Corn disease and insect pest intelligent early warning system and method based on Web and mobile terminal

Abstract

The invention discloses an intelligent early warning system and method for corn diseases and insect pests based on Web and a mobile terminal, and the system comprises a data collection module, a data processing and analysis module, an early warning generation module, a user interaction and self-learning module, and a data safety module. And a clear model training and iteration mechanism is adopted, so that the system can stably adapt to the actual monitoring requirements of large-scale corn planting, and timely and reliable decision support is provided for prevention and control of diseases and insect pests.

Description

Technical Field

[0001] This invention relates to the field of pest early warning system technology, specifically to a web-based and mobile-based intelligent early warning system and method for corn pests and diseases. Background Technology

[0002] In agricultural production, corn, as one of my country's main food crops, directly impacts national food security and agricultural economic development through its yield and quality. However, corn pests and diseases are characterized by their sudden onset and high transmissibility. Traditional control methods rely on manual inspections and identification, resulting in low efficiency, delayed response, and high misjudgment rates, and are difficult to implement for large-scale, all-weather dynamic monitoring. Furthermore, existing pest and disease early warning systems generally suffer from limitations such as single data sources, insufficient model recognition accuracy, weak multi-terminal collaboration capabilities, and a lack of closed-loop optimization mechanisms, failing to meet the demands of modern agricultural precision control. With the rapid development of technologies such as the Internet of Things (IoT), BeiDou positioning, deep learning, edge computing, and 5G communication, technological support has been provided for the intelligent upgrading of agriculture. Multi-source sensor networks can achieve real-time acquisition of environmental parameters; UAV remote sensing combined with BeiDou positioning technology can overcome the spatial limitations of ground monitoring; improved deep learning models can enhance the accuracy of image recognition; edge computing and cloud-based collaborative processing can balance data real-time performance and analytical depth; and cross-platform interaction between the Web and mobile terminals can achieve precise delivery of early warning information. Against this backdrop, a multi-technology integrated intelligent early warning system for corn diseases and pests has emerged, aiming to address the pain points of traditional early warning models and build a fully intelligent system encompassing "collection-analysis-early warning-optimization". Summary of the Invention

[0003] To solve the above-mentioned technical problems, the technical solution provided by the present invention is: a smart early warning system for corn diseases and pests based on Web and mobile terminals, comprising:

[0004] The data acquisition module is used to acquire multi-dimensional data of the corn planting area through multiple sources. The multi-dimensional data includes corn plant image data, environmental parameter data, corn growth cycle data, historical pest and disease occurrence data, and agricultural operation data. The multiple sources include deployed high-definition cameras, sensor networks, UAV remote sensing equipment, and user input units.

[0005] The data processing and analysis module includes an edge computing node and a cloud processing unit, which are connected to the data acquisition module via a data bus. The edge computing node is used to perform local preprocessing on the real-time acquired environmental data, and the cloud processing unit is used to perform in-depth processing on the full amount of multi-dimensional data and integrate deep learning models and hybrid machine learning algorithms for intelligent analysis.

[0006] An early warning generation module connected with the data processing and analysis module, configured to generate multi-level early warning information based on the analysis results combined with an early warning rule base; the early warning rule base is constructed based on historical pest data statistics, agricultural expert knowledge modeling and confidence rule reasoning, and supports dynamic updating; the early warning generation module further comprises an early warning trace unit configured to record the data basis, model analysis results and operation logs of each early warning trigger;

[0007] A user interaction and self-learning module respectively connected with the early warning generation module and a Web terminal and a mobile terminal, configured to push early warning information and provide visual services and bidirectional interaction functions, and collect user feedback data; the user interaction and self-learning module dynamically optimizes model parameters and early warning thresholds through a reinforcement learning algorithm based on a reward function; the reward function takes early warning accuracy, false alarm rate and user feedback satisfaction as core evaluation indexes;

[0008] A data security module connected with the data collection module and the data processing and analysis module, configured to realize identity authentication, data encryption and integrity verification in the data transmission and storage process by combining a hardware encryption module with an AES symmetric encryption algorithm.

[0009] Preferably, the sensor network comprises a field sensor unit including an air temperature and humidity sensor, a soil humidity sensor, an illumination intensity sensor and a rainfall sensor, configured to continuously collect environmental parameter data; the user input unit is embedded in the Web terminal and the mobile terminal, supporting user manual reporting of pest symptoms, cultivation management measures and corn growth stage information.

[0010] Preferably, the deep learning model is a serial combination of a convolutional neural network with a channel attention mechanism and a YOLOv5 model, which first extracts deep image features through the convolutional neural network, and then inputs the YOLOv5 model to realize accurate identification of pest types and severity; the hybrid machine learning algorithm is an algorithm that fuses random forests and long short-term memory networks, which first sorts the feature importance of environmental parameters through random forests, filters key environmental factors, and then inputs the key environmental factors and time series data into the long short-term memory network to predict pest occurrence risks.

[0011] Preferably, the dynamic updating of the early warning rule base comprises: based on newly added pest occurrence data, user feedback information and agricultural expert revision opinions every quarter, the confidence parameters of the rules are optimized through an evidence reasoning algorithm to ensure the adaptability of the rules to actual planting scenarios.

[0012] Preferably, the user interaction with the visualization service of the self-learning module includes displaying the pest and disease distribution, early warning history and environmental parameter trend in the form of heat map, statistical chart, text and image combination on the web end and mobile end; the two-way interaction function supports users to upload field photos for instant diagnosis, obtain personalized prevention and control suggestions and consult experts online.

[0013] The application also discloses a corn pest and disease intelligent early warning method based on a web and a mobile end.

[0014] S1, multi-source data acquisition: environmental parameter data are collected by a sensor network at a frequency of once per minute, and are received and preliminarily processed by an edge computing node in real time; multi-spectral images and visible light images are obtained by a preset flight route of a Beidou positioning-based unmanned aerial vehicle (UAV) three times per week; corn plant images are captured by a high-definition camera at regular time intervals, and combined with manually reported agricultural operation data and pest and disease symptom information to form a complete data set; the collected data are transmitted to a cloud server by an AES encryption algorithm;

[0015] S2, data standardization preprocessing: the cloud server cleans, converts formats and removes outliers of the collected multi-source heterogeneous data, performs image registration by scale-invariant feature transformation, converts the data to the [0, 1] interval by min-max or Z-score standardization, and forms a structured data set suitable for model analysis;

[0016] S3, intelligent analysis and prediction: the preprocessed image data are input into an improved deep learning model, deep layer features are extracted by a CNN with channel attention mechanism, and pest and disease species, severity and feature parameters are output by a YOLOv5 model; environmental, growth cycle and historical data are input into a hybrid machine learning algorithm, key environmental factors are selected by a random forest, and the future 72-hour pest and disease occurrence risk level is predicted by an LSTM network training;

[0017] S4, multi-level early warning generation: based on the analysis and prediction results, corresponding level early warning is triggered by combining the judgment logic of the early warning rule library, early warning content including pest and disease information, risk level, personalized prevention and control suggestions and contact information of responsible personnel is generated;

[0018] S5, information pushing and interaction: early warning information is accurately pushed to users by three ways of a web end, a mobile end, a short message and an application, and visual query service and decision support are provided;

[0019] S6, closed-loop optimization: prevention and control effects and field verification data fed back by users are collected, model parameters and early warning thresholds are adjusted based on a reinforcement learning algorithm; model iteration is completed once every half month, and system self-adaptive optimization is realized.

[0020] Compared with the prior art, the application has the advantages that:

[0021] (1) The application integrates high-definition cameras, multiple types of sensors, Beidou positioning unmanned aerial vehicles, and multi-source collection methods reported by users to realize all-around acquisition of multi-dimensional data such as corn plant images, environmental parameters, growth cycles, and historical pest and disease information. The breakpoint continuation and intelligent obstacle avoidance functions of the unmanned aerial vehicle, the layered transmission of multiple communication protocols, ensure the stability and real-time performance of data acquisition, and provide a solid data foundation for accurate analysis.

[0022] (2) The application uses a CNN combined with a YOLOv5 serial combination model to effectively extract deep image features, with a pest and disease identification accuracy of not less than 90%. A hybrid algorithm combining random forests and LSTM is used to accurately filter key environmental factors and realize 72-hour risk prediction, with a model recall rate of more than 88%. An edge computing and cloud collaborative processing architecture ensures the local rapid preprocessing of real-time data and the deep analysis of full data, balancing efficiency and accuracy.

[0023] (3) The application constructs a dynamic early warning rule base based on historical data, expert knowledge, and confidence rule reasoning, which is iteratively optimized quarterly to adapt to actual scenarios. Four early warning levels are divided, combined with an expert online confirmation mechanism and a warning traceability unit to ensure the accuracy and traceability of early warning information. The dynamic threshold setting for specific pests and diseases further improves the relevance of early warning.

[0024] (4) The application uses mobile applications compatible with Android and iOS and a Web terminal to display data in various visual forms, supports offline viewing, voice interaction, and instant diagnosis, and realizes seamless connection of "early warning-prevention" through personalized recommendations based on a green prevention knowledge base, online expert consultation, and other functions, fully meeting the actual production needs of users.

[0025] (5) The application uses a reinforcement learning algorithm based on a reward function to dynamically optimize model parameters with indicators such as early warning accuracy and false alarm rate, and iterates every half month to form a self-adaptive closed-loop mechanism. The application uses a combination of hardware encryption and AES symmetric encryption to realize identity authentication, encryption, and integrity verification of data transmission and storage, and fully guarantees data security.

[0026] (6) The application's standardized multi-source data acquisition frequency, scientific preprocessing process, and clear model training and iteration mechanism ensure that the system can stably adapt to the actual monitoring needs of large-scale corn planting and provide timely and reliable decision support for pest and disease control. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1is a framework schematic diagram of a corn disease and pest intelligent early warning system based on Web and mobile terminal.

[0028] Figure 2 is a framework schematic diagram of a data processing and analysis module in a corn disease and pest intelligent early warning system based on Web and mobile terminal.

[0029] Figure 3 is a framework schematic diagram of a warning generation module in a corn disease and pest intelligent early warning system based on Web and mobile terminal. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0031] EMBODIMENT

[0032] A corn disease and pest intelligent early warning system based on Web and mobile terminal comprises:

[0033] A data acquisition module is used to acquire multi-dimensional data of a corn planting area through a multi-source mode, wherein the multi-dimensional data comprises corn plant image data, environmental parameter data, corn growth cycle data, historical disease and pest occurrence data and farming operation data, and the multi-source mode comprises a deployed high-definition camera, a sensor network, a unmanned aerial vehicle remote sensing device and user manual reporting; the unmanned aerial vehicle remote sensing device realizes preset route cruising and breakpoint continuous flying based on Beidou positioning technology, and is equipped with an intelligent obstacle avoidance module.

[0034] The sensor network comprises: a field sensor unit comprising an air temperature and humidity sensor, a soil humidity sensor, an illumination intensity sensor and a rainfall sensor, which is used to continuously collect environmental parameter data; a unmanned aerial vehicle image acquisition unit equipped with a multispectral camera and a high-definition camera, which is used to regularly cruise and shoot multispectral images and visible light images of corn fields; a user input unit embedded in a mobile application, which supports user manual reporting of disease and pest symptoms, cultivation management measures and corn growth stage information; and a data communication unit supporting 4G / 5G, LoRa and Wi-Fi multiple communication protocols, which realizes layered transmission of collected data to edge computing nodes and cloud servers.

[0035] The data processing and analysis module includes an edge computing node and a cloud processing unit, which are connected with the data acquisition module through a data bus; the edge computing node is used for local preprocessing of the real-time collected environmental data, the cloud processing unit is used for deep processing of full-volume multidimensional data, and an improved deep learning model and a hybrid machine learning algorithm are integrated for intelligent analysis; the improved deep learning model is a serial combination of a convolutional neural network with a channel attention mechanism and a YOLOv5 model, the deep features of the image are extracted through the convolutional neural network first, and then the YOLOv5 model is input to realize accurate identification of the type and severity of the plant diseases and insect pests; the hybrid machine learning algorithm is an algorithm fusing random forest and long short-term memory network, the feature importance of the environmental parameters is sorted through the random forest first, the key environmental factors are screened, and then the key environmental factors and time series data are input into the long short-term memory network to predict the occurrence risk of the plant diseases and insect pests;

[0036] The preprocessing of the data processing and analysis module includes that the edge computing node adopts a sliding window filtering method to eliminate environmental sensor data noise, and processes image data through Gaussian filtering and histogram equalization; the cloud processing unit adopts a Laplace criterion or a Z-score standardization method to eliminate outliers, performs format conversion, min-max normalization, and time label and geographic coordinate labeling on all data to form a structured data set; multi-spectral images are geometrically and radiometrically corrected, and combined with RGB images to realize accurate positioning of the plant disease and insect pest area.

[0037] The early warning generation module is connected with the data processing and analysis module, and is used for generating multi-level early warning information based on the analysis results and a pre-warning rule base; the pre-warning rule base is constructed based on historical plant disease and insect pest data statistics, agricultural technology expert knowledge modeling and confidence rule reasoning, and supports dynamic updating; the dynamic updating of the pre-warning rule base includes that based on newly added plant disease and insect pest occurrence data, user feedback information and agricultural technology expert revision opinions every quarter, the confidence parameters of the rules are optimized through an evidence reasoning algorithm to ensure the adaptability of the rules to actual planting scenarios; the early warning levels are divided into a blue observation level, a yellow warning level, an orange danger level and a red serious level; the early warning generation module further includes a pre-warning tracing unit, which is used for recording the data basis, model analysis results and operation logs of each pre-warning triggering, and the pre-warning information needs to be confirmed online by regional agricultural technology experts before being published, to ensure the traceability and accuracy of the pre-warning;

[0038] A user interaction and self-learning module is in communication connection with the early warning generation module and a web terminal and a mobile terminal respectively, used for pushing early warning information and providing visual service and a two-way interaction function, and collecting user feedback data; a reward function-based reinforcement learning algorithm is used to dynamically optimize model parameters and early warning thresholds, forming a closed-loop optimization mechanism; the reward function takes early warning accuracy, false alarm rate and user feedback satisfaction as core evaluation indexes; the visual service of the user interaction and self-learning module includes displaying pest and disease distribution, early warning history and environmental parameter trends in the form of a heat map, statistical charts, text and images on the web terminal and the mobile terminal; the two-way interaction function supports users to upload field photos for instant diagnosis, obtain personalized prevention and control suggestions and consult experts online; the personalized prevention and control suggestions are generated based on pest and disease species, severity and real-time environmental data, and in association with a preset green prevention and control knowledge base; the mobile application is compatible with Android and iOS systems, and has the ability of offline viewing, voice interaction and early warning information voice broadcast.

[0039] A data security module is in communication connection with the data collection module and the data processing and analysis module, and adopts a combination of a hardware encryption module and an AES symmetric encryption algorithm to realize identity authentication, data encryption and integrity check in the process of data transmission and storage.

[0040] A corn pest and disease intelligent early warning method based on a web terminal and a mobile terminal includes the following steps:

[0041] S1, multi-source data collection: environmental parameter data is collected by a sensor network at a frequency of once per minute, and is received and preliminarily processed by an edge computing node in real time; multi-spectral images and visible light images are obtained by a preset flight route based on Beidou positioning of an unmanned aerial vehicle cruising three times a week; corn plant images are taken by a high-definition camera at regular intervals, combined with user-manually-reported farming operation data and pest and disease symptom information, to form a complete data set; the collected data is transmitted to a cloud server by an AES encryption algorithm;

[0042] S2, data standardization preprocessing: the cloud server cleans, converts formats, and removes outliers of the collected multi-source heterogeneous data, uses scale-invariant feature transformation for image registration, and converts the data to the [0, 1] interval through min-max or Z-score standardization, to form a structured data set suitable for model analysis;

[0043] S3, intelligent analysis and prediction: input the preprocessed image data into an improved deep learning model, first extract deep features through the CNN with channel attention mechanism, then output the pest and disease species, severity and feature parameters through the YOLOv5 model, the model recognition accuracy is not less than 90%; input the environment, growth cycle and historical data into the hybrid machine learning algorithm, filter the key environmental factors through the random forest, then input the LSTM network for training, and predict the pest and disease risk level within 72 hours; the training process of the improved deep learning model includes: constructing an image sample library containing more than 20 common corn diseases and pests and 5 severity levels, and performing data enhancement through image rotation, cropping, noise addition and brightness adjustment; the sample library is divided into training set, validation set and test set in the ratio of 7:2:1, Adam optimizer is used, learning rate is set to 0.001, and iteration number is 100 rounds, after training and parameter adjustment, the model recall rate is not less than 88%;

[0044] S4, multi-level early warning generation: based on the analysis and prediction results, combined with the judgment logic of the early warning rule library, trigger the corresponding level early warning, generate early warning content containing pest and disease information, risk level, personalized prevention and control suggestions and contact information of responsible personnel; the early warning information is published after being confirmed online by regional agricultural technology experts, if the experts do not respond within 30 minutes, the system automatically triggers the early warning and marks the "to be reviewed" state; the early warning threshold of different pests and diseases is dynamically set, among which: when the continuous humidity exceeds 85% and lasts for more than 24 hours, the early warning of early blight is triggered; when the number of corn borer adults captured per day exceeds 5 per trap, the orange early warning is triggered; when the predicted incidence of pests and diseases exceeds 30%, the red severe level early warning is directly triggered;

[0045] S5, information pushing and interaction: push the early warning information to the user through Web, mobile phone message and application pushing, provide visual query service and decision support;

[0046] S6, closed loop optimization: collect user feedback on prevention and control effect and field verification data, adjust model parameters and early warning threshold based on reinforcement learning algorithm; the reinforcement learning takes the improvement of early warning accuracy by 10% and the reduction of false alarm rate by 15% as the goal, completes model iteration every half month, and realizes self-adaptive optimization of the system; the user feedback data includes prevention and control measure execution, pest and disease control effect score and field actual pest and disease occurrence, which are collected through the star rating and text input functions of the mobile application as the reward signal input model of reinforcement learning.

[0047] I. Application example

[0048] Taking the main corn production area of Shangshui County, Zhoukou City, Henan Province as an example, the area has a planting area of about 120,000 mu, and common diseases and pests include corn borer, early blight, large spot, aphid, etc. The implementation period is from June to September 2024 during the corn growing season.

[0049] 1. Multi-source acquisition device deployment

[0050] High-definition camera: Deploy Hikvision DS-2CD3T47WD-L high-definition cameras in the corn planting core area of 12 towns in the county. Install on 3-meter-high field poles, lens focal length 8mm, take visible light images of corn plants every 2 hours from 6:00 to 18:00 every day, focusing on capturing leaf and stem features.

[0051] 2. Sensor network:

[0052] Field sensor unit: Deploy 1 set every 20 mu, including SHT30 air temperature and humidity sensor, soil moisture sensor, light intensity sensor, and rain sensor. Collect environmental parameters every minute and store locally.

[0053] Unmanned aerial vehicle image acquisition unit: Configure DJI M300RTK unmanned aerial vehicle, based on Beidou positioning to preset the cruise route, cruise from 9:00 to 11:00 every Monday, Wednesday and Friday morning, support breakpoint resume flight, intelligent obstacle avoidance module to avoid obstacles such as power poles and trees.

[0054] User input unit: Through custom mobile application, support farmers / agronomists to manually report disease and pest symptoms, farming operations, and growth stages;

[0055] 3. Construction and dynamic update of early warning rule library

[0056] Blue observation level: Corn borer adult capture amount 1-3 per trap per day, or air humidity 75%-85% for 12 hours;

[0057] Yellow warning level: Early blight detected with 2nd level symptoms, and air humidity >85% for 24 hours; or aphid density 10-20 per plant;

[0058] Orange danger level: Corn borer adult capture amount >5 per trap per day, or large spot 3rd level symptoms >10%;

[0059] Red severe level: Predict any disease and pest incidence rate >30%, or detect 5th level symptoms.

[0060] Through the above implementation, the system issued 37 effective warnings during the corn growing season in Shangshui County in 2024, and the disease and pest control response time was shortened to within 4 hours. The corn disease and pest loss rate in the county decreased from 8.5% last year to 3.2%, verifying the practicality and effectiveness of the system.

[0061] The above describes the present application and its embodiments, which are not limited, and the drawings only show one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired thereby, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution are not creative, and should belong to the protection scope of the present application.

Claims

1. A web- and mobile-based intelligent early warning system for corn diseases and pests, characterized in that, include: The data acquisition module is used to acquire multi-dimensional data of the corn planting area through multiple sources. The multi-dimensional data includes corn plant image data, environmental parameter data, corn growth cycle data, historical pest and disease occurrence data, and agricultural operation data. The multiple sources include deployed high-definition cameras, sensor networks, UAV remote sensing equipment, and user input units. The data processing and analysis module includes an edge computing node and a cloud processing unit, which are connected to the data acquisition module via a data bus. The edge computing node is used to perform local preprocessing on the real-time acquired environmental data, and the cloud processing unit is used to perform in-depth processing on the full amount of multi-dimensional data and integrate deep learning models and hybrid machine learning algorithms for intelligent analysis. The early warning generation module, connected to the data processing and analysis module, is used to generate multi-level early warning information based on the analysis results and the early warning rule base. The early warning rule base is constructed based on historical pest and disease data statistics, agricultural expert knowledge modeling, and confidence rule reasoning, and supports dynamic updates. The early warning generation module also includes an early warning tracing unit, which is used to record the data basis, model analysis results and operation logs for each early warning trigger; The user interaction and self-learning module is connected to the warning generation module and the web and mobile terminals respectively. It is used to push warning information, provide visualization services and two-way interactive functions, and collect user feedback data. The user interaction and self-learning module dynamically optimizes model parameters and warning thresholds through a reinforcement learning algorithm based on a reward function. The reward function uses warning accuracy, false alarm rate and user feedback satisfaction as core evaluation indicators. The data security module is communicatively connected to the data acquisition module and the data processing and analysis module. It uses a combination of hardware encryption module and AES symmetric encryption algorithm to realize identity authentication, data encryption and integrity verification during the data acquisition, transmission and storage process.

2. The intelligent early warning system for corn diseases and pests based on Web and mobile terminals according to claim 1, characterized in that, The sensor network includes field sensor units, including air temperature and humidity sensors, soil moisture sensors, light intensity sensors, and rainfall sensors, for continuously collecting environmental parameter data; the user input unit is embedded in the web terminal and mobile terminal, supporting users to manually report pest and disease symptoms, cultivation management measures, and corn growth stage information.

3. The intelligent early warning system for corn diseases and pests based on Web and mobile terminals according to claim 1, characterized in that, The deep learning model is a sequential combination of a convolutional neural network with channel attention mechanism and a YOLOv5 model. It first extracts deep features of the image through the convolutional neural network, and then inputs them into the YOLOv5 model to achieve accurate identification of the types and severity of pests and diseases. The hybrid machine learning algorithm is an algorithm that integrates random forest and long short-term memory network. It first sorts the environmental parameters by feature importance through random forest, selects key environmental factors, and then inputs the key environmental factors and time series data into long short-term memory network to predict the risk of pest and disease occurrence.

4. The intelligent early warning system for corn diseases and pests based on Web and mobile terminals according to claim 1, characterized in that, The dynamic updates to the early warning rule base include: every quarter, based on newly added pest and disease occurrence data, user feedback information, and revision opinions from agricultural experts, optimizing the confidence parameters of the rules through evidence reasoning algorithms to ensure the adaptability of the rules to actual planting scenarios.

5. The intelligent early warning system for corn diseases and pests based on Web and mobile terminals according to claim 1, characterized in that, The visualization services of the user interaction and self-learning module include displaying the distribution of pests and diseases, historical warnings, and environmental parameter trends on both web and mobile devices using heat maps, statistical charts, text, and images. The two-way interactive function allows users to upload field photos for instant diagnosis, obtain personalized prevention and control suggestions, and consult experts online.

6. A method for intelligent early warning of corn diseases and pests based on Web and mobile terminals, characterized in that, Includes the following steps: S1. Multi-source data acquisition: Environmental parameter data is collected once per minute through a sensor network, and received and preliminarily processed in real time through edge computing nodes; multispectral and visible light images of the field are acquired by drones based on Beidou positioning and pre-set routes three times a week; high-definition cameras are used to take images of corn plants at regular intervals, and combined with agricultural operation data and pest and disease symptom information manually reported by users to form a complete dataset; the collected data is transmitted to the cloud server through AES encryption algorithm. S2. Data standardization preprocessing: The cloud server cleans, converts, and removes outliers from the collected multi-source heterogeneous data. Image registration is performed using scale-invariant feature transformation. The data is transformed to the [0,1] interval through min-max or Z-score standardization to form a structured dataset suitable for model analysis. S3. Intelligent Analysis and Prediction: The preprocessed image data is input into an improved deep learning model. First, deep features are extracted by a CNN with a channel attention mechanism. Then, the YOLOv5 model outputs the types, severity, and feature parameters of pests and diseases. Environmental, growth cycle, and historical data are input into a hybrid machine learning algorithm. Key environmental factors are selected by random forest and then input into an LSTM network for training to predict the risk level of pests and diseases in the next 72 hours. S4. Multi-level early warning generation: Based on the analysis and prediction results, combined with the judgment logic of the early warning rule base, the corresponding level of early warning is triggered to generate early warning content that includes pest and disease information, risk level, personalized prevention and control suggestions and contact information of responsible personnel. S5. Information Push and Interaction: Accurately push early warning information to users through three methods: web, mobile SMS and application push, and provide visual query services and decision support; S6. Closed-loop optimization: Collect user feedback on prevention and control effects and field validation data, and adjust model parameters and early warning thresholds based on reinforcement learning algorithms; complete a model iteration every half month to achieve adaptive optimization of the system.

Images