A precise pesticide application method and system based on pest situation forecasting and intelligent path planning

Abstract

The application discloses a precise pesticide application method and system based on pest situation forecasting and intelligent path planning, and belongs to the agricultural technology field in digital rural areas. The steps include crop pest data collection and processing, pest density evaluation, pesticide knowledge base matching, and pesticide application unmanned aerial vehicle path planning. Based on the aforementioned precise pest density distribution, firstly, through a structured knowledge base and a dynamic adjustment mechanism, intelligent and precise recommendation of pesticide types and doses is realized, so that pesticide application decisions can comprehensively consider real-time pest pressure, crop characteristics and environmental factors. Then, in the path planning link, multiple targets such as pesticide dosage, coverage rate, operation time and energy consumption are cooperatively optimized, and an intelligent algorithm is introduced to realize dynamic adaptation, so as to generate an efficient and energy-saving flight path and perform variable spraying, effectively avoiding pesticide abuse, and realizing the balance between prevention and treatment effects and environmental protection.

Description

Technical Field

[0001] This application belongs to the field of agricultural technology, and specifically relates to a precision pesticide application method based on insect pest monitoring and intelligent path planning. Background Technology

[0002] The prevention and control of agricultural pests and diseases is a crucial link in ensuring crop yield and quality. Currently, common pest monitoring methods mainly rely on manual field inspections or the use of simple trapping devices, estimating pest occurrence through visual observation and manual counting. In the pesticide application stage, with technological advancements, aerial spraying has begun to be used. These aerial sprayers typically follow pre-planned flight paths based on farmland boundaries or are equipped with basic obstacle avoidance features, thus achieving a degree of automation in pesticide application.

[0003] However, the aforementioned existing technical solutions still have significant shortcomings in practical applications. On the one hand, monitoring methods relying on manual labor or simple equipment make it difficult to achieve large-scale, high-frequency, and high-precision pest data collection. The information obtained lacks comprehensiveness, real-time accuracy, and precision, resulting in an inability to accurately grasp the spatial distribution details of pests. On the other hand, existing aerial spraying operation modes fail to deeply integrate path planning with real-time and accurate pest distribution information. Spraying decisions are also largely based on fixed experience, making it impossible to dynamically adjust pesticide types and dosages according to pest density, crop type, and environmental conditions. This easily leads to problems such as uneven pesticide coverage, excessive pesticide use, or poor control effects, and improvements are needed in terms of operational efficiency, economic cost, and environmental friendliness. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a precision pesticide application method based on pest monitoring and intelligent path planning. This method has the advantages of comprehensively considering real-time pest pressure, crop characteristics, and environmental factors to make pesticide application decisions, and intelligently and accurately recommending pesticide types and dosages.

[0005] This application provides a precise pesticide application method based on insect pest monitoring and intelligent path planning, including the following steps: Several monitoring points were set up in the target farmland, monitoring data were collected from each monitoring point, and information on the types of pests corresponding to each monitoring point was obtained. The monitoring data included at least pest image data and pest quantity data. The monitoring points were grouped according to the pest species information. The insect pest image data and insect quantity data are weighted and fused to obtain the insect density per unit area at the monitoring point; For each monitoring point of each pest species, spatial interpolation calculations are performed based on the insect density per unit area and geographical location to generate a continuous density distribution layer for that pest species. By overlaying continuous density distribution layers of various pest species, a comprehensive pest density distribution map of the target farmland is generated. Based on the pest species information, the corresponding pesticides and pesticide dosage benchmarks in the preset pesticide knowledge base are matched, and the pesticide dosage benchmarks are dynamically adjusted and calculated based on the pest density per unit area, the crop varieties of the target farmland, and environmental data to obtain the final spraying dosage per unit area. An optimization objective is set, and a multi-objective path planning solution is performed based on the comprehensive pest density distribution map, the final unit area spraying dose, the geographical information of the target farmland and the performance parameters of the aircraft to obtain the flight path. Control the aircraft to perform pesticide application operations along its flight path and the final spray dosage per unit area.

[0006] Furthermore, based on pest species information, the monitoring points were grouped according to pest species, including: After obtaining information on the main pest species corresponding to each monitoring point, a label is attached to each monitoring point to indicate the corresponding main pest type. Monitoring points with the same main pest type label are grouped together; Based on the main pest type labels corresponding to each monitoring point, the insect density per unit area of ​​each monitoring point is classified into the pest category indicated by the label.

[0007] Furthermore, the weighted fusion process includes: Assign confidence weights to the number of insects labeled for image recognition and the number of insects counted by the sensor, respectively; The effective insect count is obtained by summing the number of insects labeled by image recognition after the confidence weight is assigned and the number of insects counted by the sensor.

[0008] Furthermore, after generating a continuous pest density distribution layer, high-risk areas are identified based on a preset density threshold.

[0009] Furthermore, the pesticide knowledge base records at least the pest type, recommended pesticide, and corresponding dosage reference information.

[0010] Furthermore, the calculation of the final spraying dose per unit area includes: Based on pest species information, the corresponding recommended pesticides and baseline dosages are matched from the pesticide knowledge base; Obtain the pest density per unit area corresponding to the pest species, and calculate the density influence factor by combining it with the preset historical average pest density value. Obtain pesticide sensitivity parameters for the target crop, as well as wind speed from real-time environmental data; Based on density influencing factors, pesticide sensitivity parameters of the target crop, and wind speed in real-time environmental data, the baseline dose is comprehensively adjusted to obtain the final spraying dose per unit area.

[0011] Furthermore, a collaborative optimization method combining a decomposition-based multi-objective evolutionary algorithm and reinforcement learning is employed to solve the multi-objective path planning problem with the optimization objective. This method includes: Initialize the multi-objective evolutionary algorithm population, define the optimization objective, and decompose the optimization objective into multiple single-objective sub-problems. The optimization objective includes at least minimizing the total amount of pesticides used, maximizing the coverage of high-risk areas in the comprehensive density map, minimizing the total operation time of the UAV, and minimizing the flight energy consumption of the UAV. Using a pre-trained reinforcement learning strategy, each candidate path in the population is executed and dynamically fine-tuned in a dynamic simulation environment, and its fitness is evaluated based on the actual performance metrics generated. Based on fitness-driven population evolution, through iterative optimization, a set of Pareto optimal flight paths is finally output, which constitutes a candidate set of flight paths.

[0012] Furthermore, after obtaining the final spraying dose per unit area, a variable application prescription map is generated. The variable application prescription map correlates the final spraying dose per unit area with the geographical location of the target farmland, and is used to guide the aircraft to perform variable spraying operations along its flight path.

[0013] On the one hand, it provides a precision pesticide application system based on insect pest monitoring and intelligent path planning. The system includes: The data acquisition module is used to collect monitoring data from each monitoring point in the target farmland and obtain pest species information corresponding to each monitoring point. The monitoring data includes at least pest image data and insect quantity data. The data processing module is used to group each monitoring point according to the main pest species based on the pest species information; and to perform weighted fusion processing on the pest image data and insect quantity data to obtain the insect density per unit area of ​​the monitoring point. The density modeling module is used to perform spatial interpolation calculations for each major pest species, for each monitoring point of each major pest species, and for the unit area insect density and geographical location, to generate a continuous density distribution layer of the pest species. The continuous density distribution layers of each pest species are then superimposed to generate a comprehensive pest density distribution map of the target farmland. The pesticide matching and dosage calculation module is used to match the corresponding pesticides and pesticide dosage benchmarks in the preset pesticide knowledge base according to the pest species information, and dynamically adjust and calculate the pesticide dosage benchmarks based on the pest density per unit area, the crop variety of the target farmland, and environmental data to obtain the final spraying dosage per unit area. The path planning module is used to set optimization objectives and, based on the comprehensive pest density distribution map, the final unit area spraying dose, the geographical information of the target farmland and the performance parameters of the aircraft, to solve multi-objective path planning to obtain the flight path. The pesticide application control module is used to control the aircraft to perform pesticide application operations along the flight path and the final unit area spraying dosage.

[0014] An electronic device includes at least one processor and at least one memory, the memory being data-connected to the processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform any of the methods described above.

[0015] A computer-storeable medium storing computer instructions, which, when executed by a processor, specifically perform the steps of any of the methods described above.

[0016] A computer program product includes computer instructions that, when executed by a processor, specifically perform the steps in any of the methods described above.

[0017] Compared with the prior art, this application has the following advantages: This application effectively improves the reliability of a single data source by integrating multi-source data from insect monitoring lamp images and auxiliary sensors and performing weighted processing, thus obtaining a more accurate effective insect count. Furthermore, it uses a spatial interpolation algorithm to transform the density data of these discrete monitoring points into a continuous and visualized farmland insect density distribution map, thereby achieving a precise and intuitive characterization of the spatial heterogeneity of insect pests and providing a solid data foundation for subsequent differentiated management.

[0018] To address the issues of disconnect between path planning and pest infestation and inefficient pesticide application in existing aerial spraying operations, this application, based on the aforementioned precise pest density distribution, firstly achieves intelligent and accurate recommendation of pesticide type and dosage through a structured knowledge base and dynamic adjustment mechanism. This enables spraying decisions to comprehensively consider real-time pest pressure, crop characteristics, and environmental factors. Furthermore, in the path planning stage, multiple objectives such as pesticide dosage, coverage, operation time, and energy consumption are collaboratively optimized, and intelligent algorithms are introduced to achieve dynamic adaptation. This generates efficient and energy-saving flight paths and executes variable spraying, improving the overall efficiency and economy of the operation while effectively avoiding pesticide abuse and achieving a balance between control effectiveness and environmental protection.

[0019] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart of a method according to an embodiment of this application is shown; Figure 2 A diagram showing the specific data source of the insect monitoring lamp according to an embodiment of this application is provided. Figure 3 A flowchart of a pest density assessment method according to an embodiment of this application is shown; Figure 4 A flowchart of a spatial interpolation calculation method according to an embodiment of this application is shown; Figure 5 A flowchart illustrating a pesticide knowledge base matching method according to an embodiment of this application is shown; Figure 6 A flowchart of a path planning method according to an embodiment of this application is shown; Figure 7 A system structure diagram according to an embodiment of this application is shown. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] The solutions provided in this application are mainly applied to the scenario of precision pest control in large-scale modern agricultural production. They are especially suitable for farms, agricultural cooperatives or agricultural service companies that require efficient management and pursue green and sustainable development. In this scenario, the traditional methods that rely on manual inspection and experience-based application of pesticides can no longer meet the needs for rapid response to pests, precise control and reduction of agricultural non-point source pollution.

[0024] The existing technology has the following main problems: First, the fragmented nature of pest monitoring data makes it difficult to form a scientific understanding of the overall distribution of pests in farmland. Second, pesticide application decisions rely on fixed experience and cannot be dynamically adapted to real-time pest infestations, crop conditions and environmental conditions. Third, the low level of intelligence in the operation path planning of pesticide application equipment (such as aircraft) and its failure to integrate with the spatial heterogeneity of pests result in poor operation efficiency and resource utilization.

[0025] To address the aforementioned issues, this application constructs an intelligent closed loop encompassing perception, analysis, decision-making, and execution. This enhances the accuracy and comprehensiveness of pest monitoring, enables precise pesticide matching and variable spraying, and optimizes the operational path of the aircraft, thereby achieving multiple beneficial effects such as improved control efficacy, cost savings, and environmental protection.

[0026] The implementation process of this application will now be described in detail with reference to the accompanying drawings.

[0027] Please see Figure 1 The method of this application includes the following steps: S1. Data collection and processing of crop pests.

[0028] This step aims to address the issues of incomplete and inaccurate pest information collection.

[0029] Specifically, data is collected through a network of insect monitoring lamps deployed in farmland and a network of auxiliary counting sensors: S1-1, Multi-source data acquisition.

[0030] Data is collected through a network of insect monitoring lamps deployed in farmland and auxiliary counting sensors. Please refer to [link to relevant documentation]. Figure 2 The insect monitoring lamp includes an image sensor and an environmental monitoring unit. During daily operation, the insect monitoring lamp works automatically to attract pests and capture images of pests from multiple angles and under various lighting conditions. The environmental monitoring unit is used to collect environmental data such as temperature, humidity, light intensity, and air pressure.

[0031] At the same time, auxiliary counting sensors (such as infrared counting sensors) simultaneously record the number of pests passing through.

[0032] S1-2, Preprocessing of multi-source data.

[0033] The pest image data collected by the image sensor in step S1-1 is processed for clarity and classified and labeled to identify the number and type of pests. The clarity processing aims to make the image data clearer to ensure the accuracy of pest identification and labeling, and may include operations such as adjusting contrast, brightness, and sharpening.

[0034] Subsequently, a unique label is assigned to each type of pest (e.g., noctuid moth is labeled as 0, rice stem borer as 1, and planthopper as 2). Unique labels can also be assigned to images of healthy crops to build the basis for training and recognition. Classification and labeling can be done by agricultural technicians viewing the images to determine the species, or by a deep learning image recognition model deployed in the cloud for automatic classification.

[0035] S2. Insect density assessment.

[0036] This step aims to transform discrete insect monitoring data into continuous and visualized spatial distribution information, thereby addressing the problem of unscientific density assessment.

[0037] The data input for this step is mainly divided into two categories: insect monitoring lamp data and manual judgment data. The data sources for input are shown in the table below.

[0038]

[0039] The output data mainly consists of a pest density matrix, density layers by pest type, and high-risk area markers. The data output table is shown below.

[0040]

[0041] Please see Figure 2 Specifically, it includes the following steps: S2-1. Merge the multi-source monitoring data of each monitoring point and calculate its effective insect quantity.

[0042] Specifically, the number of pests in the images captured by insect monitoring lamps is identified to obtain the number of pests at each monitoring point. The measured number of insects obtained by directly counting auxiliary counting sensors in the target farmland. The effective insect population of each pest at its corresponding monitoring point was calculated using a weighted fusion formula. The formula is as follows:

[0043] in, The confidence weight coefficient is between 0 and 1, and can be dynamically adjusted or set empirically based on the accuracy of the image recognition algorithm. Indicates the monitoring duration.

[0044] S2-2, Calculate the insect density at the monitoring points.

[0045] Based on effective insect population and the known equivalent monitoring area of ​​each monitoring lamp The insect density at each monitoring point is calculated using the following formula:

[0046] in, Let be the insect density (insects / m²) at the location of the i-th monitoring lamp. This indicates the equivalent coverage area within the effective monitoring radius of the monitoring light. Let represent the number of insects at the location of the i-th monitoring light.

[0047] S2-3. Group the monitoring points according to the pest species information.

[0048] Based on the results of manual interpretation or automatic identification of pest images, the main pest species in the area corresponding to each monitoring point are determined, and each monitoring point is labeled with the corresponding main pest type.

[0049] Based on the main pest type label of each monitoring point, all monitoring points are classified and grouped according to pest type.

[0050] If a monitoring point has the highest number of noctuid moths, then that monitoring point is marked as a noctuid moth infestation. After classifying all monitoring points, all monitoring points that are labeled as noctuid moth infestations for the main pest types are grouped together.

[0051] S2-4. Based on the main pest type labels assigned to each monitoring point. The insect density calculated at this point in the previous step is then assigned to the pest category name corresponding to this label, serving as the basis for subsequent calculations of the density distribution of this type of pest. The formula is as follows:

[0052] in, This indicates z types of pests, each labeled with a major pest type.

[0053] This represents the label of the main pest type at the i-th monitoring point.

[0054] S2-5. Generate a continuous density distribution layer for this pest species.

[0055] Please see Figure 4 Spatial interpolation algorithms are used to interpolate the insect density per unit area of ​​all points in each monitoring group into a continuous density distribution map covering the entire farmland. Inverse distance weighted interpolation (IDW) or Kriging interpolation can be used.

[0056] Specifically, the inverse distance weighted interpolation method assumes that the value of the point to be estimated is influenced by neighboring known points, and the weight is inversely proportional to the distance. When using the inverse distance weighted interpolation method to establish a continuous density distribution map covering the entire farmland, the following calculation formula is used:

[0057] in, This represents the estimated pest density at any point (x, y) in the farmland. This refers to the number of insect monitoring lamps used in the interpolation. This represents the inverse distance weighting coefficient for monitoring point i. The calculation formula is as follows:

[0058] in, The decay exponent is an experimental value obtained from experiments or an empirical value selected based on experience. This represents the Euclidean distance between any point and monitoring point i.

[0059] Subsequently, a continuous density distribution map of the farmland was constructed based on the estimated pest density at any point in the farmland.

[0060] Compared to inverse distance weighted interpolation, the Kriging method is a more accurate geostatistical method. Its estimated interpolation points are obtained by solving a system of Kriging equations based on the variogram function. (See [reference needed]). Figure 2 The calculation steps are as follows: (1) Calculate the experimental variability function.

[0061] Specifically, based on monitoring point x i Calculate the insect density at any monitoring point in the target farmland, calculate the Euclidean distance between all pairs of monitoring points, and divide the distances into k intervals. For each distance interval Calculate the experimental variability function γ ( ).

[0062] The calculation formula for the theoretical variogram model is as follows:

[0063] Based on the theoretical variogram model, the experimental variogram γ ( The calculation formula for ) is as follows:

[0064] Where N(h) represents all pairs of points (x, h) whose Euclidean distance is equal to h. i , x j The quantity of ).

[0065] ρ(x) i ) is x i Insect density at the point.

[0066] ρ(x) i + ) for x i The Euclidean distance between points is Insect density.

[0067] (2) Fitting the theoretical model.

[0068] Regarding the experimental variability function value γ obtained in the previous step ( A continuous theoretical variogram model is determined by fitting using the least squares method. .

[0069] Specifically, a theoretical model is selected, and the parameters in the theoretical model are adjusted using the least squares method to fit and refine the theoretical variogram model. Theoretical value Compared with experimental value γ ( The sum of squared errors between the total errors is minimized, as shown in the following formula:

[0070] (3) Construct the Kriging equation system.

[0071] For any point to be interpolated within the farmland A system of linear equations is established to solve for the optimal weights, and its formula is as follows:

[0072] The above equation can also be expressed as:

[0073] in, For the weight vector, It is a Lagrange multiplier.

[0074] (4) Solve for the weight parameters.

[0075] Based on the fitted linear equations from the previous step, for each interpolation point... Calculate Lagrange multipliers and the weight vector of each interpolation point .

[0076] (5) Calculate the estimated value.

[0077] Based on the existing pest density per unit area at monitoring points Calculate the estimated value of the Kriging interpolation point. The calculation formula is as follows:

[0078] Calculated That is, the point to be interpolated. The optimal linear unbiased estimate of the pest density at a given location is obtained, and a pest density distribution map is constructed based on this estimate.

[0079] (6) Estimation of variance analysis.

[0080] The estimated variance is calculated by combining the Lagrange multipliers obtained from the previous steps and the weight vector of each interpolation point, as shown in the following formula:

[0081] Finally, the estimated variance is analyzed and reliability information is generated. The smaller the estimated variance, the better the interpolation point. The more reliable the density estimation results at a given location, the better.

[0082] S2-6. By traversing all grid points within the target farmland area, repeat the steps in S2-5 to generate a continuous final integrated density map with statistical reliability information. The calculation formula is as follows:

[0083] in, This represents the density of the z-th pest at geographical coordinates (x, y) in the farmland, where Z represents all pest species.

[0084] This represents the insect density at the i-th monitoring point.

[0085] in, This represents the overall insect density at the geographical coordinates (x, y) of the farmland.

[0086] S2-7, Identification of high-risk areas.

[0087] Set pest density threshold Areas exceeding this threshold are identified as high-risk areas.

[0088] The calculation formula is as follows:

[0089] in, This indicates a high-risk area.

[0090] S3, pesticide knowledge base matching.

[0091] This step aims to achieve intelligent pesticide recommendation and precise dosage calculation, solving the problem of the lack of intelligent matching mechanisms for pesticide use. A flowchart is attached. Figure 3 As shown.

[0092] The system involved in this embodiment has a built-in structured pesticide knowledge base, whose fields include pest type, preferred pesticide, alternative pesticide, dosage reference, environmental sensitive area restrictions, etc. An example of some content in the knowledge base is shown below:

[0093] The following is an example of a section from the Pesticide Knowledge Rules Table:

[0094] Please see Figure 5 The matching process is as follows: S3-1. Based on the identified main pest types, refer to the pesticide knowledge rule table and search the pesticide knowledge base for the baseline dosage of the preferred pesticide for that type of pest.

[0095] S3-2. The baseline dose Q is dynamically adjusted based on real-time insect infestation, crop conditions, and environmental factors, using the following formula:

[0096] in, The baseline dose in the pesticide knowledge rule table represents the recommended pesticide dose under standard conditions.

[0097] The historical average density value of pests is the average density of this pest type derived from historical data or experience, and is used as a reference value.

[0098] This represents the real-time monitored insect density.

[0099] The density-dose elasticity index represents the degree to which changes in pest density affect pesticide dosage, and is typically taken as 0.5 to 1.5.

[0100] This is the density influence factor.

[0101] This is the variety sensitivity index (pesticide sensitivity parameter). Different crop varieties may have different sensitivities to the same pesticide.

[0102] This is the wind speed compensation index.

[0103] These are parameters representing the sensitivity of crop varieties to pesticides, set according to the characteristics of the crop variety.

[0104] This refers to wind speed.

[0105] Through the above calculations, the system outputs the final pesticide dosage that matches the specific pest density, crop variety, and meteorological conditions for different areas of farmland, and generates a precision pesticide prescription map to guide variable-rate spraying by drones.

[0106] S4. Path planning for pesticide application drones.

[0107] To simultaneously optimize four conflicting objectives—minimizing pesticide use, maximizing coverage, minimizing operation time, and minimizing energy consumption—this application employs a hybrid optimization strategy combining a decomposition-based multi-objective evolutionary algorithm (MOEA / D) with proximal policy optimization (PPO) reinforcement learning to plan the pesticide application path for the drone. The data types and uses of the inputs for path planning are shown in the table below:

[0108] Please see Figure 6 Specifically, it includes the following steps: S4-1. Initialize the population for the Multi-Objective Evolutionary Algorithm (MOEA / D).

[0109] First, define four optimization objectives: (1) The total pesticide usage calculated based on the precision application prescription map of drone variable spraying is the minimum; (2) The coverage rate is highest for high-risk areas in the comprehensive density map; (3) The total operation time of the drone is the shortest; (4) Lowest flight energy consumption.

[0110] The degree of mutual constraint among the above four objectives is minimal.

[0111] Subsequently, several drone drug delivery paths that satisfy basic flight rules are randomly generated based on the input data of path planning, and these are used as the initial population of the multi-objective evolutionary algorithm (MOEA / D).

[0112] S4-2. Decompose a multi-objective problem into sub-problems.

[0113] Specifically, this step formalizes the pesticide application path planning into a multi-objective optimization problem, which is then decomposed into several sub-problems. Each sub-problem is defined by a weight vector, and different weight vectors represent different combinations of preferences for pesticide dosage, coverage, operation time, and energy consumption.

[0114] Each UAV candidate path in the multi-objective evolutionary algorithm (MOEA / D) population is associated with a specific weight vector.

[0115] S4-3. Train the Proximal Policy Optimization (PPO) agent as a dynamic policy.

[0116] S4-3-1. Initialize the dynamic simulation environment, load the digital simulation environment containing the input data of the above path planning, and reset the digital simulation environment to its initial state.

[0117] S4-3-2. Collect interaction experience data; Specifically, in the dynamic simulation environment, the PPO agent interacts with the dynamic simulation environment from the previous step based on a network architecture.

[0118] In the Proximal Policy Optimization Agent (PPO) framework, the Actor network (policy network) simulates a drone operating in a dynamic simulation environment to complete multiple interactions. Each interaction generates a drone trajectory, and for each state of the drone in each trajectory data... (Including drone position, speed, remaining drug / battery charge, surrounding obstacle information, and performance metrics of the current path), the Actor network (policy network) outputs actions. (such as fine-tuning of heading angle, changes in speed), among which, actions The Actor network samples the action probability distribution.

[0119] S4-3-3, Calculate the estimated value of the dominance function; Critic networks (value networks) are based on each state in each trajectory. Output its state estimate V( ), where V ( ) is a scalar.

[0120] Simultaneously, the reward function is used to calculate the state. Start to task end status The sum of the reward values ​​at each time step is used as the cumulative reward. Then, based on the advantage function, the current state is calculated. taking action Advantage estimate Used to quantify actions The relative advantages and disadvantages.

[0121] The formula for calculating the reward function is as follows:

[0122] in, This indicates the amount of pesticide applied by the drone along its application route; This indicates the pesticide coverage rate of the drone along its application route; This indicates the flight time of the drone along the pesticide application route; This indicates the energy consumption of the drone along the pesticide application route.

[0123] R represents the reward value of the reward function; FP represents the penalty for constraint violation; , , , These are all weighting coefficients, used in conjunction with the other units mentioned above to measure the amount of pesticide sprayed, coverage, flight time, and drone energy consumption.

[0124] The advantage function is as follows:

[0125] in, The advantage estimate represents the value in state . Select Real-Time Action Total revenue.

[0126] Discount factor; The reward value at time step t+k is obtained from the reward function; For the state value function in The estimated value at that location.

[0127] A value greater than 0 indicates that the expected benefit of taking the real-time action is better than remaining in the original state without performing the action, and its probability should be increased. If the value is less than 0, it means that the expected benefit of the real-time action is worse than not performing the action and remaining in the original state, and its probability should be reduced.

[0128] S4-3-4. Construct the PPO loss function and update the network parameters.

[0129] Utilizing the status in batch data ,action and their corresponding advantage estimates Constructing the PPO loss function :

[0130] The PPO loss function is a well-known technique in the prior art, wherein... The current drone route planning strategy is used in the state of the drone. Select action The probability of.

[0131] For old drone route planning strategies in the state of the drone Select action The probability of.

[0132] This is the trimming threshold, used to control the trimming range; The policy parameters of the Actor network are updated by minimizing the PPO loss function using the gradient descent algorithm.

[0133] This optimization process is carried out iteratively, which continuously improves the strategy parameters, and finally obtains an optimized PPO strategy that can output the optimal flight control actions based on real-time environmental conditions (such as wind speed and pest density).

[0134] S4-3-5, MOEA / D and PPO co-evolution.

[0135] The optimized PPO strategy in S4-3-4 is used as a dynamic fitness evaluator and embedded in the evolutionary loop of MOEA / D.

[0136] Specifically, for each candidate path in the MOEA / D population, static evaluation is no longer performed. Instead, it is used as a task framework, and the optimized PPO strategy is executed and dynamically fine-tuned in a dynamic simulation environment (such as wind deflection resistance and variable rate spraying). Finally, the actual amount of pesticide used, coverage, operation time, and flight energy consumption are used as the fitness evaluation value of the path and fed back to the MOEA / D algorithm. The MOEA / D algorithm performs population evolution and neighborhood cooperation under the dynamic feedback of PPO agents: Each subproblem exchanges information (such as crossover and mutation) among the solutions of its neighboring subproblems according to its weight vector, generating a new solution; After the new solution is adjusted and evaluated online by the PPO agent, it replaces the worse old solution among the neighbors; The above process is executed iteratively until the termination condition is met (such as reaching the maximum number of iterations or convergence), and finally a set of Pareto optimal or near-optimal flight path candidate sets is output.

[0137] S5. Select the final drone application path.

[0138] Based on preset decision preferences (such as a greater emphasis on pesticide reduction or operational efficiency), a final optimal flight path is automatically selected or recommended from the Pareto optimal path candidate set, and converted into a waypoint sequence and variable spraying control command sequence that can be executed by the UAV flight control system.

[0139] The aircraft is controlled to perform pesticide application along the final path. This involves adjusting the spraying system's rate or on / off status based on the final unit area spraying dose associated with each point along the path, thereby achieving targeted spraying of high-risk areas and economical spraying of low-risk areas.

[0140] As a specific embodiment, the method of selecting the final flight path from the Pareto optimal path candidate set may further include: Based on the weighted sum scoring method, weights corresponding to decision preferences are assigned to the optimization objectives (pesticide usage, coverage, operation time, energy consumption), the weighted comprehensive score of each path in the candidate set is calculated, and the path with the highest score is selected as the final path. Alternatively, the Pareto optimal path candidate set and its corresponding multi-objective performance indicators can be displayed through an interactive visualization interface, allowing decision-makers to manually review and select the path based on expert experience and actual on-site conditions. In summary, this solution addresses agricultural dynamic optimization problems by combining MOEA / D co-evolution with dynamic parameter tuning in relation to a crop pest identification model. Online adjustment of weight vectors enables real-time environmental adaptation, and hard constraints are directly integrated into the objective function to avoid the lag in feasibility fixes. Personalized recommendations are introduced during the Pareto selection phase to improve system usability. Compared to traditional methods, this approach offers the aforementioned process innovations and technical advantages.

[0141] In some embodiments, please refer to Figure 7 This application also discloses a precision pesticide application system based on insect pest monitoring and intelligent path planning, the system comprising: The data acquisition module is used to collect monitoring data from each monitoring point in the target farmland and obtain pest species information corresponding to each monitoring point. The monitoring data includes at least pest image data and pest quantity data. The data processing module is used to group each monitoring point according to the main pest species based on the pest species information; and to perform weighted fusion processing on the pest image data and insect quantity data to obtain the insect density per unit area of ​​the monitoring point. The density modeling module is used to perform spatial interpolation calculations for each major pest species, using the insect density per unit area and geographical location of monitoring points belonging to that group of major pest species, to generate a continuous density distribution layer for that pest species. The continuous density distribution layers of each pest species are then overlaid to generate a comprehensive pest density distribution map of the target farmland. The pesticide matching and dosage calculation module is used to match the corresponding pesticides and pesticide dosage benchmarks in the preset pesticide knowledge base according to the pest species information, and dynamically adjust and calculate the pesticide dosage benchmarks based on the pest density per unit area, the crop variety of the target farmland, and environmental data to obtain the final spraying dosage per unit area. The path planning module is used to set optimization objectives and, based on the comprehensive pest density distribution map, the final unit area spraying dose, the geographical information of the target farmland and the performance parameters of the aircraft, to solve multi-objective path planning to obtain the flight path. The pesticide application control module is used to control the aircraft to perform pesticide application operations along the flight path and the final unit area spraying dose.

[0142] An electronic device includes at least one processor and at least one memory, the memory being data-connected to the processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform any of the methods described above.

[0143] A computer-storeable medium storing computer instructions, which, when executed by a processor, specifically perform the steps of any of the methods described above.

[0144] A computer program product includes computer instructions that, when executed by a processor, specifically perform the steps of any of the methods described above.

[0145] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A precision pesticide application method based on insect pest monitoring and intelligent path planning, characterized in that, Includes the following steps: Several monitoring points are set up in the target farmland, monitoring data are collected from each monitoring point, and pest species information corresponding to each monitoring point is obtained. The monitoring data includes at least pest image data and pest quantity data. The insect pest image data and insect quantity data are weighted and fused to obtain the insect density per unit area at the monitoring point; Based on the pest species information, each monitoring point is grouped according to the preset pest species; For each set of monitoring points for each preset pest species, spatial interpolation calculations are performed to calculate the insect density per unit area and geographical location, generating a continuous density distribution layer for that pest species. By overlaying continuous density distribution layers of various pest species, a comprehensive pest density distribution map of the target farmland is generated. Based on the pest species information, the corresponding pesticides and pesticide dosage benchmarks in the preset pesticide knowledge base are matched, and the pesticide dosage benchmarks are dynamically adjusted and calculated based on the pest density per unit area, the crop varieties of the target farmland, and environmental data to obtain the final spraying dosage per unit area. An optimization objective is set, and a multi-objective path planning solution is performed based on the comprehensive pest density distribution map, the final unit area spraying dose, the geographical information of the target farmland, and the performance parameters of the aircraft to obtain the flight path. Control the aircraft to perform pesticide application operations along its flight path and the final spray dosage per unit area.

2. The method according to claim 1, characterized in that, Based on pest species information, the monitoring points were grouped according to pest species, including: After obtaining information on the main pest species corresponding to each monitoring point, a label is attached to each monitoring point to indicate the corresponding main pest type. Monitoring points with the same main pest type label are grouped together; Based on the main pest type labels corresponding to each monitoring point, the insect density per unit area of ​​each monitoring point is classified into the pest category indicated by the label.

3. The method according to claim 1, characterized in that, The weighted fusion process includes: Assign confidence weights to the number of insects labeled for image recognition and the number of insects counted by the sensor, respectively; The effective insect count is obtained by summing the number of insects labeled by image recognition after the confidence weight is assigned and the number of insects counted by the sensor.

4. The method according to claim 1, characterized in that, After generating the continuous pest density distribution layer, high-risk areas are identified based on a preset density threshold.

5. The method according to claim 1, characterized in that, The pesticide knowledge base records at least the pest type, recommended pesticides, and corresponding dosage reference information.

6. The method according to claim 1, characterized in that, The calculation of the final spraying dose per unit area includes: Based on the pest species information, the corresponding recommended pesticides and baseline dosages are obtained by matching from the pesticide knowledge base; Obtain the pest density per unit area corresponding to the pest species, and calculate the density influence factor by combining it with the preset historical average pest density value. Obtain pesticide sensitivity parameters for the target crop, as well as wind speed from real-time environmental data; Based on the density influencing factor, the pesticide sensitivity parameters of the target crop, and the wind speed in the real-time environmental data, the baseline dose is comprehensively adjusted to obtain the final spraying dose per unit area.

7. The method according to claim 4, characterized in that, The multi-objective path planning solution with optimization objective employs a collaborative optimization method combining a decomposition-based multi-objective evolutionary algorithm and reinforcement learning, including: Initialize the multi-objective evolutionary algorithm population, define the optimization objective, and decompose the optimization objective into multiple single-objective sub-problems. The optimization objective includes at least minimizing the total amount of pesticides used, maximizing the coverage of high-risk areas in the comprehensive density map, minimizing the total operation time of the UAV, and minimizing the flight energy consumption of the UAV. Using a pre-trained reinforcement learning strategy, each candidate path in the population is executed and dynamically fine-tuned in a dynamic simulation environment, and its fitness is evaluated based on the actual performance metrics generated. Based on the fitness-driven population evolution, through iterative optimization, a set of Pareto optimal flight paths is finally output, constituting a candidate set of flight paths.

8. The method according to claim 1, characterized in that, After obtaining the final unit area spraying dose, a variable application prescription map is generated. The variable application prescription map associates the final unit area spraying dose with the geographical location of the target farmland and is used to guide the aircraft to perform variable spraying operations on the flight path.

9. A precision pesticide application system based on insect pest monitoring and intelligent path planning, characterized in that, The system includes: The data acquisition module is used to collect monitoring data from each monitoring point in the target farmland and obtain pest species information corresponding to each monitoring point. The monitoring data includes at least pest image data and pest quantity data. The data processing module is used to group each monitoring point according to the main pest species based on the pest species information; and to perform weighted fusion processing on the pest image data and insect quantity data to obtain the insect density per unit area of ​​the monitoring point. The density modeling module is used to perform spatial interpolation calculations for each major pest species, using the insect density per unit area and geographical location of each monitoring point belonging to each major pest species, to generate a continuous density distribution layer for that pest species. The continuous density distribution layers of each pest species are then overlaid to generate a comprehensive pest density distribution map of the target farmland. The pesticide matching and dosage calculation module is used to match the corresponding pesticides and pesticide dosage benchmarks in the preset pesticide knowledge base according to the pest species information, and dynamically adjust and calculate the pesticide dosage benchmarks based on the pest density per unit area, the crop variety of the target farmland, and environmental data to obtain the final spraying dosage per unit area. The path planning module is used to set optimization objectives and, based on the comprehensive pest density distribution map, the final unit area spraying dose, the geographical information of the target farmland and the performance parameters of the aircraft, to solve multi-objective path planning to obtain the flight path. The pesticide application control module is used to control the aircraft to perform pesticide application operations along the flight path and the final unit area spraying dosage.

10. An electronic device, characterized in that, The electronic device includes at least one processor and at least one memory, the memory being data-connected to the processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

11. A computer-storable medium, characterized in that, The storable medium stores computer instructions, which, when executed by a processor, specifically perform the steps of the method as described in any one of claims 1-8.

12. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they specifically perform the steps in the method as described in any one of claims 1-8.

Images