Crop root growth simulation and three-dimensional irrigation decision system and method thereof

Abstract

The invention relates to the technical field of agricultural irrigation, in particular to a crop root growth simulation and three-dimensional irrigation decision system and method. The crop root system intelligent irrigation system integrates data acquisition, neural network processing, irrigation control and decision analysis. Three-dimensional root structure data is obtained by using a micro CT scanning technology, and a high-precision basis is provided for growth simulation; hydrodynamics and deep learning are fused, a multi-scale model is established, and the root system water absorption process is accurately simulated; a telescopic drip irrigation capillary array is configured, a pressure oscillation sensor and a reinforcement learning algorithm are combined, irrigation parameters are dynamically adjusted, and automatic conversion of irrigation modes is achieved; the system realizes non-destructive observation and precise irrigation control of root morphology, improves crop growth efficiency and water resource utilization, and provides powerful support for intelligent agriculture development.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural irrigation, and particularly to a crop root growth simulation and three-dimensional irrigation decision-making system and method thereof. Background Art

[0002] With the increasing global water shortage and the continuous growth of agricultural water demand, the research and application of precision irrigation technology have received extensive attention. Traditional irrigation methods often adopt a unified standard, ignoring the differential water demand characteristics of crops at different growth stages, resulting in serious water resource waste. In addition, existing irrigation systems mainly focus on surface water supply, lacking an in-depth understanding of the crop root growth and water absorption processes, and unable to achieve an accurate response to root development.

[0003] Currently, root research mainly uses manual excavation methods to obtain root morphology data, which not only damages the root integrity but also makes it difficult to monitor the root growth dynamics in real time. Existing irrigation decision-making systems are mostly based on empirical models, lacking an in-depth analysis of root morphology, growth laws, and water absorption mechanisms, and it is difficult to achieve precise and intelligent irrigation. Especially at different growth stages, due to significant differences in root distribution and water absorption characteristics, there is an urgent need for a system that can automatically adjust irrigation strategies according to the dynamic growth of roots.

[0004] Although in recent years, artificial intelligence and sensor technologies have made great progress in the agricultural field, the research on combining root growth simulation with irrigation decision-making is still in its infancy. Therefore, developing a system that can accurately simulate crop root growth and formulate three-dimensional irrigation decisions based on this is of great significance for improving water resource utilization efficiency and promoting sustainable agricultural development. Summary of the Invention

[0005] The present invention aims to overcome the problems existing in the prior art, and provides a crop root growth simulation and three-dimensional irrigation decision-making system and method thereof. By accurately simulating the dynamic growth of crop roots, it realizes precise irrigation of crops at different growth stages, improves water resource utilization efficiency, and promotes the healthy growth of crops.

[0006] The present invention proposes a crop root growth simulation and three-dimensional irrigation decision-making system, including:

[0007] A data acquisition module for collecting three-dimensional structure data of crop roots;

[0008] A neural network processing module connected to the data acquisition module for constructing a root growth simulation model based on the three-dimensional structure data;

[0009] An irrigation control module connected to the neural network processing module for controlling a drip irrigation system based on the root growth simulation model;

[0010] The decision-making analysis module, connected to the neural network processing module and the irrigation control module, is used to generate a three-dimensional irrigation plan according to crop growth indicators.

[0011] Preferably, the data acquisition module includes:

[0012] A scanning unit, used to scan the crop roots using a micro-CT device;

[0013] An image processing unit, connected to the scanning unit, used to process the scanned image;

[0014] A database construction unit, connected to the image processing unit, used to establish a three-dimensional root system structure database based on the processed image.

[0015] Preferably, the neural network processing module includes:

[0016] A network architecture unit, used to construct a long short-term memory artificial neural network architecture;

[0017] A training unit, connected to the network architecture unit, used to train the neural network using the three-dimensional structure data;

[0018] A calculation unit, connected to the training unit, used to calculate the root water absorption rate and capillary root water potential. Preferably, the calculation unit calculates the root water absorption rate and capillary root water potential through the following formula:

[0019] P = w0 + ∑w i x i ,

[0020] where P represents the calculated pressure value, w i represents the fitting parameter, and x i represents the root water absorption rate.

[0021] Preferably, the irrigation control module includes:

[0022] A drip irrigation configuration unit, used to configure a retractable drip irrigation lateral array;

[0023] A sensing and monitoring unit, connected to the drip irrigation configuration unit, used to monitor the root development status through a pressure oscillation sensor;

[0024] A reinforcement learning unit, connected to the sensing and monitoring unit, used to dynamically adjust the irrigation layer depth and water volume distribution.

[0025] Preferably, the reinforcement learning unit realizes automatically switching between surface irrigation and subsurface infiltration irrigation modes according to the crop growth stage.

[0026] Preferably, the decision-making analysis module includes:

[0027] A growth index calculation unit for calculating leaf surface growth indexes and underground plant growth indexes;

[0028] A water requirement analysis unit, connected to the growth index calculation unit, for calculating the water requirement of crops based on a root-shoot ratio threshold;

[0029] A scheme generation unit, connected to the water requirement analysis unit, for generating a three-dimensional irrigation scheme for crops by combining rainfall information and soil moisture conditions.

[0030] Preferably, the scheme generation unit realizes the judgment of crop irrigation amounts under different growth periods and different climates through fuzzy control and neural network theory.

[0031] Preferably, the three-dimensional irrigation scheme includes a combination of surface irrigation in the early growth stage, underground irrigation in the middle growth stage, and surface irrigation in the late growth stage.

[0032] A method for simulating crop root growth and making three-dimensional irrigation decisions, comprising the following steps:

[0033] S1. Obtain a root growth simulation model: Use micro-CT scanning to construct a three-dimensional root structure database, develop a neural network based on physical information to simulate the water absorption process, fuse the hydrodynamic equation and the deep learning architecture, and establish a multi-scale model considering capillary action and root hair growth;

[0034] S2. Construct an underground infiltration irrigation system: Configure a retractable drip irrigation capillary array, monitor the root development status through a pressure oscillation sensor, and use a reinforcement learning algorithm to dynamically adjust the irrigation layer depth and water volume distribution to realize the automatic conversion of surface irrigation and underground infiltration irrigation modes according to the crop growth period;

[0035] S3. Calculate crop growth indexes: Calculate leaf surface growth indexes and underground plant growth indexes according to the crop growth conditions adjusted by the underground infiltration irrigation mode;

[0036] S4. Establish an irrigation decision-making system: Combine crop growth indexes, use fuzzy control and neural network theory to realize the judgment of crop irrigation amounts under different growth periods and different climates, and propose a three-dimensional irrigation scheme for crops according to rainfall information and soil moisture conditions.

[0037] The present invention has the following beneficial effects:

[0038] 1. By using micro-CT scanning technology to construct a three-dimensional root structure database, accurate non-destructive observation of root morphology is realized, providing a high-precision data basis for root growth simulation;

[0039] 2. By fusing the hydrodynamic equation and the deep learning architecture, a multi-scale model considering capillary action and root hair growth is established, realizing the accurate simulation of the root water absorption process;

[0040] 3. Configure a scalable drip irrigation lateral array, monitor the root system development status through a pressure oscillation sensor, and use a reinforcement learning algorithm to dynamically adjust irrigation parameters to achieve automatic conversion of irrigation modes according to the crop growth period;

[0041] 4. Apply fuzzy control and neural network theories, combine crop growth indicators, rainfall information, and soil moisture conditions to construct an intelligent irrigation decision-making system, improving the accuracy of irrigation decision-making;

[0042] 5. Through a combination of surface irrigation in the early growth stage, underground irrigation in the middle growth stage, and surface irrigation in the late growth stage, a three-dimensional irrigation plan is formed, significantly improving the water resource utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a structural block diagram of the crop root growth simulation and three-dimensional irrigation decision-making system of the present invention;

[0044] Figure 2 is a structural block diagram of the data acquisition module of the present invention;

[0045] Figure 3 is a structural block diagram of the neural network processing module of the present invention;

[0046] Figure 4 is a structural block diagram of the irrigation control module of the present invention;

[0047] Figure 5 is a structural block diagram of the decision analysis module of the present invention;

[0048] Figure 6 is a flowchart of the crop root growth simulation and three-dimensional irrigation decision-making method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] Please refer to the attached Figure 1-6 , and the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0050] As Figure 1 shown, the crop root growth simulation and three-dimensional irrigation decision-making system provided by the present invention includes a data acquisition module 1, a neural network processing module 2, an irrigation control module 3, and a decision analysis module 4.

[0051] The data acquisition module 1 is used to acquire three-dimensional structure data of crop roots. The neural network processing module 2 is connected to the data acquisition module 1 and is used to construct a root growth simulation model based on the three-dimensional structure data. The irrigation control module 3 is connected to the neural network processing module 2 and is used to control the drip irrigation system based on the root growth simulation model. The decision analysis module 4 is connected to the neural network processing module 2 and the irrigation control module 3 and is used to generate a three-dimensional irrigation plan according to crop growth indicators.

[0052] As Figure 2 shown, in an embodiment of the present invention, the data acquisition module 1 includes a scanning unit 11, an image processing unit 12, and a database construction unit 13. The scanning unit 11 is used to scan crop roots using a micro-CT device. Preferably, the spatial resolution of the micro-CT device can reach 5 microns, enabling clear capture of the microscopic structure at the root hair level. The image processing unit 12 is connected to the scanning unit 11 and is used to process the scanned images. Specifically, the image processing unit 12 first imports the root system images into Adobe Acrobat software to generate PDF files, and then uses Avizo Flow V9.3 software to perform segmentation, filtering, and enhancement processing on the images to improve the image quality and recognition accuracy. The database construction unit 13 is connected to the image processing unit 12 and is used to establish a three-dimensional root system structure database based on the processed images. This database not only stores the static morphological structure of the root system but also records the changes in the distribution characteristics of the root system at different growth stages, providing data support for subsequent model construction.

[0053] As Figure 3 shown, in an embodiment of the present invention, the neural network processing module 2 includes a network architecture unit 21, a training unit 22, and a computing unit 23. The network architecture unit 21 is used to construct a long short-term memory artificial neural network architecture. This architecture models the relationship between three-dimensional structure data and soil moisture distribution, evaporation loss, and root water absorption rate as a neural network, expressed by the following formula:

[0054] h(t) = f(U(x(t), y(t), z(t))),

[0055] where x(t), y(t), and z(t) represent soil water content, capillary root water potential, and root water absorption rate respectively; U represents the neuron weight, f represents the neuron activation function, and h represents the hidden layer state. Preferably, the activation function f uses the Softplus function, which has better smoothness and continuity compared to the traditional ReLU function, helping to improve the generalization ability of the model.

[0056] The training unit 22 is connected to the network architecture unit 21 and is used to train the neural network using the three-dimensional structure data. Specifically, the training unit 22 uses the Adam optimizer and the mean square error loss function to verify the fitting effect of the neural network. Preferably, the learning rate of the Adam optimizer is initially set to 0.001 and a dynamic adjustment strategy is adopted. A larger learning rate is used at the beginning of training to accelerate convergence, and the learning rate is reduced in the later stage of training to improve accuracy. The reason for this setting is based on experimental verification that this parameter combination can achieve high model accuracy while ensuring the convergence speed.

[0057] The calculation unit 23 is connected to the training unit 22 and is used to calculate the root water absorption rate and the capillary root water potential. Specifically, the calculation unit 23 calculates through the following formula:

[0058] P = w0 + ∑w i x i ,

[0059] where P represents the calculated pressure value, with the unit of MPa, and its physical meaning is the driving force for water movement in the root system; w0 represents the basic water potential value, that is, the basic pressure of the root system without water absorption activity, reflecting the inherent characteristics of plant species and physiological states; w i represents the water absorption contribution weight coefficient of different regions or different levels of roots in the root system, and x i represents the water absorption rate of each part of the root system, with the unit of mm 3 / h·mm 2 , indicating the water absorption volume per unit root surface area per unit time. In practical applications, the fitting parameter w i is usually automatically obtained during the training process. For field crops such as corn, the empirical value of w0 is usually 0.05 - 0.15 MPa, while the value range of w i is generally 0.01 - 0.20, and the specific value depends on the crop type and growth environment. These parameters are automatically obtained through the training process of the neural network. The specific training mechanism is as follows:

[0060] First, use the three-dimensional root system structure data obtained by micro-CT scanning as the input;

[0061] Input the measured soil moisture changes, root pressure changes, and root morphology parameters into the neural network together;

[0062] Perform backpropagation training through the Adam optimizer to minimize the mean square error between the predicted value and the measured value;

[0063] After the training is completed, the network automatically extracts the w0 and wi parameters, and these parameters contain the contribution degree of different parts of the root system to water absorption;

[0064] The correlation between the parameters and the pressure is reflected as:

[0065] w0 represents the basic pressure value inherent in plant species, reflecting the basic water potential characteristics of the roots of different crops;

[0066] wi represents the contribution weight of different levels of roots (such as main roots, lateral roots, capillary roots) to the total water absorption pressure;

[0067] When the soil moisture decreases, the xi value decreases, resulting in an increase (more negative) in the calculated P value, reflecting that plants need to overcome a greater water potential difference to absorb water under drought conditions;

[0068] Root systems with different depths and diameters have different wi values, accurately reflecting the impact of root architecture on water absorption capacity;

[0069] The root water absorption rate (xi) is measured and quantified by the following method:

[0070] Use a pressure oscillation sensor array to monitor the dynamic changes of soil moisture at different positions around the root system;

[0071] Based on Darcy's law, calculate the amount of water reduction in the soil per unit area per unit time;

[0072] Based on the root density distribution obtained by CT scanning, allocate the total water absorption to different regions and different levels of roots;

[0073] Through time series analysis, determine the dynamic change law of root water absorption at different growth stages;

[0074] The specific relationships between the parameter values and the crop type and growth environment are as follows:

[0075] For shallow-rooted crops (such as wheat), the w0 value is usually small (0.05 - 0.08 MPa), indicating a lower basic water potential of the root system;

[0076] For deep-rooted crops (such as corn), the w0 value is usually large (0.10 - 0.15 MPa), reflecting its ability to overcome a greater gravitational potential;

[0077] For crops growing in arid environments, the wi value is usually large (0.10 - 0.20), indicating that the impact of unit water absorption rate on the total water potential is more significant;

[0078] For crops growing in humid environments, the wi value is usually small (0.01 - 0.08), indicating that the environmental moisture is sufficient and the root water absorption resistance is small;

[0079] During different growth stages of the growth period, there are also significant differences in the parameter values: the w0 value is small during the seedling stage, reaches the maximum during the mid-growth stage, and slightly decreases during the maturity stage;

[0080] These parameter relationships are obtained based on a large number of field experiments and laboratory studies, reflecting the essential differences in the root structures and functions of different crops, and providing a theoretical basis for formulating precise irrigation strategies for different crops.

[0081] Such as Figure 4As shown, in one embodiment of the present invention, the irrigation control module 3 includes a drip irrigation configuration unit 31, a sensing and monitoring unit 32, and a reinforcement learning unit 33. The drip irrigation configuration unit 31 is used to configure a retractable drip irrigation capillary array. Preferably, the arrangement density of the drip irrigation capillary array is optimized according to the crop root distribution. For crops with a relatively wide horizontal root distribution (such as wheat), the capillary spacing can be set to 20 - 30 cm; for crops with a relatively deep vertical root distribution (such as corn), the capillary depth can reach 40 - 60 cm. The sensing and monitoring unit 32 is connected to the drip irrigation configuration unit 31 and is used to monitor the root development status through a pressure oscillation sensor. Preferably, the sensitivity of the pressure oscillation sensor can reach 0.01 MPa, and the sampling frequency is 1 - 5 times per minute, which can capture the minute pressure fluctuations caused by soil moisture changes and root activities in real time.

[0082] The reinforcement learning unit 33 is connected to the sensing and monitoring unit 32 and is used to dynamically adjust the irrigation layer depth and water volume distribution. Specifically, the reinforcement learning unit 33 abstracts the irrigation environment as a Markov decision process, defines the state space, action space, and reward function. The state space includes parameters such as soil water content and root development status; the action space includes irrigation layer depth adjustment and water volume distribution; the reward function takes into account water resource utilization efficiency and crop growth status. Preferably, the reinforcement learning algorithm uses a deep Q - network (DQN), and the Q - value update formula is:

[0083]

[0084] where Q(s t ,a t ) represents the value function of performing action a t in state s t , α is the learning rate, r t is the immediate reward, γ is the discount factor, and max a Q(s t + 1,a) represents the maximum Q - value of the next state s t+1 . In practical applications, the learning rate α is usually set to 0.01 - 0.05, and the discount factor γ is set to 0.9 - 0.99. These parameter settings enable the algorithm to balance short - term rewards and long - term benefits.

[0085] Furthermore, the reinforcement learning unit 33 realizes the automatic conversion between surface irrigation and subsurface infiltration irrigation modes according to the crop growth stage. Specifically, when the pressure sensor monitors that the roots grow to a specific depth (usually when the crop enters the mid - growth stage and the root depth reaches 20 - 30 cm), the system automatically converts the irrigation mode from surface irrigation to subsurface infiltration irrigation to improve water use efficiency. When the root pressure exceeds a preset threshold (usually 20 MPa), it indicates that the roots have grown to the permeable layer, and at this time, the deep infiltration irrigation mode is activated.

[0086] As Figure 5 shown, in an embodiment of the present invention, the decision analysis module 4 includes a growth index calculation unit 41, a water requirement analysis unit 42, and a scheme generation unit 43. The growth index calculation unit 41 is used to calculate the leaf surface growth index and the underground plant growth index. Preferably, the leaf surface growth index is obtained by identifying the leaf area at each growth stage through image processing technology and according to the leaf area index calculation formula; the underground plant growth index takes into account the total root length and the root cross-sectional area of different root radii.

[0087] The water requirement analysis unit 42 is connected to the growth index calculation unit 41 and is used to calculate the crop water requirement based on the root-shoot ratio threshold. Preferably, the root-shoot ratio threshold includes the leaf-canopy ratio threshold and the total root-shoot ratio threshold. For field crops such as corn, the root-shoot ratio thresholds in the early, middle, and late growth stages are 0.2 - 0.3, 0.5 - 0.7, and 0.3 - 0.5 respectively. The setting of this threshold range is based on a large amount of field test data and can accurately reflect the root and canopy development ratio of crops at different growth stages.

[0088] The scheme generation unit 43 is connected to the water requirement analysis unit 42 and is used to generate a three-dimensional irrigation scheme for crops by combining rainfall information and soil moisture conditions. Specifically, the scheme generation unit 43 realizes the judgment of crop irrigation amount under different climates at different growth stages through fuzzy control and neural network theory. The fuzzy controller calculates the membership function between the crop water requirement and the rainfall, and the membership threshold is usually set to 0.6 - 0.8. When the membership is lower than this threshold, the system judges that supplementary irrigation is needed. The neural network is used to process the leaf canopy evaporation data and predict the future water demand.

[0089] Furthermore, the three-dimensional irrigation scheme includes a combination of surface irrigation in the early growth stage, underground irrigation in the middle growth stage, and surface irrigation in the late growth stage. Preferably, in the early growth stage, the goal is to promote the downward growth of the roots, and surface irrigation is mainly used; in the middle growth stage, the root system is relatively well-developed, and underground irrigation is used to reduce evaporation loss; in the late growth stage, the crop water requirement increases, and a combination of surface and underground irrigation is used to meet the demand. This combination of irrigation modes fully considers the crop growth and development law and can significantly improve the water resource utilization efficiency.

[0090] As Figure 6 shown, the present invention also provides a method for simulating crop root growth and making three-dimensional irrigation decisions, including the following steps:

[0091] S1. Obtain a root growth simulation model: Use micro-CT scanning to construct a three-dimensional root structure database, develop a neural network based on physical information to simulate the water absorption process, fuse the fluid dynamics equation and the deep learning architecture, and establish a multi-scale model considering capillary action and root hair growth.

[0092] Specifically, step S1 includes: First, use micro-CT to scan the root system images, and based on the root system distribution characteristics at different growth stages, obtain a three-dimensional structure database of the root system, and establish a root system physical model based on physical mechanics information; Second, model the relationships among the three-dimensional structure database, soil water distribution, evaporation loss, and root water absorption rate as a long short-term memory artificial neural network, and determine the neural network architecture; Then, adjust the network hyperparameters, and train the neural network through the backpropagation algorithm to obtain a root system growth simulation model; Finally, calculate the root water absorption rate and capillary root water potential after the neural network training. In practical applications, the gradient descent step size of the backpropagation algorithm is usually set to 0.01 - 0.05, and the number of iterations is 500 - 1000 times. These parameter settings can achieve a good balance between computational efficiency and model accuracy.

[0093] S2. Construct an underground drip irrigation system: Configure a retractable drip irrigation capillary array, monitor the root system development status through a pressure oscillation sensor, and use a reinforcement learning algorithm to dynamically adjust the irrigation layer depth and water volume distribution to achieve automatic switching between surface irrigation and underground drip irrigation modes according to the crop growth stage.

[0094] Specifically, step S2 includes: First, configure the drip irrigation capillary array based on the principle of mutual restriction among flow rate, dripper, and water output to obtain a retractable drip irrigation model; Second, use a pressure oscillation sensor to measure the soil water potential and water content of different layers of soil and roots under different drippers in real time, and judge the leakage loss corresponding to the current growth stage; Then, with the number of drippers, dripper distribution, and dripping pressure at the moment of the lowest leakage loss as a reference, achieve the three-dimensional decoupling relationship between the retractable drip irrigation pipe array and soil water potential - water content - plant; Finally, use the reinforcement learning algorithm to iterate and dynamically adjust the irrigation layer depth and water volume distribution to achieve automatic switching between surface irrigation and underground drip irrigation modes according to the crop growth stage. Preferably, the exploration rate of the reinforcement learning algorithm is initially set to 0.3 and gradually reduced to 0.05 as the training progresses to balance the relationship between exploration and exploitation.

[0095] S3. Calculate crop growth indicators: Calculate the leaf surface growth indicators and underground plant growth index according to the crop growth conditions adjusted by the underground drip irrigation mode.

[0096] Specifically, step S3 includes: Use image processing technology to identify the leaf area at each growth stage, and obtain the leaf surface growth indicators of the crop at each growth stage according to the leaf area index calculation formula; Calculate the underground plant growth index using the following formula:

[0097] GI = ∑ i L·S i ,

[0098] where, $L represents the total root length, S iThe cross-sectional area of the root system represents different root radii, and GI represents the underground plant growth index. For different crops, the ideal value range of this index varies. For example, the optimal underground plant growth index of corn during the mid-growing period is usually 30 - 50, while that of rice is 20 - 40.

[0099] S4. Establish an irrigation decision-making system: Combine crop growth indicators, and use fuzzy control and neural network theories to realize the judgment of crop irrigation amounts under different climates in different growth periods. Based on rainfall information and soil moisture conditions, propose a three-dimensional irrigation plan for crops.

[0100] Specifically, step S4 includes: First, calculate the root-shoot ratio of the crop in different growth periods and determine the root-shoot ratio thresholds, which include the leaf-shoot ratio threshold and the total root-shoot ratio threshold; Second, calculate the crop water requirement according to rainfall information, soil moisture conditions, combined with the crop root-shoot ratio threshold, crop water requirement law and three-dimensional irrigation mode; Finally, realize the judgment of crop irrigation amounts in different growth periods through fuzzy control and neural network theories, and obtain an optimized irrigation mode. This optimized irrigation mode includes a combination of surface irrigation in the early growth period, underground irrigation in the mid-growing period, and surface irrigation in the late growth period.

[0101] Preferably, the calculation formula for crop water requirement is:

[0102] W = A t - A a + A evaport + A bloss ,

[0103] where W is the crop water requirement, A t is the irrigation amount during the growth period, A a is the precipitation, A evaport is the transpiration amount, and A bloss represents the irrigation leakage amount. For different crops and climate conditions, the system sets different parameter thresholds. For example, in temperate regions, the transpiration threshold during the mid-growing period of summer corn is usually 3 - 5 mm / day, and the irrigation leakage threshold is 10 - 15% of the total irrigation amount.

[0104] In summary, the crop root system growth simulation and three-dimensional irrigation decision-making system and method provided by the present invention obtain three-dimensional root structure data through micro-CT scanning technology, fuse fluid dynamics and deep learning to construct a multi-scale model, configure a scalable drip irrigation system and combine a reinforcement learning algorithm to realize intelligent irrigation control, and use fuzzy control and neural network theories to establish an irrigation decision-making system, forming a complete set of crop precision irrigation solutions. This system can not only accurately simulate the dynamic growth of the root system, but also automatically adjust the irrigation strategy according to the water requirement characteristics of different growth periods of crops, significantly improving the water resource utilization efficiency, promoting the healthy growth of crops, and having important significance for promoting agricultural modernization and sustainable development.

[0105] Example 1: Application of the Three-dimensional Irrigation System in Maize Fields

[0106] In this embodiment, the crop root growth simulation and three-dimensional irrigation decision-making system of the present invention is applied to the maize planting area in the North China Plain. The annual precipitation in this area is about 500 - 600 mm, but the precipitation distribution is uneven. There is concentrated rainfall in summer and a large evaporation amount, and periodic droughts often occur, affecting the normal growth and development of maize.

[0107] First, maize plants at different growth stages are selected from the experimental field, and their roots are non-destructively scanned using micro-CT scanning technology. The spatial resolution of the scanning device is set to 5 micrometers to ensure that the structure at the root hair level can be clearly captured. Five representative plants are selected for scanning at each growth stage (seedling stage, jointing stage, heading stage, filling stage, maturity stage) to obtain complete three-dimensional root structure data.

[0108] After the scanning is completed, the images are processed through Avizo Flow V9.3 software, including image segmentation, filtering enhancement, and three-dimensional reconstruction. In view of the characteristics of maize roots with many branches and complex morphology, an adaptive threshold segmentation algorithm is particularly adopted to make the segmentation accuracy reach more than 95%. Finally, a database containing more than 1000 groups of three-dimensional structures of maize roots at different growth stages is constructed.

[0109] Based on the obtained three-dimensional structure database, a long short-term memory neural network specifically for maize roots is constructed. The network architecture includes 3 input nodes (soil water content, capillary root water potential, root water absorption rate), 2 hidden layers (the first layer contains 8 neurons, and the second layer contains 6 neurons), and 2 output nodes (predicted root water absorption rate and water potential). The activation function uses the Softplus function to improve the smoothness and continuity of the model.

[0110] The neural network is trained with the data collected from the actual maize experimental field, using the Adam optimizer (initial learning rate 0.001, dynamically adjusted) and the mean square error loss function. The ratio of the training data set to the validation data set is 7:3, and the number of training iterations is 800 times. With this setting, the root mean square error (RMSE) of the model on the validation set reaches 0.08 MPa, and the correlation coefficient (R 2 ) reaches 0.92, indicating that the model has a high prediction accuracy.

[0111] In the maize experimental field, a retractable drip irrigation capillary array is configured according to the root distribution characteristics of maize. The spacing of the surface irrigation pipelines is set to 30 cm, and the spacing of the deep irrigation pipelines (depth 40 cm) is set to 50 cm. The drip head spacing is 25 cm, so that the irrigation uniformity reaches more than 90%.

[0112] Five pressure oscillation sensors are installed in each test area, distributed at different depths (surface layer, 15 cm, 30 cm, 45 cm, 60 cm), with a sensitivity of 0.01 MPa, and the sampling frequency is set to once every 2 minutes. Based on the data of these sensors, the system can accurately monitor the development status and activities of the roots at different depths.

[0113] The reinforcement learning unit adopts the Deep Q-Network (DQN) algorithm. The state space includes the soil water content and root activity intensity at 5 different depths, and the action space includes 6 irrigation strategies (3 irrigation intensities × 2 irrigation positions). The reward function comprehensively considers the water use efficiency (weight 0.6) and the crop growth status (weight 0.4). The learning rate is set to 0.03, and the discount factor is 0.95 to balance short-term and long-term benefits.

[0114] Through the operation of this system, the system automatically implements differential irrigation strategies at different growth stages of corn:

[0115] From the emergence stage to the jointing stage (early growth stage): Surface irrigation is mainly adopted, with an irrigation amount of 10 - 15 mm per time, at intervals of 5 - 7 days, to promote the downward growth of the roots.

[0116] From the jointing stage to the filling stage (mid - growth stage): When the root depth reaches 35 cm (usually 15 days after the jointing stage), the system automatically switches to the underground infiltration irrigation mode, with an irrigation amount of 20 - 25 mm per time, at intervals of 7 - 10 days, to reduce evaporation loss and improve water use efficiency.

[0117] From the filling stage to the maturity stage (late growth stage): A combination of surface and underground irrigation is adopted, dynamically adjusted according to the soil moisture condition and the predicted evaporation amount, with an irrigation amount of 15 - 20 mm per time, at intervals of 5 - 8 days.

[0118] Compared with the traditional irrigation method, the application effect of this system in corn planting is remarkable: The irrigation water volume is reduced by 25.6%, the water use efficiency is increased by 34.2%, and the corn yield is increased by 12.3%. Especially under the condition of drought stress in the mid - growth stage, the system can accurately identify the water demand of the roots and adjust the irrigation parameters, significantly reducing the impact of drought on the crops.

[0119] Example 2: Precision irrigation application in vineyards

[0120] In this example, the system of the present invention is applied to a grape planting park in the Shandong Peninsula. This area has high temperature and abundant rainfall in summer and cold and dry in winter. Grapes have obvious differences in water demand at different growth stages and are extremely sensitive to the timing and amount of irrigation.

[0121] In view of the characteristics of high lignification degree and deep vertical distribution of grape roots, the micro-CT scanning parameters were specifically optimized. The scanning voltage was set at 120 kV, the current was 100 μA, the acquisition angle was 360°, and the slice thickness was 0.5 mm. Since the grape root system has a large distribution range, a method of segmented scanning and stitching was adopted to obtain the complete three-dimensional root structure.

[0122] For different growth stages of grapes (germination stage, flowering stage, fruit development stage, maturity stage), the root data of 5-8 representative plants were collected respectively to construct a dedicated database containing more than 800 sets of three-dimensional structure data. When processing images, especially for the age stratification characteristics of grape roots, a root classification algorithm based on density analysis was developed to divide the roots into three levels: primary lignified roots, secondary lignified roots, and absorbing roots, providing more refined data support for the subsequent model.

[0123] Considering the particularity of grape roots, a hierarchical structure was adopted in the neural network design, including a main network and three branch networks. The main network processes the overall root structure and water relationship, and the three branch networks process the water absorption characteristics of different types of roots (primary lignified roots, secondary lignified roots, and absorbing roots) respectively.

[0124] The network training adopted the transfer learning method. First, the network was pre-trained with a large amount of general crop root data, and then fine-tuned with grape-specific data. This method overcame the problem of relatively limited grape root sample quantity and improved the model performance. The optimizer adopted the stochastic gradient descent method with momentum (momentum SGD), the initial learning rate was 0.005, the momentum coefficient was 0.9, and a learning rate decay strategy was adopted. After 1200 iterations of training, the average relative error of the model on the validation set was less than 7%.

[0125] In view of the fact that the terrain of vineyards is mostly sloping land, a scalable drip irrigation system adaptable to terrain changes was designed. Pressure-compensating drippers were installed at different altitude positions to ensure the irrigation uniformity at each position under sloping land conditions. The drip irrigation pipes adopted a tree-like layout. The main pipe was laid along the row, and the branch pipes extended to the roots of each grape plant.

[0126] The underground irrigation pipes were installed at two depths of 45 cm and 80 cm. Four pressure oscillation sensors were arranged around each plant position to monitor the soil moisture status and root activities at different depths (20 cm, 40 cm, 60 cm, 90 cm) respectively. The sensor data was collected every 30 minutes and transmitted to the central control system in real time through a wireless transmission network.

[0127] The reinforcement learning algorithm has been improved to meet the special needs of grapes. The Double Q Network (Double DQN) structure is adopted to increase the ability to suppress the over-estimation problem of the model. In addition, the experience replay mechanism (ExperienceReplay) is added to improve learning efficiency and stability. The reward function specifically considers grape quality factors, including sugar content growth rate (weight 0.3), fruit size uniformity (weight 0.2), water use efficiency (weight 0.3) and plant growth status (weight 0.2).

[0128] The application of the system in the vineyard has produced significant results, especially in the following aspects:

[0129] 1. Precise water management: During the fruit development period, the system automatically adjusts the irrigation layer according to the root growth depth, greatly reducing surface evaporation losses and improving water utilization efficiency by 41.7%.

[0130] 2. Significant improvement in quality: By precisely controlling the degree and timing of drought stress, the system moderately controls water before maturity, increasing the glucose content by 2.3°Brix and the content of fruit flavor substances by 15.6%, significantly improving the quality of the grapes.

[0131] 3. Enhanced stress resistance: Under continuous high temperature weather, the system can automatically increase the frequency of deep irrigation at night, reduce the surface temperature, alleviate the adverse effects of high temperature on grapes, and increase the photosynthetic efficiency of leaves by 18.9%.

[0132] 4. Obvious water-saving effect: Compared with traditional drip irrigation technology, this system saves 32.5% of water during the entire growth period. Under the premise of ensuring yield and quality, it saves about 35 cubic meters of irrigation water per mu.

[0133] This embodiment shows that the system of the present invention is not only suitable for field crops, but can also be effectively applied to horticultural crops with higher economic value. By accurately simulating root growth and making intelligent irrigation decisions, the dual goals of efficient use of water resources and significant improvement in crop quality are achieved.

[0134] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. Crop root growth simulation and three-dimensional irrigation decision-making system, characterized in that, Comprising: A data acquisition module, configured to acquire three-dimensional structure data of crop roots; A neural network processing module, connected to the data acquisition module, and configured to construct a root growth simulation model based on the three-dimensional structure data; An irrigation control module, connected to the neural network processing module, and configured to control a drip irrigation system based on the root growth simulation model; A decision analysis module, connected to the neural network processing module and the irrigation control module, and configured to generate a three-dimensional irrigation plan according to crop growth indicators.

2. The system according to claim 1, characterized in that The data acquisition module includes: A scanning unit, configured to scan crop roots using a micro-CT device; An image processing unit, connected to the scanning unit, and configured to process the scanned image; A database construction unit, connected to the image processing unit, and configured to establish a three-dimensional root structure database based on the processed image.

3. The system according to claim 1, wherein The neural network processing module includes: A network architecture unit, configured to construct a long short-term memory artificial neural network architecture; A training unit, connected to the network architecture unit, and configured to train a neural network using the three-dimensional structure data; A calculation unit, connected to the training unit, and configured to calculate root water absorption rate and capillary root water potential.

4. The system according to claim 3, wherein The calculation unit calculates the root water absorption rate and capillary root water potential through the following formula: P=w0+∑w i x i , Among them, P represents the calculated pressure value, w i represents the fitting parameter, x i represents the water absorption rate of the root system.

5. The system according to claim 1, characterized in that The irrigation control module includes: A drip irrigation configuration unit, configured to configure a retractable drip irrigation lateral array; A sensing and monitoring unit, connected to the drip irrigation configuration unit, and configured to monitor the root development status through a pressure oscillation sensor; A reinforcement learning unit, connected to the sensing and monitoring unit, and configured to dynamically adjust the irrigation layer depth and water volume distribution.

6. The system according to claim 5, wherein The reinforcement learning unit realizes automatically switching between surface irrigation and subsurface infiltration irrigation modes according to the crop growth stage.

7. The system according to claim 1, wherein The decision analysis module includes: A growth index calculation unit, configured to calculate leaf surface growth indicators and underground plant growth indices; A water demand analysis unit, connected to the growth index calculation unit, and configured to calculate the crop water demand based on the root-shoot ratio threshold; A plan generation unit, connected to the water demand analysis unit, and configured to generate a three-dimensional irrigation plan for the crop in combination with rainfall information and soil moisture conditions.

8. The system according to claim 7, characterized in that, The plan generation unit realizes the judgment of crop irrigation amount under different climates in different growth stages through fuzzy control and neural network theory.

9. The system according to claim 7, wherein The three-dimensional irrigation plan includes a combination mode of surface irrigation in the early growth stage, subsurface irrigation in the middle growth stage, and surface irrigation in the late growth stage of the crop.

10. A method for simulating crop root growth and making a three-dimensional irrigation decision, using the system according to any one of claims 1-9, comprising the following steps: S1. Obtain a root growth simulation model: Construct a three-dimensional root structure database by using micro-CT scanning, develop a neural network based on physical information to simulate the water absorption process, fuse the fluid dynamics equation and the deep learning architecture, and establish a multi-scale model considering capillary action and root hair growth; S2. Construct a subsurface infiltration irrigation system: Configure a retractable drip irrigation lateral array, monitor the root development status through a pressure oscillation sensor, and use a reinforcement learning algorithm to dynamically adjust the irrigation layer depth and water volume distribution, so as to realize automatically switching between surface irrigation and subsurface infiltration irrigation modes according to the crop growth stage; S3. Calculate crop growth indicators: Calculate the leaf surface growth indicators and the underground plant growth index based on the crop growth conditions adjusted according to the subsurface drip irrigation mode; S4. Establish an irrigation decision-making system: Combine the crop growth indicators, and use the fuzzy control and neural network theories to realize the judgment of the crop irrigation amount under different climates in different growth periods. According to the rainfall information and soil moisture content, propose a three-dimensional irrigation plan for the crops.

Images