Intelligent agricultural monitoring device and method

Abstract

The invention relates to an intelligent agricultural monitoring device and method. The device comprises a data acquisition module, a transmission module, a processing module, an alarm module and a power module. The data acquisition module collects environment information by using various sensors, and the data transmission module transmits data through a wireless technology. And a multi-modal data fusion unit, a deep learning optimization unit, a real-time trend prediction unit and an abnormal data diagnosis unit in the data processing module work cooperatively, data are analyzed, and a basis is provided for decision making. The alarm module notifies a manager when data is abnormal, and the power module ensures continuous operation of the device. The system overcomes the defects of a traditional monitoring mode, can comprehensively and accurately monitor the agricultural environment and the crop growth condition, optimizes disease and pest recognition through deep learning, predicts the environmental parameter trend and the disease and pest risk, guarantees the data accuracy, improves the intelligent level of agricultural production, reduces the labor cost, and has an important application value in modern agricultural production.

Description

Technical Field

[0001] The present invention relates to the field of agricultural monitoring, and specifically to an intelligent agricultural monitoring device and method. Background Art

[0002] With the development of agricultural modernization, the demand for accurate monitoring of the agricultural production environment and the growth status of crops is becoming increasingly urgent. Most traditional agricultural monitoring methods rely on manual inspections, which not only consume a large amount of manpower, material resources and time, but also have great subjectivity and lag in monitoring results. Some existing automated monitoring devices have deficiencies in aspects such as the comprehensiveness and accuracy of data collection and the adaptability to complex agricultural environments, and it is difficult to meet the requirements of modern agricultural refined and intelligent management. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent agricultural monitoring device and method to solve the problems raised in the above background art.

[0004] To achieve the above purpose, the present invention provides the following technical solutions: a data acquisition module, a data transmission module, a data processing module, an alarm module and a power supply module. The data acquisition module includes a temperature sensor, a humidity sensor, a light intensity sensor, a soil pH sensor, and a pest and disease image acquisition camera, and collects agricultural environment information through the above sensors;

[0005] The data transmission module is responsible for transmitting the data collected by the data acquisition module to the data processing center, and specifically uses ZigBee or Wi-Fi wireless transmission technology;

[0006] The data processing module specifically includes a multi-modal data fusion unit, a deep learning optimization unit, a real-time trend prediction unit and an abnormal data diagnosis unit. Among them, the multi-modal data fusion unit integrates the data transmitted by the data transmission module, and establishes a multi-modal data fusion model through environmental parameter and pest and disease image data to provide a basis for decision-making; the deep learning optimization unit uses transfer learning and GAN technology to optimize the pest and disease image recognition algorithm, improve the adaptability and recognition accuracy of the model, and is used to accurately identify pests and diseases; the real-time trend prediction unit uses time series algorithms to predict the trend of environmental parameters, and combines the correlation between pests and diseases and the environment to estimate risks to assist in early prevention; the abnormal data diagnosis unit checks the sensor data and pest and disease recognition results by setting thresholds to ensure the accuracy of the data and provide information for subsequent decision-making.

[0007] When the environmental parameters or the growth status of crops analyzed by the data processing module exceed the preset normal range, the alarm module will be immediately activated and send alarm information to agricultural managers via text messages or voice to enable timely corresponding measures to be taken;

[0008] The power supply module provides power support for the entire monitoring device and adopts a combination of a solar panel and a rechargeable battery. The solar panel converts solar energy into electrical energy and stores it in the rechargeable battery when there is light, and the rechargeable battery supplies power to the device at night or when the light is insufficient to ensure the continuous and stable operation of the device.

[0009] Preferably, the temperature sensor is used to collect the air temperature in the agricultural environment in real time; the humidity sensor monitors the air humidity and soil humidity; the light intensity sensor obtains the light intensity data; the soil pH sensor detects the soil pH; the pest and disease image acquisition camera periodically takes images of crop plants for subsequent pest and disease analysis.

[0010] Preferably, the specific working logic of the multi-modal data fusion unit is as follows:

[0011] Data acquisition: Receive the environmental parameter data of temperature, humidity, light intensity, and soil pH, as well as the pest and disease image data from the data transmission module, and the above data carry their respective acquisition time information.

[0012] Preprocessing of environmental parameter data: Time alignment: For the environmental parameter data with time series characteristics, perform precise alignment according to the data acquisition time; Construct a data matrix: After completing the time alignment, integrate the environmental parameter data of different types to construct an environmental parameter data matrix, that is, combine the temperature, humidity, light intensity, and soil pH data in the same data structure.

[0013] Association of pest and disease image data with environmental parameters: Time range screening. Since the acquisition time of pest and disease images is not completely synchronized with the acquisition time of environmental parameter data, when the pest and disease image acquisition camera captures a pest and disease image and identifies the pest and disease type, the multi-modal data fusion unit automatically searches for the environmental parameter data within 15 minutes before and after the image capture time.

[0014] Correlation analysis and model construction

[0015] Correlation analysis: Use the Pearson correlation coefficient method to measure the linear correlation between pest and disease data and environmental parameter data. The value range of the Pearson correlation coefficient is between -1 and 1. For each type of pest and disease, calculate its Pearson correlation coefficient with temperature, humidity, light intensity, and soil pH respectively.

[0016] Model construction: Based on the above-mentioned correlation analysis of pest and disease data and environmental parameter data, data feature extraction and selection are carried out. Specifically, the random forest algorithm is used to screen out environmental parameter features and pest and disease features, and the above features are used as the key inputs for model construction. The logistic regression model is selected as the basic model to construct a multi-modal data fusion model. Specifically, the screened environmental parameter features and pest and disease features are combined as the input variables of the model, and whether pests and diseases occur is used as the output variable. The logistic regression model is trained with historical pest and disease data, and the parameters of the model are determined by the maximum likelihood estimation method, so that the model can accurately fit the relationship between the occurrence of pests and diseases and environmental parameters.

[0017] Preferably, the specific working steps of the deep learning optimization unit are as follows:

[0018] Data reception and preprocessing: Receive environmental parameter data such as temperature, humidity, light intensity, and soil pH, as well as pest and disease image data from the data transmission module; adjust the size, normalize, and enhance the collected pest and disease images. At the same time, standardize the environmental parameter data;

[0019] Transfer learning optimization

[0020] Pre-trained model selection: Select a suitable pre-trained model from the model library. The pre-trained model is trained based on general agricultural pest and disease data and has basic pest and disease recognition capabilities; Model freezing and fine-tuning: Freeze the underlying network layer of the pre-trained model. The underlying network layer learns to input the pest and disease image samples and corresponding environmental parameter data in the data reception and preprocessing step into the model, encode the environmental parameters into vectors, splice them with the image feature vectors, and input them into the upper network layer of the model. Fine-tune the upper network layer. During the fine-tuning process, use the stochastic gradient descent method to continuously adjust the parameters of the model so that the model can adapt to the characteristics of pests and diseases under specific environmental conditions in the new scenario; Model evaluation and update: Use the test data set in the new scenario to evaluate the fine-tuned model, and calculate the accuracy and recall rate indicators of the model; If the evaluation result meets the expectation, update the fine-tuned model to the system; If the expectation is not met, adjust the fine-tuning strategy, such as increasing the amount of training data, adjusting the learning rate, etc., and re-perform fine-tuning;

[0021] Generative Adversarial Network to Augment the Dataset: GAN Initialization: Initialize the generative adversarial network, including the generator and the discriminator. The generator generates simulated pest and disease images, and the discriminator determines whether the input image is a real pest and disease image or a simulated image generated by the generator; Adversarial Training: The generator and the discriminator perform adversarial training. The generator generates more realistic simulated images to deceive the discriminator; The discriminator improves its judgment ability to accurately distinguish real images and simulated images; During the training process, continuously adjust the parameters of the generator and the discriminator to continuously improve the performance of both; Dataset Augmentation: When the GAN training reaches the preset level, the generator generates high-quality simulated pest and disease images, and combines the above simulated images with real pest and disease images to form a new training dataset for retraining the deep learning model to enhance the model's learning ability of pest and disease characteristics in different environments.

[0022] Preferably, the specific working logic of the real-time trend prediction unit is as follows:

[0023] Data Collection: Receive environmental parameter data of temperature, humidity, light intensity, and soil pH from the data transmission module. The environmental parameter data has an accurate timestamp to ensure the integrity of the time series characteristics of the data. At the same time, collect relevant data on past pest and disease occurrences, specifically including the types of pests and diseases, occurrence times, severity levels, and the corresponding environmental parameter situations at that time;

[0024] Data Preprocessing: Clean the collected environmental parameter data, and use the moving average method to smooth the data to reduce the impact of data fluctuations on subsequent analysis. For the historical pest and disease data, perform standardization to obtain preprocessed data;

[0025] Preliminary Trend Analysis: Use the above preprocessed data as input and input it into the ARIMA-LSTM hybrid model to perform preliminary analysis on the environmental parameter data;

[0026] Pest and Disease Risk Assessment: Combine the historical correlation between pest and disease occurrences and environmental factors. When a specific trend appears in the preliminary analysis of the environmental parameter data, match the predicted environmental parameter trend with the historical pest and disease occurrence data to assess the possible occurrence risk of pests and diseases. If the risk exceeds the preset threshold, send an alarm message to the alarm module.

[0027] Preferably, the specific working steps of using the preprocessed data as input and inputting it into the ARIMA-LSTM hybrid model to perform preliminary analysis on the environmental parameter data are as follows:

[0028] Step A, Data Preparation: Divide the preprocessed data into a training set and a test set in a ratio of 4:1 according to the time sequence;

[0029] Step B, ARIMA Model Construction

[0030] Determine the order: Analyze the characteristics of the training set with the help of the autocorrelation function and partial autocorrelation function to determine the autoregressive order p, the differencing order d, and the moving average order q. At the same time, use the Akaike information criterion and Bayesian information criterion to select the combination of p, d, and q with the smallest criterion value;

[0031] Model training: According to the determined order, use the training set to train the ARIMA model, and determine the parameters through the maximum likelihood estimation method to make it fit the short-term periodic changes of agricultural environment parameters;

[0032] Preliminary prediction: Use the trained ARIMA model to predict the training set, calculate the mean square error and mean absolute error indicators, and evaluate its ability to capture short-term fluctuations;

[0033] Step C, Build the LSTM network

[0034] Data reshaping: Reshape the data after preliminary processing by ARIMA in the format of the number of samples, time step, and number of features;

[0035] Build the network: Construct a network structure including an input layer, an LSTM layer, a fully connected layer, and an output layer. The input layer receives the reshaped data, the LSTM layer learns the long-term dependence relationship of environmental parameters, the fully connected layer integrates and transforms features, and the output layer generates the prediction result;

[0036] Train the model: Use the training set to train the LSTM network, adjust the weights and biases through the backpropagation algorithm, measure the prediction error with the mean square error loss function, use the Adam optimization algorithm to improve the performance, and regularly evaluate on the validation set during training. If the loss on the validation set no longer decreases, stop training;

[0037] Step D, Model fusion and final prediction: Combine the advantages of ARIMA in capturing short-term changes and LSTM in learning long-term trends, determine the weighting coefficient according to the characteristics of environmental parameters and the historical prediction effect of agricultural scenarios, and obtain the final prediction value through weighted averaging to provide a basis for pest and disease risk assessment; Evaluate the fused ARIMA-LSTM hybrid model with the test set, and calculate the MSE, MAE, and mean absolute percentage error indicators.

[0038] Preferably, an intelligent agricultural monitoring method includes the following steps:

[0039] Step 1, Data collection step: According to the preset time interval, various sensors in the data collection module collect agricultural environment parameters and crop growth image data in real time;

[0040] Step 2, Data transmission step: The data transmission module sends the data collected by the data collection module to the data processing module through wireless transmission;

[0041] Step 3: Data processing: After receiving the data, the data processing module first processes the pest and disease image data, uses a deep learning algorithm to pre-process the image and classify the pests and diseases, and at the same time, performs statistical analysis on other environmental parameter data to calculate the parameter change trend;

[0042] Step 4: Alarm step: According to the analysis results of the data processing module, when the environmental parameters or crop growth conditions are abnormal, the alarm module notifies relevant personnel according to the preset alarm method.

[0043] Compared with existing technologies, the present invention offers the following advantages: Comprehensive and accurate monitoring: The device is equipped with multiple sensors that can collect real-time data on environmental parameters such as temperature, humidity, light intensity, and soil pH, as well as image data on crop pests and diseases. A multimodal data fusion unit integrates this data, builds a fusion model, and conducts in-depth analysis of the relationship between environmental parameters and the occurrence of pests and diseases, providing a comprehensive and accurate basis for agricultural production decision-making, while overcoming the one-sidedness and lag of traditional monitoring methods. For example, correlation analysis revealed that aphid infestations are closely associated with specific combinations of temperature, humidity, light intensity, and soil pH, enabling early warning and prevention.

[0044] Efficient and accurate pest and disease identification: The deep learning optimization unit uses transfer learning and generative adversarial network (GAN) technology to optimize pest and disease image recognition algorithms. Transfer learning enables the model to quickly adapt to new planting areas or crop varieties, improving recognition efficiency by adjusting the model using a small number of samples. GAN expands the dataset, enhancing the model's ability to learn pest and disease characteristics in different environments, improving recognition accuracy, and facilitating timely detection and treatment of pests and diseases, thereby reducing crop losses.

[0045] Timely risk prediction and early warning: The real-time trend prediction unit uses an ARIMA-LSTM hybrid model to predict environmental parameter trends and assess risk based on historical correlations between pests and diseases and the environment. When an abnormal environmental parameter is predicted to potentially trigger a pest or disease, the alarm module promptly sends a notification via text message or voice message, enabling management to take preventive measures, reduce the risk of pest and disease outbreaks, and ensure healthy crop growth.

[0046] High Data Reliability: The Abnormal Data Diagnosis Unit sets thresholds to check sensor data and pest identification results, promptly identifying and addressing abnormal data to ensure the accuracy and reliability of monitoring data. Analysis of abnormal data also helps troubleshoot sensor failures, maintain the stable operation of the monitoring system, and provide a reliable data foundation for subsequent decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the device structure of the present invention;

[0048] Figure 2Schematic diagram of the working process of the multi-modal data fusion unit of the present invention;

[0049] Figure 3 Schematic diagram of the working process of the real-time trend prediction unit of the present invention;

[0050] Figure 4 Schematic diagram of the monitoring method process of the present invention. Specific embodiments

[0051] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0052] Embodiment 1

[0053] Please refer to Figures 1-3 , the present invention provides a technical solution: a data acquisition module, a data transmission module, a data processing module, an alarm module, and a power module. The data acquisition module includes a temperature sensor, a humidity sensor, a light intensity sensor, a soil pH sensor, and a pest and disease image acquisition camera, and collects agricultural environment information through the above sensors; the temperature sensor is used to collect the air temperature in the agricultural environment in real time; the humidity sensor monitors the air humidity and soil humidity; the light intensity sensor obtains light intensity data; the soil pH sensor detects the pH of the soil; the pest and disease image acquisition camera regularly takes images of crop plants for subsequent pest and disease analysis

[0054] The data transmission module is responsible for transmitting the data collected by the data acquisition module to the data processing center, specifically using ZigBee or Wi-Fi wireless transmission technology;

[0055] The data processing module specifically includes a multi-modal data fusion unit, a deep learning optimization unit, a real-time trend prediction unit, and an abnormal data diagnosis unit. Among them, the multi-modal data fusion unit integrates the data transmitted by the data transmission module, and establishes a multi-modal data fusion model through environmental parameter data and pest and disease image data to provide a basis for decision-making; the specific working logic is:

[0056] Data acquisition: Receive environmental parameter data such as temperature, humidity, light intensity, and soil pH, and pest and disease image data from the data transmission module, and the above data carry their respective acquisition time information;

[0057] Environmental parameter data preprocessing: Time alignment: For environmental parameter data with time series characteristics, precise alignment is carried out according to the data collection time. For example, for temperature, humidity, and light intensity data collected every 15 minutes, and soil pH data collected at the same or similar time intervals, they are matched through timestamps to ensure that different types of environmental parameter data correspond one by one in the time dimension; Constructing a data matrix: After completing time alignment, different types of environmental parameter data are integrated to construct an environmental parameter data matrix, that is, temperature, humidity, light intensity, and soil pH data are combined in the same data structure;

[0058] Association between pest and disease image data and environmental parameters: Time range screening, since the collection time of pest and disease images is not completely synchronized with the collection time of environmental parameter data, when the pest and disease image collection camera captures a pest and disease image and identifies the type of pest and disease, the multimodal data fusion unit automatically searches for environmental parameter data within 15 minutes before and after the image capture time. For example, if an aphid pest is identified in the pest and disease image captured at 10:30 am, the unit will extract various environmental parameter data during the period from 10:15 am to 10:45 am;

[0059] Correlation analysis and model construction

[0060] Correlation analysis: The Pearson correlation coefficient method is used to measure the linear correlation between pest and disease data and environmental parameter data. The value range of the Pearson correlation coefficient is between -1 and 1. The closer the absolute value is to 1, the stronger the linear correlation between the two variables. For each type of pest and disease (taking aphid pest as an example), the Pearson correlation coefficients between it and temperature, humidity, light intensity, and soil pH are calculated respectively. At the same time, considering that there may be a non-linear relationship between the occurrence of pests and diseases and environmental parameters, the Spearman rank correlation coefficient is also introduced for supplementary analysis. The Spearman rank correlation coefficient is calculated based on the ranks of the data and can better reflect the monotonic relationship between variables, not limited to linear relationships;

[0061] Model construction: Based on the above-mentioned association analysis of pest and disease data and environmental parameter data, data feature extraction and selection are carried out. Specifically, the random forest algorithm is used to screen out environmental parameter features and pest and disease features, and the above features are used as the key inputs for constructing the model; The logistic regression model is selected as the basic model to construct the multimodal data fusion model. Specifically, the screened environmental parameter features and pest and disease features are combined as the input variables of the model, and whether pests and diseases occur is used as the output variable. The logistic regression model is trained with historical pest and disease data, and then the parameters of the model are determined by the maximum likelihood estimation method to enable the model to accurately fit the relationship between the occurrence of pests and diseases and environmental parameters.

[0062] The deep learning optimization unit uses transfer learning and GAN technologies to optimize the pest and disease image recognition algorithm, improve the adaptability and recognition accuracy of the model, and is used to accurately identify pests and diseases. The specific working steps are as follows:

[0063] Data reception and preprocessing: Receive environmental parameter data such as temperature, humidity, light intensity, and soil pH, as well as pest and disease image data from the data transmission module; adjust the size of the collected pest and disease images, perform normalization processing and image enhancement, and at the same time, perform standardization processing on the environmental parameter data;

[0064] Transfer learning optimization

[0065] Pre-trained model selection: Select a suitable pre-trained model from the model library. The pre-trained model is trained based on general agricultural pest and disease data and has the basic pest and disease recognition ability; Model freezing and fine-tuning: Freeze the underlying network layer of the pre-trained model. The underlying network layer learns to input the pest and disease image samples and the corresponding environmental parameter data in the data reception and preprocessing step into the model, encode the environmental parameters into vectors, splice them with the image feature vectors, and then input them into the upper network layer of the model for fine-tuning. During the fine-tuning process, use the stochastic gradient descent method to continuously adjust the parameters of the model so that the model can adapt to the characteristics of pests and diseases under specific environmental conditions in the new scenario; Model evaluation and update: Use the test data set in the new scenario to evaluate the fine-tuned model, and calculate the accuracy rate and recall rate indicators of the model; if the evaluation result meets the expectation, update the fine-tuned model to the system; if it does not meet the expectation, adjust the fine-tuning strategy, such as increasing the amount of training data, adjusting the learning rate, etc., and re-perform fine-tuning;

[0066] Generative adversarial network to expand the data set: GAN initialization: Initialize the generative adversarial network, including the generator and the discriminator. The generator generates simulated pest and disease images, and the discriminator judges whether the input image is a real pest and disease image or a simulated image generated by the generator; Adversarial training: The generator and the discriminator perform adversarial training. The generator generates more realistic simulated images to deceive the discriminator; The discriminator improves its judgment ability to accurately distinguish real images and simulated images; During the training process, continuously adjust the parameters of the generator and the discriminator to continuously improve the performance of both; Data set expansion: When the GAN training reaches the preset level, the generator generates high-quality simulated pest and disease images, and combines the above simulated images with real pest and disease images to form a new training data set for re-training the deep learning model to enhance the model's learning ability of pest and disease characteristics in different environments

[0067] The abnormal data diagnosis unit checks the sensor data and pest and disease recognition results by setting thresholds to ensure the accuracy of the data and provide information for subsequent decision-making. The specific working logic is as follows:

[0068] Data collection: Receive environmental parameter data of temperature, humidity, light intensity, and soil pH from the data transmission module. The environmental parameter data has an accurate timestamp to ensure the integrity of the time series characteristics of the data. At the same time, collect relevant data on past pest and disease occurrences, specifically including the types of pests and diseases, occurrence time, severity, and the corresponding environmental parameter conditions at that time;

[0069] Data preprocessing: Clean the collected environmental parameter data, and use the moving average method to smooth the data to reduce the impact of data fluctuations on subsequent analysis. For the historical pest and disease data, perform standardization to obtain preprocessed data;

[0070] Preliminary trend analysis: Use the above preprocessed data as input and input it into the ARIMA-LSTM hybrid model to perform preliminary analysis on the environmental parameter data;

[0071] Pest and disease risk assessment: Combine the historical correlation between pest and disease occurrences and environmental factors. When a specific trend appears in the preliminary analysis of the environmental parameter data, match the predicted environmental parameter trend with the historical pest and disease occurrence data to assess the possible occurrence risk of pests and diseases. If the risk exceeds the preset threshold, send an alarm message to the alarm module;

[0072] Among them, the specific working steps of using the preprocessed data as input and inputting it into the ARIMA-LSTM hybrid model to perform preliminary analysis on the environmental parameter data are as follows:

[0073] Step A. Data preparation: Divide the preprocessed data into a training set and a test set according to a ratio of 4:1 in chronological order;

[0074] Step B. ARIMA model construction

[0075] Determine the order: Analyze the characteristics of the training set with the help of the autocorrelation function and partial autocorrelation function to determine the autoregressive order p, differencing order d, and moving average order q. At the same time, use the Akaike information criterion and Bayesian information criterion to select the combination of p, d, and q with the smallest criterion value;

[0076] Model training: Based on the determined order, use the training set to train the ARIMA model, and determine the parameters through the maximum likelihood estimation method to make it fit the short-term periodic changes of agricultural environmental parameters;

[0077] Preliminary prediction: Use the trained ARIMA model to predict the training set, calculate the mean square error and mean absolute error indicators, and evaluate its ability to capture short-term fluctuations;

[0078] Step C. LSTM network construction

[0079] Data reshaping: Reshape the data after preliminary processing by ARIMA into the format of the number of samples, time step, and number of features;

[0080] Network construction: Build a network structure including an input layer, an LSTM layer, a fully connected layer, and an output layer. The input layer receives the reshaped data, the LSTM layer learns the long-term dependence relationship of environmental parameters, the fully connected layer integrates and transforms features, and the output layer generates prediction results;

[0081] Model training: Train the LSTM network with the training set, adjust the weights and biases through the backpropagation algorithm, measure the prediction error with the mean square error loss function, use the Adam optimization algorithm to improve performance, and regularly evaluate on the validation set during training. If the loss on the validation set no longer decreases, stop training;

[0082] Step D, Model fusion and final prediction: Combine the advantages of ARIMA capturing short-term changes and LSTM learning long-term trends, determine the weighting coefficient according to the characteristics of environmental parameters and the historical prediction effect of agricultural scenarios, and obtain the final prediction value through weighted average to provide a basis for pest and disease risk assessment; Evaluate the fused ARIMA-LSTM hybrid model with the test set, and calculate the MSE, MAE, and mean absolute percentage error metrics

[0083] When the environmental parameters or crop growth conditions analyzed by the data processing module exceed the preset normal range, the alarm module will be immediately activated and send alarm information to agricultural managers via text messages and voice to take corresponding measures in a timely manner;

[0084] The power supply module provides power support for the entire monitoring device, adopting a combination of solar panels and rechargeable batteries; among them, the solar panels convert solar energy into electrical energy and store it in the rechargeable batteries when there is light, and the rechargeable batteries supply power to the device at night or when the light is insufficient to ensure the continuous and stable operation of the device.

[0085] Embodiment 2

[0086] Please refer to Figure 4 , An intelligent agricultural monitoring method, including the following steps:

[0087] Step 1, Data acquisition step: According to the preset time interval, various sensors in the data acquisition module collect agricultural environmental parameters and crop growth image data in real time;

[0088] Step 2, Data transmission step: The data transmission module sends the data collected by the data acquisition module to the data processing module through wireless transmission;

[0089] Step 3, Data Processing Step: After receiving the data, the data processing module first processes the pest and disease image data, preprocesses the image and identifies and classifies pests and diseases using deep learning algorithms. At the same time, it statistically analyzes other environmental parameter data and calculates the change trend of the parameters.

[0090] Step 4, Alarm Step: According to the analysis results of the data processing module, when the environmental parameters or the growth status of crops are abnormal, the alarm module notifies relevant personnel according to the preset alarm method.

[0091] The present invention focuses on the field of agricultural monitoring. Aiming at the deficiencies of traditional monitoring methods and existing automated monitoring devices, it proposes an innovative intelligent agricultural monitoring device and method, providing strong support for the development of agricultural modernization.

[0092] Core Composition and Functions: The intelligent agricultural monitoring device consists of data acquisition, transmission, processing, alarm, and power supply modules. The data acquisition module collects agricultural environment information through various sensors; the data transmission module uses wireless technology to transmit the data to the processing center; the data processing module includes multi-modal data fusion, deep learning optimization, real-time trend prediction, and abnormal data diagnosis units to achieve in-depth analysis and processing of the data; the alarm module notifies the management personnel in a timely manner when the environmental parameters or the growth status of crops are abnormal; the power supply module combines solar energy and rechargeable batteries to ensure the continuous and stable operation of the device.

[0093] Key Technologies and Advantages: The multi-modal data fusion unit integrates environmental parameter and pest and disease image data, constructs a model to reveal their correlations, and provides a basis for decision-making; the deep learning optimization unit uses transfer learning and GAN technologies to improve the adaptability and accuracy of pest and disease image recognition; the real-time trend prediction unit uses an ARIMA-LSTM hybrid model to predict the trend of environmental parameters and estimate the pest and disease risks; the abnormal data diagnosis unit ensures the accuracy of the data. The application of these technologies makes the monitoring more comprehensive, accurate, and timely, effectively improving the intelligent level of agricultural production management and reducing labor costs.

[0094] The present invention effectively solves the problems of traditional agricultural monitoring, can grasp the agricultural production environment and the growth status of crops in real time and accurately, early warning of pests and diseases in advance, provides a scientific basis for agricultural production decision-making, helps to reduce crop losses, improve agricultural production efficiency and quality, and promotes the development of agriculture towards intelligence and refinement, and has broad application prospects in modern agricultural production.

[0095] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent agricultural monitoring device, comprising a data acquisition module, a data transmission module, a data processing module, an alarm module and a power supply module, characterized in that: The data acquisition module includes a temperature sensor, a humidity sensor, a light intensity sensor, a soil pH sensor, and a pest and disease image acquisition camera, which collect agricultural environment information through the above sensors; The data transmission module is responsible for transmitting the data collected by the data acquisition module to the data processing center, and specifically uses ZigBee or Wi-Fi wireless transmission technology; The data processing module specifically includes a multi-modal data fusion unit, a deep learning optimization unit, a real-time trend prediction unit, and an abnormal data diagnosis unit. Among them, the multi-modal data fusion unit integrates the data transmitted by the data transmission module, and establishes a multi-modal data fusion model through environmental parameter and pest and disease image data to provide a basis for decision-making; the deep learning optimization unit uses transfer learning and GAN technology to optimize the pest and disease image recognition algorithm, improve the model adaptability and recognition accuracy, and is used to accurately identify pests and diseases; the real-time trend prediction unit uses time series algorithms to predict the trend of environmental parameters, combines the correlation between pests and diseases and the environment to estimate risks, and helps with early prevention; the abnormal data diagnosis unit checks the sensor data and pest and disease recognition results by setting thresholds to ensure the accuracy of the data and provide information for subsequent decision-making. When the environmental parameters or the growth status of crops analyzed by the data processing module exceed the preset normal range, the alarm module will be immediately activated and send alarm information to agricultural managers via text messages or voice, so as to take corresponding measures in a timely manner; The power supply module provides power support for the entire monitoring device, and adopts a combination of a solar panel and a rechargeable battery; among them, the solar panel converts solar energy into electrical energy and stores it in the rechargeable battery when there is light, and the rechargeable battery supplies power to the device at night or when the light is insufficient to ensure the continuous and stable operation of the device.

2. The intelligent agricultural monitoring device according to claim 1, characterized in that: The temperature sensor is used to collect the air temperature in the agricultural environment in real time; the humidity sensor monitors the air humidity and soil humidity; the light intensity sensor obtains light intensity data; the soil pH sensor detects the pH of the soil; the pest and disease image acquisition camera regularly takes images of crop plants for subsequent pest and disease analysis.

3. An intelligent agricultural monitoring device according to claim 1, characterized in that: The specific working logic of the multi-modal data fusion unit is as follows: Data acquisition: Receive environmental parameter data such as temperature, humidity, light intensity, and soil pH, as well as pest and disease image data from the data transmission module, and the above data carry their respective acquisition time information; Preprocessing of environmental parameter data: Time alignment: For environmental parameter data with time series characteristics, perform precise alignment according to the data acquisition time; Construct a data matrix: After completing the time alignment, integrate different types of environmental parameter data to construct an environmental parameter data matrix, that is, combine temperature, humidity, light intensity, and soil pH data in the same data structure; Association between pest and disease image data and environmental parameters: Time range screening. Since the acquisition times of pest and disease image data and environmental parameter data are not exactly synchronized, when the pest and disease image acquisition camera captures a pest and disease image and identifies the type of pest and disease, the multi-modal data fusion unit automatically searches for environmental parameter data within 15 minutes before and after the image capture time. Correlation analysis and model construction Correlation analysis: The Pearson correlation coefficient method is used to measure the linear correlation between pest and disease data and environmental parameter data. The value range of the Pearson correlation coefficient is between -1 and 1. For each type of pest and disease, the Pearson correlation coefficients with temperature, humidity, light intensity, and soil pH are calculated respectively; Model construction: Based on the above-mentioned association analysis of pest and disease data and environmental parameter data, data feature extraction and selection are carried out. Specifically, the random forest algorithm is used to screen out environmental parameter features and pest and disease features, and the above features are used as the key inputs for constructing the model. The logistic regression model is selected as the basic model to construct the multi-modal data fusion model. Specifically, the screened environmental parameter features and pest and disease features are combined as the input variables of the model, and whether the pest and disease occur is used as the output variable. The logistic regression model is trained with historical pest and disease data, and then the parameters of the model are determined by the maximum likelihood estimation method, so that the model can accurately fit the relationship between the occurrence of pest and disease and environmental parameters.

4. An intelligent agricultural monitoring device according to claim 1, characterized in that: The specific working steps of the deep learning optimization unit are as follows: Data reception and preprocessing: Receive environmental parameter data of temperature, humidity, light intensity, and soil pH, as well as pest and disease image data from the data transmission module; adjust the size, normalize, and enhance the acquired pest and disease images. At the same time, standardize the environmental parameter data; Transfer learning optimization Pre-trained model selection: Select a suitable pre-trained model from the model library. The pre-trained model is trained based on general agricultural pest and disease data and has the basic pest and disease recognition ability; Model freezing and fine-tuning: Freeze the underlying network layer of the pre-trained model. The underlying network layer learns to input the pest and disease image samples and corresponding environmental parameter data in the data reception and preprocessing step into the model, encodes the environmental parameters into vectors, concatenates them with the image feature vectors, and inputs them into the upper network layer of the model. Fine-tune the upper network layer. During the fine-tuning process, use the stochastic gradient descent method to continuously adjust the parameters of the model so that the model can adapt to the characteristics of pests and diseases under specific environmental conditions in the new scenario; Model evaluation and update: Use the test data set in the new scenario to evaluate the fine-tuned model, and calculate the accuracy and recall rate indicators of the model; If the evaluation result meets the expectation, update the fine-tuned model to the system; If it does not meet the expectation, adjust the fine-tuning strategy; Generative Adversarial Network for Augmenting Datasets: GAN Initialization: Initialize the generative adversarial network, including the generator and the discriminator. The generator generates simulated pest and disease images, and the discriminator determines whether the input image is a real pest and disease image or a simulated image generated by the generator; Adversarial Training: The generator and the discriminator perform adversarial training. The generator generates more realistic simulated images to deceive the discriminator; The discriminator improves its judgment ability to accurately distinguish real images and simulated images; During the training process, continuously adjust the parameters of the generator and the discriminator to continuously improve their performance; Dataset Augmentation: When the GAN training reaches the preset level, the generator generates high-quality simulated pest and disease images, and combines the above simulated images with real pest and disease images to form a new training dataset for retraining the deep learning model to enhance the model's learning ability of pest and disease characteristics in different environments.

5. An intelligent agricultural monitoring device according to claim 1, characterized in that: The specific working logic of the real-time trend prediction unit is as follows: Data Collection: Receive environmental parameter data of temperature, humidity, light intensity, and soil pH from the data transmission module. The environmental parameter data has an accurate timestamp to ensure the integrity of the time series characteristics of the data. At the same time, collect relevant data on past pest and disease occurrences, specifically including the types of pests and diseases, occurrence times, severity levels, and the corresponding environmental parameter conditions at that time; Data Preprocessing: Clean the collected environmental parameter data, and use the moving average method to smooth the data to reduce the impact of data fluctuations on subsequent analysis. For historical pest and disease data, perform standardization to obtain preprocessed data; Preliminary Trend Analysis: Use the above preprocessed data as input and input it into the ARIMA-LSTM hybrid model for preliminary analysis of the environmental parameter data; Pest and Disease Risk Assessment: Combine the historical correlation between pest and disease occurrences and environmental factors. When a specific trend appears in the preliminary analysis of the environmental parameter data, match the predicted environmental parameter trend with the historical pest and disease occurrence data to assess the possible occurrence risk of pests and diseases. If the risk exceeds the preset threshold, send an alarm message to the alarm module.

6. The intelligent agricultural monitoring device according to claim 5, characterized in that: Using the preprocessed data as input and inputting it into the ARIMA-LSTM hybrid model for preliminary analysis of the environmental parameter data, the specific working steps are as follows: Step A, Data Preparation: Divide the preprocessed data into a training set and a test set in a ratio of 4:1 according to the time sequence; Step B, ARIMA Model Construction Determine the Orders: Analyze the characteristics of the training set with the help of the autocorrelation function and the partial autocorrelation function to determine the autoregressive order p, the differencing order d, and the moving average order q. At the same time, use the Akaike information criterion and the Bayesian information criterion to select the combination of p, d, and q with the smallest criterion value; Model Training: Based on the determined orders, use the training set to train the ARIMA model, and determine the parameters through the maximum likelihood estimation method to make it fit the short-term periodic changes of agricultural environmental parameters; Preliminary Prediction: Use the trained ARIMA model to predict the training set, calculate the mean square error and mean absolute error indicators, and evaluate its ability to capture short-term fluctuations; Step C, LSTM Network Construction Data reshaping: Reshape the data after preliminary ARIMA processing in the format of the number of samples, time step, and number of features; Building the network: Construct a network structure including an input layer, an LSTM layer, a fully connected layer, and an output layer. The input layer receives the reshaped data, the LSTM layer learns the long-term dependence relationship of environmental parameters, the fully connected layer integrates and transforms features, and the output layer generates the prediction result; Training the model: Train the LSTM network with the training set, adjust the weights and biases through the backpropagation algorithm, measure the prediction error with the mean squared error loss function, use the Adam optimization algorithm to improve performance, and regularly evaluate on the validation set during training. If the loss on the validation set no longer decreases, stop training; Step D, Model fusion and final prediction: Combine the advantages of ARIMA in capturing short-term changes and LSTM in learning long-term trends, determine the weighting coefficients according to the characteristics of environmental parameters and the historical prediction effect of agricultural scenarios, and obtain the final prediction value through weighted averaging, providing a basis for pest and disease risk assessment; Evaluate the fused ARIMA-LSTM hybrid model with the test set and calculate the MSE, MAE, and mean absolute percentage error metrics.

7. An intelligent agricultural monitoring method according to any one of claims 1-6, characterized in that, Including the following steps: Step 1, Data collection step: According to the preset time interval, various sensors in the data collection module collect agricultural environmental parameter and crop growth image data in real time; Step 2, Data transmission step: The data transmission module sends the data collected by the data collection module to the data processing module through wireless transmission; Step 3, Data processing step: After receiving the data, the data processing module first processes the pest and disease image data, preprocesses the image and identifies and classifies pests and diseases using deep learning algorithms. At the same time, it statistically analyzes other environmental parameter data and calculates the change trend of the parameters; Step 4, Alarm step: According to the analysis results of the data processing module, when the environmental parameters or the growth status of crops are abnormal, the alarm module notifies relevant personnel according to the preset alarm method.

Images