Intelligent water and fertilizer integrated irrigation system based on soil moisture content and growth of fruit trees

Abstract

The application belongs to the field of agricultural intelligent irrigation, and particularly relates to an intelligent water and fertilizer integrated irrigation system based on soil moisture content and growth of fruit trees. The system comprises a global perception device, a ground monitoring network, a digital twin construction terminal, a causal inference analysis server, a generative evolution prediction device and a water and fertilizer collaborative control terminal. A digital mirror model is constructed through deep fusion of global and ground monitoring data, the driving logic of environmental factors on physiological response is analyzed by causal inference, and virtual sample completion and scenario deduction are carried out based on generative evolution, so as to guide the control terminal to realize accurate matching of water and fertilizer delivery amount and physiological demand of fruit trees. The application can accurately identify the cause of abnormal growth, improve resource utilization rate and robustness of system decision, and promote fine management and digital transformation of orchard management.

Description

Technical Field

[0001] This invention belongs to the field of intelligent agricultural irrigation, specifically involving an intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth. Background Technology

[0002] With the rapid development of smart agriculture technologies, precision irrigation and nutrient management have become key pathways to improve the production efficiency and environmental sustainability of modern orchards. Integrated water and fertilizer management technology, through the integration of sensor monitoring and automated control, enables fine-grained regulation of the crop growth environment, which is of great significance for optimizing agricultural resource allocation and increasing crop yields. In large-scale orchard management scenarios, the coordinated perception of soil physical parameters and crop physiological states provides essential data support for scientific decision-making throughout the entire growth cycle of fruit trees.

[0003] The air-ground integrated intelligent monitoring system aims to build a comprehensive digital twin model of orchards by integrating remote sensing inspections with ground-based sensor networks. This technology typically utilizes multispectral or thermal imaging techniques to assess the overall growth of the orchard, combined with local soil moisture data acquired from ground sensors, to achieve precise delivery of irrigation instructions. In practical applications, the system needs to perform real-time mapping and in-depth analysis of cross-scale data to ensure that water and fertilizer application rates precisely match the actual physiological needs of the fruit trees in both time and space.

[0004] Existing technologies suffer from a scale gap between broad overall coverage and precise individual tree sensing. Satellite or UAV remote sensing, limited by resolution, easily produces mixed pixels, making it impossible to characterize the individual differences of individual fruit trees. Meanwhile, sparsely distributed ground-based fixed monitoring points are insufficient to represent the heterogeneous characteristics of the entire orchard. Traditional analytical models rely heavily on statistical correlation mining, lacking in-depth analysis of the causal logic between environmental factors and physiological responses, and thus failing to accurately determine the underlying driving factors of abnormal growth. Due to the lack of generative evolutionary analysis and causal inference capabilities for multi-source heterogeneous data, the system struggles to establish a diagnostic logic that coordinates overall coverage and individual tree sensing in complex and ever-changing natural environments, resulting in insufficient targeting and scientific rigor in irrigation decisions. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth, which can solve the problems in the above-mentioned background technology, such as the scale gap between the breadth of the whole area coverage and the accuracy of individual perception, the lack of causal logic analysis ability of traditional models, and the insufficient scientific nature of irrigation decision-making in complex environments.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] The intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth includes a global sensing device, a ground monitoring network, a digital twin construction terminal, a causal inference analysis server, a generative evolution prediction device, and a water and fertilizer synergistic control terminal, as follows:

[0008] The global sensing device is used to acquire macroscopic image data of the orchard at a specific scale. By being equipped with a multispectral acquisition unit and a thermal imaging detection unit, it can achieve non-contact scanning of the reflection characteristics and thermal radiation energy of the entire orchard canopy, and extract spectral feature vectors and temperature field distribution characteristics that reflect the heterogeneity of the growth of the entire area.

[0009] The ground monitoring network is used to acquire micro-environmental parameters and physiological indicators of the location of a single fruit tree in real time. It acquires soil moisture content and nutrient concentration information by setting up soil moisture detection nodes at a preset depth in the soil layer, and collects physiological activity data of a single fruit tree using chlorophyll fluorescence sensing components.

[0010] The digital twin construction terminal connects the global sensing device and the ground monitoring network to establish a mapping relationship between global data and individual plant data. It aligns the mixed pixels in the macroscopic image data with the specific point data in the ground monitoring network through spatial coordinate transformation, and constructs a multi-dimensional digital mirror model covering the entire life cycle to achieve collaborative mapping between the overall growth of the orchard and the status of individual plants.

[0011] The causal inference analysis server interacts with the digital twin construction terminal to analyze the deep logic between environmental factors and fruit tree physiological responses. It uses causal discovery logic to perform path analysis on soil moisture, environmental temperature and humidity and fruit tree physiological activity. By eliminating the interference of irrelevant confounding variables, it determines the specific driving factors that lead to abnormal growth and outputs diagnostic results on water deficit or nutrient imbalance.

[0012] The generative evolution prediction device is used to amplify data and simulate scenario evolution for the digital mirror model. It generates virtual evolution samples under different preset climate conditions based on the acquired measured data, and uses generative logic to fill the data gaps in sparse monitoring areas, providing long-term dynamic extrapolation support for irrigation decisions.

[0013] The water and fertilizer co-control terminal is used to execute irrigation and fertilization commands based on the diagnostic results and predicted evolution trends. By adjusting the opening duration of the control valves in each zone and the output power of the pump station, it can achieve precise control of irrigation water volume and fertilizer solution ratio, so that the amount of water and fertilizer applied is precisely matched with the actual physiological needs of the fruit trees in the time and space dimensions.

[0014] Preferably, the multispectral acquisition unit in the global sensing device includes multiple narrow-band sensors for acquiring spectral energy values ​​of red light, red edge, near-infrared, and specific reflection bands, and calculating normalized difference vegetation index, enhanced vegetation index, and water stress index based on the spectral energy values ​​to characterize chlorophyll content and water balance across the entire region.

[0015] Furthermore, the thermal imaging detection unit in the global sensing device is used to capture the thermal radiation signal of the fruit tree canopy. By establishing the difference logic between the canopy temperature and the air temperature, it assesses the changes in the stomatal conductance of the crop and provides early warning of potential physiological drought risks before the visible light appearance characteristics change.

[0016] Furthermore, the soil moisture detection nodes in the ground monitoring network adopt a multi-layer structure, with high-frequency magnetic permeability sensors deployed in the core layer, deep layer and surface layer of the fruit tree root system to construct a vertical profile water migration model and monitor the logical relationship between root water absorption rate and infiltration depth in real time.

[0017] Furthermore, the chlorophyll fluorescence sensing component in the ground monitoring network uses an excitation light source of a specific wavelength to irradiate the fruit tree leaves, collects fluorescence quenching curve characteristics, and analyzes the actual quantum yield of photosystem II and non-photochemical quenching parameters to determine the severity of environmental stress on the fruit tree from the perspective of photosynthesis mechanism.

[0018] Furthermore, the digital twin construction terminal adopts a multi-scale feature fusion algorithm, which uses the low-resolution spectral features acquired by the global sensing device as global constraints and the high-precision point data acquired by the ground monitoring network as local features. The mixed pixels are decomposed through residual compensation logic to reconstruct an equivalent growth map with single-plant resolution.

[0019] Furthermore, the causal inference analysis server adopts structural equation model logic, setting soil moisture content and soil electrical conductivity as endogenous variables, and light intensity and ambient temperature as exogenous variables. By calculating the path coefficients of different variables on the net photosynthetic rate of fruit trees, the contribution rates of water stress, fertility deficiency and environmental diseases to the weakening of growth are quantified, and the wilting caused by insufficient water and the yellowing caused by diseases are distinguished.

[0020] Furthermore, the causal inference analysis server also employs dual machine learning logic to estimate the conditional average treatment effect of the specific intervention behavior of changing irrigation volume on the growth increment of fruit trees, while handling nonlinear interference factors, and to determine the optimal irrigation threshold boundary.

[0021] Furthermore, the generative evolution prediction device adopts a generative adversarial network structure, uses an encoder to extract the latent space features of the orchard's historical growth, uses a generator to simulate the evolution of fruit tree growth morphology under extreme drought, continuous rainfall, or specific temperature fluctuations, and compares and verifies the data with measured physical parameters through a discriminator to achieve accurate prediction of the orchard's water demand within a preset time period in the future.

[0022] Furthermore, the water and fertilizer co-control terminal includes a proportional fertilizer applicator and an automated pulse valve, which are used to adjust the frequency of the fertilizer injection pump in real time according to the decision parameters output by the causal inference analysis server, so that the flow rate of the mother liquor injected into the main pipeline and the real-time value fed back by the main flow sensor maintain a preset proportional relationship, and ensure the irrigation uniformity of each branch end according to the preset pressure compensation protocol.

[0023] Furthermore, the digital twin construction terminal is also equipped with time series calibration logic, which dynamically adjusts the inspection frequency of the all-domain sensing device and the data sampling cycle of the ground monitoring network according to the growth rate characteristics of fruit trees at different phenological stages, so as to realize high-frequency monitoring during key growth stages.

[0024] Furthermore, the causal inference analysis server is configured with abnormal data identification logic. When the received sensor signal deviates from the preset physical logic range, it automatically calls the cross-correlation data of the neighboring nodes for consistency verification. If it is determined to be a sensor failure, it uses the virtual estimated value generated by the generative evolution prediction device to perform logical substitution, ensuring the continuity of the decision chain.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. This invention constructs a cross-scale digital twin system by deeply integrating global sensing devices with ground monitoring networks. It solves the problem of mixed pixels caused by insufficient satellite remote sensing resolution and the problem of lack of representativeness caused by sparse ground monitoring points in traditional technologies, thereby improving the spatial accuracy and coverage of sensing.

[0027] 2. By introducing a causal inference analysis server, this invention changes the traditional decision-making model of artificial intelligence that relies solely on statistical correlation. It can deeply analyze the causal driving logic between environmental factors and crop responses, accurately identify the root causes of abnormal fruit tree growth, avoid mis-irrigation, over-irrigation, or ineffective fertilization, and improve the utilization efficiency of water and fertilizer resources.

[0028] 3. This invention, through a generative evolution prediction device, can amplify data and simulate virtual scenarios using generative AI based on limited measured data, providing a scientific basis for forward-looking irrigation decisions under complex climatic conditions and enhancing the system's robustness in responding to extreme weather changes and sudden environmental risks.

[0029] 4. This invention, through a water and fertilizer co-control terminal, transforms in-depth causal diagnostic conclusions into precise valve control logic and fertilizer-liquid ratio parameters, ensuring that water, fertilizer, and pesticides are accurately applied to the root area of ​​each fruit tree. This not only improves the yield and quality of the fruit trees but also protects the orchard's ecological environment by reducing nutrient loss, thus achieving both economic and environmental benefits.

[0030] 5. The system architecture constructed by this invention has a high degree of intelligence and automation, reducing the reliance on manual inspection and subjective judgment, and providing a standardized system solution for the large-scale application of modern smart agriculture. It has important practical significance for promoting the refinement and digital transformation of orchard management. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0032] Figure 2 This is a schematic diagram of the core principle framework of the logical analysis of environmental factors and fruit tree physiological responses based on causal inference in this invention;

[0033] Figure 3 This is a logical flowchart of the multi-scale sensing data fusion and digital twin image construction in this invention;

[0034] Figure 4 This is a schematic diagram of the principle framework of data amplification and fruit tree growth scenario evolution simulation based on generative logic in this invention;

[0035] Figure 5 This is a schematic diagram of the multi-level interaction relationship and data flow of water and fertilizer synergistic control and feedback regulation based on diagnostic results in this invention. Detailed Implementation

[0036] Example 1: Please refer to the appendix Figure 1 To be continued Figure 5 To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0037] The intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth includes a global sensing device, a ground monitoring network, a digital twin construction terminal, a causal inference analysis server, a generative evolution prediction device, and a water and fertilizer synergistic control terminal.

[0038] The global sensing device is used to acquire macroscopic image data of orchards at a specific scale. Designed as a multi-dimensional integrated sensing system mounted on a multi-rotor UAV platform or a high-altitude orbital scanning rig, it controls the synchronous triggering of multiple sensing units through a built-in embedded processor. The global sensing device includes a multispectral acquisition unit and a thermal imaging detection unit. The multispectral acquisition unit is configured with multiple independent narrow-band filters and photosensitive chips to accurately capture the reflectance characteristics of the orchard vegetation canopy under controlled lighting conditions or natural light radiation. These narrow bands typically include blue light with a center wavelength around 450 nm, green light around 550 nm, red light around 660 nm, red-edge light around 730 nm, and near-infrared light around 840 nm. By performing non-contact scanning of the spectral energy of these specific bands, the global sensing device can extract spectral feature vectors reflecting the physiological state of the vegetation.

[0039] The multispectral acquisition unit is further configured with spectral index calculation logic to calculate various parameters characterizing vegetation growth status based on the acquired spectral energy values. The system is configured with normalized difference vegetation index calculation logic, defined as the ratio obtained by dividing the difference between the reflectance values ​​of the near-infrared band and the red band by the sum of the reflectance values ​​of the near-infrared band and the red band; it is also configured with enhanced vegetation index calculation logic, which corrects for atmospheric aerosol interference by introducing blue light band energy and optimizes the calculation logic using preset gain factors and background adjustment coefficients; and it also includes water stress index calculation logic, which quantitatively assesses changes in water content in leaf tissue by analyzing the proportional relationship between the short-wave infrared band and the near-infrared band. These indices constitute the core data characteristics reflecting the heterogeneity of overall growth.

[0040] The thermal imaging detection unit is configured as a long-wave infrared focal plane array detector to capture the thermal radiation signals emitted by the fruit tree canopy and convert them into spatially distributed temperature field data. This unit integrates an absolute blackbody calibration algorithm to ensure that the acquired canopy surface temperature has high-precision physical meaning under different ambient background temperatures. The thermal imaging detection unit further assesses the stomatal conductance of the crop by establishing a logical relationship between the difference between canopy temperature and ambient air temperature. When fruit trees are subjected to water stress, stomatal closure leads to reduced transpiration and a decrease in canopy heat dissipation capacity, manifested as an increase in canopy temperature relative to ambient temperature. The global sensing device utilizes this trend in thermal radiation energy to provide early warning of potential physiological drought risks before visible pathological characteristics appear.

[0041] The ground monitoring network is used to acquire real-time micro-environmental parameters and physiological indicators of individual fruit trees. The network consists of wireless sensor nodes deployed at representative locations throughout the orchard. Each node includes a power management module, a main control microprocessor, and various high-precision sensing components. The soil moisture detection nodes employ a multi-layer probe structure, designed to be vertically inserted into the soil. High-frequency permeability sensors are deployed at preset depths of 20 cm, 40 cm, 60 cm, and 80 cm from the soil surface. These sensors emit high-frequency electromagnetic waves into the soil medium, measure the change in dielectric constant of the echo signal, and inversely determine the volumetric water content and conductivity (representing nutrient concentration) of the soil at that depth profile. This multi-layer structure can construct a complete vertical profile water migration model, monitor the logical relationship between root water absorption rate and water infiltration depth in real time, and provide direct evidence for determining whether deep soil moisture is sufficient.

[0042] The chlorophyll fluorescence sensing component in the ground monitoring network is configured as a miniaturized, non-destructive optical sensing unit, installed near typical leaves of the target fruit tree. This component illuminates the fruit tree leaves with a controlled excitation light source of a specific wavelength (such as a 650 nm red LED pulse), and a photodetector records the extremely weak fluorescence quenching curve characteristics reflected back from the leaves. The signal processing logic within the component is configured to analyze the actual quantum yield, non-photochemical quenching parameters, and photochemical efficiency of the photosystem II. Through these molecular-level physiological indicators, the system can determine the severity of environmental stresses (such as extreme heat, photoinhibition, or nutrient deficiency) experienced by the fruit tree at the level of photosynthetic mechanisms, with a sensing sensitivity far exceeding that of traditional visible light visual detection.

[0043] The digital twin construction terminal connects the global sensing device and the ground monitoring network to establish a mapping relationship between global data and individual tree data. This terminal is implemented as a high-performance graphics processing workstation or cloud computing cluster. Its core function is to precisely align the high-altitude macroscopic image data acquired by the UAV with the microscopic data of specific points acquired by the ground monitoring network in the spatiotemporal dimension through spatial coordinate transformation. The digital twin construction terminal incorporates a multi-scale feature fusion algorithm. This algorithm is configured to use the low spatial resolution but high coverage spectral features acquired by the global sensing device as global constraints, and the high precision but sparsely distributed point data acquired by the ground monitoring network as local fine features. By introducing residual compensation logic, the terminal can mathematically decompose the mixed pixels in the macroscopic image, breaking down the mixed spectral reflectance values ​​containing multiple ground features such as soil, weeds, and tree canopies into pure fruit tree canopy reflectance values ​​according to a preset physical model, and reconstructing an equivalent growth map with individual tree resolution.

[0044] The digital twin construction terminal is also equipped with time-series calibration logic, which stores standard growth curves for different fruit tree varieties throughout their entire life cycle, including budding, flowering, fruit enlargement, and ripening stages. This logic dynamically adjusts the drone inspection frequency of the all-area sensing device based on currently measured growth rate characteristics. For example, during the critical water-demanding and rapidly changing fruit enlargement stage, the system automatically increases the inspection frequency; while during the relatively slow dormant period, it reduces the sampling frequency to save energy. The digital twin construction terminal constructs a multi-dimensional digital mirror model covering the entire orchard and accurate to the individual tree, achieving a coordinated mapping between the overall macroscopic situation of the orchard and the microscopic physiological state of the individual tree.

[0045] The causal inference analysis server interacts with the digital twin construction terminal to analyze the deep logic between environmental factors and the physiological responses of fruit trees. Unlike traditional statistical models that only focus on correlations, the causal inference analysis server employs structural equation modeling logic to eliminate interference from irrelevant confounding variables. The server treats variables affected by human intervention, such as soil moisture content and soil electrical conductivity, as endogenous variables, and natural environmental factors such as light intensity, ambient temperature, and relative humidity as exogenous variables. By constructing a directed acyclic graph model, it calculates the path coefficients of different variables on the net photosynthetic rate or chlorophyll fluorescence parameters of fruit trees. This path analysis can quantify the contribution of water stress, fertility deficiency, insufficient light, and potential diseases to weakened growth. For example, when the system observes leaf wilting, the causal inference logic can distinguish whether the phenomenon is caused by insufficient water in the root zone, active stomatal closure due to high temperature (physiological self-protection), or nutrient transport interruption caused by root rot, outputting accurate diagnostic results regarding water deficit or nutrient imbalance.

[0046] The causal inference analysis server also integrates dual machine learning logic. This logic is configured to estimate the conditionally averaged treatment effect of changing irrigation volume on fruit tree growth increments (such as leaf area increase or biomass accumulation rate) by comparing data differences between the experimental group (areas receiving specific intensity of irrigation intervention) and the control group (areas not receiving intervention or maintaining the original intensity) while handling nonlinear interference factors. Through this causal effect estimation, the system can determine the optimal irrigation threshold boundary that maximizes fruit tree growth benefits under current environmental conditions, providing scientific quantitative support for subsequent decision-making.

[0047] The generative evolution prediction device is used to augment data and simulate scenario evolution in the digital mirror model. This device employs a generative adversarial network (GAN) structure, including an encoder, a generator, and a discriminator. The encoder extracts latent spatial features from the historical orchard growth data stored in the digital mirror model, compressing complex spatial distribution and temporal evolution characteristics into low-dimensional feature vectors. The generator, based on these latent spatial features and combined with preset future climate change parameters (such as a preset scenario of 10 consecutive days of rainless high temperatures), simulates the evolution of fruit tree growth patterns under different environmental stresses. This simulation is not merely a simple numerical extrapolation, but rather generates highly physically realistic virtual evolution samples through generative AI. The discriminator compares and verifies the generated virtual samples with measured physical parameters. If the virtual samples deviate from the biological laws of plant growth and development, a feedback mechanism optimizes the generator's parameters until the generated simulated evolution conforms to real-world logic.

[0048] The core application of the generative evolution prediction device lies in using generative logic to fill data gaps in sparse monitoring areas. For areas without ground sensors, the device can combine measured data from neighboring areas with macroscopic features provided by the global sensing device to generate virtual monitoring values ​​for that specific area, achieving comprehensive coverage of the entire orchard. Through this scenario evolution simulation, the system can predict the orchard's water demand trend over a preset period, providing managers with forward-looking dynamic projection support, enabling them to proactively address potential drought risks.

[0049] The water and fertilizer co-control terminal is used to execute irrigation and fertilization commands based on the diagnostic results and predicted evolution trends. Physically, the terminal includes a proportional fertilizer applicator, an automated pulse valve, a pump station driver, a main flow sensor, and a pressure-compensated drip irrigation network. The terminal receives irrigation threshold commands from a causal inference analysis server and, combined with future water demand predictions output by a generative evolution prediction device, calculates the optimal irrigation duration and fertilization ratio. The proportional fertilizer applicator is configured to adjust the motor frequency of the fertilizer injection pump in real time, ensuring that the mother liquor flow rate in the main pipeline and the real-time total flow rate value fed back by the main flow sensor always maintain a preset proportional relationship.

[0050] The water and fertilizer co-control terminal also employs a precise control mechanism based on automated pulse valves. This mechanism adjusts the opening and closing time ratio of the valves to achieve minute and precise regulation of irrigation water volume, ensuring that the capillary pressure at the end of the tube for each fruit tree is within a preset stable range, eliminating irrigation uniformity deviations caused by terrain undulations or pipeline resistance. The control terminal translates deep-seated causal diagnostic conclusions into specific physical execution signals, ensuring that the water and fertilizer application coincides temporally with the peak physiological water demand of the crop and spatially with the highly active areas of the root system, achieving efficient utilization of water and fertilizer resources.

[0051] To ensure the continuity and robustness of the entire decision-making chain, the causal inference analysis server is also equipped with anomaly data identification logic. When a signal returned by a ground-based sensor (such as soil moisture showing zero or conductivity values ​​exhibiting abnormal fluctuations) deviates from a preset physical logic range, this logic automatically calls the cross-correlation data of adjacent geographical nodes for consistency verification. If it is determined that the data anomaly is caused by sensor hardware failure or communication packet loss, the system will not incorporate this erroneous data into the decision model. Instead, it automatically calls the virtual estimate generated by the generative evolution prediction device for logical substitution. This self-healing mechanism ensures that even if some nodes fail, the system can still output scientific irrigation decisions.

[0052] Example 2: Based on Example 1, this example provides an intelligent water and fertilizer integrated irrigation system architecture based on edge computing and high-frequency dynamic sampling enhancement, which aims to further improve the system's adaptability in remote mountain orchards with complex environments and poor communication conditions.

[0053] In this embodiment, each sensor node in the ground monitoring network is endowed with enhanced edge computing capabilities. Each node has a built-in ultra-low-power neural network inference acceleration chip. The nodes are not only responsible for acquiring raw data but also capable of performing preliminary data preprocessing and feature extraction tasks locally. For example, after acquiring the fluorescence quenching curve, the chlorophyll fluorescence sensing component calculates the maximum photon yield locally and only sends the calculated key indicators to the cloud server via a narrowband IoT communication protocol. This architecture reduces the data transmission load, extends the battery life of the battery-powered sensor nodes, and enables them to undertake longer-term unmanned monitoring tasks.

[0054] In this embodiment, the digital twin construction terminal incorporates terrain compensation logic. Addressing the distortion of spectral images caused by terrain undulations in hilly orchards, the terminal utilizes a pre-constructed high-precision digital surface model to perform orthorectification and radiometric compensation on each frame of image acquired by the global sensing device. This logic eliminates uneven reflectance caused by mountain shadows or slope gradients, ensuring that fruit trees located on different slope aspects and positions can be evaluated for growth based on the same physical scale.

[0055] Furthermore, the causal inference analysis server in this embodiment adds an independent environmental background benchmark assessment module. This module is used to continuously monitor the microclimate environment around the orchard, including light intensity, wind speed, rainfall, and air quality index. When performing diagnosis, the causal inference logic automatically deducts population growth fluctuations caused by macro-climate fluctuations, accurately focusing on individual differences in individual plants caused by changes in soil moisture. For example, when encountering orchard-wide strong light inhibition, the system can determine, by comparing baseline environmental parameters, that the decrease in chlorophyll fluorescence is a general response caused by natural climate, rather than a fertility deficiency in a specific area, preventing misdiagnosis and overcompensation.

[0056] In this embodiment, the generative evolution prediction device is enhanced into a multimodal generative architecture. Besides generating virtual images of growth patterns, the device can also combine historical rainfall probability models with soil infiltration characteristics to generate dynamic groundwater migration cloud maps for specific plots. These cloud maps can display the horizontal diffusion radius and vertical infiltration depth of irrigation water in the soil. Through this visualized evolutionary simulation, the water and fertilizer co-control terminal can implement a pulsed irrigation strategy, that is, by performing multiple short-duration, high-intensity irrigation operations, it utilizes the capillary action of the soil to guide water evenly distributed in the root-dense area, reducing ineffective water seepage into deeper soil layers.

[0057] In this embodiment, the water and fertilizer co-control terminal incorporates a machine learning-based fault self-diagnosis algorithm. By monitoring the dynamic response curves between the pump station's operating current, outlet pressure, and the pressure sensor at the end of the pipeline, the system can automatically identify common hardware faults such as filter blockage, pipeline rupture, or nozzle scaling. Once an abnormality in the actuator is identified, the system immediately adjusts the pressure compensation logic, temporarily remedying the situation by increasing the operating pressure of other normal branches or extending the operation time, and simultaneously pushes a maintenance work order with precise fault location to the mobile terminal.

[0058] Example 3: In this example, the present invention proposes a distributed collaborative architecture for large-scale standardized orchard clusters. Under this architecture, multiple global sensing devices (drone swarms) can work collaboratively to perform gridded scanning of orchards covering thousands of acres using a swarm control algorithm.

[0059] The digital twin construction terminals in the distributed architecture employ a hierarchical data fusion model. Edge computing gateways are set up in the orchard's zoned management units (typically each irrigation sub-plot) to aggregate data from all ground monitoring points within that area. These gateways share data using a local area network, building localized digital twin sub-models at the edge. Only when the aggregation layer requires orchard-wide coordinated scheduling do the edge gateways upload anonymized aggregated features to the centralized causal inference analysis server. This distributed processing approach not only alleviates the communication pressure on the backbone network but also achieves millisecond-level response times for decision feedback.

[0060] In this embodiment, the causal inference analysis server incorporates a transfer learning mechanism. When the system is deployed to a new orchard or a new fruit tree variety is introduced, the server can access a global fruit tree growth causal knowledge base stored in the cloud, transferring existing logical models regarding the relationship between water, nutrients, and growth increments to the new environment, and quickly fine-tuning parameters based on initial measured data from the new environment. This capability enables the system to be scalable, shortening the system optimization cycle in newly deployed orchards and rapidly achieving precision irrigation.

[0061] In this embodiment, the generative evolutionary prediction device is configured to support a virtual experimental mode. Managers can manually set virtual intervention strategies within the system, such as how fruit sugar accumulation trends will change over the next three weeks if nitrogen fertilizer application is reduced by 20% and irrigation frequency is increased by 15%. The generative logic will run this hypothetical scenario in a virtual space based on the current digital mirror model and provide quantitative prediction results and confidence intervals. This function transforms traditional experience-based trial-and-error management into scientific simulation based on physical laws and artificial intelligence, reducing production risks.

[0062] In this embodiment, the water and fertilizer co-control terminal is deeply integrated with the water energy balance logic. Besides controlling according to the physiological needs of crops, the terminal also monitors peak and off-peak electricity prices for agricultural use. While ensuring that fruit trees are not subjected to water stress, it prioritizes large-volume water storage or irrigation operations during periods of low electricity prices. Through this cross-domain collaborative optimization, the system not only improves the quality and yield of agricultural products but also reduces the overall energy consumption cost of the orchard.

[0063] To achieve comprehensive protection of the orchard's ecological environment, the system in this embodiment is also equipped with a leaching risk early warning module. This module utilizes nutrient migration path analysis output by the causal inference analysis server to assess in real time the probability of fertilizer leaching into groundwater due to over-irrigation. When an abnormal increase in deep soil nutrient concentration is detected and the irrigation intensity exceeds the soil's water-holding capacity, the system automatically triggers a forced cutoff logic to reduce irrigation intensity or adjust the fertilizer-solution ratio, ensuring that all agricultural production activities remain within the safe boundary of the ecological carrying capacity.

[0064] In summary, the intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth constructed by this invention upgrades traditional experience-based agriculture into highly intelligent precision digital agriculture through multi-dimensional perception, cross-scale digital twins, deep causal diagnosis, and generative evolutionary prediction.

[0065] The multispectral acquisition unit in the global sensing device includes multiple narrow-band sensors for acquiring spectral energy values ​​in red light, red edge, near-infrared, and specific reflection bands. Based on these spectral energy values, it calculates the Normalized Difference Vegetation Index (NDVEI), Enhanced Vegetation Index (EDI), and Water Stress Index to characterize chlorophyll content and water balance across the entire region. The Normalized Difference Vegetation Index (NDVEI) is used in this calculation. The calculation formula is as follows:

[0066]

[0067] in, This refers to the reflectance value in the near-infrared band. This represents the reflectance value in the red light band.

[0068] The thermal imaging detection unit in the all-area sensing device is used to capture the thermal radiation signal of the fruit tree canopy. By establishing a logic for the difference between the canopy temperature and the air temperature, it assesses changes in the stomatal conductance of the crop and provides early warning of potential physiological drought risks before visible light appearance characteristics change. Specifically, the process involves obtaining the true surface temperature of the fruit tree canopy through built-in calibration logic, while environmental sensors collect the ambient air temperature in real time. When the difference between the true surface temperature of the canopy and the ambient air temperature exceeds a preset temperature fluctuation threshold and lasts for more than a preset time window, the system determines that the stomatal conductance has abnormally decreased and outputs an early warning signal to the decision-making level.

[0069] The soil moisture detection nodes in the ground monitoring network adopt a multi-layer structure, with high-frequency permeability sensors deployed in the core, deep, and surface layers of the fruit tree root system. This is used to construct a vertical profile water migration model and monitor the logical relationship between root water absorption rate and infiltration depth in real time. The high-frequency permeability sensors are configured to emit electromagnetic wave signals at a preset frequency, and the complex permittivity of the soil is calculated based on the phase shift of the echo signal to determine the soil volumetric water content.

[0070] The chlorophyll fluorescence sensing component in the ground-based monitoring network illuminates fruit tree leaves with a specific wavelength excitation light source, collects fluorescence quenching curve characteristics, and analyzes the actual quantum yield of system II and non-photochemical quenching parameters to determine the severity of environmental stress on the fruit trees from the perspective of photosynthetic mechanisms. The determination logic is configured such that when the observed actual quantum yield is lower than a preset proportion of the normal growth baseline value, and the non-photochemical quenching parameters increase, the system determines that the fruit trees are under environmental stress and automatically increases the monitoring sampling frequency in the corresponding area.

[0071] The digital twin construction terminal employs a multi-scale feature fusion algorithm. It uses the low-resolution spectral features acquired by the global sensing device as global constraints and the high-precision point data acquired by the ground monitoring network as local features. Residual compensation logic decomposes the mixed pixels to reconstruct an equivalent growth map with single-plant resolution. The residual compensation logic is configured to calculate the deviation between the interpolation result of the global spectral features at a specific coordinate location and the measured value from the ground-based sensor at that location, and to perform weighted correction in local areas based on the deviation. This ensures that the reconstructed map maintains global trend consistency while possessing detailed representation capabilities at the single-plant level.

[0072] The causal inference analysis server employs structural equation modeling logic, treating soil moisture content and soil electrical conductivity as endogenous variables, and light intensity and ambient temperature as exogenous variables. By calculating the path coefficients of different variables on the net photosynthetic rate of fruit trees, it quantifies the contribution rates of water stress, fertility deficiency, and environmental diseases to weakened growth, distinguishing between wilting caused by insufficient water and yellowing caused by disease. The calculation of these path coefficients follows a pre-defined causal structure diagram, determining the weight of each path by minimizing the difference between the observed covariance matrix and the model's predicted covariance matrix. When the path weight corresponding to water stress accounts for a proportion of the total contribution rate exceeding a pre-defined significance threshold, the system prioritizes outputting irrigation recommendations.

[0073] The causal inference analysis server also employs dual machine learning logic. Under the premise of handling nonlinear interference factors, it estimates the conditional average treatment effect of a specific intervention—changing irrigation volume—on the increase in fruit tree growth, and determines the optimal irrigation threshold boundary. This logic is configured to use a first machine learning model to predict irrigation behavior through environmental variables, and a second machine learning model to predict growth increment through environmental variables. It then calculates the regression relationship between the residuals of the irrigation behavior and the residuals of the growth increment, eliminating interference from background variables and locking in the true causal effect size between irrigation intervention and growth response.

[0074] The generative evolution prediction device employs a generative adversarial network (GAN) structure. It utilizes an encoder to extract latent space features of historical orchard growth, and a generator to simulate the evolution of fruit tree growth morphology under extreme drought, continuous rainfall, or specific temperature fluctuations. The results are then compared and verified with measured physical parameters by a discriminator to achieve accurate predictions of orchard water demand within a preset future timeframe. The prediction process includes inputting predicted weather forecast data into the trained generator to generate a digital state map of the orchard over a continuous future time series. Finally, it performs an integral calculation on the water consumption of each pixel in the map to obtain the total estimated water demand.

[0075] The water and fertilizer co-control terminal includes a proportional fertilizer applicator and an automated pulse valve. Based on the decision parameters output by the causal inference analysis server, it adjusts the frequency of the fertilizer injection pump in real time to maintain a preset proportional relationship between the mother liquor flow rate in the main pipeline and the real-time value fed back by the main flow sensor. It also ensures irrigation uniformity at the ends of each branch pipeline according to a preset pressure compensation protocol. This proportional relationship is configured to be dynamically adjusted according to the current fertilizer concentration requirement. When the value of the main flow sensor fluctuates, the motor frequency of the fertilizer injection pump is compensated in real time using a preset proportional-integral-derivative algorithm to ensure that the fluctuation range of the mother liquor addition is below a preset tolerance range.

[0076] The digital twin construction terminal is also equipped with time-series calibration logic, which dynamically adjusts the inspection frequency of the all-domain sensing device and the data sampling cycle of the ground monitoring network according to the growth rate characteristics of fruit trees at different phenological stages, achieving high-frequency monitoring during key growth stages. This calibration logic stores a phenological characteristic database. When the detected leaf area index increase rate reaches a first preset slope, the system automatically determines that the tree has entered the vigorous growth stage and adjusts the sampling frequency to a high-frequency mode. When the detected growth characteristics tend to level off and meet a second preset slope, the system determines that the tree has entered the dormant or late-maturity stage and adjusts the sampling frequency to a low-power, low-frequency mode.

[0077] The causal inference analysis server is configured with abnormal data identification logic. When the received sensor signal deviates from the preset physical logic range, it automatically calls the cross-correlation data of neighboring nodes for consistency verification. If it is determined to be a sensor malfunction, a virtual estimate generated by a generative evolution prediction device is used for logical substitution to ensure the continuity of the decision-making chain. The physical logic range includes the dynamic range of the sensor output value, the limit of the rate of change, and the physical constraints with other environmental parameters. For example, if soil moisture changes drastically without rainfall or irrigation, the logic determines it to be abnormal.

[0078] During system execution, various modules exchange data in real time via industrial Ethernet, wireless sensor networks, and cloud API interfaces. Image data collected by the global sensing devices is compressed and uploaded to the digital twin construction terminal for preprocessing, while data from the ground monitoring network is aggregated through a gateway. The causal inference analysis server periodically retrieves the latest digital image model from the digital twin construction terminal and, combined with meteorological forecast data, calls the generative evolutionary prediction device for forward-looking extrapolation. The resulting irrigation decision command is then issued to the water and fertilizer co-control terminal. The actuators perform physical operations according to the command, and pressure and flow sensors feed back the executed field parameters to the respective analysis servers, forming a closed-loop adaptive management and control chain.

[0079] Those skilled in the art should understand that in this invention, all the calculation processes involved, such as the determination of path coefficients, the extraction of feature vectors, the residual compensation of pixel decomposition, and the adjustment of scaling coefficients, have been transformed into specific logical steps executed by the server, terminal, and processor. These steps together constitute the functional architecture of the system.

[0080] The physical deployment of the system's components is flexible and diverse. In some implementation scenarios, the control module of the all-domain perception device can be deployed on a drone ground station, while the core causal inference server is deployed on a remote public cloud platform with powerful computing resources. In other application scenarios with higher requirements for privacy and security, all computing logic can be integrated into a private server cluster deployed within the orchard, communicating with distributed ground-based sensor nodes and water and fertilizer control terminals via a local area network to achieve fully enclosed intelligent management.

[0081] In this invention, the sensor nodes can be non-uniformly distributed according to the actual terrain and row spacing of fruit trees. In areas with significant soil texture heterogeneity, the system is equipped with logic to suggest increasing the density of sensor nodes. By analyzing the coefficient of variation of soil conductivity distribution, it automatically prompts managers to add soil moisture detection nodes in specific areas. This data-driven optimization of sensor placement further ensures the accuracy of the digital mirror model in representing the real physical world.

Claims

1. An intelligent integrated water and fertilizer irrigation system based on soil moisture and fruit tree growth, characterized in that, This includes all-domain sensing devices, ground monitoring networks, digital twin construction terminals, causal inference analysis servers, generative evolution prediction devices, and water and fertilizer synergistic control terminals; The global sensing device is used to acquire macroscopic image data of the orchard at a specific scale, and to extract spectral feature vectors and temperature field distribution characteristics that reflect the heterogeneity of growth across the entire area. The ground monitoring network is used to acquire micro-environmental parameters and physiological indicators of the location of individual fruit trees in real time, and to acquire information on soil moisture content, nutrient concentration and physiological activity data. The digital twin construction terminal connects the global sensing device and the ground monitoring network, and is used to align the mixed pixels in the macroscopic image data with the point data in the ground monitoring network through spatial coordinate transformation, so as to construct a multi-dimensional digital mirror model covering the entire life cycle. The causal inference analysis server interacts with the digital twin construction terminal to perform path analysis on environmental factors and fruit tree physiological responses using causal discovery logic, determine the specific driving factors that lead to abnormal growth, and output diagnostic results on water deficiency or nutrient imbalance. The causal inference analysis server also integrates dual machine learning logic, which is used to estimate the conditional average treatment effect of the intervention behavior of changing irrigation amount on the growth increment of fruit trees by comparing the data differences between the experimental group and the control group, under the premise of processing nonlinear interference factors. The dual machine learning logic includes a first machine learning model and a second machine learning model, wherein: the first machine learning model is used to predict irrigation behavior through environmental variables; The second machine learning model is used to predict growth increment through environmental variables; the causal inference analysis server is used to calculate the regression relationship between the residual of the irrigation behavior and the residual of the growth increment, thereby eliminating the interference of background variables and determining the optimal irrigation threshold boundary that maximizes the growth benefits of fruit trees under the current environmental conditions. The generative evolution prediction device is used to amplify data and simulate scenario evolution of the multidimensional digital mirror model, generating virtual evolution samples under different preset climate conditions; the generative evolution prediction device adopts a generative adversarial network structure, including an encoder, a generator, and a discriminator; The encoder is used to extract the latent spatial features of the orchard historical growth data stored in the digital mirror model, and compress the spatial distribution and temporal evolution features into feature vectors. The generator is used to simulate the evolution of fruit tree growth patterns under different environmental stresses based on the feature vector and in combination with preset future climate change parameters, and to generate virtual evolution samples. The discriminator is used to compare and verify the generated virtual evolution samples with the measured physical parameters; The generative evolution prediction device is used to input the predicted meteorological forecast data into the generator to generate a digital state map of the orchard over a future continuous time series, and to accumulate the water consumption of each pixel in the map to obtain the total estimated water demand. The causal inference analysis server is also configured with abnormal data identification logic. When the sensor data is determined to be abnormal, the virtual estimated value generated by the generative evolution prediction device is automatically called for logical replacement. The water and fertilizer co-control terminal is used to execute irrigation and fertilization commands based on the diagnostic results and predicted evolution trends. By adjusting the opening duration of the control valves in each zone and the output power of the pump station, it can regulate the irrigation water volume and fertilizer solution ratio.

2. The intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth according to claim 1, characterized in that, The global sensing device includes a multispectral acquisition unit and a thermal imaging detection unit; The multispectral acquisition unit includes multiple independent narrow-band filters and photosensitive chips, used to capture the spectral energy values ​​of the blue light band with a center wavelength of around 450 nm, the green light band with a center wavelength of around 550 nm, the red light band with a center wavelength of around 660 nm, the red edge band with a center wavelength of around 730 nm, and the near-infrared band with a center wavelength of around 840 nm. The multispectral acquisition unit is equipped with spectral index calculation logic, which is used to calculate the normalized difference vegetation index, enhanced vegetation index and water stress index based on the spectral energy value. Wherein, the normalized difference vegetation index is equal to the ratio obtained by dividing the difference between the reflectance value of the near-infrared band and the reflectance value of the red band by the sum of the reflectance values ​​of the near-infrared band and the red band. The enhanced vegetation index is configured to introduce the blue light band energy to correct atmospheric aerosol interference and to optimize it using a preset gain factor and background adjustment coefficient. The water stress index is configured to quantitatively assess changes in water content in leaf tissue by analyzing the ratio between the short-wave infrared band and the near-infrared band.

3. The intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth according to claim 2, characterized in that, The thermal imaging detection unit includes a long-wave infrared focal plane array detector, which is used to capture the thermal radiation signal emitted by the fruit tree canopy and convert it into spatially distributed temperature field data. The thermal imaging detection unit integrates an absolute blackbody calibration algorithm to obtain the canopy surface temperature under different ambient background temperatures. The thermal imaging detection unit assesses the stomatal conductance of crops by establishing a logical relationship between the difference between canopy temperature and ambient air temperature. The global sensing device is equipped with early warning logic. When the difference between the actual temperature of the canopy surface and the ambient air temperature collected in real time by the environmental sensor is greater than the preset temperature fluctuation threshold and the duration exceeds the preset time window, the system determines that the stomatal conductance has decreased and outputs an early warning signal to the decision-making level.

4. The intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth according to claim 3, characterized in that, The ground monitoring network includes multiple wireless sensor nodes deployed at orchard locations, each of which includes a power management module, a main control microprocessor, and sensing components. The sensing component includes a soil moisture detection node with a multi-layer probe structure. The soil moisture detection node is vertically inserted into the soil and high-frequency magnetic permeability sensors are deployed at preset depths of 20 cm, 40 cm, 60 cm and 80 cm from the soil surface. The high-frequency permeability sensor is used to transmit high-frequency electromagnetic waves into the soil medium. By measuring the change in dielectric constant of the echo signal, the volumetric water content of the soil at that depth profile and the conductivity representing nutrient concentration information are obtained. The ground monitoring network is equipped with a vertical profile water migration model, which is used to monitor the logical relationship between root water absorption rate and water infiltration depth in real time, providing a direct basis for judging the sufficiency of deep water.

5. The intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth according to claim 4, characterized in that, The sensing component also includes a chlorophyll fluorescence sensing component, which is configured as an optical sensing unit installed near the leaves of the target fruit tree. The chlorophyll fluorescence sensing component includes a wavelength-controlled excitation light source and a photodetector. The controlled excitation light source is used to irradiate fruit tree leaves with a 650 nm red light pulse, and the photodetector is used to record the fluorescence quenching curve characteristics reflected back from the leaves. The signal processing logic integrated within the chlorophyll fluorescence sensing component is used to analyze the actual quantum yield, non-photochemical quenching parameters, and photochemical efficiency of the photosystem II. The ground monitoring network is equipped with stress determination logic. When the observed actual quantum yield is lower than a preset ratio of the normal growth benchmark value, and the non-photochemical quenching parameter rises to a preset upper limit, the system determines that the fruit tree is under environmental stress and automatically increases the monitoring sampling frequency of the corresponding area.

6. The intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth according to claim 5, characterized in that, The digital twin construction terminal has a built-in multi-scale feature fusion algorithm. The multi-scale feature fusion algorithm is used to take the spectral features acquired by the global sensing device as global constraints and the point data acquired by the ground monitoring network as local fine features. The digital twin construction terminal is equipped with residual compensation logic, which is used to calculate the deviation between the interpolation result of the global spectral features at the coordinate position and the measured value of the ground sensor at the coordinate position, and to perform weighted correction in the local area based on the deviation, to perform mathematical decomposition on the mixed pixels in the macro image, to decompose the mixed spectral reflectance value containing soil, weeds and canopy features into pure fruit tree canopy reflectance value, and to reconstruct an equivalent growth map with single tree resolution; The digital twin construction terminal is also equipped with time series calibration logic, which stores standard growth curves of different varieties of fruit trees during the budding, flowering, fruit enlargement and ripening stages, and dynamically adjusts the inspection frequency of the global sensing device and the data sampling cycle of the ground monitoring network according to the measured growth rate characteristics.

7. The intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth according to claim 6, characterized in that, The causal inference analysis server uses structural equation modeling logic to construct a directed acyclic graph model, and sets soil moisture content and soil electrical conductivity as endogenous variables, and light intensity, ambient temperature and relative humidity as exogenous variables. The causal inference analysis server is used to calculate the path coefficients of different variables on the net photosynthetic rate or chlorophyll fluorescence parameters of fruit trees. The weight of each path is determined by minimizing the difference between the observed correlation distribution matrix between variables and the correlation distribution matrix predicted by the model, thereby quantifying the contribution of water stress, fertility deficiency, insufficient light and potential diseases to the weakening of growth. The causal inference analysis server is also equipped with an environmental background benchmark assessment module, which is used to continuously monitor the micro-meteorological environment around the orchard and automatically deduct the population growth fluctuations caused by the fluctuations of the macro-climate during diagnosis, so as to distinguish between wilting caused by insufficient water and yellowing caused by disease.

8. The intelligent water and fertilizer integrated irrigation system based on soil moisture and fruit tree growth according to claim 7, characterized in that, The water and fertilizer co-control terminal includes a proportional fertilizer applicator, an automated pulse valve, a pump station driver, a main flow sensor, and a pressure-compensated drip irrigation network. The water and fertilizer co-control terminal is used to receive irrigation threshold instructions from the causal inference analysis server and, in conjunction with the future water demand prediction output by the generative evolution prediction device, calculate irrigation duration and fertilizer ratio. The proportional fertilizer applicator is used to adjust the motor frequency of the fertilizer injection pump in real time, so that the flow rate of the mother liquor injected into the main pipeline and the real-time total flow rate value fed back by the main flow sensor always maintain a preset proportional relationship. The motor frequency of the fertilizer injection pump is compensated in real time by a preset proportional-integral-derivative algorithm to ensure that the fluctuation range of the mother liquor addition is lower than the preset tolerance range. The automated pulse valve is used to regulate the irrigation water volume by adjusting the ratio of valve opening and closing time, and to ensure the irrigation uniformity at the end of each branch according to a preset pressure compensation protocol. The water and fertilizer co-control terminal is also equipped with a leaching risk early warning module, which is used to analyze and assess the risk probability of fertilizer leaching into groundwater due to over-irrigation by using nutrient migration path analysis, and to trigger a forced cutoff logic when the safety boundary is exceeded.

Images