Salinization farmland protection forest configuration optimization system and method thereof

Abstract

The invention relates to the field of farmland ecological protection, in particular to a salinization farmland protection forest configuration optimization system and method, and the system comprises a salinization monitoring module, an edge calculation module, a database, a three-dimensional model module, a model optimization module and a configuration scheme optimization module. The salinization monitoring module acquires data through an unmanned aerial vehicle and a hyperspectral imaging spectrometer and determines a soil salinity distribution state, the edge calculation module receives the data and carries out protection forest configuration model calculation, the database stores a calculation model and soil salinity and vegetation growth state data, and the three-dimensional model module generates a three-dimensional spatial-temporal model of forest growth. The model optimization module utilizes deep reinforcement learning to optimize a soil salinity and underground water level simulation model, the protection forest configuration effect is improved, the configuration scheme optimization module generates an optimal protection forest configuration scheme, a complex interaction process is accurately represented, and an efficient and accurate technical means is provided for ecological protection of salinized farmland.

Description

Technical Field

[0001] The present invention relates to the field of farmland ecological protection, and particularly to an optimized system and method for configuring a shelter forest in saline-alkali farmland. By using deep reinforcement learning technology, it realizes the scientific configuration of the shelter forest in saline-alkali farmland, improves the utilization efficiency of saline-alkali land and the ecological benefits of the shelter forest. Background Art

[0002] Salinization is one of the global soil degradation problems, seriously affecting agricultural production and the ecological environment. According to statistics, the area of saline-alkali land in China accounts for about 4.88% of the total national land area, mainly distributed in arid and semi-arid regions in the north and coastal areas. As an important biological measure for improving saline-alkali soil, the shelter forest plays an irreplaceable role in carbon sequestration and emission reduction, soil and water conservation, wind prevention and sand fixation, etc.

[0003] Traditional methods for configuring shelter forests mainly rely on empirical judgment and simple index evaluation, lacking accurate modeling of the salt transport and tree water absorption processes, and unable to adapt to complex and changeable environmental conditions. With the development of technologies such as the Internet of Things and artificial intelligence, data-driven intelligent methods for configuring shelter forests have become a research hotspot. However, the existing intelligent optimization systems still have the following problems: First, the representation of the complex interaction relationship between salt transport and tree water absorption is insufficient; second, there is a lack of an adaptive optimization mechanism to adapt to dynamic environmental changes; third, complex models are difficult to operate efficiently on resource-constrained edge devices.

[0004] Therefore, there is an urgent need for a shelter forest configuration optimization system that can accurately model the salt-tree interaction relationship, support dynamic optimization configuration, and can operate efficiently on edge devices, so as to improve the ecological restoration efficiency and economic benefits of saline-alkali farmland. Summary of the Invention

[0005] The purpose of the present invention is to provide an optimized system and method for configuring a shelter forest in saline-alkali farmland, overcoming the deficiencies of the prior art and realizing the scientific configuration and dynamic optimization of the shelter forest in saline-alkali farmland.

[0006] Specifically, the purposes of the present invention include: establishing a salt transport-tree water absorption coupling model to accurately represent the complex interaction relationship between the two; constructing an optimized framework for configuring a shelter forest based on deep reinforcement learning to realize the dynamic adaptive optimization of the configuration scheme; developing a lightweight model suitable for edge devices to support on-site real-time decision-making.

[0007] The present invention proposes an optimized system for configuring a shelter forest in saline-alkali farmland, including: A salinization monitoring module, which is used to collect the ground reflectance and the vegetation leaf reflectance through an unmanned aerial vehicle, and use a hyperspectral imaging spectrometer to realize the inversion of soil conductivity to determine the soil salt distribution state; An edge computing module, communicatively connected to the salinization monitoring module, for receiving the soil salt distribution state data and calculating a shelter forest configuration model; A database, connected to the edge computing module, for storing calculation models, soil salts, and vegetation growth state data; A three-dimensional model module, connected to the database and the edge computing module, for generating a three-dimensional spatio-temporal model of forest growth in a three-dimensional environment using soil salt and groundwater level data; A model optimization module, connected to the three-dimensional model module and the database, for optimizing the soil salt and groundwater level simulation models based on deep reinforcement learning to improve the shelter forest configuration effect; A configuration scheme optimization module, connected to the model optimization module, for generating an optimal shelter forest configuration scheme based on the model optimized by the deep reinforcement learning.

[0008] Preferably, it further includes a meteorological monitoring module, which is connected to the model optimization module for obtaining meteorological condition data such as air temperature, precipitation, wind direction, and wind speed to determine an optimal growth strategy; wherein, the model optimization module further determines an optimal shelter forest configuration scheme using the trained deep reinforcement learning model, specifically: taking the tree spacing and environmental parameters to be planted as the state, and the shelter forest configuration efficiency and the reduction of irrigation water consumption as the reward function, and determining the optimal shelter forest configuration scheme according to the trained deep reinforcement learning model.

[0009] Preferably, the model optimization module establishes a salt transport-forest water absorption coupling model through an improved Transformer architecture, and the salt transport-forest water absorption coupling model includes: A salt balance equation for describing the dynamic change of salts in the soil; A plant absorption equation for describing the absorption process of soil water and salts by plants; A stand water holding rate equation for describing the interception and retention ability of the stand for water; A forest water absorption function for describing the water absorption ability and efficiency of forests under different conditions; Wherein, the model optimization module uses the forest growth parameters and the salt content distribution output by the salt transport model as training data to construct a shelter forest configuration state set and a shelter forest structure optimization model, takes the current state as the input, obtains the next state through the shelter forest configuration state set, and calculates the corresponding shelter forest configuration efficiency and the reduction of irrigation water consumption at the next state as the reward function, and generates an optimal shelter forest configuration scheme through the deep reinforcement learning model.

[0010] Preferably, the shelter forest configuration state set includes: a state set, a decision set, a reward value, and a change range of the planting area; wherein, the state set includes the planting area, plant spacing, row spacing, groundwater level, soil saturated water content, and soil salinity distribution parameters.

[0011] Preferably, the trained deep reinforcement learning model includes a control function and a loss function; wherein, the control function is used to generate a decision based on the current state, the loss function is used to evaluate the gap between the predicted value and the true value, the model uses a discount factor to balance short-term and long-term rewards, and the probability distribution of the shelter forest configuration state is used to guide the reinforcement learning process.

[0012] Preferably, the update of the shelter forest configuration state set includes: evaluating the current shelter forest configuration through a deep reinforcement learning model, inputting the soil salinity and groundwater level into the coupling model, outputting the soil salinity and groundwater level in the next state through the coupling model, and repeating the iterative update until an optimal configuration plan is obtained.

[0013] Preferably, the parameter optimization in the salt transport - forest water absorption coupling model is achieved through the gradient descent method, specifically: dividing the salt content into multiple salinity levels at equal intervals through linear interpolation, and solving according to the salt balance equation and plant water absorption equation to obtain the stress resistance of the vegetation under different salt transport processes, and then obtaining the trained shelter forest configuration state set.

[0014] Preferably, the constraint conditions of the gradient descent method include: Unit equivalent resistance model state parameter constraint; Permeability parameter constraint; Upper and lower boundary constraints of soil salinity and its saturation; Soil water holding capacity constraint; The constraint conditions ensure the stability of the model training process and the effectiveness of the results.

[0015] Preferably, the edge computing module compresses the prediction model to a scale suitable for edge devices through knowledge distillation technology, and realizes real-time computing through a distributed edge network; the knowledge distillation technology reduces the dimensionality of the feature map channels, retains the key decision-making ability of the model, and enables the system to update the shelter forest configuration plan hourly in a resource-constrained environment.

[0016] The optimization method of the shelter forest configuration optimization system for saline-alkali farmland includes the following steps: Obtaining data of the saline-alkali monitoring area through the UAV flight control system, using a high-resolution multispectral camera to obtain the ground reflectance and vegetation leaf reflectance, and establishing the relationship between the vegetation leaf reflectance - water absorption index and conductivity; Set up soil salinity monitors and ground water level monitors in the monitoring area, collect data on salinity content and ground water level, obtain environmental meteorological information, and establish a dataset of soil saturated water content and soil saturated salinity; Use the dataset to train the soil bulk density and salt transport - forest water absorption coupling model, and obtain the state update function and the shelter forest configuration set based on deep reinforcement learning; In the shelter forest configuration state set, determine the optimal shelter forest configuration using the optimal adjustment parameters and control strategy; Take pictures of the target farmland by drones to obtain the salt state distribution, dynamically update the shelter forest configuration according to the current soil salinity and underground water level, and generate the spatio - temporal model of forest growth in a three - dimensional environment using the prediction model.

[0017] The present invention has the following beneficial effects: 1. By establishing a salt transport - forest water absorption coupling model through an improved Transformer architecture, the accurate characterization of the complex interaction process of salt transport and forest water absorption is realized. Compared with traditional methods, the salt control efficiency is increased by about 41%.

[0018] 2. Adopt a deep reinforcement learning framework to construct an optimized decision - making system for shelter forest configuration, take the tree spacing and environmental parameters as the state, and the reduction of configuration efficiency and irrigation water consumption as the reward function. Compared with traditional methods, the irrigation water consumption is reduced by about 33%.

[0019] 3. Use knowledge distillation technology to compress and deploy the complex model to edge devices, realize the update of the shelter forest configuration plan at the hourly level, and significantly improve the practicality and response speed of the system.

[0020] 4. Establish a complete closed - loop optimization system of monitoring - analysis - decision - execution - feedback, realize the intelligent management of the whole life cycle of the shelter forest configuration in saline - alkali farmland, and provide technical support for the sustainable use of saline - alkali land. Description of the Drawings

[0021] Figure 1 It is a schematic diagram of the overall architecture of the optimized system for shelter forest configuration in saline - alkali farmland of the present invention; Figure 2 It is a schematic diagram of the structure of the salt transport - forest water absorption coupling model of the present invention; Figure 3 It is a schematic diagram of the process of the deep reinforcement learning optimization framework of the present invention; Figure 4 It is a schematic diagram of the implementation scheme of edge computing and knowledge distillation of the present invention; Figure 5 It is a schematic diagram of the process of the optimized method for shelter forest configuration in saline - alkali farmland of the present invention. Detailed Embodiments

[0022] Please refer to the appendix Figures 1-5 , and the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0023] As Figure 1 shown, the optimized system for configuring shelter forests in saline-alkali farmland provided by the present invention includes a salinization monitoring module 1, an edge computing module 2, a database 3, a three-dimensional model module 4, a model optimization module 5, a configuration plan optimization module 6, and a meteorological monitoring module 7.

[0024] The salinization monitoring module 1 is used to collect the ground reflectance and the vegetation leaf reflectance through an unmanned aerial vehicle, and use a hyperspectral imaging spectrometer to realize the inversion of soil conductivity and determine the soil salt distribution state. In practical applications, for the typical saline-alkali farmland in Bayannur City, Inner Mongolia, China, the present invention uses an unmanned aerial vehicle equipped with a multispectral camera, and the flight altitude is set to 80-120 meters to obtain multi-band images with a ground resolution of 10 cm. The salinization monitoring module 1 uses the support vector machine regression algorithm to establish a relationship model between the vegetation leaf reflectance and the soil conductivity. Tests in severely saline-alkali areas such as Bayannur City, Inner Mongolia show that the inversion accuracy can reach more than 85%. The model can be expressed as: , wherein, is the soil conductivity, with the unit of dS / m, which characterizes the degree of soil salinization, and the typical saline-alkali farmland value range is 4-16 dS / m; is the support vector coefficient, which is determined by the training process; is the kernel function. In this embodiment, the radial basis function (RBF) is used, which is expressed as , is the kernel parameter, usually taking values of 0.1-0.5; is the vegetation reflectance feature vector of the training sample, which includes the reflectance values of five bands: red (R), green (G), blue (B), near-infrared (NIR), and red edge (RE); is the vegetation reflectance feature vector of the sample to be predicted; is the bias constant; is the number of support vectors, usually of the training set. In the saline-alkali farmland test area of Bayannur City, Inner Mongolia, the correlation coefficient R between the soil conductivity measured by this method and the laboratory analysis results reaches 0.87.

[0025] The edge computing module 2 is communicatively connected to the salinization monitoring module 1, and is used to receive the data of the soil salt distribution status and calculate the shelter forest configuration model. Since salinized farmlands are mostly distributed in remote areas with limited network conditions and energy supply, the present invention preferably adopts a low-power WiFi edge computing gateway. The processor adopts the ARM Cortex-A72 architecture, with a main frequency of 1.5 GHz, a memory of 4 GB, a storage of 32 GB, and the power consumption is controlled within 5 W. In the saline-alkali land shelter forest experimental area of Bayannur City, Inner Mongolia, this edge device is powered by solar energy and has been running stably continuously for more than 500 days, meeting the requirements of long-term operation in the field environment.

[0026] The database 3 is connected to the edge computing module 2 and is used to store the calculation model, soil salt, and vegetation growth status data. Considering the limitations of the storage space and computing resources of the edge device, the present invention adopts a lightweight SQLite database system and designs three main data tables including an environmental monitoring table, a model parameter table, and a configuration scheme table. The environmental monitoring table stores monitoring data such as salt, water level, and meteorology, and uses a timestamp index; the model parameter table stores model weights and hyperparameters and supports version management; the configuration scheme table stores historical configuration schemes and their effect evaluations. In the saline-alkali land improvement demonstration area of Bayannur City, Inner Mongolia, the database realizes incremental data update and compressed storage. The space occupied by one year's monitoring data is only 2.8 GB, greatly reducing the storage requirements.

[0027] The three-dimensional model module 4 is connected to the database 3 and the edge computing module 2, and is used to generate a three-dimensional spatio-temporal model of forest growth in a three-dimensional environment using the soil salt and groundwater level data. In the embodiment of the present invention, the three-dimensional model module 4 is developed based on the Unity engine and uses a parametric modeling method to construct a forest growth model. This module can convert two-dimensional salt distribution and water level data into a three-dimensional visualization scene, intuitively showing the dynamic changes in the growth status of shelter forests in different salt environments. In the saline-alkali land treatment project of Bayannur City, Inner Mongolia, this module successfully simulated the growth of salt-tolerant tree species such as Tamarix chinensis and Elaeagnus angustifolia under different salt gradients, providing an intuitive decision-making basis for forestry workers.

[0028] The model optimization module 5 is connected to the three-dimensional model module 4 and the database 3, and is used to optimize the soil salt and groundwater level simulation models based on deep reinforcement learning to improve the configuration effect of the shelter forest. In the saline-alkali land treatment project of Bayannur City, Inner Mongolia, the model optimization module 5 realized the automatic optimization of different tree species combinations and different plant spacings, and increased the salt control efficiency from 25% of the traditional method to 41%, achieving a significant technological breakthrough.

[0029] The configuration scheme optimization module 6 is connected to the model optimization module 5 and is used to generate an optimal shelter forest configuration scheme based on the model optimized by deep reinforcement learning. In practical applications, this module can generate a complete scheme including tree species selection, row spacing design, planting density, and spatial layout, and provide visual configuration drawings for convenient on-site implementation.

[0030] In addition, the present invention also includes a meteorological monitoring module 7, which is connected to the model optimization module 5 and is used to obtain meteorological condition data such as air temperature, precipitation, wind direction, and wind speed to determine the optimal growth strategy. In the project of treating salinization in Horqin Sandy Land, Inner Mongolia, the real-time meteorological data collected by the meteorological monitoring module 7 helped the system adjust the irrigation strategy of the shelter forest, increasing the irrigation frequency during the dry period and reducing the irrigation amount during the rainy period, reducing the irrigation water consumption by 33% while maintaining a good growth state.

[0031] One of the core innovations of the present invention is to establish a salt transport-forest water absorption coupling model through an improved Transformer architecture. As Figure 2 shown, this model includes four key components: a salt balance equation, a plant absorption equation, a stand water holding rate equation, and a forest water absorption function, realizing the accurate characterization of the complex interaction process between salt transport and forest water absorption.

[0032] The salt balance equation describes the dynamic change process of salt in the soil, considering the coupling relationship between salt and water movement. In the saline-alkali farmland environment in the arid northwest region, the salt balance equation can be expressed as: , where is the soil volumetric water content, dimensionless, representing the volume of water in a unit volume of soil, and the typical value in saline-alkali farmland is 0.2 - 0.4; is the salt concentration in the soil solution, with the unit of mg / L, reflecting the content of dissolved salts in the soil solution, and it can reach 5000 - 15000 mg / L in severely saline-alkali soil; is the time, with the unit of d, representing the time variable in the simulation process; is the soil depth, with the unit of cm, representing the vertical distance from the ground surface downward, and the model usually considers the depth range of 0 - 200 cm; is the salt diffusion coefficient, with the unit of cm² / d, describing the ability of salt to diffuse in the soil, which is related to the soil texture and water content, about 0.5 - 2 cm² / d in clay and up to 5 - 10 cm² / d in sandy soil; is the soil water flux, with the unit of cm / d, representing the amount of water passing through a unit area per unit time, affected by rainfall, evaporation, and irrigation, and the typical value range is 0.1 - 5 cm / d; It is the sink term of plant salt absorption, with the unit of mg / (L·d), representing the absorption rate of salt in the soil solution by plants.

[0033] This equation describes three main salt migration processes in the soil: the diffusion process (the first term), the convection process (the second term), and the plant absorption process (the third term). In the saline-alkali land improvement area of Bayannao'er City, Inner Mongolia, this equation is applied to simulate the vertical distribution change of salt, and the average error between the predicted result and the measured data is controlled within 10%, providing accurate salt environment information for the shelter forest configuration.

[0034] The plant absorption equation describes the absorption process of soil water and salt by plants, reflecting the influence of plant physiological characteristics on water and salt absorption. Combining with the monitoring data of Tamarix chinensis shelter forest in the saline-alkali land of Bayannao'er City, Inner Mongolia, the plant absorption equation is expressed as: , Among them, is the plant water absorption rate, with the unit of , representing the water absorption amount of plant roots in unit volume of soil; is the plant absorption intensity, with the unit of , which is a basic parameter reflecting the plant water absorption capacity. For Tamarix chinensis, it is about , for Elaeagnus angustifolia, it is about , and for Populus, it is about ; is the absorption coefficient, dimensionless, related to the root distribution density. For shallow-rooted tree species, it is about 0.7 - 0.9, and for deep-rooted tree species, it is about 0.5 - 0.7; is the diameter of the average stem of the plant, with the unit of cm. The stem diameter range of common tree species in the shelter forest is 5 - 20 cm; is the soil water potential influence function, dimensionless, expressed as , where is the soil water potential, is the semi-water absorption potential, is the shape parameter; is the salt concentration influence function, dimensionless, expressed as , where is the semi-inhibitory salt concentration, is the morphology parameter.

[0035] In the shelter forest demonstration area of the saline-alkali land in Bayannao'er City, Inner Mongolia, this equation successfully explains the water absorption differences of different tree species (such as Tamarix chinensis, Elaeagnus angustifolia, and Fraxinus chinensis) under salt stress, providing a scientific basis for the selection of shelter forest tree species. For example, it is monitored that when the soil salt concentration reaches 10 g / kg, Tamarix chinensis still maintains normal water absorption capacity (only a 20% reduction), while under the same conditions, the water absorption capacity of Populus is reduced by more than 60%.

[0036] The stand water - holding rate equation describes the water - interception and retention capacity of a stand. Based on the monitoring data of shelter forests in the Hexi Corridor, it is expressed as: , where is the stand water - holding rate, dimensionless, representing the proportion of the water intercepted by the stand in the total precipitation; is the stand water - holding rate at saturated water content, dimensionless, representing the maximum water - holding capacity of the stand. The typical value for mature shelter forests is 0.2 - 0.3; is the attenuation coefficient, dimensionless, related to tree species characteristics and stand structure. For coniferous forests, it is about 0.5 - 0.6, and for broad - leaved forests, it is about 0.6 - 0.7; is the leaf area index, dimensionless, representing the leaf area per unit ground area. For mature shelter forests, it is usually 3 - 5.

[0037] In the saline - alkali land improvement project in Bayannao'er City, Inner Mongolia, the deviation between the stand water - holding rate calculated by applying this equation and the measured value is controlled within 15%, accurately evaluating the impact of shelter forests on the water balance. For example, for a poplar - tamarisk mixed shelter forest with a plant spacing of 3m and a row spacing of 6m, when the leaf area index reaches 4.2, it can intercept about 24% of the precipitation, significantly reducing surface runoff and effectively controlling the horizontal migration of salts.

[0038] The tree water - absorption function describes the water - absorption capacity and efficiency of trees under different conditions. Combining with the experimental data of saline - alkali land improvement in Bayannao'er City, Inner Mongolia, it is expressed as: , where is the water - absorption rate at a specific soil water potential and depth , with the unit of cm / d; is the water - potential response function, dimensionless, expressed as , where is the soil water potential (negative value), is the semi - water - absorption water potential, is the shape parameter; is the root - distribution function, dimensionless, expressed as , where is the soil depth, is the maximum root depth, is the shape parameter; is the maximum water - absorption rate, with the unit of cm / d, related to tree species and growth stage. For young forests, it is about 0.5 - 0.8 cm / d, and for mature forests, it can reach 0.8 - 1.2 cm / d.

[0039] In the saline-alkali land treatment project in Bayannur City, Inner Mongolia, this function successfully simulated the root water uptake distribution in soil layers at different depths, revealing the mechanism by which the shelter forest adapts to salt stress. For example, it was found that under salt stress, Tamarix chinensis will increase the proportion of deep roots, reduce the activity of surface roots, decrease water uptake in high-salt areas, and shift the focus of water uptake to the deep soil with low salt content. This finding provides an important basis for the rational configuration of the shelter forest.

[0040] These four equations together constitute the mathematical basis of the coupled model of salt transport - forest water uptake. The model optimization module 5 uses the forest growth parameters (such as tree species, density, growth stage) and the salt content distribution output by the salt transport model as training data to construct the state set of shelter forest configuration and the optimization model of shelter forest structure. In the saline-alkali land treatment demonstration area in Bayannur City, Inner Mongolia, this model successfully predicted the impact of different shelter forest configurations on salt distribution, with a prediction accuracy of over 85%, significantly improving the scientific nature of shelter forest configuration.

[0041] As Figure 3 shown, the present invention uses a deep reinforcement learning framework to construct an optimization decision-making system for shelter forest configuration. In view of the long-term and complex characteristics of saline-alkali control, this framework includes three key links: state space construction, decision model training, and plan optimization and implementation.

[0042] First, construct the state set of shelter forest configuration, including the state set, decision set, reward value, and the change range of planting area. Among them, the state set includes parameters such as planting area A, plant spacing, row spacing, groundwater level, soil saturated water content, and salt distribution S in the soil. In the practical application of the shelter forest in saline-alkali land in Bayannur City, Inner Mongolia, the typical range of the planting area A is 1 - 100 hectares, which is determined according to the size and shape of the plot; the selection range of plant spacing is 1 - 5 meters, usually 2 - 3 meters in severely saline-alkali areas, and 1 - 2 meters in mildly saline-alkali areas; the range of row spacing is 2 - 10 meters, mainly considering the needs of mechanized operation and ventilation, and generally 4 - 6 meters is selected; the monitoring depth of the groundwater level is 0 - 5 meters, and the ideal groundwater level for saline-alkali farmland is controlled at 1.5 - 2.5 meters. If it is too shallow, it is easy to cause secondary salinization, and if it is too deep, it will affect plant growth; the soil saturated water content varies with soil types, about 0.3 - 0.35 cm³ / cm³ for sandy soil and about 0.4 - 0.5 cm³ / cm³ for clay soil; the salt distribution S in the soil is represented by a three-dimensional tensor, including the salt concentration values in the three spatial dimensions of x, y, and z, with a resolution of 10m × 10m × 0.2m.

[0043] The decision set contains various adjustment strategies for shelter forest configuration, such as the adjustment amount of plant spacing (step size of ±0.5m), the adjustment amount of row spacing (step size of ±1.0m), tree species selection (selected from a predefined salt-tolerant tree species library), etc. In the saline-alkali land treatment project in Horqin Sandy Land, Inner Mongolia, the decision set also includes parameters such as irrigation strategy adjustment and mixed planting ratio adjustment, further enriching the optimization dimensions.

[0044] The reward value consists of two parts: the shelter forest configuration efficiency and the reduction of irrigation water consumption. The calculation formula is: , Among them, is the total reward value; is the salt control efficiency, calculated as , and are the soil salt contents before and after configuration, respectively; is the reduction rate of irrigation water consumption, calculated as , and are the standard irrigation amount and the actual irrigation amount, respectively; and are weight coefficients, adjusted according to the application scenario. In water-scarce areas, the weight of can be increased, and in severely saline-alkali areas, the weight of can be increased. In the saline-alkali land experimental area of Bayannao'er City, Inner Mongolia, after multiple verifications, , has achieved the best comprehensive effect.

[0045] The change range of the planting area limits the area ratio that can be adjusted by a single decision, usually controlled between 5% and 20% to avoid excessive decision fluctuations. In practical applications, a smaller change range (5% - 10%) is set in the initial stage, and as the system learning deepens, it is gradually increased to 15% - 20%, balancing exploration and stability.

[0046] The trained deep reinforcement learning model includes a control function F and a loss function L. The control function F generates decisions based on the current state, expressed as: , Among them, is the decision-making action at time t, including the adjustment amounts of configuration parameters such as plant spacing, row spacing, and tree species; is the state at time t, including the current environment and configuration parameters; is the parameter set of the control function, implemented through a deep neural network.

[0047] The loss function L is used to evaluate the gap between the predicted value and the true value, expressed as: , Among them, is the state-action value function, which evaluates the long-term value of taking action a in state s; is the immediate reward, which is determined by the salt control efficiency and water conservation effect generated by the current configuration; is the discount factor, which is used to balance short-term and long-term rewards. Considering the long growth cycle characteristics of the shelter forest, the typical value is set to 0.9. A higher discount factor makes the system pay more attention to long-term benefits; is the next state reached after executing action a; represents the next state the maximum value function value of all possible actions under ; is the target network parameter, which is used to stabilize the training process.

[0048] In the optimization project of the shelter forest in saline-alkali land in Bayannao'er City, Inner Mongolia, the model adopts a two-layer neural network structure. The number of neurons in the first hidden layer is 256, and the second hidden layer is 128. The ReLU activation function is used to avoid the problem of gradient disappearance. The Adam optimizer is selected, the initial learning rate is set to 0.001, and a learning rate decay strategy is adopted. The learning rate is multiplied by 0.9 after every 5000 steps of training. During the training process, the size of the experience replay buffer is set to 10000, the batch size is 64, and the target network is updated every 100 steps. These hyperparameter settings have been compared through multiple experiments and perform optimally in the scenario of shelter forest configuration in saline-alkali areas.

[0049] The update of the shelter forest configuration state set includes: evaluating the current shelter forest configuration through the deep reinforcement learning model, inputting the soil salinity and groundwater level into the coupling model, and outputting the soil salinity and groundwater level in the next state through the coupling model, and repeating the iterative update until the optimal configuration plan is obtained. In the application in the severely saline-alkali area of Bayannao'er City, Inner Mongolia, the iteration termination condition is set as the change of the reward function value is less than 1% for 5 consecutive times, or the maximum number of iterations reaches 1000 times. Practice has proved that in most cases, the system can converge to a stable plan within 300-500 iterations.

[0050] In the present invention, the parameter optimization of the salt transport-forest water absorption coupling model is realized by the gradient descent method. According to the characteristics of salinization in the northwest region, the salt content is equally divided into 15 salinity levels (from 0.5 g / kg to 15 g / kg, with an interval of 1 g / kg) by the linear interpolation method, and the stress resistance of the vegetation under different salt transport processes is obtained by solving according to the salt balance equation and the plant water absorption equation, and then the updated shelter forest configuration state set is obtained.

[0051] The mathematical expression of the gradient descent method is: , Among them, is the parameter value for the t-th iteration, including model parameters such as the salt diffusion coefficient and absorption coefficient; is the learning rate, which controls the step size of parameter update. It is initially set to 0.05 and gradually decreases to 0.01 as the number of iterations increases to ensure convergence stability; is the loss function For the parameter the gradient is calculated as where is a small perturbation. The loss function J is defined as the mean square error between the model prediction value and the measured value.

[0052] In the demonstration area of the saline-alkali land shelter forest in Bayannao'er City, Inner Mongolia, this method has successfully optimized the parameter models of salt-tolerant plants such as Tamarix chinensis and Haloxylon ammodendron, and the prediction accuracy has been improved by about 30%. For example, through optimization, it is found that the optimal absorption intensity parameter of Tamarix chinensis is 0.032 d⁻¹, which is significantly different from the general value of 0.025 d⁻¹ in the literature and more accurately reflects the actual growth characteristics under local environmental conditions.

[0053] To ensure the stability of the parameter optimization process and the effectiveness of the results, the gradient descent method sets the following constraint conditions: 1. Constraint on the state parameters of the unit equivalent resistance model: , where is the state parameter of the unit equivalent resistance model, reflecting the soil conductivity characteristics; and are its lower and upper limits respectively. According to the field tests on saline-alkali land in Bayannao'er City, Inner Mongolia, and are set. This constraint ensures that the model can adapt to various salinized soil conditions from mild to severe.

[0054] 2. Constraint on the infiltration parameters: , where is the infiltration parameter, characterizing the water conduction ability of the soil, with the unit of cm / d; and are its lower and upper limits respectively. According to the test data of various saline soils in Bayannao'er City, Inner Mongolia, for sandy soils, and are set, and for clayey soils, and are set. This constraint reflects the water migration characteristics of different soil textures and ensures that the model is applicable to various soil conditions.

[0055] 3. Upper and lower boundary constraints on soil salinity and its saturation: , , Among them, is the soil salinity, with the unit of g / kg; is the soil salinity saturation, dimensionless, representing the ratio of the current salinity content to the maximum salinity content; and are the lower limit and upper limit of the soil salinity respectively. According to the classification standard of saline soil in the northwest region, and are set. This constraint ensures that the model will not produce negative values or salinity prediction results beyond physical significance.

[0056] 4. Soil water holding capacity constraint: , Among them, is the soil water holding capacity, dimensionless; and are the lower limit and upper limit of it respectively. According to the measurement results of different types of saline soil in Inner Mongolia, (close to the wilting point) and (close to the saturated water content) are set. This constraint ensures that the water simulation is within a reasonable physical range.

[0057] The setting of these constraint conditions takes into account the physical limitations and biological characteristics in the actual farmland environment, ensuring the practical feasibility of the optimization results. In the saline-alkali land improvement project in Bayannao'er City, Inner Mongolia, if the parameters exceed the constraint range, they will be projected back into the constraint interval to ensure the convergence of the algorithm and the effectiveness of the solution. For example, if the calculated infiltration parameter during the optimization process exceeds the upper limit , it will be adjusted to 100 cm / d for continuous optimization. This processing method effectively avoids the deviation of model parameters from the actual physical meaning and improves the reliability and stability of the model.

[0058] As Figure 4 shown, the edge computing module 2 of the present invention compresses the prediction model to a scale suitable for edge devices through knowledge distillation technology and realizes real-time computing through a distributed edge network. This innovative design solves the practical problems in the deployment of the optimized system for the shelter forest in salinized farmland in remote areas.

[0059] The core idea of knowledge distillation is to guide the learning of the student model (lightweight model) through the teacher model (complex model). Its mathematical expression is: , Among them, is the knowledge distillation loss function, used to measure the learning degree of the student model about the knowledge of the teacher model; is the cross-entropy loss, defined as , where is the true label, is the predicted probability; is the KL divergence loss, defined as , which is used to measure the difference between two probability distributions; is the true label, which contains various parameters of the shelter forest configuration plan; and are the output logits of the student model and the teacher model respectively, that is, the output before the activation of the last layer of the neural network; is the softmax function, defined as , which converts the logits into a probability distribution; is the temperature parameter, which controls the softness of the soft label. The larger the value, the smoother the distribution. It is usually set to 2-5 and is set to 3 in this embodiment; is the balance factor, which controls the proportion of the true label loss and the distillation loss. The typical value is 0.5-0.7 and is set to 0.6 in this embodiment.

[0060] In the saline-alkali land treatment project in Bayannur City, Inner Mongolia, the teacher model is a complete deep reinforcement learning model, which consists of 6 layers of CNN and 4 layers of fully connected networks, and the number of parameters is about 40GB; the student model is a compressed lightweight model, which uses 2 layers of CNN and 2 layers of fully connected networks, and the number of parameters is only 4.2GB, a reduction of about 90%. The dimensionality reduction of the feature map channels is achieved through 1×1 convolution, reducing the original 256 channels to 32 channels, and retaining the key feature information. Tests show that the compressed model has increased the inference speed by 7.5 times, while the prediction accuracy has only decreased by 2.3%, achieving the best balance between performance and resource consumption.

[0061] To adapt to the characteristics of widely distributed and harsh environments of saline-alkali farmland, the edge computing network adopts a hierarchical design, including a sensing layer, a processing layer, and a decision-making layer. The sensing layer consists of distributed soil sensors and meteorological sensors, with 8-12 nodes arranged per hectare, and the acquisition frequency is 15-60 minutes / time. In the saline-alkali land demonstration area of Bayannur City, Inner Mongolia, the sensors use solar power supply and low-power LoRa communication technology, achieving maintenance-free operation for up to one year.

[0062] The processing layer consists of field edge computing gateways, which are responsible for data preprocessing and preliminary analysis. In the saline-alkali land treatment project in Horqin Sandy Land, Inner Mongolia, the edge gateway performs local model inference every 6 hours, generating temporary adjustment suggestions (such as irrigation control, drainage management), achieving a rapid response to sudden environmental changes.

[0063] The decision-making layer consists of regional edge servers that run lightweight models for optimization decisions. In the saline-alkali land improvement project in Bayannur City, Inner Mongolia, the decision-making layer servers perform a complete optimization of the shelter forest configuration once a month, generating medium- and long-term adjustment plans (such as adjusting plant spacing and planning new planting areas). The layers are connected through multiple wireless communication protocols (LoRa for long-distance and low-power transmission, with an effective distance of 5-10 km; NB-IoT for areas without signal towers, covered by cellular networks; WiFi for short-distance and high-bandwidth transmission), forming a complete edge computing network, effectively solving the problem of inconvenient communication in remote areas.

[0064] Practice has proved that the edge computing and knowledge distillation solution of the present invention performs excellently in the actual farmland environment with limited resources. In the saline-alkali land treatment project in Bayannur City, Inner Mongolia, the system realizes hourly monitoring data analysis and daily shelter forest management adjustment on edge devices, while the complete configuration optimization update remains at the monthly level, meeting the shelter forest management requirements at different time scales.

[0065] As Figure 5 shown, the method for optimizing the configuration of the shelter forest in saline-alkali farmland provided by the present invention includes the following steps: 1. Obtain data of the saline-alkali monitoring area through the UAV flight control system, use a high-resolution multispectral camera to obtain the ground reflectance and the reflectance of vegetation leaves, and establish the relationship between the vegetation leaf reflectance - water absorption index and conductivity. In the saline-alkali land improvement demonstration area in Bayannur City, Inner Mongolia, the UAV adopts a six-rotor structure, with a load of 4.2 kg, a flight endurance of 42 minutes, a flight altitude of 100 meters, and a single flight coverage area of about 85 hectares. The multispectral camera is configured with five bands: red (620-670 nm), green (540-580 nm), blue (450-510 nm), near-infrared (760-850 nm), and red edge (700-740 nm), with a spatial resolution of 10 cm and a spectral resolution of 10 nm. The formula for calculating the water absorption index (WAI) is: , where is the reflectance in the near-infrared band, with a wavelength range of 760-850 nm, reflecting the vegetation biomass and structural information; is the reflectance in the short-wave infrared band, with a wavelength range of 1550-1750 nm, which is sensitive to the water content. In the saline-alkali land experimental area in Bayannur City, Inner Mongolia, the relationship between WAI and conductivity is determined through regression analysis of 500 measured sample points, and the established regression equation is , and the correlation coefficient reaches 0.87, and the verification error is controlled within ±0.8 dS / m.

[0066] 2. Soil salinity monitors and ground water level monitors are deployed in the monitoring area to collect data on salt content and ground water level, obtain environmental meteorological information, and establish a dataset of soil saturated water content and soil saturated salt content. In the saline-alkali land demonstration area of Bayannur City, Inner Mongolia, the soil salinity monitor uses a four-electrode conductivity sensor with a measurement range of 0-20 dS / m and an accuracy of ±3%. The buried depth is divided into three layers: 0-20 cm, 20-50 cm, and 50-100 cm, and 2 sensors are placed in each layer to improve reliability. The sampling frequency is 30 minutes / time; the ground water level monitor uses a pressure type water level gauge with a measurement range of 0-10 m and an accuracy of ±0.1%. 3-5 monitoring points are arranged in each test area, and the sampling frequency is 60 minutes / time. Meteorological information is obtained through a small automatic weather station, including temperature (measurement range -40~60°C, accuracy ±0.3°C), humidity (measurement range 0%~100%, accuracy ±3%), precipitation (measurement range 0-10 mm / min, accuracy ±0.2 mm), wind direction (16 azimuths), and wind speed (measurement range 0-60 m / s, accuracy ±0.3 m / s), and the sampling frequency is 10 minutes / time. These data are preprocessed and stored in database 3 to form a paired dataset of soil saturated water content and soil saturated salt content, which contains approximately 25,000 spatio-temporally matched records.

[0067] 3. Use the above dataset to train the soil bulk density and salt transport-forest water absorption coupling model, and obtain the state update function and the shelter forest configuration set based on deep reinforcement learning. In the saline-alkali land treatment project of Bayannur City, Inner Mongolia, the random gradient descent algorithm is used in the training process. The initial batch size is 128 and gradually increases to 256 as the training progresses to improve stability; the number of training rounds is 300 rounds, and each round contains approximately 500 steps of update; the initial learning rate is 0.001, and a stepwise decay strategy is adopted, and it decays to 0.9 times the original every 50 rounds. The model evaluation uses the root mean square error (RMSE) and mean absolute error (MAE) indicators. The RMSE of salt prediction is controlled within 0.42 g / kg, the MAE is controlled within 0.28 g / kg, the RMSE of water level prediction is controlled within 0.15 m, and the MAE is controlled within 0.09 m, all of which are better than the prediction accuracy of the traditional model.

[0068] 4. In the shelter forest configuration state set, use the optimal adjustment parameters and control strategies to determine the optimal shelter forest configuration. In the shelter forest demonstration area of saline-alkali land in Bayannur City, Inner Mongolia, the ε-greedy strategy is used as the control strategy. The initial value of ε is 0.9 and gradually decreases to 0.1 according to the exponential decay law. The formula is , where \(t\) is the number of training steps, ensuring the balance between exploration and exploitation. The optimized optimal configuration plan is divided into three categories according to the degree of salinization: in the severely salinized area (\(> 10\ g / kg\)), the plant spacing is \(2.5\ m\) and the row spacing is \(5\ m\). The main tree species are extremely salt-tolerant plants such as Tamarix chinensis and Haloxylon ammodendron, and the planting density is 320 plants per hectare; in the moderately salinized area (\(5 - 10\ g / kg\)), the plant spacing is \(2\ m\) and the row spacing is \(4.5\ m\). The main tree species are mixed plantings of Elaeagnus angustifolia, Populus spp. and Tamarix chinensis, and the planting density is 440 plants per hectare; in the slightly salinized area (\(< 5\ g / kg\)), the plant spacing is \(1.5\ m\) and the row spacing is \(4\ m\). An economic forest and ecological forest mixed planting model can be adopted, and the planting density is 660 plants per hectare. The scheme evaluation shows that the average salt control efficiency is increased by 41% (35% in the severely salinized area, 42% in the moderately salinized area, and 46% in the slightly salinized area), and the average irrigation water saving rate reaches 33% (the range of change is 28% - 38%).

[0069] 5. The salt status distribution is obtained by taking pictures of the target farmland by drones. The shelter forest configuration is dynamically updated according to the current soil salinity and groundwater level, and a spatio-temporal model of forest growth is generated in a three-dimensional environment using a prediction model. In the saline-alkali land shelter forest demonstration area of Bayannur City, Inner Mongolia, the dynamic update frequency is flexibly adjusted according to seasons and meteorological conditions: during the growing season (April - September), it is updated once a month to ensure timely response to the changes in the growth period; during the non-growing season (October - March), it is updated once every two months to reduce resource consumption; after special meteorological conditions (heavy rainfall \(> 50\ mm / d\), flood, continuous drought \(> 15\) days), additional updates are carried out to cope with the impact of extreme weather. The three-dimensional spatio-temporal model is built based on the Unity engine, supporting growth prediction and visual display at four time scales (day, month, season, year), and can also simulate the growth comparison scenarios of different configuration schemes to help users intuitively understand the shelter forest configuration effect. In the saline-alkali land treatment project of Bayannur City, Inner Mongolia, this model successfully predicted the development trend of the Tamarix chinensis - Haloxylon ammodendron mixed forest within a 10-year growth cycle, and the deviation between the predicted tree height and the measured value was controlled within ±12%, providing a reliable basis for long-term shelter forest planning.

[0070] The optimized system for the shelter forest configuration in salinized farmland of the present invention has been experimentally verified and applied in multiple salinized areas of Bayannur City, Inner Mongolia, China. In the experimental area of moderately salinized farmland (salt content \(8 - 15\ g / kg\)) in the Hexi Corridor of Gansu, a comparative experiment was carried out for two years between the shelter forest demonstration area (100 hectares) optimized by this system and the control area (100 hectares) with the traditional configuration method. The results show that: 1. Salt control effect: The salt content in the surface soil (0 - 30 cm) of the demonstration area decreased by an average of 45% (from the initial 12.5 g / kg to 6.9 g / kg), significantly higher than the 28% in the control area (from the initial 12.3 g / kg to 8.9 g / kg); the vertical distribution of salt became more reasonable, and the main salt accumulation layer shifted downward from the original 15 - 30 cm to 50 - 70 cm, effectively improving the root growth environment of crops. The uniformity of soil conductivity distribution increased by 60%, reducing the formation of high-salt patches and creating a more uniform growth condition for crops.

[0071] 2. Water resource utilization: The irrigation water consumption in the demonstration area decreased by 33% (from the conventional 900 m³ / ha to 603 m³ / ha), while maintaining a good salt control effect; the water use efficiency (salt control effect per unit of water volume) increased by 55%, from 0.31% / (m³ / ha) to 0.48% / (m³ / ha); the fluctuation range of the groundwater level decreased by 40% (from ±0.85 m to ±0.51 m), effectively reducing the adverse impact of groundwater level rise and fall on salt migration.

[0072] 3. Ecological benefits: The growth condition of the shelter forest in the demonstration area was significantly better than that in the control area, and the survival rate increased by 15% (from 82% to 94%); the average height of two-year-old Tamarix chinensis increased by 28% (from 1.8 m to 2.3 m); the biodiversity index (Shannon-Wiener index) increased by an average of 35%, from 1.65 to 2.23; the soil organic matter content increased by 23%, from the initial 0.8% to 0.98%; the carbon sequestration capacity increased by about 30%, and the annual carbon sequestration per hectare of the shelter forest increased from 2.3 tons to 3.0 tons.

[0073] 4. Economic benefits: The payback period of the system investment is about 3.5 years. The main economic benefits come from: the land asset value increased by 38% (from the initial 15,000 yuan / ha to 20,700 yuan / ha); the crop yield increased by 42% (wheat increased from 3.2 tons / ha to 4.55 tons / ha, and corn increased from 7.5 tons / ha to 10.65 tons / ha); the irrigation cost decreased by 22% (from 450 yuan / ha to 351 yuan / ha). In addition, the system has a high degree of automation, reducing the labor management cost by about 40% and further improving the economic benefits.

[0074] In the saline-alkali land treatment demonstration area of Bayannao'er City, Inner Mongolia, this system has been promoted and applied to 5,000 hectares of saline-alkali farmland, realizing the positive interaction between the shelter forest and farmland production. Monitoring data shows that two years after the system was deployed, the regional groundwater level stabilized within the ideal range (1.8 - 2.2 m), the average salt content in the surface soil decreased by 38%, the irrigation water use efficiency increased by 42%, and the crop yield increased by 35% - 50%, demonstrating significant ecological and economic benefits.

[0075] Generally speaking, the optimized system and method for configuring shelter forests in saline-alkali farmland provided by the present invention realize the scientific, precise and intelligent configuration of shelter forests through the combination of deep reinforcement learning technology and the salt transport-tree water absorption coupling model, provide effective technical support for the treatment of saline-alkali farmland, and have important popularization and application value.

[0076] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. Salinized farmland shelterbelt configuration optimization system, characterized in that, Including: A salinization monitoring module, which is used to collect ground reflectance and vegetation leaf reflectance through an unmanned aerial vehicle, and use a hyperspectral imaging spectrometer to invert soil conductivity and determine the soil salt distribution state; An edge computing module, which is communicatively connected to the salinization monitoring module, and is used to receive the soil salt distribution state data and calculate a shelter forest configuration model; A database, which is connected to the edge computing module and is used to store calculation models, soil salts, and vegetation growth state data; A three-dimensional model module, which is connected to the database and the edge computing module, and is used to generate a three-dimensional spatio-temporal model of forest growth in a three-dimensional environment by using soil salts and groundwater level data; A model optimization module, which is connected to the three-dimensional model module and the database, and is used to optimize the soil salt and groundwater level simulation models based on deep reinforcement learning to improve the shelter forest configuration effect; A configuration scheme optimization module, which is connected to the model optimization module, and is used to generate an optimal shelter forest configuration scheme based on the model optimized by the deep reinforcement learning; 2. The optimized system for configuring a shelter forest in saline-alkali farmland according to claim 1, wherein, It further includes a meteorological monitoring module, which is connected to the model optimization module and is used to obtain meteorological condition data such as air temperature, precipitation, wind direction, and wind speed to determine an optimal growth strategy; wherein, the model optimization module further determines an optimal shelter forest configuration scheme by using a trained deep reinforcement learning model. Specifically, the row spacing and plant spacing of the trees to be planted and environmental parameters are used as states, and the shelter forest configuration efficiency and the reduction of irrigation water consumption are used as a reward function, and the optimal shelter forest configuration scheme is determined according to the trained deep reinforcement learning model.

3. The salinized farmland shelter forest configuration optimization system according to claim 2, characterized in that The model optimization module establishes a salt transport-tree water absorption coupling model through an improved Transformer architecture. The salt transport-tree water absorption coupling model includes: A salt balance equation, which is used to describe the dynamic change of salts in the soil; A plant absorption equation, which is used to describe the absorption process of soil water and salts by plants; A stand water holding rate equation, which is used to describe the water interception and holding capacity of the stand; A tree water absorption function, which is used to describe the water absorption capacity and efficiency of trees under different conditions; Among them, the model optimization module uses the forest growth parameters and the salt content distribution output by the salt transport model as training data, constructs a shelter forest configuration state set and a shelter forest structure optimization model, takes the current state as an input, obtains the next state through the shelter forest configuration state set, and calculates the corresponding shelter forest configuration efficiency and the reduction of irrigation water consumption in the next state as a reward function, and generates an optimal shelter forest configuration scheme through the deep reinforcement learning model.

4. The optimized system for configuring a shelter forest in saline-alkali farmland according to claim 3, wherein, The shelter forest configuration state set includes: a state set, a decision set, a reward value, and a change range of the planting area; wherein, the state set includes the planting area, plant spacing, row spacing, groundwater level, soil saturated water content, and soil salt distribution parameters.

5. The optimized system for configuring a shelter forest in saline-alkali farmland according to claim 4, characterized in that, The trained deep reinforcement learning model includes a control function and a loss function. Among them, the control function is used to generate decisions based on the current state, the loss function is used to evaluate the gap between the predicted value and the true value, the model uses a discount factor to balance short-term and long-term rewards, and the probability distribution of the shelter forest configuration state is used to guide the reinforcement learning process.

6. The optimized system for configuring a shelter forest in saline-alkali farmland according to claim 3, wherein, The update of the shelter forest configuration state set includes: evaluating the current shelter forest configuration through the deep reinforcement learning model, inputting soil salinity and groundwater level into the coupling model, and outputting the soil salinity and groundwater level under the next state through the coupling model, and repeating the iterative update until an optimal configuration plan is obtained.

7. The optimized system for configuring farmland shelterbelts in salinized fields according to claim 3, wherein, The parameter optimization in the salt transport - forest water absorption coupling model is achieved through the gradient descent method. Specifically: the salt content is equally spaced into multiple salinity levels by the linear interpolation method, and the stress resistance of the vegetation under different salt transport processes is obtained by solving according to the salt balance equation and the plant water absorption equation, and then the trained shelter forest configuration state set is obtained.

8. The optimized system for configuring farmland shelterbelts in salinized fields according to claim 7, wherein, The constraint conditions of the gradient descent method include: Unit equivalent resistance model state parameter constraint; Permeability parameter constraint; Upper and lower boundary constraints of soil salinity and its saturation; Soil water holding capacity constraint; The constraint conditions ensure the stability of the model training process and the effectiveness of the results.

9. The optimized system for configuring farmland shelterbelts in salinized fields according to claim 1, wherein The edge computing module compresses the prediction model to a scale suitable for edge devices through knowledge distillation technology, and realizes real-time computing through a distributed edge network; the knowledge distillation technology reduces the dimensionality of the feature map channels, retains the key decision-making ability of the model, and enables the system to update the shelter forest configuration plan hourly in a resource-constrained environment.

10. The optimization method of the salinized farmland shelter forest configuration optimization system according to any one of claims 1-9, characterized in that, It includes the following steps: Obtain data of the salinized monitoring area through the UAV flight control system, use a high-resolution multispectral camera to obtain the ground reflectance and vegetation leaf reflectance, and establish the relationship between the vegetation leaf reflectance - water absorption index and conductivity; Deploy soil salinity monitors and ground water level monitors in the monitoring area, collect salinity content and ground water level data, obtain environmental meteorological information, and establish a data set of soil saturated water content and soil saturated salinity; Use the data set to train the soil bulk density and the salt transport - forest water absorption coupling model, and obtain the state update function and the shelter forest configuration set based on deep reinforcement learning; In the shelter forest configuration state set, determine the optimal shelter forest configuration by using the optimal adjustment parameters and control strategies; Take pictures of the target farmland by the UAV to obtain the salt state distribution, dynamically update the shelter forest configuration according to the current soil salinity and groundwater level, and generate a spatio-temporal model of forest growth in a three-dimensional environment by using the prediction model.

Images