Crop disease and insect pest intelligent identification and control system based on unmanned aerial vehicle cooperation

Abstract

The invention discloses a crop pest intelligent identification and prevention system based on unmanned aerial vehicle cooperation. The system comprises a control module, a reconnaissance unmanned aerial vehicle cluster, an operation unmanned aerial vehicle cluster, a sensor module and a user interaction terminal. The reconnaissance unmanned aerial vehicle cluster collects multispectral and visible light images; the sensor module collects field microenvironment data; the control module fuses multi-source data, performs analysis through a pest and disease damage identification model, and generates a pest and disease damage hotspot distribution diagram and a prevention and control scheme; the operation unmanned aerial vehicle cluster receives the instruction and performs variable accurate pesticide application on the focus area; after real-time high-precision data acquired by the reconnaissance unmanned aerial vehicle cluster is analyzed by the control module, an operation instruction can be directly generated and the operation unmanned aerial vehicle cluster is driven to execute, so that the response speed and the prevention and control efficiency are greatly improved, the effect of discovering and processing diseases and insect pests is realized, and the diffusion of the diseases and insect pests is effectively restrained.

Description

Technical Field

[0001] This invention relates to the field of intelligent crop pest control technology, and in particular to an intelligent identification and control system for crop diseases and pests based on drone collaboration. Background Technology

[0002] Traditional pest and disease monitoring mainly relies on field surveys and experience-based judgment by agricultural technicians, while prevention and control generally involve large-scale carpet spraying of pesticides by manual backpack sprayers or tractor-trailer-driven sprayers. Currently, solutions have emerged that use single drones for aerial imaging of farmland to assist in reconnaissance, or use single drones equipped with spraying devices for automated spraying operations. At the same time, there are also research attempts to deploy fixed sensor nodes in the field to collect environmental information.

[0003] The aforementioned methods typically operate independently in the reconnaissance and control phases, resulting in identification results that cannot directly guide operations. Furthermore, reliance on a single drone for reconnaissance limits its field of view and efficiency, hindering comprehensive coverage and rapid response across large areas of farmland. Additionally, identification models based on a single data source (such as visible light images alone) are susceptible to interference in complex field environments, requiring improved accuracy. Moreover, existing drone spraying operations are largely based on preset flight paths and fixed dosages, failing to adapt to the actual distribution and severity of pests and diseases, leading to persistent problems of pesticide overuse and environmental pollution.

[0004] Therefore, in response to the problems mentioned above, this invention proposes an intelligent identification and control system for crop diseases and pests based on drone collaboration. Summary of the Invention

[0005] To overcome the problems of disconnect between reconnaissance and prevention, inaccurate identification due to single data, and waste and pollution caused by extensive operation in existing technologies, this invention proposes an intelligent identification and prevention system for crop diseases and pests based on UAV collaboration. Through multi-UAV network collaborative operation, multi-source data fusion intelligent diagnosis, and variable operation based on target diseases, it can achieve early detection, accurate identification, and precise prevention and control of diseases and pests.

[0006] The technical solution of this invention is: an intelligent identification and control system for crop diseases and pests based on unmanned aerial vehicle (UAV) collaboration, comprising: The control module is used for task planning, data fusion, model training and global decision-making. It integrates a geographic information system module, a task scheduling engine and a database. The database is used to store historical pest and disease data, crop growth model data, prevention and control strategy library and real-time multi-source data. A reconnaissance drone swarm, comprising multiple small multi-rotor drones, each equipped with a multispectral imaging sensor, a high-resolution visible light camera, and a first communication module. The multispectral imaging sensor covers at least five bands: blue, green, red, red-edge, and near-infrared, with a spectral resolution better than 10 nanometers and a spatial resolution better than 5 centimeters. The high-resolution visible light camera is a global shutter camera with at least 20 million pixels, used to capture detailed features such as leaf texture and insect bodies. The operation drone swarm includes multiple large and medium-sized multi-rotor drones with stronger payload capacity. Each operation drone is equipped with a pesticide storage tank, a precision spraying mechanism or a biological control agent release device, and a second communication module. The pesticide storage tank is equipped with a high-precision liquid level sensor and a flow meter for real-time monitoring of pesticide consumption. The sensor module, deployed in the farmland, includes multiple sensors with soil moisture, air temperature and humidity, and leaf surface humidity sensing functions. The user interaction terminal is used to receive system alarms, display pest and disease distribution maps, and allow users to confirm or modify prevention and control instructions. The digital twin module is based on historical operational data, real-time field data, and crop growth models to construct a virtual mapping of farmland. Before implementing a prevention and control plan, the digital twin module is used to simulate, predict, and evaluate the prevention and control effect, and the plan parameters are optimized and adjusted based on the simulation results. The control module is communicatively connected to the reconnaissance drone cluster, the operational drone cluster, and the sensor module is communicatively connected to the user interaction terminal. The workflow of the system is as follows: S1, the control module generates a collaborative reconnaissance path for the reconnaissance drone cluster based on farmland boundary information and preset waypoints. This path ensures no blind spots in farmland coverage and allows the drones to dynamically adjust their flight altitude and spacing based on real-time weather data. The preferred flight altitude is 10-50 meters above the crop canopy, and the spacing is dynamically calculated based on the sensor's field of view to ensure an overlap rate of no less than 20%. S2, during flight, the reconnaissance drone swarm simultaneously acquires multispectral and high-resolution visible light images of crops and transmits the image data to the control module in real time or near real time; S3, at the same time, the sensor module continuously collects field microenvironment data and uploads it to the control module; S4, the pest and disease identification model built into the control module, integrates and analyzes collected multispectral images, high-resolution visible light images and field microenvironment data. This pest and disease identification model is based on a deep convolutional neural network and attention mechanism. It can identify the type and severity level of pests and diseases and mark their geographical coordinates. Areas with a confidence level of less than 85% are marked as "to be verified" and secondary reconnaissance or manual verification is recommended. S5. Based on the identification results, the control module generates a dynamic "pest and disease hotspot distribution map" and matches the optimal control plan from the pre-set control strategy library according to the type of pest and disease, severity level and crop growth stage. The optimal control plan includes the type of pesticide, recommended dosage, spraying rate and operation path. S6, the control module sends the optimal prevention and control plan and operation path to the operation drone cluster. Then, the operation drone cluster flies to the hot spot of pests and diseases in a coordinated formation and applies pesticides to the identified diseased areas according to the operation path and spraying mechanism, while reducing or avoiding pesticide application in non-diseased areas.

[0007] Preferably, the training process of the pest and disease identification model includes: A1. Calculate multiple vegetation indices using multispectral images, including the normalized vegetation index and chlorophyll content index. A2, which fuses high-resolution visible light images with calculated vegetation indices at the data layer or feature layer to form multi-source data fusion features; A3 inputs multi-source data fusion features and field microenvironment data from the same period into a deep convolutional neural network for training. The attention mechanism is used to enhance the model’s attention to early subtle features of pests and diseases, including lesions and hidden pests.

[0008] Preferably, the collaborative reconnaissance path adopted by the reconnaissance drone swarm is generated based on the "ant colony optimization algorithm" or the "particle swarm optimization algorithm". Its optimization objectives include the shortest reconnaissance time, the maximum area coverage, and the priority reconnaissance weight for areas with a high incidence of historical pests and diseases. In the algorithm, the weight coefficient of areas with a high incidence of historical pests and diseases is set to 1.5 to 2.0 times that of normal areas, so as to guide the drone swarm to prioritize patrolling these key areas.

[0009] Preferably, the spraying mechanism of the drone swarm is a pulse width modulation (PWM) controlled nozzle array. The control module generates a corresponding PWM control signal matrix based on the severity level of each pixel in the pest hotspot distribution map. This matrix controls the switching frequency and duration of each nozzle, achieving variable spraying based on centimeter-level positioning. The spacing between the nozzle arrays is 20-30 cm, and each nozzle can be controlled independently. The PWM frequency range is adjustable between 1-10 Hz, thereby achieving flow control from a few drops per second to continuous spraying.

[0010] Preferably, during flight operations, the onboard visual sensors of the drone swarm monitor the drift of the pesticide caused by the downwash airflow in real time and feed the drift data back to the control module. The control module adjusts the flight altitude, nozzle switching timing, or formation of the drone swarm based on the real-time wind speed and direction and drift data to reduce pesticide drift. When the crosswind speed exceeds 3 m / s, the system will automatically instruct the drones to increase their flight altitude or suspend nozzle operations in the downwind area.

[0011] Preferably, the prevention and control strategy library includes chemical control and biological control. When the pest is identified as an insect pest and the insect population density is lower than the economic threshold, the biological control scheme is selected first, and the operation drone swarm releases natural enemies such as parasitic wasps and predatory mites. When the severity of the pest exceeds the preset threshold, the chemical control scheme is activated. The biological control agent release device is a temperature-controlled and shockproof insect storage chamber. Its release port is precisely controlled by a stepper motor, which can release natural enemies at fixed points and in fixed quantities during flight according to the preset release density and distribution pattern.

[0012] Preferably, both the first and second communication modules support 5G or LTE Cat-M cellular network communication. In areas with no cellular network signal coverage, a self-organizing network communication mode is adopted, in which the reconnaissance drone cluster and the operation drone cluster build a temporary multi-hop relay communication network on their own.

[0013] Preferably, in the interface provided by the user interaction terminal, the distribution map of pest and disease hotspots is overlaid on the farmland map in the form of a heat map. Users can use gestures to select or exclude specific areas, modify the operation path automatically generated by the system, or forcibly set certain areas as no-fly zones / no-spray zones.

[0014] The beneficial effects of this invention are: 1. This invention constructs a system integrating reconnaissance, decision-making, and execution, realizing the linkage from pest and disease identification to prevention and control. The real-time high-precision data acquired by the reconnaissance drone swarm is analyzed by the control module and can directly generate operation instructions and drive the operation drone swarm to execute. This eliminates the intermediate links of traditional manual data interpretation and scheme formulation, greatly improving response speed and prevention and control efficiency, achieving the effect of immediate treatment upon discovery of pests and diseases, and effectively suppressing the spread of pests and diseases.

[0015] 2. This invention integrates multispectral imagery, high-resolution visible light imagery, and field microenvironment data collected by the Internet of Things (IoT) to form a multidimensional information sensing network that combines air, space, and ground. This overcomes the shortcomings of low accuracy in identifying single data sources. Through the collaborative analysis of multi-source data, especially by utilizing crop physiological stress information reflected by multispectral data and disease occurrence conditions information provided by microenvironment data, this invention can significantly improve the ability to identify and diagnose early and hidden pests and diseases, and reduce the risk of misjudgment and missed diagnosis.

[0016] 3. Based on accurate pest and disease distribution maps and severity level assessments, this invention guides the application of pesticides by a cluster of drones. The system can precisely control the on / off state and dosage of the nozzles according to the prescription map, applying only the necessary amount of pesticide to the required locations, thereby minimizing the total amount of pesticides used, reducing agricultural non-point source pollution and pesticide residue risks, and saving production costs. Attached Figure Description

[0017] Figure 1 The diagram shown is a schematic representation of the system framework of this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 This invention provides an embodiment of an intelligent identification and control system for crop diseases and pests based on drone collaboration: In this embodiment, the control module is used to store all raw data, including TB-level image data from reconnaissance drones, time-series data streams from the ground Internet of Things, historical meteorological data, and crop growth model parameters.

[0020] This module incorporates the pest and disease identification model of this invention. The model uses ResNet-50 as its backbone network, with a bidirectional long short-term memory network module and an attention mechanism module integrated at its backend. ResNet-50 extracts spatial features from the fused image feature map, while the bidirectional long short-term memory network module processes micro-environmental data (such as continuous 72-hour temperature and humidity changes) input from ground-based IoT nodes in a time-series manner. The attention mechanism module is inserted after the last convolutional layer of ResNet-50, enabling the model to automatically learn and focus on small but crucial areas in the image, such as lesions and insect eggs, while suppressing interference from background information such as healthy leaves.

[0021] This module also includes a built-in prevention and control strategy library and a digital twin module. The prevention and control strategy library is a knowledge-based database that stores data in the form of rules such as "IF (pest / disease type = X, severity level = Y, crop growth stage = Z) THEN (recommended pesticide = A, dosage = B, operation method = C)". The digital twin module uses historical data to build a virtual model synchronized with the physical farmland. During decision-making, it simulates and extrapolates the recommended prevention and control plan in the virtual model, predicts the pesticide deposition effect and the development trend of pests and diseases, and thus optimizes the parameters of the plan.

[0022] In this embodiment, the reconnaissance drone swarm uses lightweight multi-rotor drones. Each drone is equipped with a five-channel multispectral camera (blue, green, red, red edge, and near-infrared) and a 25-megapixel global shutter visible light camera. The onboard computer is responsible for the preliminary preprocessing of the images (such as radiometric correction) and transmits the data back to the central platform via a 5G / 4G module.

[0023] For example, when the task scheduling engine of the control module receives a reconnaissance mission for a 500-acre rectangular farmland, it will activate the ant colony optimization algorithm to generate cooperative paths for 10 reconnaissance drones. The algorithm discretizes the farmland into a grid, treating each drone as an "ant." The initial pheromone value is set higher in areas with historically high incidence of pests and diseases. The optimization objective is to maximize coverage and minimize the total flight distance. After algorithm iteration, 10 paths are finally generated, ensuring that the full-area scan is completed with the highest efficiency, provided that the lateral overlap rate is not less than 25% and the forward overlap rate is not less than 70%. The drones will then automatically fly and conduct reconnaissance at a height of 30 meters above the canopy, following the planned paths.

[0024] In this embodiment, the swarm of drones uses a larger octocopter platform, equipped with a 10-liter capacity chemical tank and a precision spraying mechanism consisting of a diaphragm pump and a set of 12 solenoid valve-controlled nozzles with a nozzle spacing of 25 cm. Each nozzle can be independently controlled by receiving a PWM (pulse width modulation) signal.

[0025] After generating a "pest and disease hotspot distribution map," the control module converts it into a "prescription map" with a resolution of 0.1m x 0.1m, bound to the geographical coordinates of the farmland. Each grid cell of this prescription map contains a PWM control value (0-100%). The swarm of drones (e.g., 5 drones) allocates tasks according to the planned sub-regions. During flight, the onboard high-precision GNSS receiver (RTK mode) locates its own position in real time and queries the prescription map for the corresponding PWM value, thereby controlling the switching frequency of the spray nozzles below. For example, in a healthy area, the PWM value is 0%, and all nozzles are off; in a lightly infected area, the PWM value is 30%, and the nozzles spray intermittently; in the center of a severely infected area, the PWM value is 100%, and the nozzles are continuously on. This achieves centimeter-level precision and variable-rate application of pesticides.

[0026] In this embodiment, the sensor module is deployed in the farmland and includes multiple sensors with soil moisture, air temperature and humidity, and leaf surface humidity sensing functions. The sensor module is powered by solar energy and uses the LoRaWAN protocol to upload data to a gateway deployed at the edge of the field. The gateway forwards the aggregated data to the control module via Ethernet.

[0027] In this embodiment, the pest and disease identification model is described in detail: The training of this model is an offline process, and the specific steps are as follows: (1) Radiometric calibration and atmospheric correction are performed on the multispectral image to calculate multiple vegetation index layers such as normalized vegetation index and red edge normalized difference vegetation index. Then, the high-resolution visible light image is geometrically corrected to align it with the multispectral data space. The visible light image is input into ResNet-50 to extract high-level feature maps. At the same time, multiple vegetation index layers are used as a multi-channel input and feature extraction is also performed through a small convolutional network. Then, the channel attention mechanism is used to weight and fuse the feature maps from the two sources to generate a unified multi-source feature map rich in spectral and texture information.

[0028] (2) Process the ground IoT data (such as average temperature and average humidity) of the same geographical area and the same time period into a feature vector, and map it to the dimension that matches the image feature map through a fully connected layer.

[0029] (3) The fused multi-source feature map and the temporal environmental feature vector are concatenated in the channel dimension and fed into the subsequent fully connected layer. The model finally outputs the pest and disease category (such as wheat stripe rust and rice planthopper) and severity level (0-5) of each analysis unit. The loss function adopts weighted cross-entropy to deal with the class imbalance problem.

[0030] When new reconnaissance data is received, the control module calls the pre-trained model to perform forward propagation, quickly generating the identification results of the entire farmland and automatically labeling its geographical coordinates.

[0031] In this embodiment, the digital twin module will be described in detail: Before issuing commands to the operational drone, the control module's decision service activates the digital twin module to perform a simulation: (1) First, a virtual environment is constructed. Using the current three-dimensional model of farmland, crop canopy structure and real-time wind speed and direction data, a computational fluid dynamics (CFD) simulation environment is constructed.

[0032] (2) Then, the operation process is simulated. The flight path, altitude, nozzle switch status and PWM parameters of the operation drone are input into the simulation environment to simulate the deposition and drift process of the pesticide droplets in the canopy.

[0033] (3) Finally, the effect is evaluated and optimized. The simulation outputs two key results: one is the "agent deposition distribution map", which is used to evaluate whether the agent accurately covers the lesion; the other is the "drift risk map". If the simulation shows that the drift risk in a specific area exceeds the threshold (the deposition exceeds the standard at 50 meters downwind), the system will automatically adjust the plan, such as instructing the UAV to reduce its flight altitude in the area or temporarily shut down the outer nozzles, and then issue a new and better operation plan.

[0034] This invention provides a comparative example: This comparative study used agricultural simulation software to generate a virtual wheat farm of 100 hectares (1500 mu). Within the farm, three wheat leaf rust disease zones of varying severity and one rice leaf roller infestation zone were randomly set up. Five groups of experimental subjects were included in this comparative study. Specifically: Example 1 uses the complete system of the present invention.

[0035] Example 2 uses the detection and identification part of the present invention, but the operation part is changed to single machine conventional uniform spraying.

[0036] Example 3 uses the detection and identification part of the present invention, but the identification model only uses visible light images and does not perform multi-source data fusion.

[0037] Comparative Example 1 uses a single drone to take aerial photos in visible light. After human experts interpret the images, they guide ground tractors to carry out full-coverage spraying.

[0038] Comparative Example 2 relies solely on environmental data from the terrestrial Internet of Things to trigger early warnings and prevention measures in fixed areas.

[0039] This experiment first verifies the accuracy of the invention in identifying pests and diseases, including the accuracy of identifying pest and disease categories and severity levels.

[0040]

[0041] As shown in the table above, Example 1 has the highest recognition accuracy (95.2%), which proves that the multi-source data fusion and attention mechanism model is very effective. The accuracy of Example 3 (using only visible light) drops to 82.1%, indicating that in cloudy weather or early stages of pests and diseases, it is difficult to make accurate judgments based solely on visible light texture features. Multispectral information provides irreplaceable evidence of physiological stress. The manual interpretation in Comparative Example 1 is limited by subjectivity and fatigue, resulting in the lowest accuracy.

[0042] This experiment verifies the pesticide usage and effective pesticide deposition rate of the present invention, thereby comparing the economic benefits and environmental protection effects of the present invention.

[0043]

[0044] As shown in the table above, the pesticide usage in Example 1 (12.5 L / ha) is significantly lower than all comparative examples and Example 2 (uniform spraying, 28.7 L / ha), saving over 60% of pesticide. Simultaneously, it achieved the highest effective deposition rate (68.5%) and the lowest pollution index (0.15), directly attributable to variable precision spraying and drift optimization based on digital twins. In contrast, while Example 2 showed accurate identification, its extensive application methods resulted in substantial pesticide waste and environmental pollution.

[0045] This experiment verifies the operational efficiency of the present invention.

[0046]

[0047] As shown in the table above, Example 1 achieved a high efficiency of 8.5 ha / hr by leveraging the parallel operation capability of the drone swarm, which is more than 5 times that of Comparative Example 1 (tractor operation). This demonstrates that the present invention not only has advantages in accuracy but also has significant advantages in efficiency in dealing with large-scale farmland.

[0048] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A smart identification and control system for crop diseases and pests based on drone collaboration, characterized in that, Including: The control module is used for task planning, data fusion, model training, and global decision-making. A reconnaissance drone swarm, comprising multiple reconnaissance drones equipped with multispectral imaging sensors, high-resolution visible light cameras, and a first communication module; A cluster of operational drones, comprising multiple operational drones equipped with pesticide storage tanks, spraying mechanisms or biological control agent release devices, and a second communication module; The sensor module, deployed in the farmland, includes multiple sensors with soil moisture, air temperature and humidity, and leaf surface humidity sensing functions. The user interaction terminal is used to receive system alarms, display pest and disease distribution maps, and allow users to confirm or modify prevention and control instructions. The control module is communicatively connected to the reconnaissance drone cluster, the operational drone cluster, and the sensor module is communicatively connected to the user interaction terminal. The workflow of the system is as follows: S1, the control module generates a collaborative reconnaissance path for the reconnaissance drone cluster based on farmland boundary information and preset waypoints. This path ensures no blind spots in farmland coverage and allows the drones to dynamically adjust their flight altitude and spacing based on real-time weather data. S2, during flight, the reconnaissance drone swarm simultaneously acquires multispectral and high-resolution visible light images of crops and transmits the image data to the control module in real time or near real time; S3, at the same time, the sensor module continuously collects field microenvironment data and uploads it to the control module; S4, the pest and disease identification model built into the control module, integrates and analyzes the collected multispectral images, high-resolution visible light images and field microenvironment data. This pest and disease identification model is based on a deep convolutional neural network and attention mechanism, and can identify the type and severity level of pests and diseases and mark their geographical coordinates. S5. Based on the identification results, the control module generates a dynamic "pest and disease hotspot distribution map" and matches the optimal control plan from the pre-set control strategy library according to the type of pest and disease, severity level and crop growth stage. S6, the control module sends the optimal prevention and control plan and operation path to the operation drone cluster. Then, the operation drone cluster flies to the hot spot of pests and diseases in a coordinated formation and applies pesticides to the identified diseased areas according to the operation path and spraying mechanism, while reducing or avoiding pesticide application in non-diseased areas.

2. The intelligent identification and control system for crop diseases and pests based on UAV collaboration according to claim 1: the optimal control scheme includes pesticide type, recommended dosage, spraying rate and operation path.

3. The intelligent identification and control system for crop diseases and pests based on UAV collaboration according to claim 1, characterized in that, The training process of the pest and disease identification model includes: A1. Calculate multiple vegetation indices using multispectral images, including the normalized vegetation index and chlorophyll content index. A2, which fuses high-resolution visible light images with calculated vegetation indices at the data layer or feature layer to form multi-source data fusion features; A3 inputs multi-source data fusion features and field microenvironment data from the same period into a deep convolutional neural network for training. The attention mechanism is used to enhance the model’s attention to early subtle features of pests and diseases, including lesions and hidden pests.

4. The intelligent identification and control system for crop diseases and pests based on UAV collaboration according to claim 1, characterized in that: The collaborative reconnaissance path adopted by the reconnaissance drone swarm is generated based on the "ant colony optimization algorithm" or "particle swarm optimization algorithm". Its optimization objectives include the shortest reconnaissance time, the maximum area coverage, and the priority reconnaissance weight for areas with historically high incidence of pests and diseases.

5. The intelligent identification and control system for crop diseases and pests based on UAV collaboration according to claim 1, characterized in that: The spraying mechanism of the drone cluster is a pulse width modulation controlled nozzle array. The control module generates a corresponding pulse width modulation control signal matrix based on the severity level of each pixel in the pest hotspot distribution map. This matrix controls the switching frequency and duration of each nozzle.

6. The intelligent identification and control system for crop diseases and pests based on UAV collaboration according to claim 5, characterized in that: During flight operations, the onboard visual sensors of the drone swarm monitor the drift of the pesticide caused by the downwash airflow in real time and feed the drift data back to the control module. The control module adjusts the flight altitude, nozzle switching timing, or formation of the drone swarm based on the real-time wind speed and direction and drift data to reduce pesticide drift.

7. The intelligent identification and control system for crop diseases and pests based on UAV collaboration according to claim 1, characterized in that: The prevention and control strategy library includes pre-set prevention and control schemes, including chemical control and biological control. When a pest is identified and its population density is below the economic threshold, the biological control scheme is selected first, and the operation drone cluster releases natural enemy organisms. When the severity of the pest exceeds the preset threshold, the chemical control scheme is activated.

8. The intelligent identification and control system for crop diseases and pests based on UAV collaboration according to claim 1, characterized in that: The system also includes a digital twin module, which constructs a virtual mapping of farmland based on historical operation data, real-time field data, and crop growth models. Before implementing the prevention and control plan, the digital twin module simulates, predicts, and evaluates the prevention and control effect, and optimizes and adjusts the plan parameters based on the simulation results.

9. The intelligent identification and control system for crop diseases and pests based on UAV collaboration according to claim 1, characterized in that: Both the first and second communication modules support 5G or LTE Cat-M cellular network communication. In areas with no cellular network signal coverage, a self-organizing network communication mode is adopted, in which the reconnaissance drone cluster and the operation drone cluster build a temporary multi-hop relay communication network on their own.

10. The intelligent identification and control system for crop diseases and pests based on UAV collaboration according to claim 1, characterized in that: In the interface provided by the user interaction terminal, the distribution map of pest and disease hotspots is overlaid on the farmland map in the form of a heat map. Users can use gestures to select or exclude specific areas, modify the operation path automatically generated by the system, or forcibly set certain areas as no-fly zones / no-spray zones.

Images