Rubber tree disease and insect pest cooperative monitoring and early warning method and system based on multi-source fusion

Abstract

The invention relates to the technical field of agricultural disease and insect pest monitoring, and discloses a rubber tree disease and insect pest collaborative monitoring and early warning method and system based on multi-source fusion, and the method comprises the steps: collecting multi-source data, and obtaining remote sensing, unmanned aerial vehicle, ground sensor and manual inspection data; data fusion, wherein a fusion feature set is generated through preprocessing, space-time alignment and deep learning feature fusion; intelligent early warning is carried out, a basic model is constructed based on fusion features, and risk level, region and time early warning is generated through real-time data dynamic correction; cooperative prevention and control, pushing of early warning and scheduling of intelligent equipment, and feedback of a prevention and control effect optimization model; the system comprises a multi-source data acquisition module, a data fusion processing module, an intelligent early warning analysis module and a collaborative prevention and control linkage module. The problems of data isolation, early warning lag and prevention and control disjunction are solved, monitoring comprehensiveness, early warning accuracy and prevention and control timeliness are improved, and support is provided for sustainable development of the rubber industry.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural pest monitoring technology, more particularly to a rubber tree pest monitoring and early warning method and system based on multi-source fusion. BACKGROUND

[0002] Natural rubber is a key raw material for modern industry and high-end technology. As a key production capacity guarantee area for natural rubber, Guangdong Province is affected by high temperature and humidity, typhoons and other weather conditions, and diseases and pests occur frequently. Among them, powdery mildew, anthracnose and six-point spider mites are key bottlenecks restricting the development of the industry.

[0003] The existing rubber tree pest monitoring and early warning technology has obvious shortcomings: first, the monitoring means is isolated, the traditional manual step-by-step patrol efficiency is very low, only 50 mu per day, and is limited by the characteristics of tall rubber trees, with poor sampling representativeness and high missed detection rate. Although remote sensing, unmanned aerial vehicles, ground sensors and other technologies have been applied, each data source operates independently, and the data lacks effective integration, forming an "information island" and failing to fully reflect the disease and pest occurrence trend. Second, the early warning capability is weak, the data transmission and sharing means are backward, the monitoring data lags 3-5 days, often missing the best prevention and control period, and the early warning model relies on a single environmental factor or historical experience, without fully exploiting the correlation of multi-source data, making it difficult to accurately predict the disease and pest prevalence trend. Third, the monitoring and prevention and control are disconnected, the existing technology does not form a "monitoring-early warning-prevention and control" linkage mechanism, and the prevention and control equipment scheduling relies on manual decision-making, which cannot quickly respond to the demand for disease and pest outbreak.

[0004] Therefore, the present application provides a rubber tree pest monitoring and early warning method and system based on multi-source fusion to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the present application provides a rubber tree pest monitoring and early warning method and system based on multi-source fusion to solve the problems in the background art.

[0006] The present application provides the following technical solution: a rubber tree pest monitoring and early warning method based on multi-source fusion, comprising the following steps:

[0007] Data acquisition step: collecting remote sensing data, unmanned aerial vehicle inspection data, ground sensor data and manual inspection data of the rubber plantation by multi-source monitoring equipment; wherein the remote sensing data includes vegetation index and canopy temperature data; the unmanned aerial vehicle inspection data includes high-resolution leaf image and tree crown three-dimensional structure data; the ground sensor data includes temperature and humidity, soil moisture, and pathogen spore concentration data; the manual inspection data includes disease and pest occurrence position and damage degree record data;

[0008] Data fusion step: preprocess the collected multi-source data, realize the time and space matching of different source data through space-time alignment technology, and generate a fusion feature set by using a deep learning model to fuse the matched data;

[0009] Intelligent early warning step: based on the fusion feature set, a basic early warning model is constructed combining the historical occurrence rules of rubber tree diseases and pests, the model parameters are dynamically corrected through real-time data, and early warning information of disease and pest occurrence risk level, area and prediction time is generated;

[0010] Collaborative prevention and control step: push the early warning information to the prevention and control terminal, automatically dispatch corresponding intelligent prevention and control equipment according to the early warning level, and feed back the prevention and control effect data to the data fusion step to optimize the model.

[0011] As a further scheme of the present application: in the data fusion step, the preprocessing includes data cleaning, noise removal and format standardization; the space-time alignment technology realizes the space-time matching of multi-source data through time stamp synchronization and geographic coordinate calibration; the deep learning model includes a hybrid model of convolutional neural network and Transformer, which is used to extract the associated features of image features, environmental features and text features.

[0012] As a further scheme of the present application: in the intelligent early warning step, the basic early warning model is constructed by fusing the correlation rules of diseases and pests, temperature and humidity, light, and vegetation growth state; the dynamic correction is to adaptively adjust the model weight through the real-time disease spore concentration change rate and leaf lesion area growth rate.

[0013] As a further scheme of the present application: in the collaborative prevention and control step, the early warning level includes low risk, medium risk and high risk; the corresponding dispatched intelligent prevention and control equipment includes one or more of small remote control crawler sprayers, unmanned high-lift powder spraying machines and plant protection unmanned aerial vehicles; the prevention and control effect data includes disease and pest reduction rate after spraying and plant recovery state data.

[0014] The present application also provides a rubber tree disease and pest collaborative monitoring and early warning system based on multi-source fusion, comprising:

[0015] Multi-source data acquisition module: used for deploying remote sensing satellite receiving equipment, unmanned aerial vehicle inspection equipment, ground sensor array and manual inspection terminal, and collecting remote sensing data, unmanned aerial vehicle inspection data, ground sensor data and manual inspection data of rubber plantations;

[0016] Data fusion processing module: connected to the multi-source data acquisition module, used for preprocessing, space-time alignment and feature fusion of multi-source data, and outputting a fusion feature set; the data fusion processing module includes a data preprocessing unit, a space-time alignment unit and a feature fusion unit;

[0017] The intelligent early warning analysis module is connected with the data fusion processing module, is used for constructing a basic early warning model based on a fusion feature set and historical rules, dynamically correcting the model through real-time data and generating early warning information, and comprises a model construction unit, a parameter correction unit and an early warning generation unit.

[0018] The cooperative prevention and control linkage module is connected with the intelligent early warning analysis module, is used for receiving early warning information and scheduling intelligent prevention and control equipment, simultaneously collecting prevention and control effect data and feeding back to the data fusion processing module, and comprises an early warning pushing unit, an equipment scheduling unit and an effect feedback unit.

[0019] As a further scheme of the application, the ground sensor array comprises a temperature and humidity sensor, a soil moisture sensor and a pathogen spore capturing sensor, and the sensor nodes are networked through a wireless sensor network to support real-time data transmission.

[0020] As a further scheme of the application, the unmanned aerial vehicle inspection equipment is equipped with a high-definition camera, a multispectral camera and a laser radar, and is used for acquiring leaf lesion images, vegetation spectrum data and tree crown structure data.

[0021] As a further scheme of the application, the feature fusion unit extracts image features by using a convolutional neural network, extracts environmental time sequence features by using a long short-term memory network, and realizes multi-feature correlation fusion through an attention mechanism.

[0022] As a further scheme of the application, the model construction unit constructs a basic early warning model based on historical pest and disease occurrence data, corresponding environmental data and vegetation growth data through a gradient boosting tree algorithm.

[0023] As a further scheme of the application, the equipment scheduling unit is internally provided with a prevention and control equipment database, including the operation range, the pesticide application efficiency and the applicable pest and disease types of different equipment, and the optimal equipment combination is matched according to the early warning area and the pest and disease types.

[0024] Technical effects and advantages of the application:

[0025] (1) Solving the problem of data isolation: through multi-source data fusion of "sky-ground-people", the barriers of independent operation of each data source in traditional monitoring are broken, the pest and disease data are fully integrated, the completeness and refinement of monitoring are improved, and the defects of poor representativeness of artificial step-by-step patrol and one-sidedness of single device monitoring are overcome.

[0026] (2) Improving the accuracy and timeliness of early warning: based on the multi-source fusion features, a dynamic early warning model is constructed, and the parameters are corrected combined with real-time data, the problems of traditional early warning relying on experience and data lag are solved, the accurate prediction and timely early warning of pest and disease risks are realized, and valuable time is gained for prevention and control.

[0027] (3) Realize monitoring and prevention linkage: through the cooperation prevention and control module, the early warning information is associated with the intelligent equipment scheduling, a closed loop mechanism is formed, the problem that monitoring and prevention are disconnected in the prior art is solved, the response speed and operation efficiency of the disease and pest control are improved, and the prevention and control demand of the rapid spread of the disease and pest in the rubber garden is adapted.

[0028] (4) Adapt to complex rubber planting environment: the system design considers the characteristics of the rubber tree tall trees and the ecological conditions of high temperature and high humidity in the rubber planting area north of 18° north latitude in Guangdong and frequent typhoons, and can flexibly adapt to different terrain and climate areas, and provides general technical support for the disease and pest control in the main production area of natural rubber. BRIEF DESCRIPTION OF DRAWINGS

[0029] The application will be further described below in combination with the drawings.

[0030] Figure 1 is the method flowchart of the rubber tree disease and pest cooperative monitoring and early warning method based on multi-source fusion of the application;

[0031] Figure 2 is the system block diagram of the rubber tree disease and pest cooperative monitoring and early warning system based on multi-source fusion of the application;

[0032] Figure 3 is the block diagram of the ground sensor array in the application. DETAILED DESCRIPTION

[0033] The technical solutions of the application will be described in detail below in combination with the drawings and specific embodiments. The embodiments are only used to explain the application, and do not limit the protection scope of the application.

[0034] Please refer to Figure 1 The application provides a rubber tree disease and pest cooperative monitoring and early warning method based on multi-source fusion, which comprises the following steps:

[0035] Data acquisition step: remote sensing data, unmanned aerial vehicle inspection data, ground sensor data and manual inspection data of the rubber garden are collected through multi-source monitoring equipment; wherein the remote sensing data comprises vegetation index and canopy temperature data; the unmanned aerial vehicle inspection data comprises high-resolution leaf image and tree crown three-dimensional structure data; the ground sensor data comprises temperature and humidity, soil moisture condition and pathogen spore concentration data; the manual inspection data comprises disease and pest occurrence position and damage degree record data;

[0036] Data fusion step: the collected multi-source data is preprocessed, the time and space matching of different source data is realized through space-time alignment technology, the matched data is subjected to feature fusion through a deep learning model, and a fusion feature set is generated;

[0037] Intelligent early warning step: based on the fusion feature set, a basic early warning model is constructed combining the historical occurrence rules of rubber tree diseases and pests, the model parameters are dynamically corrected through real-time data, and early warning information of disease and pest occurrence risk level, area and prediction time is generated;

[0038] Collaborative prevention and control step: push the early warning information to the prevention and control terminal, automatically dispatch the corresponding intelligent prevention and control equipment according to the early warning level, and feed back the prevention and control effect data to the data fusion step to optimize the model;

[0039] In a rubber plantation base in Guangdong, the above method steps are as follows:

[0040] Data collection: deploy remote sensing satellite receiving equipment to obtain rubber plantation NDVI vegetation index and canopy temperature data every 3 days; use unmanned aerial vehicles equipped with 20 million pixel high-definition cameras and laser radars to conduct 1-2 times of inspection on rubber plantations every week, focusing on capturing disease characteristics of new shoots and tender leaves; ground sensor array is arranged with one monitoring node per 50 mu, and temperature and humidity sensors monitor the microclimate of rubber plantations in real time, and spore image is collected by spore capturing sensors every hour, and data is transmitted through wireless sensor network; manual inspection terminal uses mobile APP, which supports patrol personnel to input disease and pest location, damage degree and upload on-site photos;

[0041] Data fusion: mean filter is used in the preprocessing stage to remove unmanned aerial vehicle image noise and eliminate sensor outliers; time and space alignment is calibrated by GPS coordinates, and remote sensing data, unmanned aerial vehicle data, ground sensor data and manual inspection data are matched to 10m×10m spatial grid, and unmanned aerial vehicle inspection time is used as the reference synchronous timestamp; a hybrid model of convolutional neural network + Transformer is used to extract leaf disease image features, temperature and humidity time series features and manual record text features, and a feature set containing environment-image-semantic correlation is generated; among them, the CNN module extracts local features such as texture and color distribution of disease area for leaf image data collected by unmanned aerial vehicles through 3 layers of convolution and pooling layers; the Transformer module captures the long-period trend of environmental factors for temperature and humidity, spore concentration and other time series data collected by ground sensors through self-attention mechanism; both of them realize feature correlation through cross-attention mechanism, focusing on strengthening the weight of key correlation features such as 'high humidity + leaf water spot' and'sudden increase of spore concentration + new shoot disease'; finally, the fusion feature set is output;

[0042] Intelligent early warning: Based on the occurrence data of powdery mildew and anthracnose in the past 5 years and the corresponding fusion features, a basic early warning model is constructed; Real-time calculation of spore concentration change rate and leaf area growth rate, dynamically correct model parameters; Output early warning information, including risk level (low risk <20%, medium risk 20%-50%, high risk >50%), risk area (accurate to grid unit) and predicted outbreak time (3-7 days in the future);

[0043] Synergistic prevention and control: Early warning information is pushed to the base monitoring platform and management personnel's mobile terminal; 3WL-7 type remote control crawler sprayer is dispatched in low-risk areas, 3CF-18 type vehicle-mounted powder spraying machine is dispatched in medium-risk areas, and DJI T40 unmanned aerial vehicle and 3GF-20 type high-lift powder spraying machine are linked in high-risk areas; 3 days after pesticide application, the pest reduction rate is collected and fed back to the data fusion step for optimizing the feature fusion weight.

[0044] In the data fusion step, preprocessing includes data cleaning, noise removal and format standardization; Time and space alignment technology realizes the spatio-temporal matching of multi-source data through time stamp synchronization and geographic coordinate calibration; The deep learning model includes a hybrid model of convolutional neural network and Transformer, which is used to extract the associated features of image features, environmental features and text features;

[0045] In the data fusion step, the preprocessing is as follows: For the leaf images collected by the unmanned aerial vehicle, Gaussian filtering is used to remove salt and pepper noise, and the effective leaf area is retained by image cropping; For ground sensor data, 3σ criterion is used to remove outliers (such as data with temperature and humidity exceeding the reasonable range), and the format is unified to "time stamp-coordinate-value" structure; For manual inspection data, the text description of damage degree (such as "leaf curling" and "dense spots") is standardized to 0-5 level numerical label;

[0046] The time and space alignment technology is as follows: The geographic coordinates are obtained through the built-in GPS module of each device, and the remote sensing data, unmanned aerial vehicle data, ground sensor data and manual data are mapped to the unified WGS84 coordinate system; The remote sensing data (according to the shooting date), sensor data (according to the collection time) and manual data (according to the input time) are time-synchronized with the unmanned aerial vehicle take-off time as the reference timestamp, with an error of ±5 minutes;

[0047] The deep learning model is as follows: ResNet-50 convolutional neural network is used to extract features such as lesion edge and color distribution of leaf images; Transformer model is used to encode environmental time series data such as temperature and humidity, spore concentration and manual record text data, and cross-attention mechanism is used to capture the association between image features and environmental / text features (such as "high humidity + leaf yellowing spots" corresponding to high risk of anthracnose), and the output dimension is a fusion feature vector of 512;

[0048] In the intelligent early warning step, the basic early warning model is constructed by fusing the correlation rules of pests and diseases and temperature and humidity, light, and vegetation growth state; dynamic correction is performed by adaptively adjusting the model weight according to the real-time change rate of pathogen spore concentration and the growth rate of leaf lesion area.

[0049] In the construction of the basic early warning model, first, the correlation rules of pests and diseases and the environment in the historical data are analyzed: for example, powdery mildew occurs at a temperature of 20-25℃, a humidity of 80%-90%, and a new shoot growth period, and anthracnose is strongly correlated with high temperature and high humidity environment after continuous rainfall (daily rainfall > 50 mm); environmental characteristics such as temperature and humidity, light duration, and vegetation index (NDVI < 0.6 indicating weak growth) are fused with growth characteristics such as leaf lesion degree and crown density, and the XGBoost algorithm is used to train the basic model to output the disease incidence probability prediction value.

[0050] The dynamic correction is as follows:

[0051] The real-time calculation of the change rate of pathogen spore concentration is as follows:

[0052]

[0053] wherein, R spore is the change rate of pathogen spore concentration in 24 hours, C current is the current spore concentration (pieces / m 3 ), and C 24h is the spore concentration 24 hours ago (pieces / m 3 ). When R spore > 20%, the weight of environmental characteristics is increased from 0.3 to 0.45.

[0054] The leaf lesion area growth rate is calculated by unmanned aerial vehicle image, and the formula is as follows:

[0055]

[0056] wherein, R area is the 3-day growth rate of leaf lesion area, S current is the current lesion area (cm 2 ), and S 3d is the lesion area 3 days ago (cm 2 ). When R area > 40%, the weight of image characteristics is increased from 0.5 to 0.6. After the weight adjustment, the disease incidence probability is recalculated to ensure that the model adapts to the rapid development trend of pests and diseases.

[0057] In the cooperative prevention and control step, the early warning levels include low risk, medium risk and high risk; the corresponding intelligent prevention and control equipment for dispatching includes one or more of a small remote control crawler spraying machine, an unmanned high-lift powder spraying machine and a plant protection unmanned aerial vehicle; the prevention and control effect data includes a disease and pest reduction rate after spraying and plant recovery state data;

[0058] The early warning level division standard is that low risk refers to a predicted disease area < 10% and a diffusion rate < 5% / day; medium risk refers to a predicted disease area of 10%-30% or a diffusion rate of 5%-15% / day; and high risk refers to a predicted disease area > 30% or a diffusion rate > 15% / day;

[0059] The equipment dispatching rule is that a 3WL-7 small remote control crawler spraying machine (operation efficiency 25 mu / hour, suitable for small plot block precision spraying) is dispatched in a low risk area; a 3CF-18 small vehicle-mounted powder spraying dual-purpose machine (operation efficiency 75 mu / hour, suitable for hilly area mobile spraying) is dispatched in a medium risk area; a DJI T40 plant protection unmanned aerial vehicle (responsible for crown layer spraying, efficiency 150 mu / hour) and a 3GF-20 unmanned high-lift powder spraying machine (responsible for middle and lower layer branch and leaf spraying, efficiency 200 mu / hour) are jointly dispatched in a high risk area, forming a three-dimensional prevention and control network;

[0060] The prevention and control effect data collection includes that a disease and pest reduction rate is identified through an unmanned aerial vehicle image at 3 days and 7 days after spraying, and the formula is:

[0061]

[0062] Wherein, R reduction is the disease and pest reduction rate, N before is the number of diseased plants (plants) before spraying, and N after is the number of diseased plants (plants) after spraying; the plant recovery state is recorded through ground sampling, such as new shoot length and leaf chlorophyll content (SPAD value > 40 is regarded as good recovery), and the data is standardized and then fed back to a data fusion module.

[0063] Please refer to Figure 2 The application also provides a rubber tree disease and pest cooperative monitoring and early warning system based on multi-source fusion, comprising:

[0064] A multi-source data collection module is used to deploy remote sensing satellite receiving equipment, unmanned aerial vehicle inspection equipment, ground sensor arrays and manual inspection terminals, and collect remote sensing data, unmanned aerial vehicle inspection data, ground sensor data and manual inspection data of a rubber plantation;

[0065] A data fusion processing module is connected to the multi-source data collection module, used for pre-processing, time and space alignment and feature fusion of the multi-source data, and outputs a fusion feature set; the data fusion processing module includes a data preprocessing unit, a time and space alignment unit and a feature fusion unit.

[0066] Intelligent early warning analysis module: connected to the data fusion processing module, used to build a basic early warning model based on the fusion feature set and historical rules, dynamically correct the model through real-time data and generate early warning information; the intelligent early warning analysis module includes a model construction unit, a parameter correction unit and an early warning generation unit;

[0067] Collaborative prevention and control linkage module: connected to the intelligent early warning analysis module, used to receive early warning information and dispatch intelligent prevention and control equipment, while collecting prevention and control effect data and feeding back to the data fusion processing module; the collaborative prevention and control linkage module includes an early warning push unit, an equipment dispatching unit and an effect feedback unit;

[0068] The system hardware and software are deployed as follows:

[0069] Multi-source data acquisition module: deploy remote sensing satellite data receiving terminal (support Landsat-8, Sentinel-2 satellite data analysis); unmanned aerial vehicle inspection equipment selects multi-rotor agricultural unmanned aerial vehicle (carrying 800 million pixel high-definition camera, multi-spectral sensor and GPS module); ground sensor array consists of 20 monitoring nodes, each node contains temperature and humidity sensor (SHT30), soil moisture sensor (EC-5), and spore trap (built-in 100 times optical microscope), which transmits data through LoRa wireless networking; artificial inspection terminal is an Android phone installed with special APP (supports GPS positioning, image uploading and form input);

[0070] Data fusion processing module: data preprocessing unit is deployed in edge computing gateway, realizing data cleaning (eliminating invalid values), format conversion (unified to JSON format); spatio-temporal alignment unit is deployed in server, realizing coordinate conversion (WGS84 coordinate system) and time synchronization (NTP protocol) through Python; feature fusion unit is deployed in GPU server, running "convolutional neural network + Transformer" hybrid model, realizing feature extraction and fusion through TensorFlow framework, and outputting fusion feature set;

[0071] Intelligent early warning analysis module: model construction unit trains XGBoost basic model based on historical data (2018-2023 pest records) and stores in model database; parameter correction unit receives sensor data in real time, calculates dynamic correction coefficient and adjusts model weight; early warning generation unit converts model output into visual early warning information (risk heat map, time prediction curve) and stores in early warning database;

[0072] The cooperative prevention and control linkage module: the early warning pushing unit pushes the early warning information to the Web monitoring platform and the mobile phone APP through the MQTT protocol; the equipment scheduling unit is internally provided with an equipment database (storing the operation range and efficiency parameters of each equipment), and is linked with the equipment control system through an API interface; the effect feedback unit collects effect data through manual input combined with unmanned aerial vehicle image recognition, and transmits the standardized data to the data fusion processing module.

[0073] Please refer to Figure 3 The ground sensor array includes temperature and humidity sensors, soil moisture sensors, and pathogen spore capture sensors, and the sensor nodes are networked through a wireless sensor network to support real-time data transmission.

[0074] The ground sensor array is distributed in the rubber plantation in a "grid" shape, with one sensor node arranged every 50 mu, and a total of 30 nodes deployed to cover 1500 mu of rubber plantation. Among them: the temperature and humidity sensor adopts SHT31 model, the measurement range is temperature -40℃~125℃, humidity 0~100%RH, the accuracy is ±0.3℃ / ±2%RH, and the data is collected every 10 minutes; the soil moisture sensor adopts GS3 model, measures soil volume water content (0~100%) and electrical conductivity (0~20000μS / cm), and is inserted into the soil at a depth of 20 cm, and the data is collected every 30 minutes; the pathogen spore capture sensor is internally provided with a hundred-fold optical microscope and an image acquisition module, which automatically captures pathogen spores in the air and shoots high-definition images (resolution 2048x1536) every hour; each sensor node is integrated with a LoRa wireless communication module, accesses the gateway through a star network, the gateway is connected with the server through a 4G network, realizes real-time data transmission (delay <30 seconds), and supports edge computing preprocessing (such as outlier rejection).

[0075] The unmanned aerial vehicle inspection equipment is equipped with a high-definition camera, a multispectral camera and a laser radar, which are used to obtain leaf lesion images, vegetation spectrum data and tree crown structure data.

[0076] The unmanned aerial vehicle inspection device selects a six-rotor agricultural unmanned aerial vehicle (maximum take-off weight 10 kg) and carries multiple task payload modules: a 20 million pixel CMOS camera (focal length 24 mm, aperture f / 2.8) installed on a three-axis stabilized gimbal, with a shooting resolution of 5472x3648, used to capture details such as leaf surface powdery mildew spots and anthracnose spots; a multispectral camera includes blue (450 nm), green (550 nm), red (650 nm), red edge (730 nm), and near-infrared (850 nm) five bands, with a spectral resolution of 10 nm, used to obtain vegetation reflectivity data (calculate NDVI, RVI, and other index to reflect growth status); a 16-line mechanical LiDAR with a point cloud density of ≥200 points / m2 and a measurement distance of 0.5-120 m, used to scan three-dimensional structure data such as tree height and crown diameter. The unmanned aerial vehicle is pre-set to fly at a height of 50 m and a cruising speed of 8 m / s, covering 200 mu per inspection, and the data is transmitted back to the server in real time through the 5G module or stored on the local SD card (capacity 128 GB).

[0077] The feature fusion unit extracts image features using a convolutional neural network and extracts environmental time series features using a long short-term memory network, and achieves multi-feature correlation fusion through an attention mechanism.

[0078] The algorithm flow of the feature fusion unit is as follows:

[0079] Image feature extraction: a convolutional neural network (CNN) is used to process the unmanned aerial vehicle leaf image, the input image is processed by 3 convolutional layers (convolution kernel size 3x3, step 1) to extract edge and texture features, and is reduced in dimension by 2 pooling layers (max pooling, 2x2 step), and finally a 256-dimensional image feature vector (including lesion area position, area, and color features) is output by a fully connected layer;

[0080] Environmental time series feature extraction: a long short-term memory network (LSTM) is used to process ground sensor data, with input being near 7-day temperature, humidity, and spore concentration time series data (1 data point per hour, a total of 168 time steps), and 3 LSTM units (hidden layer dimension 128) are used to capture data trend features (such as continuous humidity rise and sudden spore concentration increase), and an 128-dimensional environmental time series feature vector is output;

[0081] Multi-feature association fusion: Introduce attention mechanism to calculate the association weight between image features and environmental features (such as the weight coefficient of "high humidity" feature and "leaf water-stained lesion" image feature), and fuse the two types of feature vectors by weighted summation, then concatenate with artificial inspection text features (encoded by Word2Vec into 128 dimensions), finally output a 512-dimensional fusion feature set, where the attention weight of key features (such as spore concentration, lesion area) is dynamically enhanced (up to 0.7).

[0082] The model construction unit constructs a basic warning model based on historical pest and disease occurrence data, corresponding environmental data, and vegetation growth data through a gradient boosting tree algorithm.

[0083] The implementation process of the model construction unit is as follows:

[0084] Data preparation: Collect pest and disease occurrence data (including the occurrence time, location, and damage level of powdery mildew and anthracnose) in rubber plantations from 2018 to 2023, corresponding environmental data (temperature, humidity, rainfall, soil moisture, and pathogen spore concentration), and vegetation growth data (NDVI index, tree canopy density, and new shoot growth volume), a total of 100,000 samples, each sample containing 30 feature variables.

[0085] Feature engineering: Derive features from environmental data, such as calculating "continuous high humidity days" (the number of consecutive days with humidity > 85%) and "spore concentration weekly growth rate"; calculate "NDVI decline rate" for vegetation data; use disease and pest damage levels (1-5 levels) as labels, where levels 3 and above are defined as "disease events";

[0086] Model training: Use the gradient boosting tree algorithm (XGBoost) with 100 decision trees, a maximum depth of 8, and a learning rate of 0.1; divide the training set and test set in a 7:3 ratio, and optimize the parameters through 5-fold cross-validation to make the model's accuracy on the test set > 85% and F1 score > 0.8; the model outputs the probability of occurrence of level 3 and above disease and pest within the next 7 days as the basis for basic warning;

[0087] Model storage and update: The trained model is stored in PMML format on the model server, supporting incremental updates based on new data every month to maintain the model's adaptability.

[0088] The equipment scheduling unit has a built-in prevention and control equipment database, including the operating range, application efficiency, and suitable disease and pest types of different equipment, and matches the optimal equipment combination according to the warning area and disease and pest types;

[0089] The core function implementation of the equipment scheduling unit is as follows:

[0090] Prevention and control equipment database: store 4 types of intelligent equipment parameters: 3GF-20 unmanned high-lift powder spraying machine (operation range: large flat area; efficiency: 200 mu / hour; suitable type: leaf disease such as powdery mildew and anthracnose); 3CF-18 small vehicle-mounted powder spraying dual-purpose machine (operation range: hilly area; efficiency: 75 mu / hour; suitable type: anthracnose and scale insect); 3WL-7 small remote control crawler spraying machine (operation range: small seedling area and narrow row spacing area; efficiency: 25 mu / hour; suitable type: seedling disease and six-point spider mite); DJI T40 rubber plantation plant protection unmanned aerial vehicle (operation range: all-terrain canopy; efficiency: 150 mu / hour; suitable type: powdery mildew and anthracnose rapid prevention and control);

[0091] Dispatching logic: when receiving the early warning information, first extract the early warning area (such as 500 mu), terrain type (such as "flat + hilly mixed") and disease and pest type (such as "powdery mildew high risk"); match equipment based on database: dispatch 3GF-20 powder spraying machine in flat area (responsible for 300 mu), dispatch 3CF-18 vehicle-mounted machine in hilly area (responsible for 150 mu), and dispatch unmanned aerial vehicle at the same time (responsible for 50 mu canopy intensive pesticide application); calculate the number of equipment required (such as 1 3GF-20, 1 3CF-18 and 1 unmanned aerial vehicle), and generate operation route planning (avoid un-warned areas);

[0092] Execution and feedback: issue dispatching instructions through equipment control system, and monitor equipment operation progress in real time; after operation is completed, record actual operation area and pesticide dosage, and match with early warning area with a matching degree of more than 90% to be considered as effective dispatching, and the data is stored in equipment operation log.

[0093] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for collaborative monitoring and early warning of rubber tree diseases and pests based on multi-source fusion, characterized in that, Includes the following steps: Data acquisition steps: Remote sensing data, drone inspection data, ground sensor data, and manual inspection data of the rubber plantation are collected through multi-source monitoring equipment; wherein, the remote sensing data includes vegetation index and canopy temperature data; the drone inspection data includes high-resolution leaf images and three-dimensional canopy structure data; the ground sensor data includes temperature and humidity, soil moisture, and pathogen spore concentration data; and the manual inspection data includes records of the location and severity of pests and diseases. Data fusion steps: preprocess the collected multi-source data, achieve temporal and spatial matching of data from different sources through spatiotemporal alignment technology, and use a deep learning model to perform feature fusion on the matched data to generate a fused feature set; Intelligent early warning steps: Based on the fusion feature set, a basic early warning model is constructed by combining the historical occurrence patterns of rubber tree diseases and pests. The model parameters are dynamically corrected through real-time data to generate early warning information on the risk level, region, and prediction time of disease and pest occurrence. Collaborative prevention and control steps: push early warning information to prevention and control terminals, automatically dispatch corresponding intelligent prevention and control equipment according to the early warning level, and feed back prevention and control effect data to the data fusion step to optimize the model.

2. The method for collaborative monitoring and early warning of rubber tree diseases and pests based on multi-source fusion according to claim 1, characterized in that: In the data fusion step, preprocessing includes data cleaning, noise removal, and format standardization; the spatiotemporal alignment technology achieves spatiotemporal matching of multi-source data through timestamp synchronization and geographic coordinate calibration; the deep learning model includes a hybrid model of convolutional neural networks and Transformers, used to extract associated features of image features, environmental features, and text features.

3. The method for collaborative monitoring and early warning of rubber tree diseases and pests based on multi-source fusion according to claim 1, characterized in that: In the intelligent early warning step, the basic early warning model is constructed by integrating the correlation between pests and diseases and temperature, humidity, light, and vegetation growth status; the dynamic correction adaptively adjusts the model weights by measuring the real-time change rate of pathogen spore concentration and the growth rate of leaf lesion area.

4. The method for collaborative monitoring and early warning of rubber tree diseases and pests based on multi-source fusion according to claim 1, characterized in that: In the collaborative prevention and control steps, the early warning levels include low risk, medium risk, and high risk; the corresponding intelligent prevention and control equipment includes one or more of the following: small remote-controlled tracked sprayer, unmanned high-lift duster, and plant protection drone; the prevention and control effect data includes the pest and disease reduction rate and plant recovery status data after pesticide application.

5. A multi-source fusion-based collaborative monitoring and early warning system for rubber tree diseases and pests, characterized in that, include: Multi-source data acquisition module: used to deploy remote sensing satellite receiving equipment, drone inspection equipment, ground sensor array and manual inspection terminal to collect remote sensing data, drone inspection data, ground sensor data and manual inspection data of rubber plantation; Data fusion processing module: Connected to the multi-source data acquisition module, it is used to preprocess, spatiotemporally align, and fuse features of multi-source data, and output a fused feature set; the data fusion processing module includes a data preprocessing unit, a spatiotemporal alignment unit, and a feature fusion unit; Intelligent early warning analysis module: connected to the data fusion processing module, used to build a basic early warning model based on the fusion feature set and historical patterns, dynamically correct the model through real-time data and generate early warning information; the intelligent early warning analysis module includes a model building unit, a parameter correction unit and an early warning generation unit; Collaborative Prevention and Control Module: Connects to the intelligent early warning and analysis module, used to receive early warning information and dispatch intelligent prevention and control equipment, while collecting prevention and control effect data and feeding it back to the data fusion and processing module; The collaborative prevention and control linkage module includes an early warning push unit, an equipment scheduling unit, and an effect feedback unit.

6. The rubber tree disease and pest collaborative monitoring and early warning system based on multi-source fusion according to claim 5, characterized in that: The ground sensor array includes temperature and humidity sensors, soil moisture sensors, and pathogen spore capture sensors. The sensor nodes are networked through a wireless sensor network to support real-time data transmission.

7. The rubber tree disease and pest collaborative monitoring and early warning system based on multi-source fusion according to claim 5, characterized in that: The drone inspection equipment is equipped with a high-definition camera, a multispectral camera, and a lidar, used to acquire images of leaf lesions, vegetation spectral data, and tree canopy structure data.

8. The method and system for collaborative monitoring and early warning of rubber tree diseases and pests based on multi-source fusion as described in claim 5, characterized in that: The feature fusion unit uses a convolutional neural network to extract image features, a long short-term memory network to extract environmental temporal features, and an attention mechanism to achieve multi-feature association fusion.

9. The rubber tree disease and pest collaborative monitoring and early warning system based on multi-source fusion according to claim 5, characterized in that: The model building unit constructs a basic early warning model based on historical pest and disease occurrence data, corresponding environmental data, and vegetation growth data, using a gradient boosting tree algorithm.

10. The rubber tree disease and pest collaborative monitoring and early warning system based on multi-source fusion according to claim 5, characterized in that: The equipment scheduling unit has a built-in database of prevention and control equipment, which includes the operating range, application efficiency and applicable pest types of different equipment, and matches the optimal equipment combination according to the area of ​​the warning area and the type of pests.

Images