Real-time online monitoring system based on smart agriculture

Abstract

The present application relates to the field of agricultural informatization technology, and particularly to a real-time online monitoring system based on smart agriculture, comprising: a perception layer comprising a plurality of sensor nodes distributedly arranged, for collecting agricultural environment parameters and crop growth parameters; an edge computing layer in communication connection with the perception layer, comprising at least one edge computing gateway, the edge computing gateway being used for pre-processing, feature extraction and adaptive anomaly detection of data collected by the sensor nodes; and a network transmission layer connected with the edge computing layer and a cloud service layer respectively, for uploading data processed by the edge computing layer to the cloud service layer, wherein the present application: through multi-layer cooperation of the perception layer, the edge computing layer, the network transmission layer and the cloud service layer, data pre-processing and anomaly detection are front-loaded to the edge, so that response delay and cloud load are greatly reduced.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, specifically a real-time online monitoring system based on smart agriculture. Background Technology

[0002] Smart agriculture uses technologies such as the Internet of Things, big data, and artificial intelligence to dynamically monitor and precisely regulate the growth environment and status of crops, which is a key path to improve agricultural production efficiency and resource utilization.

[0003] Most existing agricultural online monitoring systems adopt a direct "sensor-cloud platform" architecture. Sensor nodes transmit the raw data they collect directly to the cloud for processing via wireless networks. This architecture has significant shortcomings in practical applications: First, the real-time uploading of large amounts of raw data places high demands on network bandwidth and stability, often resulting in packet loss or outages in remote farmland environments due to insufficient signal coverage. Second, centralized cloud processing of all data leads to significant response latency, making it difficult to meet the needs for immediate detection and response to sudden environmental events. Third, traditional systems mostly only perform threshold comparisons on current environmental data, lacking dynamic modeling of crop growth processes and prediction of future trends, resulting in decisions such as irrigation and fertilization still relying on experience and failing to achieve precise on-demand control. In addition, field sensor nodes are usually battery-powered, and fixed-frequency sampling strategies struggle to balance monitoring accuracy and energy consumption, leading to high equipment maintenance costs.

[0004] How to build a real-time online monitoring system with edge intelligence, highly reliable transmission, digital twin prediction, and adaptive energy management has become a problem that needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time online monitoring system based on smart agriculture to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A real-time online monitoring system based on smart agriculture includes: The perception layer consists of multiple sensor nodes deployed in a distributed manner, used to collect agricultural environmental parameters and crop growth parameters; An edge computing layer, which is communicatively connected to the perception layer, includes at least one edge computing gateway, which is used to preprocess, extract features, and perform adaptive anomaly detection on the data collected by the sensor nodes. The network transport layer is connected to both the edge computing layer and the cloud service layer, and is used to upload the data processed by the edge computing layer to the cloud service layer. The cloud service layer includes a data storage module, a digital twin construction module, and an intelligent decision-making module. The digital twin construction module is used to drive crop growth models and soil hydrodynamic models based on uploaded data, generate virtual farmland mappings, and predict growth trends. The intelligent decision-making module is used to generate agricultural operation instructions. The application terminal communicates with the cloud service layer and is used to present monitoring results, forecast information, and agricultural operation instructions.

[0007] In a preferred embodiment of the present invention, the sensor node includes at least one of a soil multi-parameter sensor, a meteorological sensor, a plant spectral sensor, and an insect monitoring lamp. Each sensor node is equipped with a first near-field communication module and interacts with the edge computing gateway through the first near-field communication module.

[0008] In a preferred embodiment of the present invention, the edge computing gateway includes a data preprocessing module, a feature fusion module, and an anomaly detection module; The data preprocessing module is used to perform missing value imputation, filtering and noise reduction, and normalization on the collected data. The feature fusion module is used to map multi-source heterogeneous sensing data to a unified feature space and generate a fused feature vector. The anomaly detection module is used to identify anomalies in the fused feature vector based on dynamic baseline and sliding window statistics.

[0009] In a preferred embodiment of the present invention, the dynamic baseline is updated online by combining historical data from the same period with real-time meteorological factors, the anomaly detection module adaptively adjusts the judgment threshold according to the rate of change of environmental factors, and the window length of the sliding window statistics dynamically expands and contracts according to the data fluctuation amplitude.

[0010] In a preferred embodiment of the present invention, the network transmission layer includes a LoRa communication unit, an NB-IoT communication unit, and a network switching management module; The network switching management module is used to obtain the current signal strength, data priority tag and remaining power of the sensing layer, and dynamically switch between the LoRa communication unit and the NB-IoT communication unit according to the preset strategy, and adjust the data compression rate accordingly.

[0011] In a preferred embodiment of the present invention, the digital twin construction module of the cloud service layer acquires real-time situational data and weather forecast data uploaded by the edge computing layer, drives the pre-calibrated crop growth model and soil hydrodynamic model, generates a multi-dimensional virtual mapping of farmland including soil moisture distribution, crop canopy temperature and biomass accumulation, and outputs a growth trend prediction for a future preset period.

[0012] In a preferred embodiment of the present invention, the intelligent decision-making module includes an irrigation decision-making unit and a fertilization suggestion unit; The irrigation decision unit generates irrigation instructions for different irrigation zones based on the soil moisture content changes and crop water requirements at different growth stages predicted by the digital twin construction module. The fertilization recommendation unit generates a variable fertilization plan based on the crop nitrogen content retrieved by the plant spectral sensor and the soil nutrient data collected by the soil multi-parameter sensor, combined with the crop growth model.

[0013] In a preferred embodiment of the present invention, at least a portion of the sensor nodes in the sensing layer are powered by a composite power supply module, which includes a solar photovoltaic panel and an energy harvesting management circuit. The energy harvesting management circuit dynamically adjusts the sampling frequency and data reporting cycle of the corresponding sensor nodes according to the current light intensity and the remaining battery power.

[0014] In a preferred embodiment of the present invention, a network interruption resumption mechanism is provided between the edge computing layer and the cloud service layer; when the network transmission layer is interrupted, the edge computing gateway will sort the data to be transmitted according to timestamp and urgency priority and temporarily store it in local non-volatile memory, and complete the data retransmission after the network is restored.

[0015] In a preferred embodiment of the present invention, the cloud service layer further includes a model adaptive correction module, which periodically compares the predicted values ​​of the digital twin construction module with the actual monitored values, calculates the residuals, and uses an online incremental learning algorithm to fine-tune the parameters of the crop growth model.

[0016] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

[0017] Through multi-layered collaboration of the perception layer, edge computing layer, network transmission layer, and cloud service layer, data preprocessing and anomaly detection are moved to the edge, significantly reducing response latency and cloud load. Multi-mode switching and adaptive compression strategies in the network transmission layer ensure data accessibility under harsh communication conditions. The digital twin building block deeply integrates real-time monitoring data with crop mechanism models, achieving a leap from current state monitoring to future trend prediction, enabling scientific prediction of irrigation and fertilization decisions. Composite power supply and adaptive sampling strategies balance monitoring accuracy with node endurance. Overall, the system achieves real-time, reliable, and highly predictive smart agriculture monitoring. Attached Figure Description

[0018] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a schematic diagram of the module connection of a real-time online monitoring system based on smart agriculture in an embodiment of the present invention.

[0019] Figure 2 for Figure 1 Block diagram of edge computing gateway.

[0020] Figure 3 for Figure 1 A schematic diagram of the architecture of the cloud service layer.

[0021] Figure 4 This is a flowchart of the method for operating the system of the present invention. Detailed Implementation

[0022] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0023] This invention provides a real-time online monitoring system based on smart agriculture. See also... Figure 1 The system consists of a perception layer, an edge computing layer, a network transmission layer, a cloud service layer, and application terminals connected sequentially. These layers work collaboratively to complete a closed loop from farmland data collection to the issuance of intelligent decision-making commands.

[0024] The sensing layer comprises a large number of sensor nodes distributed across the field. Each sensor node acts as an electrical component and can integrate soil multi-parameter sensors, meteorological sensors, plant spectral sensors, or insect monitoring lamps, depending on monitoring requirements. The soil multi-parameter sensor internally includes a capacitive humidity probe, a thermistor temperature probe, and a conductivity electrode. Its working principle is as follows: the humidity probe obtains volumetric water content by measuring changes in the soil's dielectric constant; the temperature probe relies on the resistance of the thermistor material changing with temperature; and the conductivity electrode measures soil salinity through the ionic conduction current between its two electrodes. The meteorological sensor integrates an ultrasonic anemometer and wind direction measurement unit, a piezoresistive atmospheric pressure sensor, and an optical rain gauge. The plant spectral sensor uses a multi-channel narrowband photodetector to acquire reflectance in the visible and near-infrared bands, and calculates indicators reflecting crop nitrogen status, such as the normalized difference vegetation index, through built-in calculations.

[0025] The insect monitoring lamp uses a specific wavelength of ultraviolet light to attract pests, and works in conjunction with an infrared counting sensor and an image acquisition unit to achieve insect population statistics. Each sensor node is equipped with a first near-field communication module, specifically a wireless transceiver module based on the ZigBee protocol. This module transmits the digital signals collected by the sensors, after internal signal conditioning and analog-to-digital conversion, to its assigned edge computing gateway at regular intervals. Some sensor nodes deployed in areas without mains power are also equipped with a composite power supply module. This composite power supply module includes a solar photovoltaic panel and an energy harvesting management circuit. The core of the energy harvesting management circuit is a charging management chip with maximum power point tracking (MPPT) capability. It monitors the output voltage and current of the photovoltaic panel in real time, efficiently stores electrical energy in the lithium-ion battery pack, and simultaneously monitors the remaining battery power and the current light intensity.

[0026] When there is sufficient light and power, the energy harvesting management circuit sends instructions to the master controller of the sensor node to maintain the normal sampling frequency and reporting interval. When it is cloudy and rainy and the power drops to the preset threshold, the master controller reduces the sampling interval of sensors such as humidity and spectrum and extends the data reporting cycle, sacrificing temporary precision to ensure long-term monitoring of key parameters.

[0027] The edge computing layer consists of at least one edge computing gateway deployed in the field. Physically, each edge computing gateway is an embedded computing platform equipped with a multi-core ARM Cortex processor, memory, and non-volatile storage. It is connected to the perception layer in a star network topology via a ZigBee coordinator module.

[0028] See Figure 2 The edge computing gateway's internal functions are divided into a data preprocessing module, a feature fusion module, and an anomaly detection module. The data preprocessing module first performs time alignment and linear interpolation of missing values ​​on asynchronous data from different sensor nodes with different timestamps; then, it removes power frequency interference and random noise through a digital bandpass filter; finally, it performs min-max normalization to map various parameters to the same order of magnitude.

[0029] The feature fusion module is responsible for fusing heterogeneous data such as soil moisture, temperature, light intensity, and spectral index. It incorporates a pre-trained autoencoder model that reduces the dimensionality of multi-source inputs and maps them to a low-dimensional feature space, outputting a fused feature vector, thereby eliminating redundancy and extracting implicit correlations.

[0030] The anomaly detection module receives fused feature vectors and performs anomaly discrimination based on a dynamic baseline. Unlike fixed thresholds, the dynamic baseline is generated online by combining a weighted moving average of monitoring values ​​at the same time over the past several days with real-time meteorological factors. For example, the soil moisture threshold adaptively floats with changes in atmospheric temperature and wind speed: a larger decrease in humidity is allowed under high temperature and strong wind conditions, avoiding frequent false alarms. Simultaneously, this module maintains a sliding time window, calculates the statistical distribution of feature vectors within the window, and automatically adjusts the window length based on the severity of data fluctuations. When the deviation of a frame of data from the dynamic baseline exceeds the current adaptive threshold and is statistically significant within the sliding window, it is determined to be an anomaly and reported immediately, without waiting for a scheduled upload cycle.

[0031] The network transport layer is located between the edge computing gateway and the cloud service layer, and is used to provide wide area connectivity.

[0032] See Figure 1 The edge computing gateway motherboard integrates LoRa and NB-IoT communication units. The LoRa communication unit is based on spread spectrum modulation technology, which has the characteristics of long distance and low power consumption, but the data rate is relatively low; the NB-IoT communication unit is based on the licensed spectrum of cellular networks, which has higher data rate and better timeliness, but the power consumption is relatively high.

[0033] The network switching management module runs in the background of the gateway's operating system. It continuously monitors the current signal strength indication and signal-to-noise ratio of the two communication units and parses the priority tags of the data packets to be sent. For example, abnormal alarm signals are marked as high priority, while timed data in normal environments are marked as low priority. For high-priority data, the channel with the better current signal quality is selected for immediate transmission without compression. For low-priority batch data, if the signal is weak, the spreading factor is temporarily increased or the system switches to the more power-saving NB-IoT mode, and a lossy compression algorithm is used to compress the data volume before transmission.

[0034] When both signals are extremely weak or even interrupted, the automatic network interruption resumption mechanism is triggered. Data to be transmitted is sorted by timestamp and priority and written to the gateway's onboard large-capacity non-volatile memory. Once the network is restored, it is retransmitted to the cloud in sequence. This hybrid networking and adaptive strategy ensures that monitoring data is not lost and critical alarms are not delayed under various farmland communication conditions.

[0035] The cloud service layer receives and manages data across the entire domain.

[0036] See Figure 3 It includes a data storage module, a digital twin construction module, an intelligent decision-making module, and a model adaptive correction module. The data storage module adopts a combination of time-series databases and structured relational databases to classify and store real-time high-frequency data and device metadata, and sets up a data lifecycle strategy to automatically clean up expired data.

[0037] The digital twin module is the core prediction unit of this system. It incorporates a field-calibrated crop growth model and a soil hydrodynamic model. The crop growth model simulates leaf area index, biomass accumulation, and yield formation based on mechanistic equations such as radiation use efficiency and assimilate allocation. The soil hydrodynamic model, based on Darcy's law and the continuity equation, calculates the infiltration, redistribution, and root water uptake processes in the soil profile. Upon module startup, it acquires real-time situational data (soil temperature and humidity, meteorological elements, canopy spectrum, etc.) uploaded from the edge computing layer as initial state values ​​and obtains weather forecasts for the next few days from an external meteorological service interface. This drives the model to extrapolate in hourly steps, generating a multi-dimensional virtual map of farmland.

[0038] This mapping encompasses soil moisture distribution at different depths, crop canopy temperature changes, and biomass growth rates, and outputs key indicator prediction curves for the next 48 to 72 hours. The intelligent decision-making module uses this prediction to make agricultural feedforwards. Its internal irrigation decision-making unit, based on the soil moisture content changes projected by the digital twin for each future time period, combined with the water requirements of the current crop growth stage, calculates the water requirements and optimal irrigation start time for different irrigation zones, generating irrigation instructions that include the solenoid valve switching duration and flow rate.

[0039] The fertilization recommendation unit extracts the leaf nitrogen accumulation data recently retrieved from the plant spectral sensor, as well as the soil available nitrogen, available phosphorus and potassium data from the soil multi-parameter sensor. Using the target yield and nutrient absorption curve predicted by the crop growth model as a reference, it generates differentiated variable fertilization plans and sends them to the variable fertilization terminal.

[0040] The model adaptive correction module is used to ensure the long-term accuracy of the digital twin. It periodically extracts the actual monitoring time series stored in the storage module and compares it with the predicted series output by the digital twin model during the same period, calculating the residuals of various indicators. When the residuals continuously exceed the allowable range, the online correction process of model parameters is triggered.

[0041] The correction process does not retrain the model from scratch. Instead, it uses an online incremental learning algorithm to fix most of the model structure and only calculates gradients for a few sensitive parameters such as soil hydraulic properties and crop genetic parameters and fine-tunes them in the descent direction. This allows the model to gradually adapt to local soil variations and actual crop performance, thus avoiding model drift.

[0042] The application terminal can be a smartphone app, a personal computer webpage, or an agricultural large-screen display system. It connects to the cloud service layer via a secure socket encrypted channel, presenting a graphical representation of a 3D virtual map of farmland, real-time monitoring data, anomaly alarm lists, and future trend forecasts. Agricultural technicians or farmers can view irrigation and fertilization recommendations generated in the cloud at any time, and can also manually set thresholds or remotely control field equipment.

[0043] Figure 4 The diagram illustrates the overall workflow of the system. Sensor nodes in the perception layer collect data and send it to the edge computing gateway. The edge computing gateway performs preprocessing, feature fusion, and anomaly detection. If an anomaly is detected, an alarm is generated directly and reported in real time via the network transport layer; otherwise, the processed features and compressed data are reported periodically. The cloud receives and stores the data. The digital twin construction module combines weather forecasts to predict future conditions, the intelligent decision-making module generates agricultural instructions, and these instructions are presented to the user via the application terminal. Simultaneously, the model adaptive correction module periodically optimizes model parameters. This forms a continuous closed loop of data collection, analysis, prediction, decision-making, and correction.

[0044] Operation steps of a real-time online monitoring system based on smart agriculture: Step 1: System Deployment and Initialization; Within the target farmland, the sensor nodes of the sensing layer were installed according to a pre-defined spatial distribution plan. Soil multi-parameter sensors were buried at different depths in the crop root zone, meteorological sensors were erected in open areas of the field, plant spectral sensors were aligned with the crop canopy at a certain angle, and insect monitoring lamps were suspended at appropriate locations along the field ridges. The communication link between the first near-field communication module of each sensor node and its corresponding edge computing gateway was confirmed to be established normally. After the edge computing gateway was powered on, the communication networking program automatically ran, completing the registration and time synchronization of the sensor nodes. Simultaneously, the cloud service layer obtained historical meteorological data and future forecast data for the farmland area from an external meteorological service interface, completing the parameter initialization of the crop growth model and soil hydrodynamic model within the digital twin construction module.

[0045] Step 2: Real-time acquisition of multi-source data; Each sensor node periodically collects agricultural environmental parameters and crop growth parameters according to the initial sampling strategy set by the composite power supply module. Soil multi-parameter sensors simultaneously acquire volumetric water content, temperature, and conductivity of different soil layers; meteorological sensors collect wind speed, wind direction, atmospheric pressure, and rainfall; plant spectral sensors acquire multi-band reflectance and calculate the normalized vegetation index; insect monitoring lamps attract pests and count their populations. The collected raw data is transmitted to the edge computing gateway in real-time or near real-time via the first near-field communication module.

[0046] Step 3: Edge-side data processing and anomaly detection; After receiving data streams from multiple sensor nodes, the edge computing gateway performs time alignment and linear interpolation of missing values ​​on the asynchronous data. Then, it uses a digital bandpass filter to remove power frequency interference and random noise, and performs min-max normalization on each parameter. Next, the feature fusion module uses a pre-trained autoencoder model to map the normalized multi-source heterogeneous data into a low-dimensional fused feature vector. The anomaly detection module works synchronously: it generates a dynamic baseline online based on the historical weighted moving average and daily meteorological factors, and adjusts the sliding window length based on the current data fluctuation amplitude; it compares the fused feature vector with the dynamic baseline to identify an anomaly when the deviation exceeds an adaptive threshold and is statistically significant within the sliding window, generating a high-priority alarm.

[0047] Step 4: Multimode network transmission and resume transmission after network outage; Before each data upload, the network switching management module in the network transport layer obtains the signal strength indicators and signal-to-noise ratio of the LoRa and NB-IoT communication units, and reads the priority tags of the data packets to be transmitted. If the data to be transmitted is a high-priority packet such as an abnormal alarm or emergency command feedback, it is immediately transmitted through the communication channel with the best current signal quality without compression. If it is a low-priority packet of regular periodically collected data, when the signal quality is below a preset threshold, it switches to a more reliable communication mode and enables lossy compression. When both communication methods are interrupted, the network interruption resumption mechanism is triggered: the data to be transmitted is sorted according to timestamp and urgency, temporarily stored in the large-capacity non-volatile memory of the edge computing gateway, and retransmitted in order after the network is restored to ensure that no data is lost.

[0048] Step 5: Cloud-based digital twin simulation and intelligent decision-making; The data storage module of the cloud service layer classifies and stores the received real-time situational data according to time series and structured methods. The digital twin construction module initiates simulations periodically or triggered by events: using the latest uploaded soil moisture, meteorological conditions, vegetation indices, etc. as the initial state, combined with future weather forecasts, it drives the crop growth model and soil hydrodynamic model to perform hourly dynamic simulations, generating a multi-dimensional virtual map that includes changes in soil moisture profile, crop canopy temperature evolution, and biomass accumulation predictions, and outputs key indicator prediction curves for the next 48 to 72 hours. Subsequently, the irrigation decision unit in the intelligent decision module calculates the required irrigation amount and optimal irrigation time for each irrigation zone based on the predicted changes in soil moisture content and the current water requirement of the crop during its growth stage, and generates irrigation instructions; the fertilization suggestion unit generates differentiated variable fertilization plans based on the crop nitrogen content retrieved from the plant spectral sensor, soil nutrient detection values, and the target yield output by the crop growth model.

[0049] Step Six: Application Terminal Presentation and Agricultural Execution; Agricultural technicians or farmers log in to the cloud service layer via an application terminal and a secure socket encrypted channel. The application terminal displays a 3D virtual map of the farmland with a visual graphical interface, updating environmental monitoring data, pest information, and anomaly alarm lists in real time. Simultaneously, the terminal interface displays future trend curves predicted by the digital twin, irrigation suggestions, and fertilization plans. Users can view and confirm instructions, and if necessary, manually adjust parameter thresholds or directly remotely control field solenoid valves, variable-rate fertilizer applicators, and other execution equipment to deliver decision-making instructions to the field.

[0050] Step 7: Adaptive Model Adjustment; During continuous operation, the model adaptive correction module periodically extracts actual monitoring records from the data storage module and calculates residuals with the predicted values ​​of the concurrent digital twin model. When the residuals consistently exceed the allowable error range, the system automatically triggers an online incremental learning process: maintaining the main model structure unchanged, it calculates gradients and performs small-step fine-tuning only for a few sensitive parameters such as soil hydraulic properties and crop genetic parameters. The corrected parameters are updated and written into the digital twin construction module, allowing the prediction model to gradually adapt to local soil variations and actual crop growth performance, ensuring long-term prediction accuracy. The entire process requires no human intervention, ensuring continuous self-optimization during continuous operation.

[0051] Although embodiments of the invention have been shown and described, those skilled in the art will understand 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 claims and their equivalents.

Claims

1. A real-time online monitoring system based on smart agriculture, characterized in that, include: The perception layer consists of multiple sensor nodes deployed in a distributed manner, used to collect agricultural environmental parameters and crop growth parameters; An edge computing layer, which is communicatively connected to the perception layer, includes at least one edge computing gateway, which is used to preprocess, extract features, and perform adaptive anomaly detection on the data collected by the sensor nodes. The network transport layer is connected to both the edge computing layer and the cloud service layer, and is used to upload the data processed by the edge computing layer to the cloud service layer. The cloud service layer includes a data storage module, a digital twin construction module, and an intelligent decision-making module. The digital twin construction module is used to drive crop growth models and soil hydrodynamic models based on uploaded data, generate virtual farmland mappings, and predict growth trends. The intelligent decision-making module is used to generate agricultural operation instructions. The application terminal communicates with the cloud service layer and is used to present monitoring results, forecast information, and agricultural operation instructions.

2. The real-time online monitoring system based on smart agriculture according to claim 1, characterized in that, The sensor nodes include at least one of soil multi-parameter sensors, meteorological sensors, plant spectral sensors, and insect monitoring lamps. Each sensor node is equipped with a first near-field communication module and interacts with the edge computing gateway through the first near-field communication module.

3. The real time online monitoring system based on smart agriculture as claimed in claim 1 wherein, The edge computing gateway includes a data preprocessing module, a feature fusion module, and an anomaly detection module; The data preprocessing module is used to perform missing value imputation, filtering and noise reduction, and normalization on the collected data. The feature fusion module is used to map multi-source heterogeneous sensing data to a unified feature space and generate a fused feature vector. The anomaly detection module is used to identify anomalies in the fused feature vector based on dynamic baseline and sliding window statistics.

4. The real time online monitoring system based on smart agriculture as claimed in claim 3, wherein, The dynamic baseline is updated online by combining historical data from the same period with real-time meteorological factors; the anomaly detection module adaptively adjusts the judgment threshold according to the rate of change of environmental factors; and the window length of the sliding window statistics dynamically expands and contracts according to the data fluctuation amplitude.

5. The real time online monitoring system based on smart agriculture as claimed in claim 1 wherein, The network transmission layer includes a LoRa communication unit, an NB-IoT communication unit, and a network switching management module; The network switching management module is used to obtain the current signal strength, data priority tag and remaining power of the sensing layer, and dynamically switch between the LoRa communication unit and the NB-IoT communication unit according to the preset strategy, and adjust the data compression rate accordingly.

6. The real time online monitoring system based on smart agriculture as claimed in claim 1 wherein, The digital twin building module of the cloud service layer obtains real-time situational data and weather forecast data uploaded by the edge computing layer, drives the pre-calibrated crop growth model and soil hydrodynamic model, generates a multi-dimensional virtual mapping of farmland including soil moisture distribution, crop canopy temperature and biomass accumulation, and outputs a growth trend prediction for a future preset period.

7. The real time online monitoring system based on smart agriculture as claimed in claim 6 wherein, The intelligent decision-making module includes an irrigation decision-making unit and a fertilization suggestion unit; The irrigation decision unit generates irrigation instructions for different irrigation zones based on the soil moisture content changes predicted by the digital twin construction module and the crop water requirement patterns at different growth stages. The fertilization recommendation unit generates a variable fertilization plan based on the crop nitrogen content retrieved by the plant spectral sensor and the soil nutrient data collected by the soil multi-parameter sensor, combined with the crop growth model.

8. The real time online monitoring system based on smart agriculture as claimed in claim 1 wherein, At least some of the sensor nodes in the sensing layer are powered by a composite power supply module, which includes a solar photovoltaic panel and an energy harvesting management circuit. The energy harvesting management circuit dynamically adjusts the sampling frequency and data reporting cycle of the corresponding sensor nodes according to the current light intensity and the remaining battery power.

9. The real time online monitoring system based on smart agriculture as claimed in claim 1 wherein, A network interruption resumption mechanism is set between the edge computing layer and the cloud service layer; when the network transmission layer is interrupted, the edge computing gateway will sort the data to be transmitted according to timestamp and urgency priority and temporarily store it in local non-volatile memory, and complete the data retransmission after the network is restored.

10. The real time online monitoring system based on smart agriculture as claimed in claim 1 wherein, The cloud service layer also includes a model adaptive correction module, which periodically compares the predicted values ​​of the digital twin construction module with the actual monitored values, calculates the residuals, and uses an online incremental learning algorithm to fine-tune the parameters of the crop growth model.

Images