Crop growth environment intelligent regulation and control system and method based on Internet of Things

Abstract

The invention provides a crop growth environment intelligent regulation and control system and method based on the Internet of Things, and relates to the technical field of agricultural Internet of Things. The crop growth environment intelligent regulation and control system based on the Internet of Things comprises a multi-mode sensing layer which comprises an environment sensor, a crop physiological sensor and a spectrum acquisition device; the edge computing layer adopts edge computing nodes supporting multiple protocols; the cloud platform layer is used for constructing a crop growth model library, training a regional customization regulation and control model by utilizing historical data, and carrying out long-term data analysis and model optimization; the execution equipment layer comprises an irrigation system, ventilation equipment and a light supplementing device, and adjusts the growth environment of crops according to the edge computing node or a cloud instruction; and the user terminal layer realizes remote monitoring and personalized setting through a Web application or a mobile App. According to the invention, accurate and real-time regulation and control of the crop growth environment are realized through technologies of dynamic adaptive regulation and control, multi-modal data fusion, combination of edge computing and cloud computing and the like.

Description

Technical Field

[0001] This invention relates to the field of agricultural Internet of Things (IoT) technology, specifically to an intelligent control system and method for crop growth environment based on IoT. Background Technology

[0002] Modern agricultural production places increasingly stringent demands on environmental conditions, particularly in the precise management of temperature, humidity, light, and soil composition. Traditional manual monitoring methods are inefficient and struggle to guarantee data accuracy. With the development of IoT technology, improving the crop growing environment through real-time monitoring and automatic control is becoming a trend.

[0003] Chinese patent CN116449738A discloses a smart agricultural crop monitoring and control system based on the Internet of Things. By setting up cameras and alarms around the crops, the cameras can monitor and record the surrounding conditions of the crops. If animals are found to be stealing or damaging the crops, the alarm will sound to drive them away, thereby protecting the crops. However, the existing technology generally uses equal-interval sampling during data collection, which results in insufficient information collection during abnormal time periods, making the collected crop information incomplete.

[0004] Secondly, existing IoT-based crop growth environment control systems mostly rely on fixed threshold control, such as monitoring parameters like soil moisture and temperature through sensors, triggering irrigation or ventilation equipment when these parameters exceed preset thresholds. However, these systems have the following shortcomings:

[0005] Single data modality: Most systems rely solely on environmental sensor data (such as temperature and humidity, soil moisture), lacking monitoring of crop physiological indicators (such as chlorophyll content, stem diameter changes), resulting in incomplete regulatory decisions. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides an intelligent control system and method for crop growth environment based on the Internet of Things, which solves the problems mentioned in the background.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention provides the following technical solution: an intelligent control system for crop growth environment based on the Internet of Things, comprising:

[0010] The multimodal sensing layer includes environmental sensors (such as temperature, humidity, light, and CO2 concentration), crop physiological sensors (such as chlorophyll meters and stem diameter sensors), and spectral acquisition equipment, which are used to collect crop growth environment parameters and physiological state data in real time.

[0011] The edge computing layer uses edge computing nodes that support multiple protocols (such as the Ruibian F35 edge computer), with built-in NPUs to process multi-channel sensor data in parallel, and runs lightweight deep learning models (such as YOLOv5n) to identify pests and diseases and analyze crop growth, thereby achieving data cleaning, anomaly detection and real-time control decisions.

[0012] At the cloud platform layer, a crop growth model library is built, and regional customized regulation models (such as LSTM neural networks to predict irrigation demand) are trained using historical data. Long-term data analysis and model optimization are carried out, and user management, data visualization and remote control interfaces are provided.

[0013] The execution equipment layer includes irrigation systems, ventilation equipment, and supplemental lighting devices, which adjust the crop growth environment according to instructions from edge computing nodes or the cloud.

[0014] At the user terminal layer, remote monitoring, personalized settings, and agricultural expert system recommendations are achieved through web applications or mobile apps.

[0015] Preferably, the multimodal perception layer uses the temporal-spatial synchronization optimization algorithm (ST-FFA) to perform time alignment and feature extraction on multi-source data, and uses contrastive learning and attention mechanisms to map multimodal data to the same semantic space to generate crop growth state feature vectors.

[0016] Preferably, the edge computing layer uses deep reinforcement learning (DDRL) and human feedback-based reinforcement learning (RLHF) to generate real-time control strategies, and dynamically adjusts the irrigation cycle and water allocation based on crop water requirements and soil moisture.

[0017] Preferably, the cloud platform layer combines climate prediction data and a spectral-based crop water and fertilizer diagnostic model to optimize long-term regulation strategies and implements hierarchical execution through edge nodes. The edge nodes prioritize real-time commands such as pest and disease warnings, while the cloud handles long-term strategies such as seasonal irrigation plans.

[0018] Preferably, the execution device layer uses various sensors to provide feedback on the control effect, forming a closed loop of "perception-decision-execution-feedback". For example, irrigation equipment provides feedback on the actual irrigation amount, and cameras provide feedback on changes in crop growth.

[0019] Preferably, the edge computing node supports LoRaWAN / NB-IoT hybrid networking and solar power supply, and the hardware-level security protection adopts the TEE trusted execution environment to ensure the security of data transmission and device control.

[0020] This invention further discloses an intelligent control method for crop growth environment based on the Internet of Things (IoT), which, based on the aforementioned intelligent control system for crop growth environment based on the IoT, includes the following steps:

[0021] Step 1: Multimodal Data Acquisition and Fusion

[0022] Multi-source data is acquired through environmental sensors, crop physiological sensors, and spectral acquisition equipment. Feature fusion is performed using spatiotemporal synchronization optimization algorithms (such as ST-FFA) and cross-modal transfer technology to generate crop growth status feature vectors.

[0023] Step 2: Generation of dynamic control strategy:

[0024] Edge computing nodes use deep reinforcement learning (DDRL) and human feedback-based reinforcement learning (RLHF) to generate real-time control instructions, while the cloud combines crop growth models (such as spectral-based crop water and fertilizer diagnostic models) and climate prediction data to optimize long-term strategies.

[0025] Step 3: Tiered Implementation and Feedback

[0026] Edge computing nodes trigger real-time adjustments to irrigation, ventilation, and supplemental lighting equipment. Seasonal strategies (such as seasonal irrigation plans) are distributed from the cloud, and the results are fed back to the system via sensors to achieve closed-loop optimization.

[0027] Preferably, during the multimodal data acquisition and fusion process, the weight of thermal imaging data is automatically increased under complex lighting conditions to reduce the influence of RGB data and improve decision-making accuracy.

[0028] Preferably, in the generation of the dynamic regulation strategy, spectral data, image data and physiological index data are fused through comparative learning and attention mechanisms to improve the prediction accuracy of crop water and fertilizer requirements.

[0029] Preferably, during the hierarchical execution and feedback process, the edge computing node response speed reaches the millisecond level, and the cloud model library provides regional customized control solutions. In the application of more than 1.27 million mu of farmland in Inner Mongolia, the system response speed is improved by more than 5 times, and the leakage rate is controlled within 5%.

[0030] (III) Beneficial Effects

[0031] This invention provides an intelligent control system and method for crop growth environment based on the Internet of Things, which has the following beneficial effects:

[0032] 1. Through technologies such as dynamic adaptive regulation, multimodal data fusion, and the combination of edge computing and cloud computing, precise and real-time regulation of the crop growth environment can be achieved.

[0033] 2. By dynamically adjusting control strategies based on real-time crop status and environmental changes, water-saving rates are improved and crop yields are increased compared to traditional fixed threshold control. Integrating multi-source data such as environmental, physiological, and spectral data, and employing spatiotemporal synchronous optimization and cross-modal transfer technologies, decision-making accuracy is enhanced. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the control system structure of the present invention;

[0035] Figure 2 This is a schematic diagram of the control method of the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Example 1:

[0038] like Figure 1 As shown, this embodiment of the invention provides an intelligent control system for crop growth environment based on the Internet of Things, comprising:

[0039] The multimodal perception layer includes environmental sensors (such as temperature, humidity, light intensity, and CO2 concentration), crop physiological sensors (such as chlorophyll meters and stem diameter sensors), and spectral acquisition equipment, used to collect crop growth environment parameters and physiological state data in real time. Specifically, the multimodal perception layer uses the temporal-spatial synchronization optimization algorithm (ST-FFA) to perform time alignment and feature extraction on multi-source data, and uses contrastive learning and attention mechanisms to map multimodal data to the same semantic space to generate crop growth state feature vectors.

[0040] The edge computing layer employs multi-protocol edge computing nodes (such as the Ruibian F35 edge computer), with a built-in NPU for parallel processing of multi-channel sensor data. It runs lightweight deep learning models (such as YOLOv5n) for pest and disease identification and crop growth analysis, enabling data cleaning, anomaly detection, and real-time control decisions. Specifically, the edge computing layer uses deep reinforcement learning (DDRL) and human feedback-based reinforcement learning (RLHF) to generate real-time control strategies, dynamically adjusting irrigation cycles and water ratios based on crop water requirements and soil moisture.

[0041] At the cloud platform layer, a crop growth model library is built, and regional customized regulation models (such as LSTM neural networks to predict irrigation demand) are trained using historical data. Long-term data analysis and model optimization are performed, and user management, data visualization, and remote control interfaces are provided. Specifically, the cloud platform layer combines climate prediction data and spectral-based crop water and fertilizer diagnostic models to optimize long-term regulation strategies, and implements hierarchical execution through edge nodes. Edge nodes prioritize real-time commands such as pest and disease warnings, while the cloud handles long-term strategies such as seasonal irrigation plans.

[0042] The execution device layer, including irrigation systems, ventilation equipment, and supplemental lighting devices, adjusts the crop growth environment according to instructions from edge computing nodes or the cloud. Specifically, the execution device layer uses various sensors to provide feedback on the control effects, forming a closed loop of "perception-decision-execution-feedback." For example, irrigation equipment provides feedback on the actual irrigation amount, and cameras provide feedback on changes in crop growth. Secondly, the edge computing nodes support LoRaWAN / NB-IoT hybrid networking and solar power supply. Hardware-level security protection adopts a TEE trusted execution environment to ensure the security of data transmission and device control.

[0043] At the user terminal layer, remote monitoring, personalized settings, and agricultural expert system recommendations are achieved through web applications or mobile apps.

[0044] like Figure 2 As shown, this embodiment of the invention also provides a method for intelligent regulation of crop growth environment based on the Internet of Things (IoT). Based on the aforementioned IoT-based intelligent regulation system for crop growth environment, the method includes the following steps:

[0045] Step 1: Multimodal Data Acquisition and Fusion

[0046] Multi-source data is acquired through environmental sensors, crop physiological sensors, and spectral acquisition equipment. Feature fusion is performed using spatiotemporal synchronization optimization algorithms (such as ST-FFA) and cross-modal transfer technology to generate crop growth status feature vectors.

[0047] During multimodal data acquisition and fusion, thermal imaging data weights are automatically increased under complex lighting conditions to reduce the influence of RGB data and improve decision-making accuracy.

[0048] Step 2: Generation of dynamic control strategy:

[0049] Edge computing nodes use deep reinforcement learning (DDRL) and human feedback-based reinforcement learning (RLHF) to generate real-time control instructions, while the cloud combines crop growth models (such as spectral-based crop water and fertilizer diagnostic models) and climate prediction data to optimize long-term strategies.

[0050] In the generation of dynamic regulation strategies, spectral data, image data, and physiological index data are integrated through comparative learning and attention mechanisms to improve the prediction accuracy of crop water and fertilizer requirements.

[0051] Step 3: Tiered Implementation and Feedback

[0052] Edge computing nodes trigger real-time control of irrigation, ventilation equipment and supplemental lighting devices, while seasonal strategies (such as seasonal irrigation plans) are distributed from the cloud. The execution results are fed back to the system through sensors to achieve closed-loop optimization.

[0053] During the hierarchical execution and feedback process, the edge computing nodes have a response speed of milliseconds, and the cloud model library provides regional customized control solutions. In the application of more than 1.27 million mu of farmland in Inner Mongolia, the system response speed has been improved by more than 5 times, and the leakage rate has been controlled within 5%.

[0054] Example 2:

[0055] The following uses greenhouse tomato cultivation as an example to elaborate on the specific implementation of the present invention, covering the entire process details such as equipment deployment, data processing, strategy generation, and execution feedback, demonstrating the operability of the system in real-world scenarios:

[0056] I. Initial Deployment and Equipment Configuration

[0057] 1. Sensor and Actuation Device Layout

[0058] Environmental perception layer: divided into 50m intervals 2 The greenhouse area is a single monitoring unit, deploying a hybrid sensor network. Each node includes:

[0059] Soil parameters: A SEN0193 soil moisture sensor (measurement range 0-100%RH, accuracy ±2%) was used, buried at a depth of 15cm (tomato root active layer), and data was collected once every 30 seconds; a matching soil EC sensor (monitoring nutrient concentration, range 0-2000μS / cm) was used, and data was collected once every 10 minutes.

[0060] Air parameters: SHT30 temperature and humidity sensor (temperature -40~125℃, humidity 0-100%RH, accuracy ±0.3℃ / ±2%RH), S-1100 CO2 sensor (range 0-5000ppm, accuracy ±50ppm), installed 30cm above the plant canopy, collecting data once per minute.

[0061] Light parameters: TSL2591 light sensor (range 0.1-40000 lux) and multispectral camera (bands 450nm / 550nm / 650nm / 750nm, used for chlorophyll and nutrient inversion) are installed on the sliding rail at the top of the greenhouse. The camera moves and scans the entire area once every 2 hours, and a single scan covers 100㎡.

[0062] Crop physiological sensing: An LY-60 stem diameter sensor (measurement range 0-20mm, resolution 0.001mm) was attached to the main stem of the tomato plant at a height of 30cm above the ground to record the amount of stem shrinkage / expansion during the day and night (reflecting the degree of water stress); one plant was randomly selected from every 10 plants, and the SPAD value of the leaves was collected at regular intervals using a handheld chlorophyll meter (SPAD-502) (3 times a week, with the average value of 3 functional leaves taken each time), and the data was synchronized to the edge node via Bluetooth.

[0063] Equipment used: Intelligent drip irrigation system (with electromagnetic valves, control accuracy ±5mL / h), one drip irrigation tape (30cm spacing) is laid per row of tomatoes; ventilation equipment consists of a top-mounted axial flow fan (0.75kW power, adjustable speed) and electric push rods for side windows; supplemental lighting is LED plant growth lights (red to blue light ratio 7:3, power 30W / m²). 2 Adjustable light intensity); a small spraying robot for pest and disease control (load capacity 5L, positioning accuracy ±10cm).

[0064] Edge and network configuration: Each greenhouse is equipped with one Ruibian F35 edge computer (with NPU computing power of 8 TOPS, supporting LoRaWAN / NB-IoT dual mode), which connects to sensor nodes through a LoRa gateway (communication distance of 1km, power consumption <10mA), and uploads key data to the cloud through NB-IoT; the edge nodes are powered by solar energy and lithium batteries (battery life ≥72 hours on cloudy days), and have a built-in TEE trusted execution environment to store encryption keys.

[0065] II. Multimodal Data Acquisition and Real-time Edge Layer Processing

[0066] 1. Data preprocessing and spatiotemporal synchronization

[0067] Edge nodes perform real-time cleaning of the collected data:

[0068] Outlier removal: Soil moisture abrupt changes (such as instantaneous values ​​>80% or <10% due to sensor burial depth offset) are filtered using the 3σ rule, and missing data are filled by linear interpolation (interpolation error <3%).

[0069] Spatiotemporal alignment: Based on the ST-FFA algorithm (spatiotemporal fast alignment), multi-source data are anchored to a unified time axis (based on the local clock of the edge node, with timestamp error controlled within ±50ms); spatial interpolation is performed on sensor data from different locations (e.g., generating a soil moisture heat map of the entire greenhouse area with a resolution of 1m×1m using the inverse distance weighting method).

[0070] 2. Real-time crop status diagnosis

[0071] Growth analysis: The edge node runs a lightweight YOLOv5n model (inputting RGB images captured by the camera on the top of the greenhouse, frame rate 10fps) to identify the number of tomato leaves and plant height (error ±2cm), combined with stem diameter sensor data (e.g., the normal range of daily stem thickening during the seedling stage is 0.2-0.3mm, and if it is less than 0.1mm, it is judged as slow growth).

[0072] Physiological state assessment: Integrating spectral data and chlorophyll values, nitrogen content (R0.05) was inverted using a pre-trained PLSR (Partial Least Squares Regression) model. 2 =0.92), if the nitrogen value is <3.5% (the suitable range for tomatoes is 4.0%-5.0%), it is marked as "potentially nitrogen deficient".

[0073] Initial screening of diseases and pests: Feature extraction (such as lesion area and color moment) is performed on leaf images, and early blight and aphids are identified by an SVM classifier (91% accuracy) and aphids (95% accuracy). An early warning is triggered when the proportion of diseased leaves is greater than 5% or the number of aphids per plant is greater than 3.

[0074] III. Cloud-based model optimization and dynamic control strategy generation

[0075] 1. Dynamic adaptation during growth stages

[0076] The cloud-based LSTM growth stage segmentation model is trained based on historical data (3 years of greenhouse tomato planting records) and automatically determines the current stage by plant height, number of leaves, and number of flowers.

[0077] Seedling stage (0-30 days): Key control points include temperature (daytime 25±2℃, nighttime 18±1℃) and light intensity (≥200μmol / m²). 2 / s), soil moisture should be maintained at 60%-65%.

[0078] Flowering and fruit setting period (31-60 days): CO2 concentration needs to be increased to 800-1000ppm and soil moisture increased to 65%-70% to avoid drought causing flower drop.

[0079] During the fruiting period (61 days to harvest): increase potassium fertilizer supply, maintain soil moisture at 70%-75%, and control the diurnal temperature range at 10-12℃ (to promote sugar accumulation).

[0080] 2. Regulation Strategy Generation (Edge-Cloud Collaboration)

[0081] Real-time emergency control (edge-driven): When an edge node detects the following conditions, it generates a command within 100ms:

[0082] Sudden environmental changes: If the greenhouse temperature suddenly rises to 35℃ (the upper limit for tomato tolerance is 32℃), immediately start the top fan (adjust the wind speed to 8m / s) and open the side windows (at an angle of 45°), while reducing the supplemental light intensity to 50%.

[0083] Pest and disease early warning: Identification of localized aphid clusters (e.g., 3m) 2 If the area has more than 10 insects, instruct the spraying robot to locate the area and spray a low-toxicity insecticide (at a rate of 0.1 L / m²). 2 (Atomized particle size 50μm), and record the spraying location for review 24 hours later.

[0084] Long-term optimization strategy (cloud-led): The cloud combines 7-day weather forecasts (e.g., rainfall is expected on day 3, outdoor humidity > 90%) with a reinforcement learning model (DDRL) to optimize the irrigation plan.

[0085] Adjust the irrigation cycle: Change the original once a day (30 minutes each time) to once every other day, and extend each irrigation to 45 minutes (to ensure that the soil moisture meets the standard and avoid waterlogging);

[0086] Dynamic lighting adaptation: Automatically adjusts the LED light on-time based on the weather forecast light intensity (e.g., turn on 1 hour earlier on cloudy days, maintaining a total lighting time of 12 hours / day).

[0087] Nutrient regulation recommendations: Based on the nitrogen inversion results, a topdressing plan is generated (it is recommended to spray a 0.3% urea solution 3 days later, at a dosage of 2L / m³). 2 The message is pushed to the user's terminal.

[0088] IV. Execution Layer Linkage and Closed-Loop Feedback

[0089] 1. Tiered Implementation Mechanism

[0090] Edge commands are executed first: emergency control (such as pest spraying, overheating ventilation) is directly sent to the execution device via LoRa, and the execution results (such as valve opening status, fan operating current) are fed back to the edge node within 10 seconds.

[0091] Cloud-based strategies are executed on a timed basis: Non-urgent instructions such as irrigation and supplemental lighting are sent from the cloud to edge nodes for caching according to time windows (e.g., 2 a.m. and 10 a.m. daily). The edge nodes dynamically adjust the execution time according to the crop growth stage (e.g., irrigation during the fruiting period avoids the midday heat and is scheduled for 6 a.m.).

[0092] 2. Feedback and Model Iteration

[0093] Short-term feedback: Collect key parameters 1 hour after regulation (such as whether the soil moisture reaches the target value ±2% after irrigation, and whether the temperature drops below 30℃ after ventilation), and record the deviation value at the edge nodes (such as actual humidity 72% vs target 75%, deviation 3%).

[0094] Long-term feedback: Data is summarized weekly in the cloud, and control effect indicators are calculated (such as water saving rate = (traditional irrigation amount - system irrigation amount) / traditional irrigation amount, tomato yield increase rate = (system planting yield - traditional yield) / traditional yield). The model weights are optimized through the RLHF algorithm (such as adding attention weight to the "water sensitivity during the fruiting period" feature).

[0095] Implementation results:

[0096] In this greenhouse tomato growing scenario, after the system has been running for 6 months (one growing cycle), the key indicators are as follows:

[0097] Resource utilization: Irrigation water saving rate 42% (traditional irrigation 350m³) 3 / mu, system 203m 3 / acre), reducing electricity consumption by 30% (supplementary lighting and ventilation equipment);

[0098] Crop performance: The number of fruits per plant increased by 1.8 (traditional 5.2, system 7.0), the average single fruit weight increased by 15% (traditional 180g, system 207g), and the total yield increased by 18%.

[0099] Management efficiency: The incidence of pests and diseases is reduced to 3% (compared to 12% in the past), and the cost of manual inspection is reduced by 60% (no need for daily on-site monitoring).

[0100] Through the above specific implementation, the effectiveness of the "perception-decision-execution-feedback" closed loop of this system in greenhouse crop cultivation has been verified. It can be extended to other greenhouse crops such as strawberries and cucumbers, or applied to open field scenarios through parameter adaptation.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent control system for crop growth environment based on the Internet of Things, characterized in that: include: The multimodal sensing layer includes environmental sensors, crop physiological sensors, and spectral acquisition equipment, which are used to collect crop growth environment parameters and physiological state data in real time. The edge computing layer uses edge computing nodes that support multiple protocols, with a built-in NPU to process multi-channel sensor data in parallel, and runs a lightweight deep learning model to identify pests and diseases and analyze crop growth, thereby achieving data cleaning, anomaly detection and real-time control decisions. At the cloud platform layer, a crop growth model library is built, historical data is used to train regional customized regulation models, long-term data analysis and model optimization are carried out, and user management, data visualization and remote control interfaces are provided. The execution equipment layer includes irrigation systems, ventilation equipment, and supplemental lighting devices, which adjust the crop growth environment according to instructions from edge computing nodes or the cloud. At the user terminal layer, remote monitoring, personalized settings, and agricultural expert system recommendations are achieved through web applications or mobile apps.

2. The intelligent control system for crop growth environment based on the Internet of Things according to claim 1, characterized in that: The multimodal perception layer uses a spatiotemporal synchronization optimization algorithm to perform time alignment and feature extraction on multi-source data, and employs contrastive learning and attention mechanisms to map multimodal data to the same semantic space to generate crop growth state feature vectors.

3. The intelligent control system for crop growth environment based on the Internet of Things according to claim 1, characterized in that: The edge computing layer uses deep reinforcement learning and human feedback-based reinforcement learning to generate real-time control strategies, and dynamically adjusts the irrigation cycle and water allocation based on crop water requirements and soil moisture.

4. The intelligent control system for crop growth environment based on the Internet of Things according to claim 1, characterized in that: The cloud platform layer combines climate prediction data and a spectral-based crop water and fertilizer diagnostic model to optimize long-term regulation strategies and implements hierarchical execution through edge computing nodes. The edge computing nodes prioritize real-time commands such as pest and disease warnings, while the cloud handles long-term strategies such as seasonal irrigation plans.

5. The intelligent control system for crop growth environment based on the Internet of Things according to claim 1, characterized in that: The execution device layer uses various sensors to provide feedback on the control effect, forming a closed loop of "perception-decision-execution-feedback".

6. The intelligent control system for crop growth environment based on the Internet of Things according to claim 5, characterized in that: The edge computing node supports LoRaWAN / NB-IoT hybrid networking and solar power supply, and its hardware-level security protection adopts the TEE trusted execution environment.

7. A method for intelligent regulation of crop growth environment based on the Internet of Things, based on the intelligent regulation system for crop growth environment based on the Internet of Things as described in any one of claims 1-6, characterized in that: Includes the following steps: Step 1: Multimodal Data Acquisition and Fusion Multi-source data is acquired through environmental sensors, crop physiological sensors, and spectral acquisition equipment. Feature fusion is performed using spatiotemporal synchronization optimization algorithms and cross-modal transfer technology to generate crop growth status feature vectors. Step 2: Generation of dynamic control strategy: Edge computing nodes use deep reinforcement learning and human feedback-based reinforcement learning to generate real-time control instructions, while the cloud combines crop growth models and climate prediction data to optimize long-term strategies. Step 3: Tiered Implementation and Feedback Edge computing nodes trigger real-time adjustments to irrigation, ventilation, and supplemental lighting equipment. Seasonal strategies are distributed from the cloud, and the results are fed back to the system via sensors, achieving closed-loop optimization.

Images