spatiotemporal fusion processing includes data alignment based on sensor geographic location information and outlier correction based on timestamps; a comparison module, which inputs the standardized environmental status signals into a pre-trained multimodal diagnosis model to generate an environment-crop coupling diagnosis signal, and the multimodal diagnosis model is constructed by fusing an environmental knowledge graph with a crop growth rule library to identify the root cause of environmental stress; a rehearsal module, which generates an agronomic operation decision signal based on the environment-crop coupling diagnosis signal, and the agronomic operation decision signal includes quantitative parameters of irrigation amount, fertilization formula or disaster prevention and control instructions; and an analysis module, which converts the agronomic operation decision signal into an equipment control instruction signal and drives the actuator to complete precision agriculture operations.

[0040] like Figure 2As shown in the figure, a multi-objective optimization decision-making mechanism was introduced. Before the agronomic response strategy library outputs quantitative parameters, an optimization model with three constraints and two objectives was constructed: Constraint one is water resource utilization efficiency, requiring the cumulative dry matter per unit of irrigation water to be no less than 90% of the historical optimal value; Constraint two is fertilizer absorption rate, with a lower limit of 40% for nitrogen utilization efficiency; and Constraint three is energy consumption, requiring the irrigation system to consume no more than 10 kWh of electricity per operation. Objective function one is to maximize expected crop yield, using a modified EPIC model to simulate grain yield under different decision scenarios; Objective function two is to optimize quality indicators, using a comprehensive evaluation of sugar-acid ratio and color index for fruit crops. The optimization process utilizes an improved NSGA-II algorithm: First, an initial decision population is generated (e.g., 100 irrigation and fertilization parameter combinations). The objective function values ​​for each scenario are calculated through crop growth model simulation. Next, non-dominated sorting and crowding calculation are performed to screen the Pareto front solution set. Finally, the offspring population is generated through simulated binary crossover and polynomial mutation. After multiple generations of evolution, the system outputs a Pareto-optimal solution set for the user to choose from: In water-saving mode, the solution with the highest water resource efficiency is selected; in quality-prioritized mode, the solution with the highest sugar-acid ratio is chosen. Ultimately, the system encodes the selected solution into an agronomic decision signal. Irrigation parameters include time-segmented flow control curves (e.g., high flow for root infiltration in the first 30 minutes, followed by low flow for deep penetration in the last 90 minutes), while fertilization parameters are precisely defined, down to the liquid inflow ratio and mixing sequence for each raw material tank. This optimization mechanism ensures that the decision signal maximizes agronomic benefits within resource constraints.

[0041] The agricultural environment monitoring method and system based on digital agriculture of the present invention collect multi-dimensional environmental data through a heterogeneous sensor network, eliminate data fragmentation through spatiotemporal fusion processing to generate standardized signals; utilize a multimodal diagnostic model that integrates a knowledge graph and a growth rule library to convert environmental signals into diagnostic signals for the root causes of crop stress; dynamically generate quantitative agronomic decision signals based on the diagnostic results, and automatically convert them into equipment control instructions to drive the actuator, realizing a "perception-diagnosis-decision-execution" closed loop, breaking through the bottlenecks of traditional monitoring systems such as data isolation, reliance on manual experience, and slow response.

[0042] Therefore, the agricultural environment monitoring method and system based on digital agriculture of the present invention can solve the problems of data fragmentation, extensive decision-making and delayed execution in agricultural environment monitoring.

[0043] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. The agricultural environment monitoring method based on digital agriculture is characterized by: include: S1: Collecting raw environmental data signals through a heterogeneous sensor network deployed in the farmland area. The raw environmental data signals include soil parameter signals, meteorological parameter signals, and crop physiological parameter signals; S2: performing spatiotemporal fusion processing on the original environmental data signal to generate a standardized environmental status signal, wherein the spatiotemporal fusion processing includes data alignment based on sensor geographic location information and outlier correction based on timestamps; S3: Inputting the standardized environmental state signal into a pre-trained multimodal diagnostic model to generate an environment-crop coupled diagnostic signal, wherein the multimodal diagnostic model is constructed by integrating an environmental knowledge graph and a crop growth rule library to identify the root cause of environmental stress; S4: generating an agronomic operation decision signal based on the environment-crop coupled diagnostic signal, wherein the agronomic operation decision signal includes quantitative parameters of irrigation amount, fertilization formula or disaster prevention and control instructions; S5: Convert the agronomic operation decision signal into an equipment control command signal, and drive the execution mechanism to complete the precision agriculture operation.

2. The agricultural environment monitoring method based on digital agriculture according to claim 1, characterized in that: In step S1, the heterogeneous sensor network includes a profile moisture sensor buried deep in the soil, a photosynthetically active radiation sensor installed in the crop canopy, and a stem flow meter attached to the crop stem; the soil parameter signal forms a multi-layer soil moisture gradient signal through the profile moisture sensor, the crop physiological parameter signal forms a crop transpiration rate continuous monitoring signal through the stem flow meter, and the meteorological parameter signal forms a photosynthesis potential assessment signal through the canopy sensor; wherein each sensor node is deployed according to a preset farmland grid coordinate to form a spatial topologically associated raw data acquisition matrix.

3. The agricultural environment monitoring method based on digital agriculture according to claim 1, characterized in that: In step S2, the data alignment based on the sensor geographic location information includes: mapping the discretely collected original environmental data signals to a unified geographic grid coordinate system according to the latitude and longitude coordinates and elevation data of each sensor node, and generating a spatially continuously distributed interpolated data field; the timestamp-based outlier correction includes: for sensor data with time synchronization deviations within the same geographic grid, using a dynamic threshold filtering algorithm within a sliding time window to eliminate transient interference signals, thereby forming a standardized environmental status signal with temporal and spatial consistency.

4. The agricultural environment monitoring method based on digital agriculture according to claim 1, characterized in that: In step S3, the environmental knowledge graph includes a rule chain of pest and disease occurrence conditions with crop species as entities, a nutrient deficiency symptom feature library, and a climate stress response pattern library; the crop growth rule library includes water demand curves, accumulated temperature accumulation thresholds, and photoperiod sensitivity coefficients for different growth periods; the multimodal diagnostic model uses a graph neural network to perform similarity matching between standardized environmental state signals and knowledge graph nodes, while triggering the constraint verification in the crop growth rule library, and finally outputting a coupled diagnostic signal that characterizes the type of environmental stress and the probability of occurrence.

5. The agricultural environment monitoring method based on digital agriculture according to claim 4 is characterized in that: When generating the environment-crop coupling diagnostic signal, the multimodal diagnostic model simultaneously performs a traceability analysis of the root cause of the stress: according to the activation path of the rule chain in the knowledge graph, the core parameter combination that causes the environmental stress is located, and a root cause tracing signal containing the stress contribution weight is generated; this root cause tracing signal and the coupling diagnostic signal together constitute the diagnostic result output.

6. The agricultural environment monitoring method based on digital agriculture according to claim 1, characterized in that: In step S4, the generation of the agronomic operation decision signal includes: calling a preset agronomic response strategy library according to the stress type identifier in the environment-crop coupling diagnostic signal; the agronomic response strategy library quantifies the irrigation amount into a combination of the soil target moisture content range and the irrigation duration, quantifies the fertilization formula into the nitrogen, phosphorus and potassium element ratio and the water-fertilizer mixing concentration parameters, and quantifies the disaster prevention and control instructions into the equipment start-stop time window and the operation intensity level; wherein the quantitative parameters are dynamically generated by inversion calculation of the crop growth model.

7. The agricultural environment monitoring method based on digital agriculture according to claim 6, characterized in that: Before outputting quantitative parameters, the agronomic response strategy library performs multi-objective optimization decision-making: using water resource utilization efficiency, fertilizer absorption rate and energy consumption as constraints, and expected crop yield and quality indicators as objective functions, a multi-objective evolutionary algorithm is used to solve the optimal operation parameter combination, and the combination is encoded into an agronomic operation decision signal.

8. The agricultural environment monitoring method based on digital agriculture according to claim 1, characterized in that: In step S5, the conversion of the equipment control command signal includes: generating a water pump start-stop status signal and a solenoid valve opening adjustment signal according to the irrigation amount parameter in the agronomic operation decision signal; generating a raw material feed valve control sequence and a mixed liquid concentration feedback calibration signal of the fertilizer liquid proportioning tank according to the fertilization formula parameters; and generating a fan speed step control signal or a sunshade net extension and retraction stroke position signal according to the disaster prevention and control instruction.

9. The agricultural environment monitoring method based on digital agriculture according to claim 8, characterized in that: Before driving the actuator, a control instruction safety verification step is added: the device control instruction signal and the actuator status feedback signal are logically interlocked and verified. When a conflict is detected between the instruction signal and the actual working condition of the equipment, the preset degradation control strategy is triggered and an equipment safety operation instruction signal is generated to replace the original control instruction to drive the actuator.

10. An agricultural environment monitoring system using the agricultural environment monitoring method based on digital agriculture according to any one of claims 1 to 9, characterized in that: include: A collection module, wherein the collection module collects raw environmental data signals through a heterogeneous sensor network deployed in the farmland area; a coordination module, wherein the coordination module performs spatiotemporal fusion processing on the original environmental data signal to generate a standardized environmental status signal, wherein the spatiotemporal fusion processing includes data alignment based on sensor geographic location information and outlier correction based on timestamps; a comparison module, which inputs the standardized environmental state signal into a pre-trained multimodal diagnostic model to generate an environment-crop coupled diagnostic signal. The multimodal diagnostic model is constructed by fusing an environmental knowledge graph with a crop growth rule library to identify the root cause of environmental stress; a preview module, wherein the preview module generates an agronomic operation decision signal based on the environment-crop coupled diagnostic signal, wherein the agronomic operation decision signal includes quantitative parameters of irrigation amount, fertilization formula, or disaster prevention and control instructions; An analysis module converts the agronomic operation decision signal into an equipment control instruction signal and drives an execution mechanism to complete precision agriculture operations.

Images