Intelligent irrigation system for fruit tree planting greenhouse

Abstract

The invention relates to the technical field of intelligent irrigation, and discloses an intelligent irrigation system for a fruit tree planting greenhouse. The system comprises a multi-dimensional data acquisition module, a data fusion processing module, an irrigation decision module and an irrigation control execution module which are connected in sequence. The system obtains a root soil image, a leaf surface temperature distribution map and stem micro-deformation data, and performs multi-dimensional fusion calculation to generate a spatial water potential distribution map capable of representing the overall moisture condition of crops. And calculating specific regional water demand and preferential irrigation sites based on the map, and further driving an execution terminal to realize precise irrigation. According to the method, the physiological water demand signal of the crop can be responded, the spatial difference of water stress can be identified, the limitation that the traditional technology depends on single soil information is overcome, the crossing from environment monitoring to crop body sensing is realized, and the irrigation accuracy and the water resource utilization efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent irrigation technology, specifically to an intelligent irrigation system for fruit tree greenhouses. Background Technology

[0002] In existing greenhouse fruit tree cultivation, irrigation decisions primarily rely on monitoring soil environmental parameters. This approach uses soil moisture as the sole or primary basis for determining crop water requirements, triggering irrigation by setting humidity thresholds. Some more advanced systems incorporate meteorological data or crop growth models for supplementary assessment.

[0003] Conventional technologies relying on soil information have limitations. Soil moisture only reflects the state of the water supply side and cannot directly and in real-time reveal the crop's own physiological water requirements. In the early stages of water stress, physiological processes such as root water absorption, stem water conduction, and leaf transpiration change, but this crucial information is severely overlooked under traditional technological frameworks. This leads to irrigation timing that may lag behind the crop's actual needs, or irrigation volume that does not match actual consumption, failing to achieve precise, on-demand supply.

[0004] Current technologies need to address how to move beyond a single dimension of soil information and directly and comprehensively perceive the overall water physiological state of crops. Simultaneously, they need to solve the problem of effectively integrating and analyzing physiological sensory information from different parts and types of crops to form a decision-making basis that can directly guide precision spatial operations, thereby achieving direct response and precise regulation of the crop's own water needs. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent irrigation system for fruit tree greenhouses to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent irrigation system for fruit tree greenhouses, the system comprising:

[0007] The multidimensional data acquisition module acquires multidimensional growth data of crops during their growth cycle, including root soil images, leaf temperature distribution maps, and stem microdeformation sensing data.

[0008] The data fusion processing module performs multi-dimensional fusion calculations on the multi-dimensional growth data to generate a spatial water potential distribution map of the crop. The spatial water potential distribution map is composed of a root region water potential matrix, a leaf region water potential matrix, and a stem conduction water potential matrix.

[0009] The irrigation decision module performs irrigation decision calculations based on the spatial water potential distribution map. The output of the irrigation decision calculation includes a set of irrigation instructions for regional water demand and the coordinates of priority irrigation sites.

[0010] The irrigation control execution module inputs the irrigation instruction set to the irrigation control network, which parses the irrigation instruction set into an executable set of drive signals based on the physical layout of the irrigation execution terminals in the greenhouse.

[0011] Preferably, the step of performing multi-dimensional fusion calculations on the multi-dimensional growth data to generate a spatial water potential distribution map of the crop includes the following sub-steps:

[0012] The root soil image is subjected to texture segmentation and humidity inversion calculation to extract the actual water content and water migration rate of each soil layer, forming a dynamic profile of root water.

[0013] Simultaneously, thermal infrared feature analysis is performed on the leaf surface temperature distribution map. Combined with the preset leaf surface stomatal aperture model, the leaf surface transpiration water loss rate and leaf surface cell water potential value are inverted to generate a leaf surface water potential state map.

[0014] Simultaneously, by analyzing the micro-deformation sensing data of the stem, and calculating the micron-level periodic changes in the stem diameter, the water column tension data of the xylem inside the stem and the water flow of the vascular bundles are derived, and a water conduction status diagram of the stem is constructed.

[0015] The root water dynamic profile, the leaf water potential state map, and the stem water conduction state map are registered and superimposed in a unified crop three-dimensional spatial coordinate system.

[0016] Based on the principle of continuous water transport among roots, stems and leaves, water potential balance iterative calculations are performed on the registered root water dynamic profile, leaf water potential state map and stem water conduction state map until the root-stem-leaf water potential gradient reaches stability.

[0017] The root region water potential matrix, the leaf region water potential matrix, and the stem conduction water potential matrix, after being output and stabilized through iteration, together constitute the spatial water potential distribution map.

[0018] Preferably, the step of performing texture segmentation and humidity inversion calculation on the root soil image specifically includes:

[0019] The image of the root soil was acquired using dual-spectral information in the visible and near-infrared bands.

[0020] Image registration and fusion of dual-spectral information enhances the texture contrast of soil particles and pores;

[0021] The watershed segmentation algorithm was used to process the fused image to distinguish the soil aggregate region, the pore region and the root contact region;

[0022] For each segmented region, based on its near-infrared spectral reflectance, the pre-established calibration curve between soil spectral reflectance and volumetric water content is queried to retrieve the volumetric water content of each region.

[0023] The water migration rate is calculated by tracking the movement speed of the water front in specific soil pores using continuous time-series images of the root soil.

[0024] Preferably, the step of tracking the movement speed of the water front in specific soil pores using continuous time-series images of the root soil and calculating the water migration rate involves the following steps:

[0025] In a continuous sequence of root soil images, a representative soil pore area is selected as the observation area.

[0026] By using image difference technology, the gray value changes of the observed area in images at adjacent time points are compared to identify the image dark area expansion front caused by water infiltration;

[0027] The positional movement of the image dark area expansion front in the pixel coordinate system is tracked, and the pixel movement distance is converted into actual distance by combining the time interval of image capture and the spatial calibration parameters of the image.

[0028] The actual distance the image dark area expansion front moves per unit time is calculated, which is the water migration rate at the soil pore scale;

[0029] The above tracking and calculation process was repeated for multiple soil pores of different sizes and locations. Finally, the average value of the water migration rate of all pores was taken as the representative water migration rate of the soil layer.

[0030] Preferably, the step of performing irrigation decision calculations based on the spatial water potential distribution map includes the following processing steps:

[0031] Read the water potential matrix of the root system region, identify all grid cells whose water potential value is lower than the critical water potential threshold of the root system, and mark the center coordinates of these grid cells as primary water-deficient points;

[0032] Obtain the water potential matrix of the leaf surface region, find the region where the leaf surface water potential value is lower than the stomatal closure water potential threshold, calculate the geometric centroid coordinates and area percentage of the region, and mark it as the leaf stress region;

[0033] The stem conduction water potential matrix is ​​retrieved, and the locations of abnormal abrupt changes in water potential gradient are analyzed. The coordinates of stem nodes where the water potential gradient exceeds the normal conduction threshold are marked as potential vascular blockage points.

[0034] A multi-objective decision-making model is established with the primary water shortage point as the irrigation target, the leaf stress area as the stress correction factor, and the potential obstruction point of the vascular bundle as the conduction constraint.

[0035] In the multi-objective decision-making model, the distance between each primary water shortage point and the water shortage severity of the soil layer in which it is located is calculated, and a comprehensive score is given by combining the influence of the leaf stress area on its transpiration pull and the degree of obstruction of the potential blockage point of the xylem on its water transport.

[0036] Based on the comprehensive scoring results, all the primary water shortage points are sorted, and the points with scores higher than the irrigation initiation threshold are selected as the coordinates of the priority irrigation sites.

[0037] Based on the soil layer properties and comprehensive scores corresponding to the coordinates of each priority irrigation site, the amount of water required to restore the water potential of the site to a safe value is calculated, and the water demand of the sub-regions is obtained by summarizing.

[0038] Preferably, the calculation process of establishing a multi-objective decision model with the primary water shortage point as the irrigation target, the leaf stress region as the stress correction factor, and the potential obstruction point of the vascular bundle as the conduction constraint includes:

[0039] A decision vector is established for each of the primary water-deficient points. The dimensions of the decision vector include the water potential value of the water-deficient point itself, the pipeline path length to the nearest water source, and the type of the soil layer to which it belongs.

[0040] By introducing the area ratio and centroid coordinates of the leaf stress region, the transpiration pull weight of the leaf stress region on each primary water shortage point is calculated. The water shortage point that is closer to the leaf stress region has a higher transpiration pull weight coefficient.

[0041] The coordinates of the potential blockage points of the conduit are introduced, and the resistance coefficients to the upward transport of water at each of the primary water shortage points are calculated. The resistance coefficients are determined based on the increase in hydraulic resistance of the conduit path between the blockage point and the water shortage point.

[0042] The transpiration pull weighting coefficient is used as a positive incentive factor, and the resistance coefficient is used as a negative constraint factor, and they are weighted and fused with the decision vector.

[0043] The comprehensive score for each point is obtained by solving an objective function that achieves the most balanced improvement in the overall water shortage situation of all the primary water shortage points.

[0044] Preferably, after the multi-objective decision model is run, the following model feedback correction step is also performed:

[0045] After irrigation is completed, a new spatial water potential distribution map is collected and generated again.

[0046] By comparing the root zone water potential matrix before and after irrigation, the actual water potential increase value of each priority irrigation site coordinate is calculated.

[0047] The actual water potential increase is compared with the increase predicted by the multi-objective decision model to obtain the water potential correction error at the point.

[0048] Based on the water potential correction error at all points, adjust the key parameters in the multi-objective decision model used to calculate the regional water demand. The key parameters include the soil moisture diffusion coefficient and the crop water absorption resistance coefficient.

[0049] The adjusted key parameters are used to update the multi-objective decision model for the next irrigation decision calculation.

[0050] Preferably, inputting the irrigation instruction set into the irrigation control network includes the following steps:

[0051] The irrigation control network includes a pre-recorded 3D map of the greenhouse and a spatial coordinate database of all irrigation execution terminals;

[0052] The coordinates of the priority irrigation sites are matched with the three-dimensional map of the greenhouse to determine one or more target irrigation execution terminals covering each of the priority irrigation site coordinates;

[0053] Based on the water demand of the different regions, and combined with the water pressure loss model of the irrigation network and the rated flow of each target irrigation execution terminal, a specific irrigation water volume and irrigation duration are allocated to each target irrigation execution terminal;

[0054] Considering the overall efficiency of irrigation operations, the scheduling optimization of the start-up time of multiple target irrigation execution terminals is carried out to avoid drastic fluctuations in pipeline pressure, and an operation sequence table containing terminal identifier, water volume, duration and start-up time is generated.

[0055] The operation timing table is encoded into the drive signal set, which includes a pulse signal sequence for controlling the opening and closing of the solenoid valve and an analog control signal for adjusting the speed of the variable frequency water pump.

[0056] Preferably, the method for scheduling and optimizing the start-up times of multiple target irrigation execution terminals includes:

[0057] The main optimization objective is to minimize the total irrigation operation time, with the hard constraint being to avoid the pressure at any pipeline node falling below the minimum working pressure.

[0058] Obtain the topology, pipe diameter data, and friction coefficient of each pipe segment of the irrigation network, and establish a hydraulic calculation model of the network.

[0059] The planned flow rates of all target irrigation execution terminals in the operation sequence table are used as inputs to simulate the network pressure distribution under different start-up sequences in the network hydraulic calculation model.

[0060] A heuristic search algorithm is used to find the startup order and delay interval that minimize the time span for all terminals to complete the irrigation task, while satisfying the pressure constraints.

[0061] Based on the optimized startup sequence and delay interval, the precise startup time of each target irrigation execution terminal in the operation timing table is finally determined.

[0062] Preferably, a heuristic search algorithm is used to find the starting order and delay interval that minimizes the time span for all terminals to complete the irrigation task, under the condition of satisfying pressure constraints. This is specifically achieved in the following way:

[0063] Establish an optimization objective function with the goal of minimizing the total irrigation time span;

[0064] Set hydraulic constraints for the pipeline network, requiring that the pressure value of each pipeline node during irrigation is never lower than the preset minimum working pressure threshold.

[0065] Genetic algorithm was chosen as the heuristic search algorithm, and the algorithm parameters were initialized, including population size, crossover probability, and mutation probability.

[0066] An initial population is randomly generated, and each individual represents a combination of the start order of irrigation execution terminals and the delay interval between adjacent starts;

[0067] An iterative search is performed to simulate the pipeline pressure distribution under each combination in each generation of the population and to calculate the objective function value.

[0068] The fitness of individuals is evaluated based on the objective function value and the constraint satisfaction, and a new generation of population is generated through selection, crossover and mutation operations.

[0069] Repeat the iterative process until the convergence condition is met, and output the startup order and delay interval corresponding to the individual with the highest fitness as the optimization result.

[0070] Compared with the prior art, the beneficial effects of the present invention are:

[0071] By integrating leaf surface temperature distribution maps and stem microdeformation sensing data, and synchronizing them with root soil images, a multi-source direct sensing of crop water status is achieved. Leaf surface temperature distribution, through infrared thermal imaging, reflects the intensity and uniformity of canopy transpiration, directly indicating areas of stomatal closure and temperature rise caused by water deficit. Stem microdeformation sensing captures the minute expansion and contraction of stem diameter caused by water transport in the vascular bundles, reflecting in real time the dynamics and resistance of water transport from the roots to the canopy. This technical approach shifts the focus from solely monitoring the soil environment to simultaneously monitoring key physiological signals of the crop itself, enabling earlier and more direct detection of water stress and transport anomalies within the plant, providing direct evidence from the crop itself for irrigation decisions.

[0072] Multi-source heterogeneous data were collected and fused in multiple dimensions to generate a spatial water potential distribution map composed of root region water potential matrix, leaf region water potential matrix, and stem conduction water potential matrix. This calculation process quantifies and spatially maps images, temperature fields, and deformation signals with different physical meanings into water potential values ​​that characterize the water potential energy state, respectively depicting the water availability at the soil-root interface, the transpiration driving force at the canopy-atmosphere interface, and the water transport efficiency within the stem. This map integrates discrete sensing information into a holistic model reflecting the water movement potential energy and bottlenecks within the crop's three-dimensional space. Based on the regional water demand and priority irrigation site coordinates output by this model, irrigation commands can accurately target specific spatial locations of water potential imbalance or water conduction obstruction, realizing a leap from large-area uniform irrigation or regional irrigation based on soil moisture to targeted regulation based on spatial differences in the water status within the crop. Attached Figure Description

[0073] Figure 1 This is a timing diagram of the intelligent irrigation system for fruit tree planting greenhouses described in this invention;

[0074] Figure 2 A flowchart for processing root soil images;

[0075] Figure 3 A flowchart for calculating a multi-objective decision model;

[0076] Figure 4 A graph showing the correlation between irrigation terminal start-up time distribution and pipeline network load;

[0077] Figure 5 A comparison chart of performance scores across various dimensions before and after optimization of the intelligent irrigation system for fruit tree greenhouses. Detailed Implementation

[0078] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0079] Please see Figure 1 This invention provides an intelligent irrigation system for fruit tree greenhouses. The system includes: First, a multi-dimensional data acquisition module acquires multi-dimensional growth data of the crop during its growth cycle. This multi-dimensional growth data includes root soil images acquired by visible light and near-infrared cameras, leaf temperature distribution maps acquired by an infrared thermal imager, and stem micro-deformation sensing data acquired by a micro-strain sensor attached to the stem. Subsequently, a data fusion processing module performs multi-dimensional fusion calculations on the acquired multi-dimensional growth data. This calculation process integrates information from different parts of the roots, stems, and leaves. Through a series of image processing, model inversion, and physical calculations, a spatial water potential distribution map characterizing the overall water status of the crop is finally generated. This map is composed of a root region water potential matrix, a leaf region water potential matrix, and a stem conduction water potential matrix. An irrigation decision module performs irrigation decision calculations based on the generated spatial water potential distribution map. This calculation analyzes the water potential matrix of each region, identifies water shortage points and stress conditions, and outputs an irrigation instruction set containing regional water requirements and coordinates of priority irrigation sites through a decision model. Finally, the irrigation control execution module receives the irrigation instruction set. Based on the pre-stored 3D map of the greenhouse and the layout of the irrigation terminals, the irrigation control network parses the abstract instruction set into a specific set of drive signals that can drive the actuators, thereby controlling the corresponding irrigation execution terminals to perform precise irrigation operations.

[0080] Example 1: Texture segmentation and moisture inversion calculations were performed on root soil images to extract the actual water content and water migration rate of each soil layer, forming a dynamic profile of root moisture. Simultaneously, thermal infrared feature analysis was performed on leaf temperature distribution maps. Combined with a pre-defined stomatal aperture model, the leaf transpiration rate and leaf cell water potential were inverted to generate a leaf water potential state map. Simultaneously, stem micro-deformation sensor data were analyzed. By calculating the micrometer-level periodic changes in stem diameter, the water column tension data of the xylem and the water flow of the xylem vessels inside the stem were derived, constructing a stem water conduction state map. The dynamic profile of root moisture, the leaf water potential state map, and the stem water conduction state map were registered and overlaid in a unified crop three-dimensional spatial coordinate system. Based on the principle of continuous water transport among roots, stems, and leaves, iterative calculations of water potential balance are performed on the registered root water dynamic profile, leaf water potential state map, and stem water conduction state map. This calculation simulates the dynamic equilibrium process of water absorption from the soil through the roots, transport through the stem xylem to the leaves, and transpiration loss, until the root-stem-leaf water potential gradient reaches stability. The iteratively stabilized root region water potential matrix, leaf region water potential matrix, and stem conduction water potential matrix are output, which together constitute the spatial water potential distribution map.

[0081] In practical implementation, taking the afternoon irrigation decision cycle of cherry trees grown in a greenhouse as an example, multidimensional growth data is fused and calculated to generate a spatial water potential distribution map. The multidimensional data acquisition module simultaneously acquires a set of multidimensional growth data, including soil images of the target fruit tree roots, leaf temperature distribution maps, and stem micro-deformation sensing data. The data fusion processing module starts the processing flow, performs texture segmentation and humidity inversion calculation on the root soil image, extracts the actual water content and water migration rate of each soil layer in the 0-30 cm soil layer of the main root zone of the fruit tree, and forms a dynamic profile map of root water using contour lines to represent the soil moisture status at different depths. Simultaneously, thermal infrared feature analysis is performed on the leaf temperature distribution map, and combined with the preset leaf stomatal aperture model, the leaf transpiration water loss rate and leaf cell water potential value are inverted to generate a leaf water potential state map marked with different water potential intervals. Simultaneously, by analyzing the micro-deformation sensing data of the stem, and calculating the micron-level periodic changes in the stem diameter, the water column tension data of the xylem inside the stem and the water flow of the vascular bundles were derived, and a stem water conduction status map reflecting the relationship between water potential and flow at different nodes of the stem was constructed.

[0082] In practice, the root water dynamic profile, leaf water potential state map, and stem water conduction state map are registered and overlaid in a unified crop three-dimensional spatial coordinate system. This unified coordinate system is established with the base of the fruit tree trunk as the origin, and the registration process transforms the spatial coordinates of data from different sources to this unified coordinate system. Based on the principle of continuous water transport among roots, stems, and leaves, water potential balance iterative calculations are performed on the registered root water dynamic profile, leaf water potential state map, and stem water conduction state map. These iterative calculations are based on the soil-plant-atmosphere continuum theory, updating the water potential estimates for the root region, stem nodes, and leaf units in each iteration step. The iterative calculations continue until the root-stem-leaf water potential gradient reaches stability. A convergence criterion for determining system stability is that the sum of the relative errors of water flux in all computational units in the system is less than a minimum value. An optional convergence criterion formula is expressed as:

[0083] ;

[0084] in: This represents the total number of computing units in the system. Indicates the first In the nth iteration, the first value calculated based on the current water potential distribution is... Water flux per computing unit Indicates the first The corresponding flux calculated in the next iteration. This is a preset, dimensionless, minimal convergence tolerance. When the above conditions are met, the water potential gradient is considered to have reached stability. In some embodiments, the water potential balance iterative calculation uses the finite element method to discretize and solve the water transport control equations. The output consists of the root region water potential matrix, the leaf region water potential matrix, and the stem conduction water potential matrix after iterative stabilization. The root region water potential matrix is ​​a two-dimensional array, where each element represents the water potential value of a soil grid cell. The leaf region water potential matrix is ​​a two-dimensional array mapped onto the leaf surface. The stem conduction water potential matrix is ​​a one-dimensional array distributed along the stem axis. These three matrices together constitute a spatial water potential distribution map.

[0085] Example 2: See Figure 2 The study acquired dual-spectral information of root soil images in the visible and near-infrared bands. Image registration and fusion were performed on the dual-spectral information to enhance the texture contrast of soil particles and pores. A watershed segmentation algorithm was used to process the fused image, distinguishing soil aggregate regions, pore regions, and root contact regions. For each segmented region, based on its near-infrared spectral reflectance, a pre-established calibration curve between soil spectral reflectance and volumetric water content was consulted to deduce the volumetric water content of each region. By tracking the movement speed of the water front in specific soil pores using continuous time-series root soil images, the water migration rate is calculated. The specific steps are as follows: In a continuous sequence of root soil images, a representative soil pore is selected as the observation area; using image differencing techniques, the grayscale value changes of the observation area in images at adjacent time points are compared to identify the image dark area expansion front caused by water infiltration; the positional movement of the image dark area expansion front in the pixel coordinate system is tracked, and the pixel movement distance is converted into actual distance by combining the image capture time interval and the image spatial calibration parameters; the actual distance the image dark area expansion front moves per unit time is calculated, which is the water migration rate at the soil pore scale; the above tracking and calculation process is repeated for multiple soil pores of different sizes and locations, and finally, the average water migration rate of all pores is taken as the representative water migration rate of the soil layer.

[0086] In the specific implementation, taking a monitoring period of six hours after irrigation in an area planted with apple trees as an example, texture segmentation and moisture inversion calculations were performed on the root soil images. Dual-spectral information of the root soil images in the visible and near-infrared bands was acquired. A simultaneous dual-spectral imager was used to obtain soil profile images containing the main distribution area of ​​the target fruit tree roots at a height of one meter above the soil surface. The visible light image resolution was 2048×1536 pixels, and the near-infrared image center wavelength was 970 nanometers, consistent with the visible light image resolution. Image registration and fusion were performed on the dual-spectral information. Image registration was based on a scale-invariant feature transform algorithm, aligning common feature points in the two images. Image fusion used a weighted average method, fusing the texture details of the registered visible light image with the moisture-sensitive band information of the near-infrared image to generate a fused image with enhanced texture contrast. The edge clarity of soil aggregates, pores, and moist areas in the fused image was improved by approximately 30% compared to the original single-band image. The watershed segmentation algorithm is used to process the fused image. This algorithm uses the image's grayscale gradient as the terrain elevation, starting from pre-defined markers to grow regions, ultimately segmenting the entire image into disconnected regions. It distinguishes between soil aggregates with uniform color and texture, dark-colored porous regions, and root contact regions that correspond to plant root morphology. In a typical processing run, an image is segmented into over two hundred independent regions. For each segmented region, a pre-established calibration curve between soil spectral reflectance and volumetric water content is used, based on its near-infrared spectral reflectance. This calibration curve is obtained through laboratory calibration of similar soil samples. A polynomial relationship is established between near-infrared reflectance values ​​and volumetric water content measured by the drying method, thus retrieving the volumetric water content of each region. For example, a typical soil porous region with a near-infrared reflectance of 15% has a volumetric water content of 18% determined by consulting the calibration curve.

[0087] In some embodiments, the movement speed of a water front in a specific soil pore is tracked using continuous time-series root soil images to calculate the water migration rate. The specific steps are as follows: In a continuous sequence of root soil images, a soil pore with a clear shape and suitable area is selected as the observation area. The observation area is defined by a set of pixel coordinates in the image coordinate system. Using image differencing techniques, the grayscale value changes of the observation area in adjacent time-series images are compared. Image differencing calculates the absolute value of the grayscale difference between each pixel in the observation area of ​​two consecutive frames. When the absolute value exceeds a set threshold, the pixel is considered to have undergone a water change. By connecting the outer edges of these changed pixels, the image dark area expansion front caused by water infiltration is identified. The positional movement of the image dark area expansion front in the pixel coordinate system is tracked. The tracking algorithm calculates the centroid displacement of the dark area front contour between adjacent frames. Combining the fixed time interval between image captures and the spatial calibration parameters of the image, the pixel movement distance is converted into an actual distance by multiplying the pixel displacement by the spatial calibration parameters. The actual distance the image dark area expansion front moves per unit time is calculated as the water migration rate at the soil pore scale. The formula for calculating the water migration rate is:

[0088] ;

[0089] in: This represents the rate of water migration at the soil pore scale. This represents the actual distance the image dark area expansion front moves within a time interval; This represents the time interval between consecutive image captures. The above tracking and calculation process is repeated for multiple soil pores of different sizes and locations. In one analysis, five representative pores distributed at different locations in the image are selected for tracking. Finally, the arithmetic mean of the water migration rates of all pores is taken as the representative water migration rate of the soil layer. For example, if the calculated rates of the five pores are 0.45 mm, 0.52 mm, 0.48 mm, 0.50 mm, and 0.55 mm per hour, respectively, then the average value yields a representative water migration rate of 0.50 mm per hour.

[0090] Example 3: See Figure 3The root region water potential matrix is ​​read, and all grid cells with water potential values ​​below the root critical water potential threshold are identified. The center coordinates of these grid cells are marked as primary water shortage points. The leaf region water potential matrix is ​​obtained, and regions with leaf water potential values ​​below the stomatal closure water potential threshold are located. The geometric centroid coordinates and area percentage of these regions are calculated and marked as leaf stress regions. The stem conduction water potential matrix is ​​retrieved, and the locations of anomalous abrupt changes in water potential gradients are analyzed. The coordinates of stem nodes with water potential gradients exceeding the normal conduction threshold are marked as potential vascular blockage points. A multi-objective decision-making model was established, with primary water-deficient points as irrigation targets, leaf stress areas as stress correction factors, and potential obstruction points in the vascular system as conduction constraints. The calculation process includes: establishing a decision vector for each primary water-deficient point, with dimensions including the water potential of the water-deficient point itself, the pipe path length to the nearest water source, and the type of the soil layer to which it belongs; introducing the area proportion and centroid coordinates of the leaf stress area to calculate the transpiration pull weight of the leaf stress area on each primary water-deficient point, with higher transpiration pull weight coefficients for water-deficient points closer to the leaf stress area; introducing the coordinates of potential obstruction points in the vascular system to calculate their resistance coefficients for upward water transport to each primary water-deficient point, with the resistance coefficients determined based on the increase in hydraulic resistance along the vascular path between the obstruction point and the water-deficient point; weighting and fusing the transpiration pull weight coefficients as positive incentive factors and the resistance coefficients as negative constraint factors with the decision vectors; and obtaining a comprehensive score for each point by solving an objective function that achieves the most balanced improvement in the overall water-deficient situation of all primary water-deficient points. Based on the comprehensive scoring results, all primary water-deficient points are sorted, and points with scores higher than the irrigation initiation threshold are selected as priority irrigation sites. Based on the soil layer properties and comprehensive score corresponding to each priority irrigation site, the amount of water required to restore the water potential of the site to a safe value is calculated, and the regional water requirements are summarized.

[0091] After the multi-objective decision model is run, a model feedback correction step is also performed: after irrigation is completed, a new spatial water potential distribution map is collected and generated again; the root zone water potential matrix before and after irrigation is compared, and the actual water potential increase value of each priority irrigation site coordinate is calculated; the actual water potential increase value is compared with the increase value predicted by the multi-objective decision model to obtain the water potential correction error of the site; based on the water potential correction error of all sites, the key parameters used in the multi-objective decision model to calculate the regional water demand are adjusted, including the soil moisture diffusion coefficient and the crop water absorption resistance coefficient; the adjusted key parameters are used to update the multi-objective decision model for the next irrigation decision calculation.

[0092] In practical implementation, taking the irrigation decision cycle of a vineyard greenhouse during the afternoon high-temperature period as an example, irrigation decision calculations are performed based on the spatial water potential distribution map. The root region water potential matrix is ​​read from the spatial water potential distribution map. This matrix is ​​a two-dimensional array containing 200×200 grid cells, each representing the water potential value of a 1 square centimeter soil area. All grid cells with water potential values ​​below a preset root critical water potential threshold are identified, and their center coordinates are marked as primary water shortage points. In one analysis, 35 primary water shortage points were identified, distributed at different depths around the plant. The leaf region water potential matrix is ​​then obtained from the spatial water potential distribution map. This matrix maps the water potential values ​​of various parts of the grape leaf. Regions with leaf water potential values ​​below the stomatal closure water potential threshold are identified, and their geometric centroid coordinates and area percentages are calculated. These regions are marked as leaf stress areas. For example, the geometric centroid coordinates of a leaf stress area are calculated as follows: The area accounts for 20% of the total leaf surface area. The stem conduction water potential matrix from the spatial water potential distribution map is retrieved. This matrix contains water potential values ​​at each node along the main vine and branches. Abnormal abrupt changes in water potential gradients are analyzed. The water potential gradient is obtained by calculating the ratio of the water potential difference between adjacent nodes to their distance. Stem node coordinates where the water potential gradient exceeds the normal conduction threshold are marked as potential vascular blockage points. For example, a potential vascular blockage point is identified at the connection between the main vine and a lateral branch.

[0093] In some embodiments, a multi-objective decision-making model is established, with primary water shortage points as irrigation targets, leaf stress regions as stress correction factors, and potential vascular blockage points as conduction constraints. The specific construction method of the multi-objective decision-making model is based on a comprehensive analysis of crop water status. The model first constructs a decision vector for each identified primary water shortage point, which includes key parameters such as the water potential value of the water shortage point itself, the pipe path length to the nearest water source, and the type of the soil layer to which it belongs. Next, the model incorporates information about leaf stress regions, calculating the area proportion and geometric centroid coordinates of these regions to determine their transpiration pull weights on each primary water shortage point, with water shortage points closer to the stress region receiving higher weight coefficients. Simultaneously, the coordinates of potential vascular blockage points are used to calculate the resistance coefficient, which quantifies the degree to which the blockage point hinders the upward transport of water from the water shortage point, based on the increase in hydraulic resistance along the vascular path between the blockage point and the water shortage point. Subsequently, the model uses the transpiration pull weight coefficient as a positive incentive factor and the resistance coefficient as a negative constraint factor, and weights them together with the decision vector. A decision vector is established for each primary water-deficient point. The dimensions of the decision vector include the water potential value of the water-deficient point itself, the pipe path length to the nearest water source, and the type of the soil layer to which it belongs. The area proportion and centroid coordinates of the leaf stress region are introduced. These are used to calculate the transpiration pull weight of the leaf stress region on each primary water-deficient point. The closer the water-deficient point is to the geometric centroid coordinates of the leaf stress region, the higher its transpiration pull weight coefficient; the weight coefficient is inversely proportional to the distance. The coordinates of potential obstruction points in the xylem are introduced. These coordinates are used to calculate the resistance coefficient to upward water transport from each primary water-deficient point. The resistance coefficient is determined based on the increase in hydraulic resistance along the xylem path between the obstruction point and the water-deficient point. The increase in hydraulic resistance is characterized by estimating the product of the xylem path length and the outlier of the water potential gradient at the obstruction point. The transpiration pull weight coefficient is used as a positive incentive factor, and the hindering coefficient as a negative constraint factor. These are then weighted and fused with the decision vector. The weighted fusion involves assigning fusion weight parameters with corresponding physical dimension conversion functions to each dimension of the decision vector, the transpiration pull weight coefficient, and the hindering coefficient, ensuring that all contributions are converted into dimensionless standardized values ​​before summation. A comprehensive score for each point is obtained by solving an objective function that achieves the most balanced improvement in the overall water shortage situation of all primary water shortage points. An optional, dimensionally consistent comprehensive score calculation formula is expressed as follows:

[0094] ;

[0095] in: Indicates the first The overall score of a primary water shortage point indicates that the higher the score, the higher the priority of irrigation. These are pre-defined, dimensionless weighting coefficients used to balance the relative importance of different factors in decision-making. It is a water potential reference value, such as the critical water potential threshold of the root system; Indicates the first The water potential value of a primary water-deficient point; Indicates the first The transpiration pull weighting coefficient obtained from each primary water-deficient point; Indicates the first The length of the pipeline path from a primary water shortage point to the nearest water source; It is a length reference value, for example, taking the characteristic length of the entire irrigation area; Indicates the potential obstruction point of the catheter to the first The resistance coefficient caused by a primary water shortage point; It is a reference value for the obstacle coefficient.

[0096] In practice, all primary water-deficient points are sorted according to the comprehensive scoring results. All 35 primary water-deficient points are ranked from highest to lowest comprehensive score, and points with scores higher than the irrigation initiation threshold are selected as priority irrigation sites. For example, the 12 highest-scoring points are selected. Based on the soil layer attributes and comprehensive score corresponding to each priority irrigation site coordinate (including soil texture and field capacity), the amount of water required to restore the water potential at that site to a safe value is calculated. The calculation process is based on the soil moisture characteristic curve and the planned wetting layer depth. The water requirements for all priority irrigation site coordinates are summarized to obtain the regional water demand for different areas. For example, the total water demand for the area where the 12 priority irrigation site coordinates are located is calculated to be 15 liters. After irrigation is completed, a new spatial water potential distribution map is collected and generated. The root zone water potential matrix before and after irrigation is compared, and the actual water potential increase value for each priority irrigation site coordinate is calculated. For example, if the water potential at a certain coordinate point increases from -0.8 MPa before irrigation to -0.4 MPa, the actual water potential increase value is 0.4 MPa. The actual water potential elevation is compared with the elevation predicted by the multi-objective decision model. The predicted elevation is derived from the model's calculation of regional water demand and soil moisture movement, resulting in the water potential correction error for that location. The water potential correction error is the difference between the actual and predicted elevation. Based on the water potential correction errors for all locations, the key parameters used in the multi-objective decision model to calculate regional water demand are adjusted. These key parameters include the soil moisture diffusion coefficient and the crop water absorption resistance coefficient. The adjustment method uses an error feedback approach; for example, if the actual elevation is generally lower than the predicted value, the soil moisture diffusion coefficient is reduced proportionally. Using the adjusted key parameters, the multi-objective decision model is updated for the next irrigation decision calculation, making the model's predictions closer to the actual system response.

[0097] Example 4: The irrigation control network includes a pre-recorded 3D map of the greenhouse and a spatial coordinate database of all irrigation execution terminals. Priority irrigation site coordinates are matched with the greenhouse 3D map to identify one or more target irrigation execution terminals covering each priority irrigation site coordinate. Based on regional water demand, and combined with the water pressure loss model of the irrigation network and the rated flow rate of each target irrigation execution terminal, specific irrigation water volume and irrigation duration are allocated to each target irrigation execution terminal. Considering the overall efficiency of irrigation operations, the start-up times of multiple target irrigation execution terminals are scheduled and optimized to avoid drastic fluctuations in network pressure, generating an operation sequence table containing terminal identification, water volume, duration, and start-up time. The operation sequence table is encoded into a set of drive signals, including a pulse signal sequence for controlling the opening and closing of solenoid valves and analog control signals for adjusting the speed of variable frequency pumps.

[0098] In practical implementation, taking the irrigation command execution process of a strawberry greenhouse as an example, the steps of inputting the irrigation command set into the irrigation control network are explained. The irrigation control network includes a pre-entered 3D map of the greenhouse and a spatial coordinate database of all irrigation execution terminals. The 3D map of the greenhouse is generated by a 3D laser scanner and includes the geometric coordinates of the three-dimensional space within the greenhouse and the distribution information of crop furrows. The spatial coordinate database of all irrigation execution terminals records the unique identifier of each drip irrigation head or micro-sprinkler and its precise 3D coordinates in the 3D map of the greenhouse. The coordinates of priority irrigation sites in the irrigation command set are matched with the 3D map of the greenhouse. The matching algorithm calculates the Euclidean distance between the coordinates of each priority irrigation site and the coordinates of all irrigation execution terminals in the database, determining one or more target irrigation execution terminals covering each priority irrigation site coordinate. The coverage rule is set as a circular area centered on the irrigation execution terminal and with an effective wetting radius as the range of action. For example, a coordinate... The priority irrigation sites were determined by the identifier " "and" The two drip heads cover the area together.

[0099] Based on the regional water demand in the irrigation instruction set, and combined with the water pressure loss model of the irrigation network and the rated flow of each target irrigation execution terminal, specific irrigation water volume and irrigation duration are allocated to each target irrigation execution terminal. The water pressure loss model of the irrigation network uses the Hassen-Williams formula to calculate the head loss along the network and in local areas. The rated flow of each target irrigation execution terminal is obtained from the equipment nameplate parameters. Considering the overall efficiency of irrigation operations, the start-up time of multiple target irrigation execution terminals is scheduled and optimized to avoid drastic fluctuations in network pressure, generating an operation sequence table containing terminal identification, water volume, duration, and start-up time. The operation sequence table is encoded as a set of drive signals, which includes a pulse signal sequence for controlling the opening and closing of the solenoid valve and an analog control signal for adjusting the speed of the variable frequency pump. The pulse signal sequence defines the width modulation signal for opening the solenoid valve, and the analog control signal is a voltage signal corresponding to the target pump speed. In a typical irrigation control scenario, the operation sequence table contains the following information, as shown in Table 1:

[0100] Table 1: Irrigation Schedule for Strawberry Greenhouse Zones

[0101]

[0102] In some embodiments, encoding the job timing table into a set of drive signals involves conversion logic. For the pulse signal sequence of solenoid valve opening and closing, the irrigation control network generates a high-level signal with a specific timing according to the start time and irrigation duration in the job timing table. The duration of the pulse signal sequence is equal to the irrigation duration, and the amplitude and level standards of the pulse signal sequence conform to the industrial fieldbus protocol. For the analog control signal for adjusting the speed of the variable frequency water pump, its target value is determined by the total flow demand of all target irrigation execution terminals that need to be activated at the same time, and the calculation formula is:

[0103] ;

[0104] in: This indicates the target voltage value of the analog control signal output to the variable frequency water pump; This represents a conversion factor that maps total flow demand to a voltage signal; its value is determined by the pump's rated parameters and the control system's range. This represents the sum of instantaneous flow rates of all target irrigation execution terminals that are in an active irrigation state at a given moment. The central controller of the irrigation control network continuously monitors the execution progress of the job sequence table and dynamically calculates and outputs real-time analog control signals based on the above formula.

[0105] See Figure 4In the optimization of terminal start-up scheduling in the irrigation control network, the distribution of the number of irrigation terminals and the corresponding changes in total flow demand within different start-up time intervals are presented intuitively. Specifically, the purple bars in the figure represent the number of start-up terminals in each time interval (0-60s to 540-600s), and the orange line represents the total flow demand for the corresponding interval (unit: Based on the data characteristics, the number of terminals launched in the 0-60s interval was 2, corresponding to the total traffic demand. ; The number of terminals launched within the interval remains stable at 1, and the total traffic demand will remain unchanged for the time being. Later dropped to ; The number of terminals launched in the interval remains at 1, but the total traffic demand is increased from... Continue to climb to corresponding ; The number of started terminals then dropped to 0, and the total traffic demand simultaneously decreased to 0. The core value of the graph lies in quantifying the dynamic impact of start-up time distribution on the network load: Although only one terminal was activated in the interval, the rapid increase in total flow demand reflected the high flow characteristics of the terminal during that period, making it a potential risk area for pipeline pressure fluctuations; while The multi-terminal startup corresponds to a relatively stable traffic demand, reflecting the load difference between the "multi-terminal low traffic" and "single-terminal high traffic" scenarios in scheduling optimization.

[0106] Example 5: The primary optimization objective is to minimize the total irrigation operation time, with the constraint of preventing any network node pressure from falling below the minimum operating pressure. The topology, pipe diameter data, and friction coefficients of each pipe segment of the irrigation network are obtained to establish a network hydraulic calculation model. The planned flow rates of all target irrigation execution terminals in the operation sequence table are used as input to simulate the network pressure distribution under different start-up sequences in the network hydraulic calculation model. A heuristic search algorithm is employed to find the startup sequence and delay interval that minimizes the time span for all terminals to complete the irrigation task, under pressure constraints. This is achieved through the following steps: An optimization objective function is established to minimize the total irrigation time span; hydraulic constraints on the pipeline network are set, requiring that the pressure value of each network node during irrigation remains above a preset minimum working pressure threshold; a genetic algorithm is selected as the heuristic search algorithm, and its parameters, including population size, crossover probability, and mutation probability, are initialized; an initial population is randomly generated, with each individual representing a combination of the startup sequence of irrigation terminals and the delay interval between adjacent startups; an iterative search is performed, simulating the pipeline pressure distribution under each combination and calculating the objective function value in each generation of the population; the fitness of individuals is evaluated based on the objective function value and constraint satisfaction, and a new generation of population is generated through selection, crossover, and mutation operations; the iterative process is repeated until convergence is met, and the startup sequence and delay interval corresponding to the individual with the highest fitness are output as the optimization result. Based on the optimized startup sequence and delay interval, the precise startup time of each target irrigation terminal in the work schedule is finally determined.

[0107] In practical implementation, taking a blueberry greenhouse with a complex branching pipe network as an example, this paper illustrates a method for scheduling and optimizing the start-up time of multiple target irrigation execution terminals. The main optimization objective is to minimize the total irrigation operation time, with the constraint of avoiding any pipe network node pressure falling below the minimum working pressure. The minimum working pressure threshold of the pipe network node is set at 0.25 MPa to ensure the normal operation of the end drippers. The topology, pipe diameter data, and friction coefficient of each pipe segment of the irrigation pipe network are obtained. The topology shows that the pipe network includes one main pipe, three branch pipes, and eight capillary pipes. The pipe diameter data includes a main pipe diameter of 50 mm, a branch pipe diameter of 32 mm, and a capillary pipe diameter of 16 mm. The friction coefficient of each pipe segment is obtained by referring to a table based on the pipe material. A hydraulic calculation model of the pipe network is established. The hydraulic calculation model of the pipe network describes the pipe network connection relationship based on graph theory and solves the nodal pressure equation and loop pressure drop equation in fluid mechanics simultaneously.

[0108] The planned flow rates of all target irrigation execution terminals in the job sequence table are used as input. In one optimization task, the job sequence table contains the planned flow rates of eight target irrigation execution terminals, with each terminal's planned flow rate varying between 0.5 and 1.2 cubic meters per hour. The network hydraulic calculation model simulates the network pressure distribution under different start-up sequences. The pump outlet pressure is set as a constant value during the simulation. Hydraulic calculations are used to solve for the pressure values ​​at all nodes in the network under different terminal combination start-up states. A heuristic search algorithm is used to find the start-up sequence and delay interval that minimizes the time span for all terminals to complete the irrigation task, while satisfying pressure constraints. An optimization objective function is established with the goal of minimizing the total irrigation time span. The objective function expression is to minimize the difference between the start-up time of the first terminal and the shut-off time of the last terminal. Network hydraulic constraints are set, requiring that the pressure value of each network node during irrigation is never lower than a preset minimum working pressure threshold. In the calculation, the node pressures under all simultaneously activated terminal combination conditions need to be verified.

[0109] In some embodiments, a genetic algorithm is selected as the heuristic search algorithm, and the algorithm parameters are initialized, including population size, crossover probability, and mutation probability. For example, the population size is set to 100, the crossover probability to 0.8, and the mutation probability to 0.05. An initial population is randomly generated, where each individual represents a combination of the start-up order of irrigation execution terminals and the delay interval between adjacent starts. An individual can be encoded by a real number array of length , where the first half of the array represents the sequence number of the terminal start-up, and the second half represents the delay time between adjacent starts. Iterative search is performed. In each generation of the population, the pipeline pressure distribution under each combination is simulated, and the objective function value is calculated. For the start-up order and delay interval sequence decoded by each individual, the start and stop of each terminal are simulated in time steps, and the pipeline hydraulic calculation model is called at each time step to solve for the pressure distribution, while simultaneously calculating the total irrigation time span. The fitness of individuals is evaluated based on the objective function value and constraint satisfaction. The fitness function design needs to unify the total time span and pressure constraint violation into a single scalar value. An optional, dimensionlessly consistent fitness calculation formula is expressed as:

[0110] ;

[0111] in: This represents the fitness value of an individual; a higher fitness value indicates a better scheduling scheme. This represents a preset constant time that is much larger than the total duration of any scheduling scheme; This indicates the total irrigation time span under this scheduling scheme; This represents a scaling factor that converts stress violation time into an equivalent time penalty; Indicates the time during the irrigation process At that time, the lowest pressure value among all nodes in the entire network; This indicates the preset minimum working pressure threshold; It is a unit step function, when The function value is 1 if the time interval is short, and 0 otherwise. The integral term in the denominator of the formula calculates the total irrigation time span. Internally, the pipeline pressure is below the threshold. The total duration. A new generation of the population is generated through selection, crossover, and mutation operations. Selection uses roulette wheel selection, crossover uses partial mapping crossover, and mutation slightly perturbs the delay interval gene. The iteration process is repeated until the convergence condition is met. The convergence condition is set so that the optimal fitness value of the population no longer increases after twenty consecutive generations. The starting order and delay interval corresponding to the individual with the highest fitness are output as the optimization result. It can be understood that genetic algorithms can efficiently find approximate optimal solutions in a huge combinatorial search space. Based on the optimized starting order and delay interval, the precise starting time of each target irrigation execution terminal in the job time sequence is finally determined. For example, the optimization result indicates that the terminal should start according to... The tasks are started sequentially, with adjacent start intervals of 90 seconds, 120 seconds, and 60 seconds, respectively, thus generating a job sequence table accurate to the second.

[0112] See Figure 5 In the system performance evaluation presented, time efficiency, water saving rate, energy consumption, reliability, and stability were used as the core evaluation dimensions. The scores of the intelligent irrigation system before optimization (red line) and after optimization (green line) were compared. Specifically, the scores of each dimension of the system fluctuated greatly before optimization, with an average score of 67.6. After optimization, the scores of each dimension of the system were at a high level and fluctuated more smoothly, with an average score of 88.4. Analyzing the sub-dimensions: the score for time efficiency after optimization was significantly higher than before, reflecting the improved efficiency of the irrigation decision-making and execution process; the scores for water saving rate and energy consumption after optimization remained stable at a high level, reflecting the improvement in resource consumption brought about by precision irrigation based on spatial water potential distribution maps; the scores for reliability and stability after optimization were also in a high range, verifying the enhancing effect of irrigation control network scheduling optimization (such as terminal startup timing optimization driven by genetic algorithms) on the system's operational stability.

Claims

1. An intelligent irrigation system for fruit tree greenhouses, characterized in that, The following processing steps are included: The multidimensional data acquisition module acquires multidimensional growth data of crops during their growth cycle, including root soil images, leaf temperature distribution maps, and stem microdeformation sensing data. The data fusion processing module performs multi-dimensional fusion calculations on the multi-dimensional growth data to generate a spatial water potential distribution map of the crop. The spatial water potential distribution map is composed of a root region water potential matrix, a leaf region water potential matrix, and a stem conduction water potential matrix. The irrigation decision module performs irrigation decision calculations based on the spatial water potential distribution map. The output of the irrigation decision calculation includes a set of irrigation instructions for regional water demand and the coordinates of priority irrigation sites. The irrigation control execution module inputs the irrigation instruction set to the irrigation control network, which parses the irrigation instruction set into an executable set of drive signals based on the physical layout of the irrigation execution terminals in the greenhouse.

2. The intelligent irrigation system for fruit tree greenhouses according to claim 1, characterized in that, The step of performing multi-dimensional fusion calculations on the multi-dimensional growth data to generate a spatial water potential distribution map of the crop includes the following sub-steps: The root soil image is subjected to texture segmentation and humidity inversion calculation to extract the actual water content and water migration rate of each soil layer, forming a dynamic profile of root water. Simultaneously, thermal infrared feature analysis is performed on the leaf surface temperature distribution map. Combined with the preset leaf surface stomatal aperture model, the leaf surface transpiration water loss rate and leaf surface cell water potential value are inverted to generate a leaf surface water potential state map. Simultaneously, by analyzing the micro-deformation sensing data of the stem, and calculating the micron-level periodic changes in the stem diameter, the water column tension data of the xylem inside the stem and the water flow of the vascular bundles are derived, and a water conduction status diagram of the stem is constructed. The root water dynamic profile, the leaf water potential state map, and the stem water conduction state map are registered and superimposed in a unified crop three-dimensional spatial coordinate system. Based on the principle of continuous water transport among roots, stems and leaves, water potential balance iterative calculations are performed on the registered root water dynamic profile, leaf water potential state map and stem water conduction state map until the root-stem-leaf water potential gradient reaches stability. The root region water potential matrix, the leaf region water potential matrix, and the stem conduction water potential matrix, after being output and stabilized through iteration, together constitute the spatial water potential distribution map.

3. The intelligent irrigation system for fruit tree greenhouses according to claim 2, characterized in that, The process of performing texture segmentation and humidity inversion calculation on the root soil image specifically includes: The image of the root soil was acquired using dual-spectral information in the visible and near-infrared bands. Image registration and fusion of dual-spectral information enhances the texture contrast of soil particles and pores; The watershed segmentation algorithm was used to process the fused image to distinguish the soil aggregate region, the pore region and the root contact region; For each segmented region, based on its near-infrared spectral reflectance, the pre-established calibration curve between soil spectral reflectance and volumetric water content is queried to retrieve the volumetric water content of each region. The water migration rate is calculated by tracking the movement speed of the water front in specific soil pores using continuous time-series images of the root soil.

4. The intelligent irrigation system for fruit tree greenhouses according to claim 3, characterized in that, The water migration rate is calculated by tracking the movement speed of the water front in specific soil pores using continuous time-series images of the root soil. The specific steps are as follows: In a continuous sequence of root soil images, a representative soil pore area is selected as the observation area. By using image difference technology, the gray value changes of the observed area in images at adjacent time points are compared to identify the image dark area expansion front caused by water infiltration; The positional movement of the image dark area expansion front in the pixel coordinate system is tracked, and the pixel movement distance is converted into actual distance by combining the time interval of image capture and the spatial calibration parameters of the image. The actual distance the image dark area expansion front moves per unit time is calculated, which is the water migration rate at the soil pore scale; The above tracking and calculation process was repeated for multiple soil pores of different sizes and locations. Finally, the average value of the water migration rate of all pores was taken as the representative water migration rate of the soil layer.

5. The intelligent irrigation system for fruit tree greenhouses according to claim 1, characterized in that, The irrigation decision calculation based on the spatial water potential distribution map includes the following processing steps: Read the water potential matrix of the root system region, identify all grid cells whose water potential value is lower than the critical water potential threshold of the root system, and mark the center coordinates of these grid cells as primary water-deficient points; Obtain the water potential matrix of the leaf surface region, find the region where the leaf surface water potential value is lower than the stomatal closure water potential threshold, calculate the geometric centroid coordinates and area percentage of the region, and mark it as the leaf stress region; The stem conduction water potential matrix is ​​retrieved, and the locations of abnormal abrupt changes in water potential gradient are analyzed. The coordinates of stem nodes where the water potential gradient exceeds the normal conduction threshold are marked as potential vascular blockage points. A multi-objective decision-making model is established with the primary water shortage point as the irrigation target, the leaf stress area as the stress correction factor, and the potential obstruction point of the vascular bundle as the conduction constraint. In the multi-objective decision-making model, the distance between each primary water shortage point and the water shortage severity of the soil layer in which it is located is calculated, and a comprehensive score is given by combining the influence of the leaf stress area on its transpiration pull and the degree of obstruction of the potential blockage point of the xylem on its water transport. Based on the comprehensive scoring results, all the primary water shortage points are sorted, and the points with scores higher than the irrigation initiation threshold are selected as the coordinates of the priority irrigation sites. Based on the soil layer properties and comprehensive scores corresponding to the coordinates of each priority irrigation site, the amount of water required to restore the water potential of the site to a safe value is calculated, and the water demand of the sub-regions is obtained by summarizing.

6. The intelligent irrigation system for fruit tree greenhouses according to claim 5, characterized in that, The calculation process of establishing a multi-objective decision-making model with the primary water shortage point as the irrigation target, the leaf stress region as the stress correction factor, and the potential obstruction point of the vascular bundle as the conduction constraint includes: A decision vector is established for each of the primary water-deficient points. The dimensions of the decision vector include the water potential value of the water-deficient point itself, the pipeline path length to the nearest water source, and the type of the soil layer to which it belongs. By introducing the area ratio and centroid coordinates of the leaf stress region, the transpiration pull weight of the leaf stress region on each primary water shortage point is calculated. The water shortage point that is closer to the leaf stress region has a higher transpiration pull weight coefficient. The coordinates of the potential blockage points of the conduit are introduced, and the resistance coefficients to the upward transport of water at each of the primary water shortage points are calculated. The resistance coefficients are determined based on the increase in hydraulic resistance of the conduit path between the blockage point and the water shortage point. The transpiration pull weighting coefficient is used as a positive incentive factor, and the resistance coefficient is used as a negative constraint factor, and they are weighted and fused with the decision vector. The comprehensive score for each point is obtained by solving an objective function that achieves the most balanced improvement in the overall water shortage situation of all the primary water shortage points.

7. The intelligent irrigation system for fruit tree greenhouses according to claim 6, characterized in that, After the multi-objective decision model is run, the following model feedback correction steps are also performed: After irrigation is completed, a new spatial water potential distribution map is collected and generated again. By comparing the root zone water potential matrix before and after irrigation, the actual water potential increase value of each priority irrigation site coordinate is calculated. The actual water potential increase is compared with the increase predicted by the multi-objective decision model to obtain the water potential correction error at the point. Based on the water potential correction error at all points, adjust the key parameters in the multi-objective decision model used to calculate the regional water demand. The key parameters include the soil moisture diffusion coefficient and the crop water absorption resistance coefficient. The adjusted key parameters are used to update the multi-objective decision model for the next irrigation decision calculation.

8. The intelligent irrigation system for fruit tree greenhouses according to claim 1, characterized in that, The step of inputting the irrigation instruction set into the irrigation control network includes the following steps: The irrigation control network includes a pre-recorded 3D map of the greenhouse and a spatial coordinate database of all irrigation execution terminals; The coordinates of the priority irrigation sites are matched with the three-dimensional map of the greenhouse to determine one or more target irrigation execution terminals covering each of the priority irrigation site coordinates; Based on the water demand of the different regions, and combined with the water pressure loss model of the irrigation network and the rated flow of each target irrigation execution terminal, a specific irrigation water volume and irrigation duration are allocated to each target irrigation execution terminal; Considering the overall efficiency of irrigation operations, the scheduling optimization of the start-up time of multiple target irrigation execution terminals is carried out to avoid drastic fluctuations in pipeline pressure, and an operation sequence table containing terminal identifier, water volume, duration and start-up time is generated. The operation timing table is encoded into the drive signal set, which includes a pulse signal sequence for controlling the opening and closing of the solenoid valve and an analog control signal for adjusting the speed of the variable frequency water pump.

9. The intelligent irrigation system for fruit tree greenhouses according to claim 8, characterized in that, The method for scheduling and optimizing the start-up time of multiple target irrigation execution terminals includes: The main optimization objective is to minimize the total irrigation operation time, with the hard constraint being to avoid the pressure at any pipeline node falling below the minimum working pressure. Obtain the topology, pipe diameter data, and friction coefficient of each pipe segment of the irrigation network, and establish a hydraulic calculation model of the network. The planned flow rates of all target irrigation execution terminals in the operation sequence table are used as inputs to simulate the network pressure distribution under different start-up sequences in the network hydraulic calculation model. A heuristic search algorithm is used to find the startup order and delay interval that minimize the time span for all terminals to complete the irrigation task, while satisfying the pressure constraints. Based on the optimized startup sequence and delay interval, the precise startup time of each target irrigation execution terminal in the operation timing table is finally determined.

10. The intelligent irrigation system for fruit tree greenhouses according to claim 9, characterized in that, The heuristic search algorithm is employed to find the startup order and delay interval that minimizes the time span for all terminals to complete the irrigation task, under the condition of satisfying pressure constraints. This is achieved in the following way: Establish an optimization objective function with the goal of minimizing the total irrigation time span; Set hydraulic constraints for the pipeline network, requiring that the pressure value of each pipeline node during irrigation is never lower than the preset minimum working pressure threshold. Genetic algorithm was chosen as the heuristic search algorithm, and the algorithm parameters were initialized, including population size, crossover probability, and mutation probability. An initial population is randomly generated, and each individual represents a combination of the start order of irrigation execution terminals and the delay interval between adjacent starts; An iterative search is performed to simulate the pipeline pressure distribution under each combination in each generation of the population and to calculate the objective function value. The fitness of individuals is evaluated based on the objective function value and the constraint satisfaction, and a new generation of population is generated through selection, crossover and mutation operations. Repeat the iterative process until the convergence condition is met, and output the startup order and delay interval corresponding to the individual with the highest fitness as the optimization result.

Images