Intelligent precision irrigation system driven by low-altitude remote sensing data

Abstract

The invention discloses an intelligent precision irrigation system driven by low-altitude remote sensing data, including a low-altitude unmanned aerial vehicle, a ground sensor network module, a meteorological data integration module, a spatiotemporal self-evolution feature graph network module, a spatiotemporal adaptive attention network module, a contextual multi-armed tiger algorithm module and an automatic irrigation control module; the low-altitude unmanned aerial vehicle is used to obtain remote sensing data of crops; the ground sensor network module is used to collect real-time environmental information; the meteorological data integration module is used for multimodal data integration; the spatiotemporal self-evolution feature graph network module is used to generate a water demand distribution map; the spatiotemporal adaptive attention network is used to optimize irrigation strategies; the contextual multi-armed tiger algorithm module is used to dynamically select the optimal irrigation strategy; the automatic irrigation control module is used to execute the optimal irrigation strategy and implement differentiated intelligent irrigation management. The invention uses the spatiotemporal self-evolution feature graph network and the contextual multi-armed tiger algorithm to achieve precise intelligent irrigation.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural intelligent irrigation, and in particular to an intelligent precision irrigation system driven by low-altitude remote sensing data. Background Art

[0002] Existing agricultural irrigation systems have made great progress in increasing crop yields and saving water resources, especially irrigation systems based on automation and intelligent technology have been widely used. However, the current mainstream irrigation technology still has many problems and cannot fully cope with the complex environment and resource management needs in modern agricultural production.

[0003] Traditional irrigation systems usually rely on manual operation or simple rule-based automation equipment. For example, by installing humidity sensors or temperature sensors to monitor the environmental conditions of farmland, irrigation equipment is turned on or off according to set thresholds. Although such systems reduce human intervention to a certain extent, they lack accurate data analysis capabilities. Especially when faced with climate change, differences in soil structure, and different needs of crop growth stages, the response speed and accuracy of such systems are obviously insufficient. In addition, traditional irrigation systems usually rely on a single data source and lack the ability to comprehensively process multimodal data (such as soil moisture, temperature, weather forecasts, etc.), resulting in irrigation strategies that are often simple and one-sided, making it difficult to achieve refined management of the complex ecological environment of farmland.

[0004] There are some irrigation management systems based on sensor networks and drone data in the existing technology. For example, drone remote sensing technology can carry multispectral sensors and thermal infrared cameras to conduct regular flights over farmland to obtain data such as crop spectral reflectance and surface temperature, thereby monitoring crop health and soil moisture conditions. However, these technologies usually only stay at the data collection stage and lack the intelligent processing capabilities for in-depth analysis of data, making it difficult to adjust irrigation strategies according to dynamically changing environments and crop needs. In addition, these technologies are also unable to effectively combine real-time weather forecast data with historical irrigation feedback information, resulting in irrigation decisions lagging behind in terms of dynamic adjustments to weather changes and historical data.

[0005] Furthermore, existing automated irrigation systems often lack an in-depth understanding of the crop growth cycle, making it difficult to make real-time adjustments based on the needs of crops at different growth stages. The amount of water required by crops during growth is not constant. In certain growth stages, such as flowering or fruit development, crops require much more water than in other stages. Most current irrigation systems make irrigation decisions based only on current environmental data, without fully considering the changes in crop water requirements at different growth stages. This not only leads to a waste of water resources, but may also have an adverse effect on crop growth.

[0006] In addition, although some intelligent irrigation systems in the existing technology have introduced machine learning or artificial intelligence technology, most of them use simple algorithms, such as decision-making methods based on fixed rules or thresholds. These algorithms are difficult to cope with the complex relationship between multiple factors such as the environment, crop demand and water resources. In particular, when faced with uncertain factors such as weather changes and fluctuations in crop health, existing algorithms cannot make accurate predictions and dynamic adjustments. In addition, the current system has a low efficiency in the use of historical data, resulting in slow updates of irrigation strategies and an inability to quickly adapt to changes in the environment or crop demand.

[0007] Therefore, how to provide an intelligent precision irrigation system driven by low-altitude remote sensing data is an urgent problem that technicians in this field need to solve. Summary of the invention

[0008] One purpose of the present invention is to propose an intelligent precision irrigation system driven by low-altitude remote sensing data. The present invention combines multimodal data fusion, spatiotemporal self-evolving feature graph network and contextual multi-armed bandit algorithm. Through the multi-dimensional information integrated by low-altitude remote sensing, ground sensors and meteorological data, it dynamically generates accurate water demand distribution maps, and adaptively adjusts irrigation strategies to optimize water resource utilization. The system can not only monitor crop growth and environmental changes in real time, but also predict future water demand through short-term and long-term spatiotemporal dependencies to ensure that crops obtain the optimal water supply. It has the significant advantages of high water-saving efficiency, high irrigation accuracy and intelligent crop management.

[0009] According to an embodiment of the present invention, an intelligent precision irrigation system driven by low-altitude remote sensing data includes a low-altitude drone, a ground sensor network module, a meteorological data integration module, a spatiotemporal self-evolution feature map network module, a spatiotemporal adaptive attention network module, a contextual multi-armed tiger algorithm module, and an automatic irrigation control module;

[0010] The low-altitude drone is equipped with a multispectral sensor and a thermal infrared camera for regular flights over farmland areas to obtain remote sensing data of crops;

[0011] The ground sensor network module is used to collect real-time environmental information;

[0012] The meteorological data integration module is used to receive meteorological forecast information from an external data source and perform multimodal data integration;

[0013] The spatiotemporal self-evolving characteristic graph network module is used to process the acquired multimodal data, dynamically construct and evolve the dependency graph between regions, and generate a water demand distribution map;

[0014] The spatiotemporal adaptive attention network is used to process the water demand distribution map and meteorological forecast data, dynamically capture short-term and long-term spatiotemporal dependencies, and optimize irrigation strategies;

[0015] The contextual multi-armed bandit algorithm module dynamically selects the optimal irrigation strategy and adjusts the irrigation parameters in real time according to the current crop status, environmental characteristics and historical irrigation feedback data;

[0016] The automatic irrigation control module is used to execute the optimal irrigation strategy, control the switch, flow rate and irrigation duration of the irrigation equipment, and implement differentiated intelligent irrigation management according to different areas.

[0017] According to an embodiment of the present invention, an intelligent precision irrigation method based on low-altitude remote sensing data driving includes the following steps:

[0018] S1. Use low-altitude drones equipped with multispectral sensors and thermal infrared cameras to regularly fly over farmland areas to obtain remote sensing data of crops;

[0019] S2, using ground sensor networks to collect real-time environmental information, while obtaining weather forecast data from external data sources for multimodal data integration;

[0020] S3, process the acquired multimodal data through the spatiotemporal self-evolving feature map network, dynamically construct and evolve the dependency relationship map between regions, adjust the dependency weights between regions according to the water demand and environmental changes of different regions, and generate a water demand distribution map;

[0021] S4. Combined with spatiotemporal dynamic feature extraction, the self-evolving feature graph network is used to adaptively analyze the water demand of different regions, generate regional differentiated irrigation strategies, and make real-time adjustments according to different crop growth stages;

[0022] S5. Input the water demand distribution map and meteorological forecast data into the spatiotemporal adaptive attention network to dynamically capture short-term and long-term spatiotemporal dependencies, predict the changes in water demand at different time scales in the future, and optimize irrigation strategies.

[0023] S6, through the contextual multi-armed bandit algorithm, dynamically select the optimal irrigation strategy based on the current crop status, environmental characteristics and historical irrigation feedback data, and adjust the irrigation parameters in real time;

[0024] S7. Through the automated irrigation equipment control module, the optimal irrigation strategy is implemented to control the switch, flow rate and irrigation duration of the irrigation equipment, and differentiated intelligent irrigation management is implemented according to different regions.

[0025] Optionally, the real-time environmental information includes soil moisture, temperature, air humidity and wind speed, and the external data source includes satellite remote sensing and weather stations.

[0026] Optionally, the S3 specifically includes:

[0027] S31. Preprocess the multimodal data obtained from low-altitude drones and ground sensor networks, normalize the data according to time series and spatial dimensions, and generate a spatiotemporal data matrix , the spatiotemporal data matrix It includes crop spectral reflectance, soil moisture, temperature, air humidity and wind speed parameters in each area, and divides the time window according to the crop growth cycle;

[0028] S32. Based on the spatiotemporal self-evolving feature graph network, an initial dependency graph is established for each area of ​​the farmland. , where nodes represent farmland areas, edges represent dependencies between areas, and edge weights are calculated:

[0029] ;

[0030] in, Indicates at time At time t, the initial dependency weight between farmland area i and farmland area j is, represents the spatial distance between farmland area i and farmland area j, represents the environmental parameters of farmland area i, represents the environmental parameters of farmland region j, represents the environmental regulating factor, represents the temperature adjustment factor, represents the difference in vegetation index between farmland area i and farmland area j, and represents the control factor, represents the temperature of farmland area i, represents the temperature of farmland area j;

[0031] The above formula introduces the interaction of multiple environmental parameters and combines nonlinear processing. The addition of makes the weight not only affected by spatial distance, but also combines temperature changes and vegetation growth differences, comprehensively considers the actual demand changes between regions, and improves the accuracy and dynamic response ability of the model;

[0032] S33, through the spatiotemporal convolution operation, the spatiotemporal data matrix Perform feature extraction to generate dynamic water demand feature vectors for each farmland area , the dynamic water demand characteristic vector Including the changing trends of water demand at current and historical moments;

[0033] S34. Dynamically update the dependency graph based on real-time feedback data from the sensor network , and calculate the updated edge weights:

[0034] ;

[0035] in, represents the dynamically updated weight dependency weight between farmland area i and farmland area j at time t, Indicates at time At time t, the dependency weight between farmland area i and farmland area j after the previous time step update, represents the update rate, represents the water demand of farmland area i at time t, represents the water demand of farmland area j at time t, represents the time decay factor, and represents the control adjustment factor, represents the crop growth status index of farmland area i, represents the crop growth status index of farmland area j, represents the spatiotemporal feature vector of farmland area i at time t, represents the spatiotemporal feature vector of farmland area j at time t;

[0036] The above formula can better reflect the dynamic adjustment of dependency by considering the changes in spatiotemporal characteristics between regions and the growth status of crops. and crop growth index The addition of improves the sensitivity of the model, making the update of dependencies more refined and adaptable to complex environmental changes;

[0037] S35. Continuously evolving dependency graph , adaptively adjust the dependency weights between regions in the spatiotemporal dimension, and optimize the dependency relationship by combining historical environmental data and real-time sensor feedback;

[0038] S36, inputting the evolved dependency graph into a water demand distribution generation module to generate a water demand distribution graph.

[0039] Optionally, the S4 specifically includes:

[0040] S41. Extract spatiotemporal dynamic features from the multimodal data of each region to generate a spatiotemporal feature matrix , and divide the features into time windows to establish multi-level feature data sets for different time periods;

[0041] S42. The spatiotemporal dynamic characteristics of each region are processed through the self-evolving feature graph network, and adaptive analysis is performed based on crop growth stages, environmental changes and historical data to calculate the water demand change rate of each region:

[0042] ;

[0043] in, represents the rate of change of water demand in farmland area i, represents the water demand of farmland area i, represents the environmental parameters of farmland area i, represents the environmental parameters of farmland region j, represents the temperature of farmland area i, represents the temperature of farmland area j, Indicates the time difference, represents the smoothing term, represents the soil moisture of farmland area i, represents the soil moisture in farmland area j, represents the difference in vegetation index between farmland area i and farmland area j, represents the difference in water demand changes between farmland area i and farmland area j;

[0044] The above formula complicates the interaction of each parameter through nonlinear functions, including time, space, environmental differences, water demand and temperature changes, and takes into account the influence of various environmental conditions and vegetation index changes, making the calculation more accurate;

[0045] S43, based on the change rate of regional water demand , generate the regional priority matrix , calculate each priority value:

[0046] ;

[0047] in, represents the relative priority between farmland area i and farmland area j at time t, represents the hyperbolic cosine function, represents the exponential function, represents the water demand of farmland area j, represents the crop growth status index of farmland area i, represents the crop growth status index of farmland area j, represents the rate of change of water demand in farmland region j, represents the maximum difference in the rate of change of water demand between regions, represents the weight factor for adjusting the influence of temperature and humidity, represents the weight coefficient for regulating water demand between regions, represents the weight coefficient for regulating the impact of environmental differences among regions;

[0048] The above formula combines multiple factors such as water demand, crop growth status, vegetation index, environmental characteristics and temperature changes through complex multi-level nonlinear combinations, enhancing the adaptive analysis capabilities for different environments and crop conditions. The priority between regions not only depends on water demand, but also takes into account crop growth differences, temperature and humidity changes, making irrigation decisions more accurate and intelligent;

[0049] S44, based on regional priority matrix , adaptively adjust the irrigation strategies of different areas, so that high-priority areas get priority in water allocation; the irrigation strategies of low-priority areas are dynamically adjusted according to the overall water resource conditions and crop growth stage;

[0050] S45. Generate regional differentiated irrigation strategies by combining regional water demand characteristic matrix, historical meteorological data and future meteorological forecasts;

[0051] S46. Input regional differentiated irrigation strategies into the automated irrigation control system, continuously adjust irrigation plans based on feedback data, and respond to environmental changes and crop needs in real time.

[0052] Optionally, the S5 specifically includes:

[0053] S51, taking the water demand distribution map of the farmland area and the future weather forecast data as input and passing them into the spatiotemporal adaptive attention network;

[0054] S52, process 1-3 days of meteorological data through the short-term convolution module of the spatiotemporal adaptive attention network, dynamically capture the short-term spatiotemporal dependencies between regions, and calculate the change rate of water demand in each region in the short term:

[0055] ;

[0056] in, represents the rate of change of water demand of farmland area i in the short term at time t, represents the water demand of farmland area i at time t, represents the weather forecast data of farmland area i at time t, represents the temperature of farmland area i, represents the temperature of farmland area j, represents the soil moisture of farmland area i, represents the soil moisture in farmland area j, represents the environmental parameters of farmland area i, represents the environmental parameters of farmland region j, represents the difference in vegetation index between farmland area i and farmland area j, represents the short-term adjustment factor, Represents the temperature and humidity difference adjustment coefficient, represents the spatiotemporal dependency weight;

[0057] S53, through the long-term convolution module of the spatiotemporal adaptive attention network, the meteorological data of the next 4-30 days are processed to capture the long-term spatiotemporal dependencies and generate the long-term water demand change rate of each region:

[0058] ;

[0059] in, represents the rate of change of long-term water demand of farmland area i at time t, n represents the total number of days, represents the water demand of farmland area i in the past k days, represents the meteorological data of farmland area i over the past k days, represents the temperature of farmland area i in the past k days, represents the rate of change of water demand in farmland area i, represents the rate of change of water demand in farmland region j, represents the long-term adjustment factor, represents the long-term spatiotemporal feature adjustment factor, represents the long-term spatiotemporal dependence coefficient, represents the smoothing term;

[0060] S54. Weighted fusion of captured short-term and long-term spatiotemporal dependencies to generate a comprehensive water demand change forecast , used to optimize irrigation strategies:

[0061] ;

[0062] S55. Comprehensive water demand forecast results Input into the irrigation strategy optimization module to obtain the optimized irrigation strategy.

[0063] Optionally, the S6 specifically includes:

[0064] S61, constructing context feature vectors for each farmland area based on the current crop status, environmental characteristics, and historical irrigation feedback data through a contextual multi-armed bandit algorithm;

[0065] S62, input the context feature vector into the context multi-armed bandit model, and calculate the instant reward function of each area according to the environmental state and feedback data of each area:

[0066] ;

[0067] in, represents the instantaneous return function of farmland region i, represents the water demand of farmland area i at time t, represents the moisture change of farmland area i at time t, represents the temperature of farmland area i, represents the temperature of farmland area j, represents the soil moisture of farmland area i, represents the soil moisture in farmland area j, represents the crop growth status index of farmland area i, represents the crop growth status index of farmland area j, and represents the adjustment coefficient;

[0068] S63, the system based on context features and immediate reward function , dynamically select the optimal irrigation strategy for each area:

[0069] ;

[0070] in, represents the optimal irrigation strategy for farmland area i at time t, represents the balance factor between exploration and exploitation, The adjustment factor representing the effect of temperature change, represents the number of times irrigation strategy a is selected, represents the meteorological data of farmland area i at the current time t;

[0071] S64. Optimal irrigation strategy based on dynamic selection , adjusting irrigation parameters in real time, the irrigation parameters including water volume, irrigation frequency and irrigation duration;

[0072] S65, inputting the optimized optimal strategy into the automatic irrigation control system, executing corresponding irrigation operations, and adopting differentiated irrigation plans for different regions according to water demand, crop growth stage and historical feedback;

[0073] S66. After implementing the irrigation strategy, the crop status, soil moisture and meteorological changes are monitored in real time through the sensor network, and the new feedback data is input into the contextual multi-armed bandit model to further adjust the reward function and optimize irrigation strategy selection.

[0074] The beneficial effects of the present invention are:

[0075] First of all, the present invention realizes the comprehensive integration and processing of multimodal data. It uses low-altitude drones, multispectral sensors, thermal infrared cameras and ground sensor networks, combined with meteorological data integration modules, to obtain multi-dimensional environmental and crop growth information in real time. Unlike traditional systems that rely only on a single data source, this system dynamically generates a water demand distribution map by integrating and analyzing data such as soil moisture, temperature, air humidity, wind speed, and crop spectral reflectivity, thereby formulating a more refined irrigation plan for each farmland area. This fusion of multimodal data greatly improves the accuracy of the data, can more realistically reflect the complex conditions of farmland, and make irrigation decisions more scientific and reasonable.

[0076] Secondly, the spatiotemporal self-evolving feature graph network in the system can accurately capture the changes in spatiotemporal characteristics between regions by dynamically constructing and evolving dependency graphs between regions. This dynamic dependency is not only based on spatial distance and environmental parameters, but also combines factors such as crop growth status and meteorological changes to comprehensively analyze the differences in water demand between regions. Through this modeling method based on spatiotemporal characteristics, the system can adjust irrigation strategies in real time according to crop growth cycles, environmental changes, and historical data, effectively avoiding the problem of untimely response to changes in crop growth stage demand in traditional irrigation systems, and ensuring that water supply is closely matched to actual crop demand.

[0077] At the same time, the spatiotemporal adaptive attention network introduced in the system further optimizes the process of formulating irrigation strategies. By capturing short-term and long-term spatiotemporal dependencies, the system can predict changes in water demand at different time scales in the future. This feature enables the system to not only make real-time decisions based on the current environmental status, but also optimize irrigation strategies in advance in combination with future meteorological forecast data. Compared with the passive response of traditional irrigation systems in response to climate change, this system can adjust irrigation plans in advance when weather changes (such as rainfall, temperature fluctuations, etc.) in the next few days are anticipated to avoid wasting water resources, greatly improving the intelligence level and water-saving effect of the irrigation system.

[0078] In addition, the present invention realizes dynamic adaptive optimization of irrigation strategies through the contextual multi-armed bandit algorithm. The algorithm comprehensively considers the current crop growth status, environmental characteristics and historical irrigation feedback data, and can dynamically select the optimal irrigation strategy according to the actual needs of different regions, and adjust the irrigation parameters in real time. The introduction of this algorithm enables irrigation decisions to not only have the ability to respond in real time, but also to continuously self-optimize through feedback iteration during system operation. The system can learn and improve strategies based on the actual effect of each irrigation, so that the irrigation process gradually tends to the optimal state. This dynamic adjustment and feedback learning mechanism significantly improves the accuracy of irrigation and the efficiency of water resource utilization, ensuring that the water supply is adapted to the crop growth status and environmental changes.

[0079] Finally, the introduction of the automatic irrigation control module realizes the full automation of the irrigation process, completely reducing the need for human intervention. The system automatically controls the switch, flow rate and irrigation duration of the irrigation equipment through the optimal irrigation strategy, and implements differentiated intelligent irrigation management based on the water demand of different regions. Unlike traditional fixed-rule irrigation, this system can continuously adjust irrigation parameters according to the real-time environment and crop status to ensure efficient use of water resources and healthy growth of crops. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0081] Figure 1 This is a schematic diagram of the structure of the intelligent precision irrigation system driven by low-altitude remote sensing data proposed in the present invention;

[0082] Figure 2 This is an overall flow chart of the intelligent precision irrigation method driven by low-altitude remote sensing data proposed in the present invention. DETAILED DESCRIPTION

[0083] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0084] refer to Figure 1 , an intelligent precision irrigation system driven by low-altitude remote sensing data, including low-altitude drones, ground sensor network modules, meteorological data integration modules, spatiotemporal self-evolving feature map network modules, spatiotemporal adaptive attention network modules, contextual multi-armed bandit algorithm modules and automated irrigation control modules;

[0085] The low-altitude drone is equipped with a multispectral sensor and a thermal infrared camera for regular flights over farmland areas to obtain remote sensing data of crops;

[0086] The ground sensor network module is used to collect real-time environmental information;

[0087] The meteorological data integration module is used to receive meteorological forecast information from an external data source and perform multimodal data integration;

[0088] The spatiotemporal self-evolving characteristic graph network module is used to process the acquired multimodal data, dynamically construct and evolve the dependency graph between regions, and generate a water demand distribution map;

[0089] The spatiotemporal adaptive attention network is used to process the water demand distribution map and meteorological forecast data, dynamically capture short-term and long-term spatiotemporal dependencies, and optimize irrigation strategies;

[0090] The contextual multi-armed bandit algorithm module dynamically selects the optimal irrigation strategy and adjusts the irrigation parameters in real time according to the current crop status, environmental characteristics and historical irrigation feedback data;

[0091] The automatic irrigation control module is used to execute the optimal irrigation strategy, control the switch, flow rate and irrigation duration of the irrigation equipment, and implement differentiated intelligent irrigation management according to different areas.

[0092] refer to Figure 2 , an intelligent precision irrigation method driven by low-altitude remote sensing data, including the following steps:

[0093] S1. Use low-altitude drones equipped with multispectral sensors and thermal infrared cameras to regularly fly over farmland areas to obtain remote sensing data of crops;

[0094] S2, using ground sensor networks to collect real-time environmental information, while obtaining weather forecast data from external data sources for multimodal data integration;

[0095] S3, process the acquired multimodal data through the spatiotemporal self-evolving feature map network, dynamically construct and evolve the dependency relationship map between regions, adjust the dependency weights between regions according to the water demand and environmental changes of different regions, and generate a water demand distribution map;

[0096] S4. Combined with spatiotemporal dynamic feature extraction, the self-evolving feature graph network is used to adaptively analyze the water demand of different regions, generate regional differentiated irrigation strategies, and make real-time adjustments according to different crop growth stages;

[0097] S5. Input the water demand distribution map and meteorological forecast data into the spatiotemporal adaptive attention network to dynamically capture short-term and long-term spatiotemporal dependencies, predict the changes in water demand at different time scales in the future, and optimize irrigation strategies.

[0098] S6, through the contextual multi-armed bandit algorithm, dynamically select the optimal irrigation strategy based on the current crop status, environmental characteristics and historical irrigation feedback data, and adjust the irrigation parameters in real time;

[0099] S7. Through the automated irrigation equipment control module, the optimal irrigation strategy is implemented to control the switch, flow rate and irrigation duration of the irrigation equipment, and differentiated intelligent irrigation management is implemented according to different regions.

[0100] In this implementation, the real-time environmental information includes soil moisture, temperature, air humidity and wind speed, and the external data source includes satellite remote sensing and weather stations.

[0101] In this implementation, S3 specifically includes:

[0102] S31. Preprocess the multimodal data obtained from low-altitude drones and ground sensor networks, normalize the data according to time series and spatial dimensions, and generate a spatiotemporal data matrix , the spatiotemporal data matrix It includes crop spectral reflectance, soil moisture, temperature, air humidity and wind speed parameters in each area, and divides the time window according to the crop growth cycle;

[0103] S32. Based on the spatiotemporal self-evolving feature graph network, an initial dependency graph is established for each area of ​​the farmland. , where nodes represent farmland areas, edges represent dependencies between areas, and edge weights are calculated:

[0104] ;

[0105] in, Indicates at time At time t, the initial dependency weight between farmland area i and farmland area j is, represents the spatial distance between farmland area i and farmland area j, represents the environmental parameters of farmland area i, represents the environmental parameters of farmland region j, represents the environmental regulating factor, represents the temperature adjustment factor, represents the difference in vegetation index between farmland area i and farmland area j, and represents the control factor, represents the temperature of farmland area i, represents the temperature of farmland area j;

[0106] The above formula introduces the interaction of multiple environmental parameters and combines nonlinear processing. The addition of makes the weight not only affected by spatial distance, but also combines temperature changes and vegetation growth differences, comprehensively considers the actual demand changes between regions, and improves the accuracy and dynamic response ability of the model;

[0107] S33, through the spatiotemporal convolution operation, the spatiotemporal data matrix Perform feature extraction to generate dynamic water demand feature vectors for each farmland area , the dynamic water demand characteristic vector Including the changing trends of water demand at current and historical moments;

[0108] S34. Dynamically update the dependency graph based on real-time feedback data from the sensor network , and calculate the updated edge weights:

[0109] ;

[0110] in, represents the dynamically updated weight dependency weight between farmland area i and farmland area j at time t, Indicates at time At time t, the dependency weight between farmland area i and farmland area j after the previous time step update, represents the update rate, represents the water demand of farmland area i at time t, represents the water demand of farmland area j at time t, represents the time decay factor, and represents the control adjustment factor, represents the crop growth status index of farmland area i, represents the crop growth status index of farmland area j, represents the spatiotemporal feature vector of farmland area i at time t, represents the spatiotemporal feature vector of farmland area j at time t;

[0111] The above formula can better reflect the dynamic adjustment of dependency by considering the changes in spatiotemporal characteristics between regions and the growth status of crops. and crop growth index The addition of improves the sensitivity of the model, making the update of dependencies more refined and adaptable to complex environmental changes;

[0112] S35. Continuously evolving dependency graph , adaptively adjust the dependency weights between regions in the spatiotemporal dimension, and optimize the dependency relationship by combining historical environmental data and real-time sensor feedback;

[0113] S36, inputting the evolved dependency graph into a water demand distribution generation module to generate a water demand distribution graph.

[0114] In this implementation manner, the S4 specifically includes:

[0115] S41. Extract spatiotemporal dynamic features from the multimodal data of each region to generate a spatiotemporal feature matrix , and divide the features into time windows to establish multi-level feature data sets for different time periods;

[0116] S42. The spatiotemporal dynamic characteristics of each region are processed through the self-evolving feature graph network, and adaptive analysis is performed based on crop growth stages, environmental changes and historical data to calculate the water demand change rate of each region:

[0117] ;

[0118] in, represents the rate of change of water demand in farmland area i, represents the water demand of farmland area i, represents the environmental parameters of farmland area i, represents the environmental parameters of farmland region j, represents the temperature of farmland area i, represents the temperature of farmland area j, Indicates the time difference, represents the smoothing term, represents the soil moisture of farmland area i, represents the soil moisture in farmland area j, represents the difference in vegetation index between farmland area i and farmland area j, represents the difference in water demand changes between farmland area i and farmland area j;

[0119] The above formula complicates the interaction of each parameter through nonlinear functions, including time, space, environmental differences, water demand and temperature changes, and takes into account the influence of various environmental conditions and vegetation index changes, making the calculation more accurate;

[0120] S43, based on the change rate of regional water demand , generate the regional priority matrix , calculate each priority value:

[0121] ;

[0122] in, represents the relative priority between farmland area i and farmland area j at time t, represents the hyperbolic cosine function, represents the exponential function, represents the water demand of farmland area j, represents the crop growth status index of farmland area i, represents the crop growth status index of farmland area j, represents the rate of change of water demand in farmland region j, represents the maximum difference in the rate of change of water demand between regions, represents the weight factor for adjusting the influence of temperature and humidity, represents the weight coefficient for regulating water demand between regions, represents the weight coefficient for regulating the impact of environmental differences among regions;

[0123] The above formula combines multiple factors such as water demand, crop growth status, vegetation index, environmental characteristics and temperature changes through complex multi-level nonlinear combinations, enhancing the adaptive analysis capabilities for different environments and crop conditions. The priority between regions not only depends on water demand, but also takes into account crop growth differences, temperature and humidity changes, making irrigation decisions more accurate and intelligent;

[0124] S44, based on regional priority matrix , adaptively adjust the irrigation strategies of different areas, so that high-priority areas get priority in water allocation; the irrigation strategies of low-priority areas are dynamically adjusted according to the overall water resource conditions and crop growth stage;

[0125] S45. Generate regional differentiated irrigation strategies by combining regional water demand characteristic matrix, historical meteorological data and future meteorological forecasts;

[0126] S46. Input regional differentiated irrigation strategies into the automated irrigation control system, continuously adjust irrigation plans based on feedback data, and respond to environmental changes and crop needs in real time.

[0127] In this implementation manner, S5 specifically includes:

[0128] S51, taking the water demand distribution map of the farmland area and the future weather forecast data as input and passing them into the spatiotemporal adaptive attention network;

[0129] S52, process 1-3 days of meteorological data through the short-term convolution module of the spatiotemporal adaptive attention network, dynamically capture the short-term spatiotemporal dependencies between regions, and calculate the change rate of water demand in each region in the short term:

[0130] ;

[0131] in, represents the rate of change of water demand of farmland area i in the short term at time t, represents the water demand of farmland area i at time t, represents the weather forecast data of farmland area i at time t, represents the temperature of farmland area i, represents the temperature of farmland area j, represents the soil moisture of farmland area i, represents the soil moisture in farmland area j, represents the environmental parameters of farmland area i, represents the environmental parameters of farmland region j, represents the difference in vegetation index between farmland area i and farmland area j, represents the short-term adjustment factor, Represents the temperature and humidity difference adjustment coefficient, represents the spatiotemporal dependency weight;

[0132] S53, through the long-term convolution module of the spatiotemporal adaptive attention network, the meteorological data of the next 4-30 days are processed to capture the long-term spatiotemporal dependencies and generate the long-term water demand change rate of each region:

[0133] ;

[0134] in, represents the rate of change of long-term water demand of farmland area i at time t, n represents the total number of days, represents the water demand of farmland area i in the past k days, represents the meteorological data of farmland area i over the past k days, represents the temperature of farmland area i in the past k days, represents the rate of change of water demand in farmland area i, represents the rate of change of water demand in farmland region j, represents the long-term adjustment factor, represents the long-term spatiotemporal feature adjustment factor, represents the long-term spatiotemporal dependence coefficient, represents the smoothing term;

[0135] S54. Weighted fusion of captured short-term and long-term spatiotemporal dependencies to generate a comprehensive water demand change forecast , used to optimize irrigation strategies:

[0136] ;

[0137] S55. Comprehensive water demand forecast results Input into the irrigation strategy optimization module to obtain the optimized irrigation strategy.

[0138] In this implementation manner, S6 specifically includes:

[0139] S61, constructing context feature vectors for each farmland area based on the current crop status, environmental characteristics, and historical irrigation feedback data through a contextual multi-armed bandit algorithm;

[0140] S62, input the context feature vector into the context multi-armed bandit model, and calculate the instant reward function of each area according to the environmental state and feedback data of each area:

[0141] ;

[0142] in, represents the instantaneous return function of farmland region i, represents the water demand of farmland area i at time t, represents the moisture change of farmland area i at time t, represents the temperature of farmland area i, represents the temperature of farmland area j, represents the soil moisture of farmland area i, represents the soil moisture in farmland area j, represents the crop growth status index of farmland area i, represents the crop growth status index of farmland area j, and represents the adjustment coefficient;

[0143] S63, the system based on context features and immediate reward function , dynamically select the optimal irrigation strategy for each area:

[0144] ;

[0145] in, represents the optimal irrigation strategy for farmland area i at time t, represents the balance factor between exploration and exploitation, The adjustment factor representing the effect of temperature change, represents the number of times irrigation strategy a is selected, represents the meteorological data of farmland area i at the current time t;

[0146] S64. Optimal irrigation strategy based on dynamic selection , adjusting irrigation parameters in real time, the irrigation parameters including water volume, irrigation frequency and irrigation duration;

[0147] S65, inputting the optimized optimal strategy into the automatic irrigation control system, executing corresponding irrigation operations, and adopting differentiated irrigation plans for different regions according to water demand, crop growth stage and historical feedback;

[0148] S66. After implementing the irrigation strategy, the crop status, soil moisture and meteorological changes are monitored in real time through the sensor network, and the new feedback data is input into the contextual multi-armed bandit model to further adjust the reward function and optimize irrigation strategy selection.

[0149] Embodiment 1:

[0150] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the agricultural demonstration base in City A. In the past, the agricultural demonstration base relied on manual operation and traditional irrigation methods for farmland management. The irrigation process mainly relied on experience and rough soil moisture monitoring, resulting in waste of water resources and low crop production efficiency. Especially in the hot and dry season, the water demand of crops increases rapidly, but the traditional irrigation method is difficult to respond to the dynamic water demand of crops in time, and often over-irrigation or insufficient water supply occurs. This method not only affects the growth of crops, but also leads to serious waste of water resources.

[0151] To solve the above problems, the agricultural demonstration base in City A introduced the intelligent precision irrigation system based on low-altitude remote sensing data. The system collects multimodal data of farmland through low-altitude drones and ground sensor networks, and dynamically adjusts irrigation strategies by combining spatiotemporal self-evolving feature graph networks and contextual multi-armed bandit algorithms to achieve precision irrigation and efficient water resource management.

[0152] The application of this system is first reflected in the multi-dimensional fusion of data. Low-altitude drones are equipped with multispectral sensors and thermal infrared cameras. They fly low over farmland every two days in the early morning and dusk to collect spectral reflectance and temperature distribution maps of crops. At the same time, the ground sensor network collects environmental information such as soil moisture, air temperature, air humidity and wind speed in the farmland in real time. Through the meteorological data integration module, the system can also obtain weather forecast data for the next week from satellite remote sensing and meteorological stations, including rainfall, wind speed and temperature changes.

[0153] When processing these multimodal data, the spatiotemporal self-evolving feature map network converts them into a spatiotemporal data matrix, and dynamically evaluates the water demand of each farmland area by constructing a dependency graph between regions. The system combines the historical data of each area with the growth stage of the crop to adjust the irrigation strategy in real time. For example, the corn field in the southern part of the farmland has a higher temperature, lower soil moisture, and is in the flowering period, so it requires a large amount of water. The system quickly selects a higher irrigation priority for this area through the multi-armed tiger machine algorithm, and continuously adjusts the irrigation amount through automated irrigation equipment over the next two days to ensure that the crop has sufficient water supply. In contrast, the wheat field planted in the north of the farmland has a relatively low water demand because it has just experienced rainfall and is in the early stages of growth. The system allocates less irrigation to this area to avoid wasting water resources.

[0154] In specific applications, the system predicts changes in water demand in various areas of farmland in the next three days by analyzing short-term and long-term meteorological changes. For example, on August 15, it was predicted that there would be two showers in City A in the next three days, and the temperature would gradually drop from 35°C to 28°C. Based on this prediction, the system reduced the amount of irrigation two days later, and reduced the water supply by about 30% for the first day after the shower. Through this dynamic adjustment, the water supply in the farmland is more accurate, which not only avoids over-irrigation, but also effectively improves the growth efficiency of crops.

[0155] After two consecutive months of application, the effectiveness of the system has been significantly verified. The irrigation of corn fields in the southern part of the farmland during the flowering and fruiting periods is more accurate, the average growth of corn has increased by about 15%, and the yield per mu has increased from 980 kg last year to 1,120 kg. At the same time, the utilization rate of water resources in farmland has also improved significantly. The traditional irrigation system consumes about 620 cubic meters of water per mu of farmland per month, while after the application of the intelligent system, the water consumption has been reduced to 450 cubic meters, and the water saving rate has reached 27%. The data also shows that during the irrigation process from August to October, the prediction accuracy of the system is as high as 92%, which can not only respond to meteorological changes in a timely manner, but also perform adaptive regulation based on historical data and crop growth stages.

[0156] The application of the automatic irrigation control module has greatly reduced labor costs. In the previous irrigation process, each worker needed to spend an average of 6 hours to manually irrigate about 20 acres of farmland. After the introduction of the present invention, the system automatically controls the switch, flow rate and duration of the irrigation equipment through the preset optimized irrigation strategy, making the entire irrigation process fully automated. Workers only need to check the system operation regularly, and the average working time per person per day is shortened to 1.5 hours, greatly improving labor efficiency.

[0157] Table 1 Comparative analysis of traditional irrigation system and intelligent precision irrigation system

[0158]

[0159] The data in Table 1 above proves that the intelligent precision irrigation system has obvious advantages over the traditional irrigation system. The average growth of crops has increased by 15%, the yield per mu has increased from 980 kg to 1120 kg, the water resource utilization rate has increased significantly, and the water saving rate has reached 27%. At the same time, the prediction accuracy of the system is 92%, which can respond to meteorological changes more accurately. Labor costs have also been significantly reduced from 6 hours a day to 1.5 hours, greatly improving management efficiency.

[0160] The application of this invention in the agricultural demonstration base of City A not only solves the problems of the traditional irrigation system's slow response, single data processing, and inability to dynamically adjust irrigation strategies, but also greatly improves water resource utilization efficiency and crop yields through precise multimodal data fusion and adaptive optimization algorithms. In the future, the system is expected to be promoted and applied in a wider range of agricultural production, helping the intelligent development of modern agriculture.

[0161] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. Intelligent precision irrigation system driven by low-altitude remote sensing data, characterized by: It includes low-altitude drones, ground sensor network modules, meteorological data integration modules, spatiotemporal self-evolving feature graph network modules, spatiotemporal adaptive attention network modules, contextual multi-armed bandit algorithm modules and automated irrigation control modules; The low-altitude drone is equipped with a multispectral sensor and a thermal infrared camera for regular flights over farmland areas to obtain remote sensing data of crops; The ground sensor network module is used to collect real-time environmental information; The meteorological data integration module is used to receive meteorological forecast information from an external data source and perform multimodal data integration; The spatiotemporal self-evolving characteristic graph network module is used to process the acquired multimodal data, dynamically construct and evolve the dependency graph between regions, and generate a water demand distribution map; The spatiotemporal adaptive attention network is used to process the water demand distribution map and meteorological forecast data, dynamically capture short-term and long-term spatiotemporal dependencies, and optimize irrigation strategies; The contextual multi-armed bandit algorithm module is used to dynamically select the optimal irrigation strategy and adjust the irrigation parameters in real time according to the current crop status, environmental characteristics and historical irrigation feedback data; The automatic irrigation control module is used to implement the optimal irrigation strategy, control the switch, flow rate and irrigation duration of the irrigation equipment, and implement differentiated intelligent irrigation management according to different regions; The process of generating a water demand distribution map by the spatiotemporal self-evolution characteristic map network module specifically includes: Preprocess the multimodal data obtained from low-altitude drones and ground sensor networks, normalize the data according to time series and spatial dimensions, and generate a spatiotemporal data matrix , the spatiotemporal data matrix It includes crop spectral reflectance, soil moisture, temperature, air humidity and wind speed parameters in each area, and divides the time window according to the crop growth cycle; Based on the spatiotemporal self-evolving feature graph network, an initial dependency graph is established for each area of ​​the farmland , where nodes represent farmland areas, edges represent dependencies between areas, and edge weights are calculated: ； in, Indicates at time Time, farmland area and farmland areas The initial dependency weight between Indicates farmland area and farmland areas The spatial distance Indicates farmland area Environmental parameters, Indicates farmland area Environmental parameters, represents the environmental regulating factor, represents the temperature adjustment factor, Indicates farmland area and farmland areas The difference in vegetation index between and represents the control factor, Indicates farmland area The temperature, Indicates farmland area Temperature; Through the spatiotemporal convolution operation, the spatiotemporal data matrix Perform feature extraction to generate dynamic water demand feature vectors for each farmland area , the dynamic water demand characteristic vector Including the changing trends of water demand at current and historical moments; Dynamically update the dependency graph based on real-time feedback data from the sensor network , and calculate the updated edge weights: ； in, Indicates at time Time, farmland area and farmland areas The dynamically updated weight dependency weights between Indicates at time Time, farmland area and farmland areas The dependency weights between the last time step after update, represents the update rate, Indicates farmland area In time Moisture needs at all times, Indicates farmland area In time Moisture needs at all times, represents the time decay factor, and represents the control adjustment factor, Indicates farmland area The crop growth status index, Indicates farmland area The crop growth status index, Indicates farmland area In time The space-time feature vector at time, Indicates farmland area In time The spatiotemporal feature vector of the moment; Continuously evolving dependency graph , adaptively adjust the dependency weights between regions in the spatiotemporal dimension, and optimize the dependency relationship by combining historical environmental data and real-time sensor feedback; The evolved dependency graph is input into the water demand distribution generation module to generate a water demand distribution graph.

2. An intelligent precision irrigation method driven by low-altitude remote sensing data, applied to the intelligent precision irrigation system driven by low-altitude remote sensing data as claimed in claim 1, characterized in that: The steps include: S1. Use low-altitude drones equipped with multispectral sensors and thermal infrared cameras to regularly fly over farmland areas to obtain remote sensing data of crops; S2, using ground sensor networks to collect real-time environmental information, while obtaining weather forecast data from external data sources for multimodal data integration; S3, process the acquired multimodal data through the spatiotemporal self-evolving feature map network, dynamically construct and evolve the dependency relationship map between regions, adjust the dependency weights between regions according to the water demand and environmental changes of different regions, and generate a water demand distribution map; S4. Combined with spatiotemporal dynamic feature extraction, the self-evolving feature graph network is used to adaptively analyze the water demand of different regions, generate regional differentiated irrigation strategies, and make real-time adjustments according to different crop growth stages; S5. Input the water demand distribution map and meteorological forecast data into the spatiotemporal adaptive attention network to dynamically capture short-term and long-term spatiotemporal dependencies, predict the changes in water demand at different time scales in the future, and optimize irrigation strategies. S6, through the contextual multi-armed bandit algorithm, dynamically select the optimal irrigation strategy based on the current crop status, environmental characteristics and historical irrigation feedback data, and adjust the irrigation parameters in real time; S7, through the automatic irrigation equipment control module, execute the optimal irrigation strategy, control the switch, flow and irrigation duration of the irrigation equipment, and implement differentiated intelligent irrigation management according to different regions; The S3 specifically includes: S31. Preprocess the multimodal data obtained from low-altitude drones and ground sensor networks, normalize the data according to time series and spatial dimensions, and generate a spatiotemporal data matrix , the spatiotemporal data matrix It includes crop spectral reflectance, soil moisture, temperature, air humidity and wind speed parameters in each area, and divides the time window according to the crop growth cycle; S32. Based on the spatiotemporal self-evolving feature graph network, an initial dependency graph is established for each area of ​​the farmland. , where nodes represent farmland areas, edges represent dependencies between areas, and edge weights are calculated: ； in, Indicates at time Time, farmland area and farmland areas The initial dependency weight between Indicates farmland area and farmland areas The spatial distance Indicates farmland area Environmental parameters, Indicates farmland area Environmental parameters, represents the environmental regulating factor, represents the temperature adjustment factor, Indicates farmland area and farmland areas The difference in vegetation index between and represents the control factor, Indicates farmland area The temperature, Indicates farmland area Temperature; S33, through the spatiotemporal convolution operation, the spatiotemporal data matrix Perform feature extraction to generate dynamic water demand feature vectors for each farmland area , the dynamic water demand characteristic vector Including the changing trends of water demand at current and historical moments; S34. Dynamically update the dependency graph based on real-time feedback data from the sensor network , and calculate the updated edge weights: ； in, Indicates at time Time, farmland area and farmland areas The dynamically updated weight dependency weights between Indicates at time Time, farmland area and farmland areas The dependency weights between the last time step after update, represents the update rate, Indicates farmland area In time Moisture needs at all times, Indicates farmland area In time Moisture needs at all times, represents the time decay factor, and represents the control adjustment factor, Indicates farmland area The crop growth status index, Indicates farmland area The crop growth status index, Indicates farmland area In time The space-time feature vector at time, Indicates farmland area In time The spatiotemporal feature vector of the moment; S35. Continuously evolving dependency graph , adaptively adjust the dependency weights between regions in the spatiotemporal dimension, and optimize the dependency relationship by combining historical environmental data and real-time sensor feedback; S36, inputting the evolved dependency graph into a water demand distribution generation module to generate a water demand distribution graph.

3. The intelligent precision irrigation method based on low-altitude remote sensing data drive according to claim 2 is characterized in that: The real-time environmental information includes soil moisture, temperature, air humidity and wind speed, and the external data sources include satellite remote sensing and weather stations.

4. The intelligent precision irrigation method based on low-altitude remote sensing data drive according to claim 2 is characterized in that: The S4 specifically includes: S41. Extract spatiotemporal dynamic features from the multimodal data of each region to generate a spatiotemporal feature matrix , and divide the features into time windows to establish multi-level feature data sets for different time periods; S42. The spatiotemporal dynamic characteristics of each region are processed through the self-evolving feature graph network, and adaptive analysis is performed based on crop growth stages, environmental changes and historical data to calculate the water demand change rate of each region: ； in, represents the rate of change of water demand in farmland area i, represents the water demand of farmland area i, represents the environmental parameters of farmland area i, represents the environmental parameters of farmland region j, represents the temperature of farmland area i, represents the temperature of farmland region j, Indicates the time difference, represents the smoothing term, represents the soil moisture of farmland area i, represents the soil moisture in farmland area j, represents the difference in vegetation index between farmland area i and farmland area j, represents the difference in water demand changes between farmland area i and farmland area j; S43, based on the change rate of regional water demand , generate the regional priority matrix , calculate each priority value: ； in, represents the relative priority between farmland area i and farmland area j at time t, represents the hyperbolic cosine function, represents the exponential function, represents the water demand of farmland area j, represents the crop growth status index of farmland area i, represents the crop growth status index of farmland area j, represents the rate of change of water demand in farmland region j, represents the maximum difference in the rate of change of water demand between regions, represents the weight factor for adjusting the influence of temperature and humidity, represents the weight coefficient for regulating water demand between regions, represents the weight coefficient for regulating the impact of environmental differences among regions; S44, based on regional priority matrix , adaptively adjust the irrigation strategies of different areas, so that high-priority areas get priority in water allocation; the irrigation strategies of low-priority areas are dynamically adjusted according to the overall water resource conditions and crop growth stage; S45. Generate regional differentiated irrigation strategies by combining regional water demand characteristic matrix, historical meteorological data and future meteorological forecasts; S46. Input regional differentiated irrigation strategies into the automated irrigation control system, continuously adjust irrigation plans based on feedback data, and respond to environmental changes and crop needs in real time.

5. The intelligent precision irrigation method based on low-altitude remote sensing data drive according to claim 2 is characterized in that: The S5 specifically includes: S51, taking the water demand distribution map of the farmland area and the future weather forecast data as input and passing them into the spatiotemporal adaptive attention network; S52, process 1-3 days of meteorological data through the short-term convolution module of the spatiotemporal adaptive attention network, dynamically capture the short-term spatiotemporal dependencies between regions, and calculate the change rate of water demand in each region in the short term: ； in, represents the rate of change of water demand of farmland area i in the short term at time t, represents the water demand of farmland area i at time t, represents the weather forecast data of farmland area i at time t, represents the temperature of farmland area i, represents the temperature of farmland region j, represents the soil moisture of farmland area i, represents the soil moisture in farmland area j, represents the environmental parameters of farmland area i, represents the environmental parameters of farmland region j, represents the difference in vegetation index between farmland area i and farmland area j, represents the short-term adjustment factor, Represents the temperature and humidity difference adjustment coefficient, represents the spatiotemporal dependency weight; S53, through the long-term convolution module of the spatiotemporal adaptive attention network, the meteorological data of the next 4-30 days are processed to capture the long-term spatiotemporal dependencies and generate the long-term water demand change rate of each region: ； in, represents the rate of change of long-term water demand of farmland area i at time t, n represents the total number of days, represents the water demand of farmland area i in the past k days, represents the meteorological data of farmland area i over the past k days, represents the temperature of farmland area i in the past k days, represents the rate of change of water demand in farmland area i, represents the rate of change of water demand in farmland region j, represents the long-term adjustment factor, represents the long-term spatiotemporal feature adjustment factor, represents the long-term spatiotemporal dependence coefficient, represents the smoothness term; S54. Weighted fusion of captured short-term and long-term spatiotemporal dependencies to generate a comprehensive water demand change forecast , used to optimize irrigation strategies: ； S55. Comprehensive water demand forecast results Input into the irrigation strategy optimization module to obtain the optimized irrigation strategy.

6. The intelligent precision irrigation method based on low-altitude remote sensing data drive according to claim 2 is characterized in that: The S6 specifically includes: S61, constructing context feature vectors for each farmland area based on the current crop status, environmental characteristics, and historical irrigation feedback data through a contextual multi-armed bandit algorithm; S62, input the context feature vector into the context multi-armed bandit model, and calculate the instant reward function of each area according to the environmental state and feedback data of each area: ； in, represents the instantaneous return function of farmland region i, represents the water demand of farmland area i at time t, represents the moisture change of farmland area i at time t, represents the temperature of farmland area i, represents the temperature of farmland area j, represents the soil moisture of farmland area i, represents the soil moisture in farmland area j, represents the crop growth status index of farmland area i, represents the crop growth status index of farmland area j, and represents the adjustment coefficient; S63, the system based on context features and immediate reward function , dynamically select the optimal irrigation strategy for each area: ； in, represents the optimal irrigation strategy for farmland area i at time t, represents the balance factor between exploration and exploitation, The adjustment factor representing the effect of temperature change, represents the number of times irrigation strategy a is selected, represents the meteorological data of farmland area i at the current time t; S64. Optimal irrigation strategy based on dynamic selection , adjusting irrigation parameters in real time, the irrigation parameters including water volume, irrigation frequency and irrigation duration; S65, inputting the optimized optimal strategy into the automatic irrigation control system, executing corresponding irrigation operations, and adopting differentiated irrigation plans for different regions according to water demand, crop growth stage and historical feedback; S66. After implementing the irrigation strategy, the crop status, soil moisture and meteorological changes are monitored in real time through the sensor network, and the new feedback data is input into the contextual multi-armed bandit model to further adjust the reward function and optimize irrigation strategy selection.

Images