A rice blast monitoring and diagnosis method based on a tethered hot air balloon and drone system

Abstract

The present invention belongs to the technical field of agricultural disease monitoring and prevention and control, and specifically relates to a rice blast monitoring and diagnosis method based on a tethered hot air balloon and drone system. By constructing a dynamic observation platform of a tethered hot air balloon + drone group, multi-variety, multi-phase, and multi-dimensional multi- / hyperspectral data of rice blast diseased plants and multivariate information such as geography, soil, climate, and field management are simultaneously acquired to construct a spectral database. Utilizing self-supervised learning and deep learning techniques, an early diagnosis inversion model is constructed based on the spectral characteristics of diseased rice plants and the data characteristics of drones and tethered hot air balloons. This method gives full play to the synergistic advantages of tethered hot air balloons and drones, realizes comprehensive data collection and deep integration, improves the accuracy and timeliness of rice blast monitoring and diagnosis, provides strong technical support for the healthy development of the rice industry, and contributes to the intelligentization and sustainable development of agriculture.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural disease monitoring and prevention, and in particular to a rice blast monitoring and diagnosis method based on a tethered hot air balloon and an unmanned aerial vehicle system. Background Art

[0002] The statements in this section merely provide background technology related to the present invention and do not necessarily constitute prior art.

[0003] As one of the world's major staple crops, rice plays a crucial role in global food security. my country, a major rice producer and consumer, cultivates rice on a vast acreage and produces substantial yields. However, the rice industry faces numerous challenges, among which rice blast is a key factor severely impacting rice yield and quality. This disease significantly reduces rice production annually, posing a serious threat to food security.

[0004] While some progress has been made in rice disease control, many challenges remain. Existing monitoring methods are relatively limited, lacking effective technologies and platforms for coordinated spatiotemporal observation of rice diseases. This results in a lack of sufficient information support for rice disease detection and control management. Traditional monitoring methods struggle to comprehensively and in real time capture the occurrence and development of rice blast in different regions and at different growth stages. Regarding remote sensing data integration, existing data collection systems suffer from silos, imperfect data fusion mechanisms, and easily broken information chains, severely hindering their in-depth utilization. This makes it challenging to extract valuable information from massive amounts of remote sensing data for accurate rice blast diagnosis. In the area of ​​diagnosis and prediction, historical data suffers from trade-offs in spectral, spatial, and temporal resolution, and existing models rely heavily on big data combined with manual training, making it difficult to achieve fully automated, high-precision disease diagnosis and trend prediction. This results in the inability to accurately predict rice blast outbreak trends in real-time, making it difficult to implement effective prevention and control measures in advance.

[0005] Furthermore, while drone technology has been applied in agriculture, it still faces limitations in payload and endurance, and its operating radius and efficiency need to be improved. Tethered hot air balloons are relatively unused in agricultural remote sensing, and exploration of their collaboration with drones for rice disease monitoring is still in its infancy. Furthermore, while various research teams in my country have their own strengths in rice blast research, a lack of effective collaboration and data sharing hinders the development of a comprehensive and accurate rice blast diagnostic model. An innovative, comprehensive rice blast monitoring and diagnostic technology is urgently needed to enhance the rice industry's information technology capabilities and disease control capabilities. Summary of the Invention

[0006] To overcome the many shortcomings of current rice blast monitoring and diagnosis technologies, the present invention provides a method for monitoring and diagnosing rice blast based on a tethered hot air balloon and drone system. This method constructs a dynamic observation platform based on a tethered hot air balloon and drone to collect multi- / hyperspectral information of rice blast-infected plants from multiple varieties, multiple phases, and multiple dimensions. It also integrates multivariate data such as geography, soil, climate, and field management, and utilizes a self-supervised learning model combined with deep learning technology to construct an inversion model for early diagnosis of rice blast. This invention fully leverages the synergistic advantages of the tethered hot air balloon and drone system, effectively improving the accuracy and timeliness of rice blast monitoring and diagnosis. It provides a new technical approach for the precise prevention and control of rice blast and the guarantee of rice yield and quality, thereby promoting the intelligent and sustainable development of the rice industry.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] The first aspect of the present invention provides a method for monitoring and diagnosing rice blast based on a tethered hot air balloon and an unmanned aerial vehicle system:

[0009] Build a dynamic observation platform based on tethered hot air balloons and drone swarms, determine the coordination framework of the tethered hot air balloon and drone systems and the sensor equipment they carry, install mobile signal base stations, and build a multi-channel data interaction method to achieve fast and stable communication connections and data transmission between operating systems, ensuring that drones and tethered hot air balloons can obtain multi- / hyperspectral data in the near- and low-altitude composite airspace simultaneously;

[0010] Based on a dynamic observation platform, we collected multi- and hyperspectral information on rice blast-infected plants from multiple varieties, multiple temporal phases, and multiple dimensions. We integrated geographic, soil, climate, and field management data, optimized data preprocessing processes, including denoising, radiometric correction, and image enhancement, developed data fusion tools, enabled the complementarity of drone and tethered hot air balloon data, extracted universal key bands and disease indices sensitive to rice blast, and constructed a spectral database for rice blast disease across multiple varieties.

[0011] Based on the spectral characteristics of diseased rice plants and the data features of drones and tethered hot air balloons, a self-supervised learning model combined with deep learning technology is used to enhance the ability to identify the hallmark characteristics of rice blast. The continuous data stream provided by drones and tethered hot air balloons is used to enhance the dynamic change tracking ability of the rice blast monitoring model and construct an early diagnosis inversion model for rice blast.

[0012] As a further limitation of the first aspect of the present invention, when constructing the dynamic observation platform and the internal data transmission channel, the following details are added:

[0013] Comprehensively investigate and analyze the needs for rice blast monitoring and rice growth characteristics, identify the key rice growth stages (tillering and heading), common types of rice blast (leaf blast and panicle blast), disease causes (climatic conditions, soil fertility, and variety resistance), and key indicators of disease development (lesion area and disease index) to ensure that the platform system design closely matches actual agricultural needs.

[0014] The flight altitude of the near-low-altitude unmanned system is precisely determined based on factors such as the experimental rice field area, topography, expected experimental results, and local meteorological conditions. The drone's flight altitude is set between 30 and 60 meters, and can be flexibly adjusted according to different monitoring stages and rice varieties. It is equipped with a high-resolution multispectral camera and high-precision temperature and humidity sensors to quickly collect detailed rice field crop information.

[0015] The KX55 series tethered hot air balloons from the Chinese Academy of Sciences' Institute of Space Information Innovation are used as continuous monitoring equipment for rice fields. They fly at an altitude of 2,000-3,000 meters, stay aloft for 7-10 days, and cover a radius of 160-200 kilometers. They are equipped with wind direction and speed sensors, atmospheric pressure sensors, and high-resolution hyperspectral sensors.

[0016] The tethered hot air balloon uses a specially designed high-strength, lightweight mounting structure, with a high-performance mobile signal base station securely installed in its gondola or specific bearing area. This base station has powerful signal transmission and reception capabilities and uses multi-band antenna technology, enabling stable communication with different types of drones and ground control terminals. Its transmission power can be dynamically adjusted to meet communication needs at different distances and in different environments. The base station also has a built-in high-performance signal processor that can perform real-time encoding, decoding, and error correction on transmitted signals, effectively reducing the bit error rate during signal transmission and ensuring the accuracy of data transmission between it and the drone.

[0017] The Paxos protocol is fully implemented in the multi-channel data exchange process, assigning a unique identifier to each communication channel and establishing a distributed node status management mechanism. When a drone sends a data request or status update to the tethered hot air balloon base station, the base station first parses the request, extracting key information (request type, source node identifier, data content, etc.). It then initiates a multi-node voting process based on the Paxos protocol to ensure consensus on the same operation across multiple communication channels. For example, when a drone uploads rice blast monitoring data, the base station coordinates the various channels using the Paxos protocol, ensuring that all relevant nodes receive, store, and process the data in the same order, avoiding data conflicts and inconsistencies. This strict consistency guarantee ensures highly reliable and stable multi-channel interaction, effectively preventing monitoring interruptions or misdiagnoses caused by communication failures or data errors.

[0018] As a further limitation of the first aspect of the present invention, when processing and analyzing a variety of remote sensing data and ground monitoring data, the following details are added:

[0019] Based on the diversity of rice varieties and the complexity of rice blast, multiple sampling time points were set throughout the growth cycle of different rice varieties (common varieties such as indica rice, japonica rice, and glutinous rice, as well as local specialty varieties). These sampling time points covered key stages of rice from seeding, tillering, jointing, booting, heading, and maturity, ensuring the collection of multi-temporal data. In the spatial dimension, different monitoring areas were divided according to the layout and terrain characteristics of the rice fields. Multiple sampling points were set up in each area to form a multi-dimensional monitoring network. Utilizing the advantage of UAVs' flexible low-altitude flight, close-range and detailed multi- / hyperspectral imagery was collected for rice varieties in different regions, obtaining spectral information of diseased plants at different growth stages and different parts (leaves, stems, ears, etc.). Tethered hot air balloons conducted macroscopic monitoring of the entire rice field at high altitude, obtaining comprehensive information on rice growth conditions and disease distribution over a large area. The two systems worked together to achieve comprehensive and multi-level data collection.

[0020] A unified data integration framework was established to correlate and integrate collected geographic information, soil data, climate data, and field management data with the multi- and hyperspectral information of rice blast-infected plants. To address the noise problem in multi- and hyperspectral images, an adaptive filtering denoising algorithm was used to automatically adjust the filtering parameters based on the local characteristics of the image to effectively remove noise interference. A radiation correction method based on physical models was used, combined with field-measured atmospheric parameters (aerosol optical depth and water vapor content), to accurately correct the radiation values ​​received by the sensor and restore the true spectral reflectance characteristics of rice. For image enhancement, a combination of nonlinear stretching and histogram equalization was used to enhance the contrast between the characteristics of the diseased plants and the background, highlight the details of the disease, improve data quality, and provide a reliable foundation for subsequent analysis.

[0021] Based on the principles of multi-scale analysis and feature fusion, the wavelet transform is used to decompose the image into sub-bands of different scales and directions. The high-resolution detail information collected by the UAV and the macroscopic information obtained by the tethered hot air balloon are layered and fused, preserving the global features and local details of the image. The gray-level co-occurrence matrix is ​​extracted from the UAV image to calculate the texture parameters and spectral curve shape characteristics, and the vegetation index and color characteristics of the tethered hot air balloon image are extracted. The principal component analysis (PCA) method is used for feature fusion to construct a comprehensive feature vector, fully leveraging the advantages of the UAV and tethered hot air balloon data.

[0022] By comparing and analyzing the reflectance differences between healthy rice plants and diseased plants with different disease severity in various spectral bands, and combining statistical methods such as correlation analysis and stepwise discriminant analysis, bands with significant differences were screened out. Through verification and optimization of a large amount of sample data, the disease index that can most effectively characterize the rice blast condition was determined. Based on this, a MySQL database was established to store rice variety information, spectral data, geographical soil and climate field management data, disease information, and extracted key bands and disease index information.

[0023] As a further limitation of the first aspect of the present invention, when constructing the rice blast disease monitoring model, the following details are added:

[0024] Representative spectral data of diseased rice plants and their corresponding spectral data of healthy plants were selected from the MySQL database as the basic data set for model construction. The diseased plant data were finely annotated, and detailed information such as the location, shape, size, and severity of the lesions were recorded. Combined with the comprehensive feature vector obtained in S23, these data were used as the input parameters of the model.

[0025] A deep learning model architecture based on self-supervised learning is constructed using an encoder-decoder structure. The encoder consists of multiple convolutional and pooling layers, which encode the input multimodal feature vectors, gradually extracting abstract features from the data and reducing the data dimension. The decoder uses upsampling and deconvolution layers to map the encoded features back to the original data space, reconstructing the data. Skip connections are introduced between the encoder and decoder to preserve feature information at different levels and enhance the model's ability to capture detailed features.

[0026] The model uses a self-supervised learning strategy for model training. It uses the inherent structural information of the data to design pre-training tasks. By randomly cropping, rotating, flipping, and other transformations on the input image, the model predicts the relationship between the transformed image and the original image, thereby learning the invariant characteristics of the data. During the pre-training phase, a large amount of unlabeled data is used for training, enabling the model to automatically discover potential patterns and feature representations in the data.

[0027] After pre-training, the model was fine-tuned using labeled data to optimize model parameters for the rice blast diagnosis task. During fine-tuning, the cross-entropy loss function was used as the model's optimization objective to measure the difference between the model's predictions and the true labels. Simultaneously, the stochastic gradient descent (SGD) algorithm was combined to dynamically adjust the learning rate based on the data distribution to accelerate model convergence and improve its generalization ability. The model was evaluated using a validation set to monitor its performance in metrics such as accuracy, recall, and F1 value. Training was stopped when model performance stopped improving to prevent overfitting.

[0028] Establish a real-time data stream processing system to receive new data continuously collected by drones and tethered hot air balloons. Perform real-time preprocessing on the new data, including data cleaning, format conversion, and feature extraction similar to the training phase. The extracted dynamic features are then input into the trained model for real-time rice blast monitoring and diagnosis.

[0029] To adapt to the dynamic changes in rice growth and the evolution of rice blast, a dynamic feature update mechanism was designed. Over time and as data accumulates, the model is dynamically updated using an incremental learning method. This allows the model to continuously learn new data based on existing knowledge without having to retrain the entire model, thus avoiding the "catastrophic forgetting" problem. When new data accumulates to a certain level, the model is fine-tuned using this new data and some of its parameters are updated, allowing the model to promptly adapt to new changes and trends in rice blast.

[0030] During the model updating process, a reward mechanism is introduced in combination with reinforcement learning ideas. Corresponding rewards are given according to the model's performance in actual monitoring (accurately predicting the outbreak time of the disease and accurately judging the severity of the disease). This encourages the model to continuously optimize its own parameters and improve monitoring accuracy and reliability. At the same time, a model performance evaluation index system is established. In addition to traditional indicators such as accuracy and recall rate, the timeliness and stability factors of the model are additionally considered to comprehensively evaluate the model's performance. Targeted optimization and improvement are carried out based on the evaluation results to ensure that the constructed rice blast early diagnosis inversion model always maintains a good performance state and provides timely and effective disease warning and prevention and control decision support for rice production.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. This invention innovatively combines a tethered hot air balloon with an unmanned aerial vehicle (UAV) system, enabling the simultaneous acquisition of multi- and hyperspectral data in near- and low-altitude composite airspaces. This greatly expands the monitoring range and data collection dimensions. The tethered hot air balloon, by remaining at high altitude for extended periods, provides a large-scale macroscopic monitoring field of view, while the UAV enables detailed detection of key areas at low altitude. The collaborative operation of the two effectively overcomes the limitations of a single monitoring platform.

[0033] 2. By comprehensively integrating multi-source data such as geography, soil, climate, and field management, and deeply integrating it with the multi- / hyperspectral information of rice blast plants, a rice blast spectral database for multiple rice varieties was constructed, covering all-round information on the rice growth environment and disease characteristics, forming a comprehensive data resource library, which provides solid data support for subsequent in-depth analysis of the pathogenesis, transmission patterns, and precise prevention and control of rice blast.

[0034] 3. Based on the spectral characteristics of diseased rice plants and the data characteristics of drones and tethered hot air balloons, combined with the continuous data streams provided by drones and tethered hot air balloons, an early diagnosis inversion model for rice blast is constructed using a self-supervised learning model combined with deep learning technology. It has powerful feature learning and pattern recognition capabilities, and realizes real-time tracking of the dynamic changes of rice blast.

[0035] 4. The monitoring and diagnosis system constructed by the present invention forms a complete technical process. From the establishment of a dynamic observation platform, data collection and integration, model construction and optimization to the final real-time monitoring and early warning, each link is closely connected and mutually supportive.

[0036] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0038] Figure 1 A technical roadmap for the rice blast monitoring and diagnosis method based on a tethered hot air balloon and drone system provided by the present invention; DETAILED DESCRIPTION

[0039] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0040] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0041] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0042] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0043] The technical solutions of the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0044] A method for monitoring and diagnosing rice blast based on a tethered hot air balloon and an unmanned aerial vehicle system comprises the following steps:

[0045] S10. Build a dynamic observation platform based on a tethered hot air balloon + drone swarm, determine the coordination framework of the tethered hot air balloon and drone system and the sensor equipment carried, equip it with a mobile signal base station, build a multi-channel data interaction method, achieve fast and stable communication connection and data transmission between operating systems, and ensure that drones and tethered hot air balloons obtain synchronous multi- / hyperspectral data in the near-low altitude composite airspace;

[0046] S20. Based on a dynamic observation platform, collect multi- / hyperspectral information of rice blast-infected plants from multiple varieties, multiple temporal phases, and multiple dimensions. Integrate geographic, soil, climate, and field management data. Optimize data preprocessing processes, including denoising, radiometric correction, and image enhancement. Develop data fusion tools to complement drone and tethered hot air balloon data. Extract universal key bands and disease indices sensitive to rice blast-infected plants. Construct a multi-variety rice blast spectral database.

[0047] S30. Based on the spectral characteristics of diseased rice plants and the data features of drones and tethered hot air balloons, a self-supervised learning model combined with deep learning technology is used to enhance the ability to identify the hallmark features of rice blast. The continuous data streams provided by drones and tethered hot air balloons are used to enhance the dynamic change tracking capability of the rice blast monitoring model and to construct an early diagnosis inversion model for rice blast.

[0048] Preferably, S10 further includes the following steps:

[0049] S11. Comprehensively investigate and analyze the needs for rice blast monitoring and rice growth characteristics, clarifying the key rice growth stages (tillering stage, heading stage) that need to be monitored, common types of rice blast (leaf blast, panicle blast), disease causes (climatic conditions, soil fertility, variety resistance), and key indicators of disease development (lesion area, disease index), to ensure that the platform system design closely matches the actual needs of agriculture;

[0050] S12. Based on factors such as the experimental rice field area, topography, expected experimental results, and local meteorological conditions, the flight altitude of the near-low-altitude unmanned system is accurately determined. The flight altitude of the drone is set between 30 and 60 meters, and the altitude is flexibly adjusted according to different monitoring stages and rice varieties. The drone is equipped with a high-resolution multispectral camera and a high-precision temperature and humidity sensor to quickly collect detailed rice field crop information.

[0051] S13. Use the KX55 series tethered hot air balloon from the Chinese Academy of Sciences' Institute of Space Information Innovation as a continuous monitoring device for rice fields. The balloon will fly at an altitude of 2,000-3,000 meters, stay aloft for 7-10 days, and cover a radius of 160-200 kilometers. The balloon will be equipped with wind direction and speed sensors, atmospheric pressure sensors, and high-resolution hyperspectral sensors.

[0052] S14. The tethered hot air balloon uses a specially designed high-strength, lightweight mounting structure to securely mount a high-performance mobile signal base station on its gondola or specific load-bearing area. The base station has powerful signal transmission and reception capabilities and uses multi-band antenna technology to enable stable communication with different types of drones and ground control terminals. Its transmission power can be dynamically adjusted to meet communication requirements at different distances and in different environments. The base station also has a built-in high-performance signal processor that can perform real-time encoding, decoding, and error correction on transmitted signals, effectively reducing the bit error rate during signal transmission and ensuring the accuracy of data transmission between it and the drone.

[0053] S15. The Paxos protocol is fully implemented in multi-channel data interaction, assigning a unique identifier to each communication channel and establishing a distributed node status management mechanism. When a drone sends a data request or status update to the tethered hot air balloon base station, the base station first parses the request, extracting key information (request type, source node identifier, data content, etc.). It then initiates a multi-node voting process based on the Paxos protocol to ensure consensus on the same operation across multiple communication channels. For example, when a drone uploads rice blast monitoring data, the base station coordinates the channels using the Paxos protocol to ensure that all relevant nodes receive, store, and process the data in the same order, avoiding data conflicts and inconsistencies. This strict consistency guarantee ensures highly reliable and stable multi-channel interaction, effectively preventing monitoring interruptions or misdiagnoses caused by communication failures or data errors.

[0054] Preferably, S20 further includes the following steps:

[0055] S21. Based on the diversity of rice varieties and the complexity of rice blast, multiple sampling time points were set throughout the growth cycle of different rice varieties (common varieties such as indica, japonica, and glutinous rice, as well as local specialty varieties). These sampling time points covered key stages of rice cultivation, from seedling to tillering, jointing, booting, heading, and maturity, ensuring the collection of multi-temporal data. Spatially, different monitoring areas were divided according to the layout and topographic characteristics of the rice fields. Multiple sampling points were set within each area to form a multi-dimensional monitoring network. Utilizing the flexible low-altitude flight capabilities of drones, close-range, detailed multi- / hyperspectral imagery was acquired for rice varieties in different regions. Spectral information was obtained for diseased plants at different growth stages and different parts (leaves, stems, ears, etc.). Tethered hot air balloons were used to conduct macroscopic monitoring of the entire rice field from high altitude, obtaining comprehensive information on rice growth and disease distribution over a large area. The two systems worked together to achieve comprehensive, multi-level data collection.

[0056] S22. Establish a unified data integration framework to correlate and integrate collected geographic information, soil data, climate data, and field management data with multi- and hyperspectral information of rice blast-infected plants. To address the noise problem in multi- and hyperspectral images, an adaptive filtering denoising algorithm is used to automatically adjust the filtering parameters based on the local characteristics of the image to effectively remove noise interference. A radiation correction method based on physical models is used, combined with field-measured atmospheric parameters (aerosol optical depth and water vapor content), to accurately correct the radiation values ​​received by the sensor and restore the true spectral reflectance characteristics of rice. For image enhancement, a combination of nonlinear stretching and histogram equalization is used to enhance the contrast between the characteristics of the diseased plants and the background, highlight the details of the disease, improve data quality, and provide a reliable foundation for subsequent analysis.

[0057] S23. Based on the principles of multi-scale analysis and feature fusion, the wavelet transform is used to decompose the image into sub-bands of different scales and directions. The high-resolution detail information collected by the UAV and the macroscopic information obtained by the tethered hot air balloon are layered and fused, preserving the global features and local details of the image. The gray-level co-occurrence matrix is ​​extracted from the UAV image to calculate the texture parameters and spectral curve shape characteristics, and the vegetation index and color characteristics from the tethered hot air balloon image. The principal component analysis (PCA) method is used for feature fusion to construct a comprehensive feature vector, fully leveraging the advantages of the UAV and tethered hot air balloon data.

[0058] S24. By comparing and analyzing the reflectance differences between healthy rice plants and diseased plants with different disease severity in various spectral bands, combined with statistical methods such as correlation analysis and stepwise discriminant analysis, the bands with significant differences were screened out. Through verification and optimization of a large amount of sample data, the disease index that can most effectively characterize the rice blast condition was determined, and a MySQL database was established to store rice variety information, spectral data, geographical soil and climate field management data, disease information, and extracted key bands and disease index information.

[0059] Preferably, S30 further includes the following steps:

[0060] S31. Spectral data of representative diseased rice plants and corresponding healthy rice plants are selected from the MySQL database as the basic data set for model construction. The diseased plant data are finely labeled, and information such as the location, shape, size, and severity of the lesions is recorded in detail. The comprehensive feature vector obtained in S23 is combined and used as the input parameters of the model.

[0061] S32. Construct a deep learning model architecture based on self-supervised learning, using an encoder-decoder structure. The encoder consists of multiple convolutional layers and pooling layers, which are used to encode the input multimodal feature vector, gradually extract abstract features from the data, and reduce the data dimension. The decoder uses upsampling layers and deconvolution layers to map the encoded features back to the original data space to achieve data reconstruction. Skip connections are introduced between the encoder and decoder to retain feature information at different levels and enhance the model's ability to capture detailed features.

[0062] S33. The model uses a self-supervised learning strategy for model training. It uses the intrinsic structural information of the data to design pre-training tasks. By randomly cropping, rotating, flipping, and other transformations on the input image, the model predicts the relationship between the transformed image and the original image, thereby learning the invariant characteristics of the data. In the pre-training stage, a large amount of unlabeled data is used for training, enabling the model to automatically discover potential patterns and feature representations in the data.

[0063] S34. After pre-training, the model is fine-tuned using labeled data to optimize model parameters to suit the rice blast diagnosis task. During fine-tuning, the cross-entropy loss function is used as the model's optimization objective to measure the difference between the model's predictions and the true labels. Simultaneously, the stochastic gradient descent (SGD) algorithm is combined to dynamically adjust the learning rate based on the data distribution to accelerate model convergence and improve the model's generalization ability. The model is evaluated using a validation set to monitor its performance in terms of accuracy, recall, F1 value, and other indicators. When model performance no longer improves, training is stopped to prevent overfitting.

[0064] S35. Establish a real-time data stream processing system to receive new data continuously collected by drones and tethered hot air balloons. Perform real-time preprocessing on the new data, including data cleaning, format conversion, and feature extraction similar to the training phase. The extracted dynamic features are input into the trained model to perform real-time rice blast monitoring and diagnosis.

[0065] S36. To adapt to the dynamic changes in rice growth and the evolution of rice blast, a dynamic feature update mechanism was designed. Over time and as data accumulates, the model is dynamically updated using an incremental learning method. This allows the model to continuously learn new data based on existing knowledge without having to retrain the entire model, thus avoiding the "catastrophic forgetting" problem. When new data accumulates to a certain level, the model is fine-tuned using this new data and some of its parameters are updated, allowing the model to promptly adapt to new changes and trends in rice blast.

[0066] S37. During the model updating process, a reward mechanism is introduced in combination with reinforcement learning ideas. Corresponding rewards are given according to the model's performance in actual monitoring (accurately predicting the time of disease outbreak and accurately judging the severity of the disease). This encourages the model to continuously optimize its own parameters and improve monitoring accuracy and reliability. At the same time, a model performance evaluation index system is established. In addition to traditional indicators such as accuracy and recall rate, the timeliness and stability factors of the model are additionally considered to comprehensively evaluate the model's performance. Targeted optimization and improvement are carried out based on the evaluation results to ensure that the constructed rice blast early diagnosis inversion model always maintains a good performance state and provides timely and effective disease warning and prevention and control decision support for rice production.

Claims

1. A method for monitoring and diagnosing rice blast based on a tethered hot air balloon and an unmanned aerial vehicle system, characterized in that: The following steps are involved: S10. Build a dynamic observation platform based on a tethered hot air balloon and drone swarm. Determine the coordination framework and sensor equipment of the tethered hot air balloon and drone systems, install mobile signal base stations, and build a multi-channel data interaction method to achieve fast and stable data transmission between operating systems, ensuring that drones and tethered hot air balloons can simultaneously obtain multi- / hyperspectral data in the near- and low-altitude composite airspace. The step S10 further includes: S11. Comprehensively investigate and analyze the needs for rice blast monitoring and rice growth characteristics, identify the key rice growth stages to be monitored, common types of rice blast, causes of disease, and key indicators of disease development, to ensure that the platform system design closely matches the actual needs of agriculture; S12. Determine the operating altitude of the near-low-altitude unmanned system based on the experimental rice field area, topography, expected experimental results, and local meteorological conditions. The drone's flight altitude is set between 30m and 60m, adjusted according to different monitoring stages and rice varieties. The drone is equipped with a high-resolution multispectral camera and high-precision temperature and humidity sensors to quickly collect detailed rice field crop information. S13. Use a tethered hot air balloon as a continuous monitoring device for rice fields, with a flight altitude of 2km-3km, a flight time of 7-10 days, and a coverage radius of 160-200km. The balloon should be equipped with wind direction and speed sensors, atmospheric pressure sensors, and high-resolution hyperspectral sensors. S14. The tethered hot air balloon uses a high-strength, lightweight mounting structure, with a mobile signal base station securely mounted on its gondola or specific bearing area. The base station has powerful signal transmission and reception capabilities and uses multi-band antenna technology, enabling stable communication with different types of drones and ground control terminals. Its transmission power can be dynamically adjusted to meet communication needs at different distances and in different environments. The base station has a built-in high-performance signal processor that can perform real-time encoding, decoding, and error correction on the transmitted signal, effectively reducing the bit error rate during signal transmission and ensuring the accuracy of data transmission between it and the drone. S20. Based on a dynamic observation platform, collect multi- / hyperspectral information of rice blast-infected plants from multiple varieties, multiple temporal phases, and multiple dimensions. Integrate geographic, soil, climate, and field management data. Optimize data preprocessing processes, including denoising, radiometric correction, and image enhancement. Develop data fusion tools to complement drone and tethered hot air balloon data. Extract universal key bands and disease indices sensitive to rice blast-infected plants. Construct a multi-variety rice blast spectral database. S30. Based on the spectral characteristics of diseased rice plants and the data features of drones and tethered hot air balloons, a self-supervised learning model combined with deep learning technology is used to enhance the ability to identify the hallmark features of rice blast. The continuous data streams provided by drones and tethered hot air balloons are used to enhance the dynamic change tracking capability of the rice blast monitoring model and to construct an early diagnosis inversion model for rice blast.

2. The rice blast monitoring and diagnosis method according to claim 1, characterized in that: The step S10 further includes: S15. In the multi-channel data interaction process, the Paxos protocol is fully introduced, a unique identifier is assigned to each communication channel, and a distributed node status management mechanism is established. When the UAV sends a data request or status update to the tethered hot air balloon base station, the tethered hot air balloon base station first parses the request and extracts key information, and then initiates a multi-node voting process based on the Paxos protocol to ensure that consensus is reached on the same operation on multiple communication channels. That is, when the UAV uploads rice blast monitoring data, the tethered hot air balloon base station coordinates each channel through the Paxos protocol to ensure that all relevant nodes receive, store and process the data in the same order, avoid data conflicts and inconsistencies, and effectively prevent monitoring interruptions or misdiagnoses caused by communication failures or data errors.

3. The rice blast monitoring and diagnosis method according to claim 2, characterized in that: The step S20 further includes: S21. Based on the diversity of rice varieties and the complexity of rice blast, multiple sampling time points are set for different rice varieties throughout their growth cycle, covering the seedling stage, tillering stage, jointing stage, booting stage, heading stage to maturity stage after sowing, to ensure the collection of multi-temporal data. In the spatial dimension, different monitoring areas are divided according to the layout and terrain characteristics of the rice fields. Multiple sampling points are set in each area to form a multi-dimensional monitoring network. The advantage of UAVs' flexible low-altitude flight is utilized to conduct close-range and refined multi- / hyperspectral image collection of rice in different regions and varieties, obtaining spectral information of diseased plants at different growth stages and different parts. Tethered hot air balloons conduct macroscopic monitoring of the entire rice field at high altitude to obtain overall information on rice growth status and disease distribution over a large area. The two work together to achieve all-round and multi-level data collection. S22. Establish a unified data integration framework to correlate and integrate collected geographic information, soil data, climate data, and field management data with the multi- / hyperspectral information of rice blast diseased plants. To address the noise problem in multi- / hyperspectral images, use an adaptive filtering denoising algorithm to automatically adjust the filtering parameters based on the local characteristics of the image to effectively remove noise interference. Use a radiation correction method based on physical models, combined with field-measured atmospheric parameters, to accurately correct the radiation values ​​received by the sensor and restore the true spectral reflectance characteristics of rice. For image enhancement, use a combination of nonlinear stretching and histogram equalization to enhance the contrast between the characteristics of the diseased plant and the background, highlight the details of the disease, improve data quality, and provide a reliable foundation for subsequent analysis. S23. Based on the principles of multi-scale analysis and feature fusion, wavelet transform is used to decompose images into sub-bands of different scales and directions. High-resolution detail information collected by drones and macroscopic information obtained by tethered hot air balloons are layered and fused, preserving the global features and local details of the images. Gray-level co-occurrence matrices are extracted from drone images to calculate texture parameters and spectral curve shape features, while vegetation index and color features are extracted from tethered hot air balloon images. Principal component analysis is used for feature fusion to construct a comprehensive feature vector, fully leveraging the advantages of drone and tethered hot air balloon data. S24. By comparing and analyzing the reflectance differences between healthy rice plants and diseased plants with different disease severity in various spectral bands, combined with correlation analysis and stepwise discriminant analysis statistical methods, the bands with significant differences were screened out. Through verification and optimization of a large amount of sample data, the disease index that can most effectively characterize the rice blast condition was determined, and a MySQL database was established to store rice variety information, spectral data, geographical soil and climate field management data, disease information, and extracted key bands and disease index information.

4. The rice blast monitoring and diagnosis method according to claim 3, characterized in that: The step S30 further includes: S31. Spectral data of representative diseased rice plants and corresponding spectral data of healthy plants are selected from the MySQL database as the basic data set for model construction. The diseased plant data are finely annotated, and detailed information on the location, shape, size, and severity of the lesions is recorded. The comprehensive feature vector obtained in step S23 is combined and used as the input parameters of the model. S32. Construct a deep learning model architecture based on self-supervised learning, using an encoder-decoder structure. The encoder consists of multiple convolutional layers and pooling layers, which are used to encode the input multimodal feature vector, gradually extract abstract features from the data, and reduce the data dimension. The decoder uses upsampling layers and deconvolution layers to map the encoded features back to the original data space to achieve data reconstruction. Skip connections are introduced between the encoder and decoder to retain feature information at different levels and enhance the model's ability to capture detailed features. S33. The model uses a self-supervised learning strategy for model training. It uses the intrinsic structural information of the data to design pre-training tasks. By randomly cropping, rotating, and flipping the input image, the model predicts the relationship between the transformed image and the original image, thereby learning the invariant characteristics of the data. In the pre-training stage, a large amount of unlabeled data is used for training, enabling the model to automatically discover the potential patterns and feature representations in the data. S34. After pre-training is completed, the model is fine-tuned using labeled data to optimize model parameters to adapt to the rice blast diagnosis task. During the fine-tuning process, the cross-entropy loss function is used as the optimization target of the model to measure the difference between the model prediction results and the true labels. At the same time, combined with the stochastic gradient descent algorithm, the learning rate is dynamically adjusted according to the distribution characteristics of the data to accelerate model convergence and improve the generalization ability of the model. The model is evaluated using the validation set to monitor the performance of the model in terms of accuracy, recall rate, and F1 value indicators. When the model performance no longer improves, training is stopped to prevent overfitting; S35. Establish a real-time data stream processing system to receive new data continuously collected by drones and tethered hot air balloons. Perform real-time preprocessing on the new data, including data cleaning, format conversion, and feature extraction similar to the training phase. The extracted dynamic features are input into the trained model to perform real-time rice blast monitoring and diagnosis. S36. To adapt to the dynamic changes in rice growth and the evolution of rice blast, a dynamic feature update mechanism is designed. As time passes and data accumulates, the model is dynamically updated using an incremental learning method. This allows the model to continuously learn new data based on existing knowledge without retraining the entire model. When new data accumulates to a certain level, the model is fine-tuned using the new data and some model parameters are updated, allowing the model to promptly adapt to new changes and trends in rice blast. S37. During the model updating process, a reward mechanism is introduced in combination with reinforcement learning ideas. Corresponding rewards are given according to the model's performance in actual monitoring, which encourages the model to continuously optimize its own parameters and improve monitoring accuracy and reliability. At the same time, a model performance evaluation index system is established. In addition to traditional accuracy and recall rate indicators, the timeliness and stability factors of the model are additionally considered to comprehensively evaluate the performance of the model. Targeted optimization and improvement are carried out based on the evaluation results to ensure that the constructed rice blast early diagnosis inversion model always maintains a good performance state and provides timely and effective disease warning and prevention and control decision support for rice production.

Images