A plant monitoring-based pest warning method and system

Abstract

The application discloses a plant monitoring-based disease and pest warning method and system, and the method comprises the following steps: S1, based on the plant varieties, planting density, growth cycle and historical disease and pest atlas of the monitoring area, a digital twin virtual farm model is constructed, and the disease and pest susceptible areas and transmission paths at different growth stages are simulated, so that the application is deeply integrated with the plant growth scene through the digital twin model, the high-risk areas can be predicted in advance without waiting for visible damage caused by diseases and pests, and the early warning lag problem is effectively solved; meanwhile, rare disease and pest synthetic samples are generated by using a generative adversarial network, and advanced technologies such as a channel attention mechanism are used to make up for the shortage of real samples, and reduce the false negative and false positive probabilities; in addition, the spatiotemporal graph neural network is combined with the physical transmission model in a fusion mode, and combined with multi-source data such as plant physiological indexes, the disease and pest diffusion and infection conditions can be dynamically simulated, and the scientificity and pertinence of risk assessment are improved.

Description

Technical Field

[0001] This application relates to the technical field of plant monitoring, and in particular to a method and system for early warning of pests and diseases based on plant monitoring. Background Technology

[0002] During plant growth, diseases and pests are among the core factors affecting plant growth, survival, and quality. Timely and accurate early warning and control of plant diseases and pests are crucial to ensuring healthy plant growth and reducing growth losses. Currently, plant disease and pest early warning technology has gradually developed from traditional manual monitoring towards intelligent and automated methods, but many shortcomings still exist, making it difficult to meet the refined monitoring needs of different plant varieties and growth scenarios.

[0003] First, existing early warning methods mostly rely on visible signs that appear after plants are infected by pests and diseases, such as leaf spots, stem holes, and fruit deformities, to identify and warn. The warning is only triggered after the pests and diseases have broken out and caused significant damage to the plants, resulting in a delay in control and making it impossible to effectively curb the spread of pests and diseases, which can easily cause large-scale plant damage and yield reduction.

[0004] Meanwhile, existing plant disease and pest identification models mostly rely on training with real samples. However, due to the low frequency of occurrence and scarcity of samples of rare diseases and pests of different plants, the models cannot fully learn the characteristic damage they cause to specific plants, making it difficult to achieve effective identification and prone to missed or false alarms. In addition, existing early warning risk assessment methods mostly use a single model, which cannot dynamically simulate the spread path and infection probability of diseases and pests in different plant varieties and different growth stages, resulting in assessment results that lack specificity and scientific rigor.

[0005] Application content

[0006] This application aims to address, at least to some extent, the technical problems in the related art.

[0007] To achieve the above objectives, this application proposes a method for early warning of pests and diseases based on plant monitoring, comprising the following steps:

[0008] S1. Based on the plant varieties, planting density, growth cycle and historical pest and disease maps of the monitoring area, construct a digital twin virtual farm model to simulate the susceptible areas and transmission paths of pests and diseases at different growth stages.

[0009] S2. Deploy heterogeneous intelligent sensing terminals to simultaneously collect multi-dimensional data such as leaf texture, canopy temperature, insect vibration sound patterns, and microenvironment parameters. Use a preset model to filter out invalid images in real time, and use lightweight blockchain technology to generate timestamp hashes for key data storage.

[0010] S3. Fine-grained feature extraction is performed by combining a preset model with a channel attention mechanism. Transfer learning and contrastive learning techniques are integrated to construct a dynamic pest and disease feature library, generate synthetic samples of rare pests and diseases, and automatically trigger a manual review process for low-confidence results.

[0011] S4. Construct a risk assessment engine that integrates spatiotemporal graph neural network and physical propagation model, inputting real-time identification results, plant physiological indices, micrometeorological data and weather forecasts, and dynamically simulating the spread path and infection probability of pests and diseases.

[0012] S5. Based on user roles, intelligently generate differentiated early warnings. Driven by agricultural knowledge graphs, generate prevention and control suggestions and link them to the local agricultural input supply chain. Users upload images of prevention and control effects, and the system automatically evaluates the effectiveness of the suggestions and dynamically optimizes the early warning threshold and recommendation strategy.

[0013] In addition, the application may also include the following additional technical features:

[0014] Specifically, in step S1, the historical pest and disease occurrence map is constructed by integrating plant protection records, remote sensing monitoring reports and farmer-reported data from at least three previous growing seasons in the region, and is associated with specific geographical locations, occurrence times and climatic conditions.

[0015] Specifically, the key data in step S2 includes at least image data of the first identification of a specific pest or disease, environmental parameter data that has reached the warning threshold, and result data of manual verification.

[0016] Specifically, in step S3, rare pest and disease synthetic samples are generated through generative adversarial networks. The generated samples combine the morphological, texture and color features of the pest and disease and have a similarity of no less than 90% with the real samples. After manual verification, the synthetic samples are included in the dynamic pest and disease feature database.

[0017] Specifically, the plant physiological indices in step S4 include, but are not limited to, the normalized vegetation index, the photochemical vegetation index, and the water stress index; the physical propagation model considers wind direction, wind speed, turbulence intensity, and spore survival rate for airborne diseases, and considers irrigation water flow path and soil type for soil-borne diseases.

[0018] Specifically, in step S5, the user roles include at least large-scale farm owners, cooperative technicians, individual farmers, and agricultural regulatory personnel; the level of detail of the warning information, the frequency of push notifications, the technical complexity of the suggestions, and the related agricultural input supply chain levels are differentiated and adapted for different roles.

[0019] A plant-based pest and disease early warning system includes the following modules:

[0020] The digital twin modeling module constructs and maintains a digital twin virtual farm model based on input regional planting information and historical data.

[0021] The monitoring and evidence storage module integrates a heterogeneous intelligent sensing terminal network, edge computing unit, and blockchain evidence storage unit deployed in the field. It is responsible for multi-dimensional data collection, invalid data filtering, and trusted timestamp evidence storage of key data.

[0022] The intelligent recognition and learning module includes a convolutional neural network model with channel attention mechanism, a dynamic pest and disease feature library, a data augmentation unit, and a manual review interaction interface. It is responsible for fine-grained feature extraction, identification, sample library updates, and review management of low-confidence samples for pests and diseases.

[0023] The risk assessment module incorporates a spatiotemporal graph neural network engine and a pest and disease physical transmission model. It is used to integrate multi-source input data, perform spatiotemporal risk assessment and diffusion dynamic simulation, and generate risk heat maps.

[0024] The early warning and prevention module includes a user role management unit, an agricultural knowledge graph engine, an early warning information generation and push unit, an agricultural input supply chain interface, and a prevention and control effect feedback and optimization unit. It is responsible for generating differentiated early warnings and prevention and control suggestions, and realizing closed-loop optimization based on feedback.

[0025] In summary, the beneficial effects of the plant monitoring-based pest and disease early warning method and system proposed in this application are as follows: This application deeply integrates a digital twin model with the plant growth scenario, enabling early prediction of high-risk areas without waiting for visible damage from pests and diseases, effectively solving the problem of delayed early warning; simultaneously, it utilizes generative adversarial networks to generate synthetic samples of rare pests and diseases, combined with advanced technologies such as channel attention mechanisms, to compensate for the scarcity of real samples and reduce the probability of missed or false alarms; furthermore, by employing a fusion of spatiotemporal graph neural networks and physical propagation models, combined with multi-source data such as plant physiological indices, it can dynamically simulate the spread and infection of pests and diseases, improving the scientific rigor and relevance of risk assessment. Attached Figure Description

[0026] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0027] Figure 1 This is a flowchart of a method and system for early warning of pests and diseases based on plant monitoring, as described in this application.

[0028] Figure 2 This is a flowchart of step S3 of the method and system for early warning of pests and diseases based on plant monitoring in this application;

[0029] Figure 3This is a system flowchart of a method and system for early warning of diseases and pests based on plant monitoring, as proposed in this application. Detailed Implementation

[0030] To make the technical means, inventive features, objectives, and effects of this application easier to understand, the application is further described below with reference to specific illustrations. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0031] The present application will now be described in further detail with reference to the accompanying drawings.

[0032] like Figure 1 As shown in the figure, a method for early warning of pests and diseases based on plant monitoring according to an embodiment of this application includes the following steps:

[0033] S1. Based on the plant varieties, planting density, growth cycle and historical pest and disease maps of the monitoring area, construct a digital twin virtual farm model to simulate the susceptible areas and transmission paths of pests and diseases at different growth stages.

[0034] Specifically, based on the deep integration of digital twin technology and agricultural scenarios, the system replicates the physical characteristics and growth status of real farms by monitoring regional plant varieties, planting density, and growth cycles. At the same time, it integrates historical pest and disease maps, links plant growth patterns with the logic of pest and disease occurrence, and simulates pest and disease susceptibility areas formed by plant variety characteristics and planting layout differences at different growth stages through simulation algorithms. Combined with field environments such as airflow and irrigation, it deduces the transmission paths of pests and diseases under different conditions, and replicates the full-scenario logic of pest and disease occurrence and spread in advance.

[0035] This step enables proactive pest and disease early warning, allowing for the prediction of high-risk areas through virtual models without waiting for actual pest and disease occurrences. This provides precise guidance for subsequent monitoring deployment and prevention and control preparations. At the same time, it avoids the subjectivity of traditional manual prediction. The model, built based on standardized data, can repeatedly simulate the spread trends of pests and diseases in different scenarios, improving the scientific rigor of the early warning. In addition, the virtual model can simultaneously simulate the susceptibility characteristics of plants at different growth stages, adapting to the early warning needs of the entire plant growth cycle, making it more versatile.

[0036] S2. Deploy heterogeneous intelligent sensing terminals to simultaneously collect multi-dimensional data such as leaf texture, canopy temperature, insect vibration sound patterns, and microenvironment parameters. Use a preset model to filter out invalid images in real time, and use lightweight blockchain technology to generate timestamp hashes for key data storage.

[0037] Specifically, the deployment of intelligent sensing terminals (such as high-definition cameras, temperature sensors, voiceprint collectors, and environmental monitors) is based on the principle of multi-dimensional collaborative perception, covering plant physiological characteristics (leaf texture, canopy temperature), direct signs of pests and diseases (insect vibration voiceprints), and growth environment (microenvironmental parameters), avoiding identification bias caused by single-dimensional data; preset models filter out invalid images, lightweight blockchain technology is used for evidence storage, and timestamp hashes are generated for key data to ensure the authenticity, integrity, and timeliness of the data, prevent data from being tampered with or forged, and provide a basis for subsequent early warning, tracing, and responsibility determination.

[0038] Compared to traditional single monitoring, this multi-dimensional data collection step has higher identification accuracy and can detect early signs of pests and diseases in advance; invalid data screening can improve data processing efficiency, reduce computing power waste, and ensure the accuracy of subsequent analysis; lightweight blockchain notarization is adapted to the computing power resources of field terminals to realize key data notarization, while timestamp hashing can realize full-process data traceability and improve the credibility of early warning results.

[0039] S3. Fine-grained feature extraction is performed by combining a preset model with a channel attention mechanism. Transfer learning and contrastive learning techniques are integrated to construct a dynamic pest and disease feature library, generate synthetic samples of rare pests and diseases, and automatically trigger a manual review process for low-confidence results.

[0040] Specifically, the pre-set model combines a channel attention mechanism to focus on key features of pests and diseases (such as the subtle texture of leaf lesions and the specific morphology of insects) to achieve fine-grained feature extraction and avoid interference from irrelevant features (such as normal leaf patterns and dust spots). It also integrates transfer learning and contrastive learning techniques. Transfer learning can utilize the general features of the trained model to reduce the training cost in new scenarios, while contrastive learning can enhance the distinction between pest and disease features and normal plant features, thereby improving recognition accuracy.

[0041] The dynamic pest and disease feature database is continuously updated based on real-time data collection, ensuring the ability to identify novel pests and diseases. Synthetic samples of rare pests and diseases are generated through generative adversarial networks (GANs), which are then trained to simulate the morphological, texture, and color characteristics of rare pests and diseases, compensating for the lack of real samples. Low-confidence results automatically trigger manual review, reducing the false positive rate and ensuring the reliability of the identification results.

[0042] The channel attention mechanism in this step improves the accuracy of feature extraction, enabling the identification of early signs of minor pests and diseases. The fusion of transfer learning and contrastive learning reduces model training costs and improves model adaptability and recognition accuracy. The synthesis of dynamic feature libraries and rare samples expands the recognition range. The automatic triggering of the manual review process balances the efficiency of machine recognition with the accuracy of manual recognition, effectively reducing false alarms and false negatives and improving the reliability of the system.

[0043] S4. Construct a risk assessment engine that integrates spatiotemporal graph neural network and physical propagation model, inputting real-time identification results, plant physiological indices, micrometeorological data and weather forecasts, and dynamically simulating the spread path and infection probability of pests and diseases.

[0044] Specifically, the spatiotemporal graph neural network excels at processing spatiotemporally correlated data, capturing the temporal dynamics and spatial correlations of pest and disease spread. The physical propagation model, based on the actual propagation patterns of pests and diseases, combines plant physiological indices, micrometeorological data, and other factors to construct a propagation logic that conforms to real-world scenarios. The risk assessment engine, which integrates both, takes into account real-time identification results (pest and disease type and severity), plant physiological indices (reflecting plant resistance), micrometeorological data, and weather forecasts (key factors affecting pest and disease spread). Through dynamic calculations by the engine, it simulates the spread paths and infection probabilities of pests and diseases, transforming abstract propagation patterns into visualized risk assessment results.

[0045] The fusion of spatiotemporal neural networks and physical propagation models in this step closely matches the actual propagation scenarios of pests and diseases, resulting in more accurate and targeted risk assessments. Dynamically simulating diffusion paths and infection probabilities allows users to grasp the spread trends of pests and diseases in advance, buying time for prevention and control deployment and preventing large-scale outbreaks. The input of multi-source data ensures the comprehensiveness of the risk assessment, while the integration of weather forecasts enhances the foresight of the risk assessment.

[0046] S5. Based on user roles, intelligently generate differentiated early warnings. Driven by agricultural knowledge graphs, generate prevention and control suggestions and link them to the local agricultural input supply chain. Users upload images of prevention and control effects, and the system automatically evaluates the effectiveness of the suggestions and dynamically optimizes the early warning threshold and recommendation strategy.

[0047] Specifically, differentiated early warnings are generated based on user roles. By classifying user roles (such as farmers, technicians, and regulatory departments), the system analyzes the needs of different roles (such as farmers focusing on practical prevention and control, and regulatory departments focusing on macro risks) and generates appropriate early warning information. Agricultural knowledge graphs drive prevention and control suggestions. Based on the relationships between pests, diseases, control methods, plant varieties, and environmental conditions stored in the knowledge graph, scientific and appropriate prevention and control suggestions are generated. The system is linked to the local agricultural input supply chain to achieve integrated connection between early warning, prevention and control, and agricultural input supply. Users upload images of the prevention and control effects, and the system evaluates the effectiveness of the suggestions through image recognition algorithms. Based on the feedback results, the system dynamically optimizes the early warning threshold and recommendation strategy, forming a closed-loop logic of early warning, prevention and control, feedback, and optimization, ensuring that the system continuously adapts to actual prevention and control scenarios.

[0048] This differentiated early warning approach avoids information redundancy or insufficiency, improves user experience and the practicality of early warning information, and allows different roles to obtain the core content they need. The prevention and control suggestions driven by the agricultural knowledge graph are more targeted and scientific, which can improve the prevention and control effect. The connection with the local agricultural input supply chain realizes the seamless connection between early warning and prevention and control, improves the efficiency of prevention and control, and reduces the prevention and control cost for users. The closed-loop optimization logic allows the system to continuously iterate based on the actual prevention and control effect, continuously improve the accuracy of the early warning threshold and the adaptability of the recommendation strategy, and long-term use can significantly improve the practicality and reliability of the system.

[0049] In one embodiment of this application, the historical pest and disease occurrence map in step S1 is constructed by integrating plant protection records, remote sensing monitoring reports and farmer-reported data from at least three previous growing seasons in the region, and associated with specific geographical locations, occurrence times and climatic conditions.

[0050] It should be noted that the construction of the historical pest and disease occurrence map integrates plant protection records (monitoring data from professional plant protection agencies), remote sensing monitoring reports (large-scale macro-monitoring data), and data reported by farmers (actual field occurrence data) from at least three past growing seasons in the region. This ensures the comprehensiveness and diversity of the data, and links it to specific geographical locations, occurrence times, and climatic conditions. The occurrence of pests and diseases is highly correlated with these factors. Through correlation analysis, the patterns of pest and disease occurrence can be discovered, providing reliable historical data support for the construction of the digital twin virtual farm model and ensuring the accuracy of the model simulation.

[0051] The integration of data from at least three growing seasons avoids the randomness of data from a single growing season, ensures the representativeness of historical maps, and can reflect the high incidence patterns of pests and diseases in the region. The fusion of multi-source data takes into account professionalism, macro-level perspectives, and practicality. The correlation between geographical location, occurrence time, and climatic conditions allows the digital twin model to accurately simulate the susceptible areas and transmission paths of pests and diseases in different scenarios, improving the model's adaptability and simulation accuracy.

[0052] In one embodiment of this application, the key data in step S2 includes at least image data of the first identification of a specific pest or disease, environmental parameter data of reaching the warning threshold, and result data of manual verification.

[0053] It should be noted that the definition of key data focuses on the core nodes of early warning. Image data that identifies a specific pest or disease for the first time is evidence of the source of the pest or disease occurrence. It can serve as the initial trigger for early warning and is also key to tracing the source of the pest or disease occurrence. Environmental parameter data that reaches the early warning threshold is evidence of the inducing effect of the pest or disease outbreak. It can reflect the environmental conditions of the outbreak and provide a basis for risk assessment and prevention and control recommendations. The results data that are manually verified can verify the accuracy of the machine recognition results and are also key to subsequent system optimization and responsibility determination. By clarifying the scope of key data, the core value of the evidence data is ensured, providing reliable support for all aspects of the early warning process.

[0054] The scope of key data is clearly defined, balancing the reliability of evidence preservation with the rationality of resources. It avoids the preservation of irrelevant data, reduces the computing power and storage pressure on terminals, and the three types of key data cover the entire process from initial identification to environmental induction to final confirmation, forming a complete chain of evidence to ensure the traceability and verifiability of early warning results. The preservation of key data can provide a reliable basis for subsequent pest and disease tracing, system optimization, and control effect evaluation, while also enhancing the credibility and practicality of the system.

[0055] In one embodiment of this application, such as Figure 2 As shown, in step S3, rare pest and disease synthetic samples are generated through generative adversarial networks. The generated samples combine the morphological, texture and color features of the pest and disease and have a similarity of no less than 90% with the real samples. After manual verification, the synthetic samples are included in the dynamic pest and disease feature database.

[0056] It should be noted that the synthetic samples of rare pests and diseases are generated through generative adversarial networks. Through adversarial training, the realism of the generated samples is continuously optimized to ensure that the similarity between the synthetic samples and real samples is not less than 90%. The synthetic samples are combined with the core features of this type of pest and disease to ensure their effectiveness. They can be used for model training to improve the model's ability to identify rare pests and diseases. The synthetic samples are included in the dynamic pest and disease feature database after being manually reviewed and confirmed. This is to avoid deviations in the synthetic samples, ensure the accuracy of the feature database samples, and provide reliable support for subsequent identification work.

[0057] Generative adversarial networks (GANs) generate synthetic samples that effectively address the scarcity of real samples for rare pests and diseases, filling gaps in the feature library and enabling the model to be effectively trained on rare pests and diseases, thus improving the system's ability to identify them. A similarity requirement of at least 90% ensures the effectiveness of the synthetic samples, allowing them to be used for model training on par with real samples and avoiding identification biases caused by distortion in the synthetic samples. A manual review and confirmation process further guarantees the quality of the synthetic samples, ensuring the accuracy and reliability of the samples included in the feature library, thereby improving the accuracy and reliability of the model's recognition.

[0058] In one embodiment of this application, the plant physiological indices in step S4 include, but are not limited to, the normalized vegetation index, the photochemical vegetation index, and the water stress index; the physical propagation model considers wind direction, wind speed, turbulence intensity, and spore survival rate for airborne diseases, and considers irrigation water flow path and soil type for soil-borne diseases.

[0059] It should be noted that plant physiological indices can accurately reflect the growth status and resistance of plants (e.g., a high water stress index indicates weak plant resistance and susceptibility to pests and diseases). Incorporating these indices into risk assessment can improve the accuracy of the assessment results. The physical transmission model is designed differently for pests and diseases of different transmission types because the transmission patterns of different types of pests and diseases vary greatly (airborne diseases are greatly affected by wind direction and speed, while soil-borne diseases are greatly affected by irrigation water flow and soil type). Targeted consideration of key influencing factors can make the transmission simulation more closely resemble the actual scenario, ensuring the accuracy of the simulation of diffusion paths and infection probabilities.

[0060] The inclusion of plant physiological indices allows risk assessment to go beyond just focusing on pests and diseases themselves, while also taking into account plant resistance, resulting in more comprehensive and accurate assessments and enabling differentiated risk assessments. The differentiated design of the physical transmission model adapts to the transmission patterns of different types of pests and diseases, and the simulation of diffusion paths and infection probabilities is more scientific and realistic, providing precise guidance for the prevention and control deployment of different types of pests and diseases.

[0061] In one embodiment of this application, the user roles in step S5 include at least large-scale farm owners, cooperative technicians, individual farmers, and agricultural regulatory personnel; the level of detail of the warning information, the frequency of push notifications, the technical complexity of the suggestions, and the associated agricultural input supply chain level are differentiated and adapted for different roles.

[0062] It should be noted that the user roles (large-scale farm owners, cooperative technicians, individual farmers, and agricultural regulatory personnel) are precisely tailored to the core needs and usage scenarios of different users. Large-scale farm owners focus on the overall regional prevention and control deployment and efficiency, cooperative technicians focus on professional prevention and control techniques and guidance, individual farmers focus on simple and easy-to-understand practical suggestions, and agricultural regulatory departments focus on the overall regional pest and disease risk and the implementation of prevention and control measures. For different roles, differentiated adaptations are made based on the level of detail in the early warning information (e.g., regulatory departments need a macro overview, farmers need simple practical instructions), the frequency of push notifications (e.g., farm owners need high-frequency real-time early warnings, regulatory departments need to periodically summarize early warnings), the technical complexity of the suggestions (e.g., technicians need professional technical parameters, farmers need simple steps), and the level of the agricultural input supply chain (e.g., large-scale farm owners need bulk agricultural input supply, individual farmers need retail agricultural input), ensuring that each user can obtain content that fits their own needs.

[0063] like Figure 3As shown in the figure, a pest and disease early warning system based on plant monitoring according to an embodiment of this application includes the following modules:

[0064] The digital twin modeling module constructs and maintains a digital twin virtual farm model based on input regional planting information and historical data.

[0065] Specifically, based on digital twin technology, the system replicates the physical form and growing environment of a real farm using input regional planting information (plant varieties, planting density, growth cycle, etc.) and integrates historical data (historical pest and disease maps, etc.). Through simulation algorithms, a virtual model synchronized with the real farm is constructed. The module continuously maintains the model to ensure that the virtual model is consistent with the growth status and environmental conditions of the actual farm, providing virtual simulation support for subsequent monitoring, identification, and risk assessment. Essentially, it achieves a two-way mapping between the real farm and the virtual farm.

[0066] The monitoring and evidence storage module integrates a heterogeneous intelligent sensing terminal network, edge computing unit, and blockchain evidence storage unit deployed in the field. It is responsible for multi-dimensional data collection, invalid data filtering, and trusted timestamp evidence storage of key data.

[0067] Specifically, an integrated heterogeneous intelligent sensing terminal network enables the synchronous collection of multi-dimensional data, covering core dimensions such as plants, pests and diseases, and the environment; edge computing units are deployed in field terminals to process data on-site, reducing data transmission pressure and improving data processing efficiency; and a blockchain evidence storage unit uses lightweight blockchain technology to generate timestamp hashes for key data, ensuring the credibility, integrity, and traceability of the data. The three work together to provide reliable data support for subsequent identification, verification, and traceability.

[0068] The intelligent recognition and learning module includes a convolutional neural network model with channel attention mechanism, a dynamic pest and disease feature library, a data augmentation unit, and a manual review interaction interface. It is responsible for fine-grained feature extraction, identification, sample library updates, and review management of low-confidence samples for pests and diseases.

[0069] It should be noted that the module is equipped with a channel attention mechanism convolutional neural network, which is responsible for fine-grained feature extraction from the collected multi-dimensional data, focusing on key features of pests and diseases; the dynamic pest and disease feature library continuously receives real-time data and synthetic samples to achieve self-update; the data augmentation unit generates rare samples through generative adversarial networks to make up for the lack of real samples; the manual review interaction interface provides a manual verification channel for low-confidence results to ensure the accuracy of the recognition results. The entire module improves the recognition accuracy and adaptability of the system through the logic of recognition-learning-update-review.

[0070] The risk assessment module incorporates a spatiotemporal graph neural network engine and a pest and disease physical transmission model. It is used to integrate multi-source input data, perform spatiotemporal risk assessment and diffusion dynamic simulation, and generate risk heat maps.

[0071] Specifically, the module's built-in spatiotemporal graph neural network engine processes spatiotemporal correlation data on pest and disease spread, capturing the propagation patterns in both time and space dimensions; the pest and disease physical propagation model combines different pest and disease propagation types to adapt to actual propagation scenarios; the risk assessment engine formed by the fusion of these two elements takes multi-source data as input, calculates through algorithms, dynamically simulates pest and disease spread paths and infection probabilities, generates risk heatmaps, transforms abstract risks into visualized results, and provides users with intuitive risk references.

[0072] The early warning and prevention module includes a user role management unit, an agricultural knowledge graph engine, an early warning information generation and push unit, an agricultural input supply chain interface, and a prevention and control effect feedback and optimization unit. It is responsible for generating differentiated early warnings and prevention and control suggestions, and realizing closed-loop optimization based on feedback.

[0073] It should be noted that the user role management unit is responsible for classifying user roles and analyzing the needs of different roles; the agricultural knowledge graph engine calls on the built-in agricultural knowledge to generate appropriate prevention and control suggestions; the early warning information generation and push unit generates differentiated early warnings and pushes them accurately according to role needs; the agricultural input supply chain interface connects to local agricultural input resources; and the prevention and control effect feedback and optimization unit receives prevention and control effect images uploaded by users, evaluates the effectiveness of suggestions through image recognition, and then dynamically optimizes the early warning threshold and recommendation strategy to form a closed-loop optimization logic.

[0074] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0075] The present application and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present application. The actual structure is not limited to this. In conclusion, if a person skilled in the art is inspired by this description and designs a similar structure and embodiment without departing from the spirit of the present application, such design should fall within the protection scope of the present application.

Claims

1. A method for early warning of pests and diseases based on plant monitoring, characterized in that, Includes the following steps: S1. Based on the plant varieties, planting density, growth cycle and historical pest and disease maps of the monitoring area, construct a digital twin virtual farm model to simulate the susceptible areas and transmission paths of pests and diseases at different growth stages. S2. Deploy heterogeneous intelligent sensing terminals to simultaneously collect multi-dimensional data such as leaf texture, canopy temperature, insect vibration sound patterns, and microenvironment parameters. Use a preset model to filter out invalid images in real time and generate timestamp hashes for key data through lightweight blockchain technology. S3. Fine-grained feature extraction is performed by combining a preset model with a channel attention mechanism. Transfer learning and contrastive learning techniques are integrated to construct a dynamic pest and disease feature library, generate synthetic samples of rare pests and diseases, and automatically trigger a manual review process for low-confidence results. S4. Construct a risk assessment engine that integrates spatiotemporal graph neural network and physical propagation model, inputting real-time identification results, plant physiological indices, micrometeorological data and weather forecasts, and dynamically simulating the spread path and infection probability of pests and diseases. S5. Based on user roles, intelligently generate differentiated early warnings. Driven by agricultural knowledge graphs, generate prevention and control suggestions and link them to the local agricultural input supply chain. Users upload images of prevention and control effects, and the system automatically evaluates the effectiveness of the suggestions and dynamically optimizes the early warning threshold and recommendation strategy.

2. The method for early warning of pests and diseases based on plant monitoring according to claim 1, characterized in that, In step S1, the historical pest and disease occurrence map is constructed by integrating plant protection records, remote sensing monitoring reports and farmer-reported data from at least three previous growing seasons in the region, and is associated with specific geographical locations, occurrence times and climatic conditions.

3. The method for early warning of pests and diseases based on plant monitoring according to claim 1, characterized in that, The key data in step S2 includes at least image data of the first identification of a specific pest or disease, environmental parameter data that has reached the warning threshold, and result data of manual verification.

4. The method for early warning of pests and diseases based on plant monitoring according to claim 1, characterized in that, In step S3, rare pest and disease synthetic samples are generated through generative adversarial networks. The generated samples combine the morphological, texture and color features of the pest and disease and have a similarity of no less than 90% with the real samples. After manual verification, the synthetic samples are included in the dynamic pest and disease feature database.

5. The method for early warning of pests and diseases based on plant monitoring according to claim 1, characterized in that, The plant physiological indices in step S4 include, but are not limited to, the normalized vegetation index, the photochemical vegetation index, and the water stress index; the physical propagation model considers wind direction, wind speed, turbulence intensity, and spore survival rate for airborne diseases, and irrigation water flow path and soil type for soil-borne diseases.

6. The method for early warning of pests and diseases based on plant monitoring according to claim 1, characterized in that, In step S5, the user roles include at least large-scale farm owners, cooperative technicians, individual farmers, and agricultural regulatory personnel; the level of detail of the early warning information, the frequency of push notifications, the technical complexity of the suggestions, and the associated agricultural input supply chain levels are differentiated and adapted for different roles.

7. A pest and disease early warning system based on plant monitoring, characterized in that, Includes the following modules: The digital twin modeling module constructs and maintains a digital twin virtual farm model based on input regional planting information and historical data. The monitoring and evidence storage module integrates a heterogeneous intelligent sensing terminal network, edge computing unit and blockchain evidence storage unit deployed in the field, and is responsible for multi-dimensional data collection, invalid data filtering and trusted timestamp evidence storage of key data. The intelligent recognition learning module includes a convolutional neural network model with channel attention mechanism, a dynamic pest and disease feature library, a data augmentation unit, and a manual review interaction interface. It is responsible for fine-grained feature extraction, identification, sample library updates, and review management of low-confidence samples for pests and diseases. The risk assessment module has a built-in spatiotemporal graph neural network engine and a pest and disease physical transmission model. It is used to integrate multi-source input data, perform spatiotemporal risk assessment and diffusion dynamic simulation, and generate risk heat maps. The early warning and prevention module includes a user role management unit, an agricultural knowledge graph engine, an early warning information generation and push unit, an agricultural input supply chain interface, and a prevention and control effect feedback and optimization unit. It is responsible for generating differentiated early warnings and prevention and control suggestions, and realizing closed-loop optimization based on feedback.

Images