A real-time control method for precise pesticide application of a plant protection unmanned aerial vehicle

Abstract

The present invention provides a real-time control method for precise pesticide application of a plant protection unmanned aerial vehicle, which relates to the technical field of agricultural pesticide application. Specifically, it includes: dividing the farmland to be sprayed with pesticides into several sub-regions, collecting the plant spectral images of each region, extracting the reflectance data of the key bands, obtaining the average leaf thickness and the average photosynthetic rate, and calculating the leaf structure density index, the plant growth health index and the water-light synergy evaluation index in combination with the reflected wave data, and then generating the regional pest and disease evaluation index. Based on the historical fertilization data and the crop health status, a pesticide application prediction model is established using a long short-term memory network, and the regional pest and disease evaluation index is used as the input to predict the pesticide application amount and method. The environmental correction factor is calculated in combination with the environmental parameters to correct the pesticide application amount, ensuring that the unmanned aerial vehicle accurately completes the pesticide application task. The present invention effectively improves the accuracy and efficiency of pesticide application.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural pesticide application, and particularly to a real-time control method for precise pesticide application of a plant protection unmanned aerial vehicle (UAV). Background Art

[0002] In modern agricultural production, the prevention and control of pests and diseases not only relate to the yield and quality of crops, but also directly affect the economic income of farmers. Traditional pesticide application methods usually rely on experience and often adopt fixed pesticide application rates and frequencies. However, this method has many deficiencies. First, uneven pesticide application results in excessive amounts of pesticides in some areas, while other areas are severely affected by pests and diseases due to lack of sufficient protection. Second, the phenomenon of pesticide waste is widespread, causing waste of resources and environmental pollution. In addition, climate change and the heterogeneity of farmland make it difficult for fixed pesticide application patterns to adapt to different agricultural environments, thereby reducing the efficiency and effectiveness of pesticide use. The limitations of traditional methods pose greater challenges for farmers in pest and disease management, and there is an urgent need for a more precise and intelligent pesticide application solution.

[0003] In addition, existing technologies are also relatively weak in terms of real-time monitoring and data analysis capabilities. Although some technologies have been applied to plant growth monitoring and pest and disease identification, there are still deficiencies in integrating multiple biophysical parameters and environmental factors. Existing technologies often cannot obtain and process key data such as the spectral information, leaf thickness, and photosynthetic rate of plants in real time, resulting in reduced accuracy of pesticide application decisions. At the same time, there is a lack of effective models to evaluate and predict the health status of different crops under different environmental conditions, resulting in a lack of scientific basis for pesticide application strategies. Existing systems usually cannot achieve dynamic assessment of pest and disease risks and cannot adjust pesticide application strategies in a timely manner to cope with changing climate conditions and the development trend of pests and diseases. This technical deficiency seriously affects the sustainability and economic benefits of agricultural production, and there is an urgent need to introduce new technologies to improve the efficiency and precision of pest and disease management.

[0004] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and therefore it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a real-time control method for precise pesticide application of a plant protection UAV to solve the problems raised in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] A real-time control method for precise pesticide application of a plant protection UAV, the specific steps include:

[0008] S1. Uniformly divide the farmland to be sprayed with pesticides into several sub - regions. Use a plant protection UAV to collect the plant spectral images of each sub - region, extract the reflectance data of the key bands from the spectral images, and at the same time obtain the average leaf thickness data and average photosynthetic rate of the sub - regions;

[0009] S2. Calculate the chlorophyll content, normalized difference vegetation index (NDVI), plant water content, and leaf area index of each sub - region according to the reflection wave data, and perform an interaction term calculation on the average leaf thickness data and the leaf area index to generate a leaf structure density index; Generate a plant growth health index based on the chlorophyll content and the normalized difference vegetation index, and generate a water - light synergy evaluation index based on the plant water content and the average photosynthetic rate;

[0010] S3. Combine the plant growth health index, the water - light synergy evaluation index, and the leaf structure density index to generate a regional pest and disease assessment index;

[0011] S4. Collect a number of historical farmland fertilization data and regional pest and disease assessment indexes, map the regional pest and disease assessment index in each sample to the corresponding farmland fertilization data one by one to generate a sample data set. The farmland fertilization data includes the amount of pesticides applied and the spraying method. Establish a neural network prediction model based on the long short - term memory network (LSTM) model. Use the regional pest and disease assessment index in the sample data set as the input and the corresponding farmland fertilization data in the sample data set as the label to train the neural network prediction model to obtain a pesticide application prediction model;

[0012] S5. Input the regional pest and disease assessment index of the sub - region into the trained pesticide application prediction model. The model outputs the predicted value of the farmland fertilization data for the sub - region to be sprayed with pesticides. Obtain the environmental parameters of the farmland to be sprayed with pesticides, calculate the environmental correction factor based on the obtained environmental parameters, correct the amount of pesticides applied in the predicted value of the farmland fertilization data according to the environmental correction factor to obtain the accurate value of the amount of pesticides applied. According to the accurate value of the amount of pesticides applied and the fertilization method in the predicted value of the farmland fertilization data, operate the UAV to complete the pesticide application for each sub - region in the farmland to be sprayed with pesticides. The environmental parameters include the average temperature and the average wind speed.

[0013] Further, the key bands refer to the bands in the spectral image that can reflect the chlorophyll content, normalized difference vegetation index, plant water content, and leaf area index;

[0014] Calculate the plant chlorophyll content of the sub - region according to the reflection wave data. The formula for calculating the chlorophyll content is as follows:

[0015] ;

[0016] Where, is the chlorophyll content, is the reflectance at the 660 nm band, is the reflectivity in the 450nm band, is a constant, used to reflect the influence of chlorophyll content on the reflectivity ratio;

[0017] Calculate the plant water content based on the reflected wave data. The formula for calculating the plant water content is as follows:

[0018] ;

[0019] where, is the plant water content, is the reflectivity in the 970nm band, is the reflectivity in the 660nm band, is a constant, used to reflect the influence of plant water content on the reflectivity ratio;

[0020] Calculate the leaf area index based on the reflected wave. The formula is as follows:

[0021] ;

[0022] where, is the leaf area index, is the light attenuation index, obtained from experimental data, is the reflectivity of the area without leaves in the 660nm band, is the reflectivity of the area with leaves in the 660nm band;

[0023] Calculate the normalized difference vegetation index based on the reflected wave. The formula is as follows:

[0024] ;

[0025] where, is the normalized difference vegetation index, is the near-infrared reflectivity, is the red light reflectivity;

[0026] Collect the average photosynthetic rate. The specific logic is as follows: Collect three samples of the same mass of plants in the sub-region, place them in a closed chamber, apply the same light conditions, record the change in carbon dioxide over a certain period of time, and calculate the photosynthetic rate. Take the average of the photosynthetic rates of the three samples as the average photosynthetic rate, denoted as ;

[0027] The formula for calculating the photosynthetic rate is as follows:

[0028] ;

[0029] where, is the photosynthetic rate, For the change of concentration in the gas chamber during the measurement period, is the measurement period, is the effective photosynthetic area of the leaf;

[0030] The specific logic for collecting the average leaf thickness data is as follows: Collect the plant leaf thickness data in the sub-region multiple times and take the average value as the average leaf thickness data of the sub-region, denoted as .

[0031] Furthermore, calculate the interaction term of the dimensionless processed average leaf thickness data and the leaf area index to generate the leaf structure density index. The formula is as follows:

[0032] ;

[0033] In the formula, is the leaf structure density index, is the average leaf thickness data, is the leaf area index;

[0034] Generate the plant growth health index based on the dimensionless processed chlorophyll content and the normalized difference vegetation index. The formula is as follows:

[0035] ;

[0036] Among them, is the plant growth health index, is the chlorophyll content, is the normalized difference vegetation index, , are the weight coefficients of the chlorophyll content and the normalized difference vegetation index respectively, , and satisfy ;

[0037] Generate the water-light synergy evaluation index based on the dimensionless processed plant water content and the average photosynthetic rate. The formula is as follows:

[0038] ;

[0039] Among them, is the water-light synergy evaluation index, is the average photosynthetic rate, is the photosynthetic rate reference threshold, is the plant water content, is the water content reference threshold.

[0040] ​Further, by combining the plant growth health index, the water-light synergy evaluation index, and the leaf structure density index, an initial regional pest and disease evaluation index is generated, and the formula is as follows:

[0041] ;

[0042] Among them, is the regional pest and disease evaluation index, is the natural constant, is the leaf structure density index, is the plant growth health index, is the water-light synergy evaluation index, , and are the weight coefficients of the leaf structure density index, the plant growth health index, and the water-light synergy evaluation index respectively, , and satisfy .

[0043] Further, a neural network prediction model is established based on the long short-term memory network (LSTM) model, and an activation function and an optimization algorithm are selected. Among them, the Tanh function is selected as the activation function, and Adam is selected as the optimization algorithm for the LSTM model; the formula of the Tanh function is:

[0044] ;

[0045] In the formula, represents the Tanh function, and the independent variable represents the weighted sum of the inputs of the neuron, that is, the result after the inputs received by the neuron from the previous layer are weighted and summed; at the same time, the hyperparameters of the LSTM model are set. The hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the batch processing quantity, and the number of neurons in the hidden layer; among them, the number of network layers is set to a three-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the batch processing quantity is set to 256, and the number of neurons in the hidden layer is 32;

[0046] The regional pest and disease evaluation index of the sub-region is input into the trained pesticide application prediction model, and the model outputs the predicted value of the farmland fertilization data of the sub-region to be sprayed with pesticides. The predicted value of the farmland fertilization data includes the amount of pesticides and the fertilization method .

[0047] Further, the environmental parameters of the farmland to be sprayed with pesticides are obtained. The specific logic is as follows: The environmental temperature and wind speed data of the farmland to be sprayed with pesticides are collected multiple times, and the average environmental temperature is used as the average temperature of the farmland, denoted as , the average wind speed is taken as the average wind speed of the farmland and denoted as , and the environmental temperature refers to the atmospheric temperature at a vertical height of 1.5 - 2 m from the ground;

[0048] After dimensionless processing of the obtained environmental parameters, the environmental correction factor is calculated according to the following formula:

[0049] ;

[0050] Wherein, is the environmental correction factor, is the average temperature, is the average wind speed.

[0051] Furthermore, according to the environmental correction factor, the amount of pesticide applied in the predicted value of farmland fertilization data is corrected to obtain the accurate value of the amount of pesticide applied. The specific formula is:

[0052] ;

[0053] Wherein, is the accurate value of the amount of pesticide applied, is the amount of pesticide applied in the predicted value of farmland fertilization data, e is the natural constant, is the environmental correction factor, is the set weight value, and ;

[0054] According to the accurate value of the amount of pesticide applied and the fertilization method in the predicted value of farmland fertilization data , operate the drone to complete the pesticide application for each sub - area in the farmland to be sprayed.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] By conducting detailed regional division and health assessment on the farmland to be sprayed, the present invention can ensure that the amount of pesticide applied in each sub - area matches the actual demand, reduce the waste of pesticides, and reduce the risk of environmental pollution. Secondly, the prediction model constructed based on the long - short - term memory network can accurately predict the amount of pesticide applied and the spraying method, making the pesticide application process more scientific and systematic. This method not only improves the management efficiency of crop health but also provides a basis for farmers to optimize pesticide application decisions, ultimately achieving the improvement of crop yield and quality. Through the correction of environmental parameters, the pesticide application process can better adapt to changing environmental conditions, further improving the accuracy and reliability of crop health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a schematic diagram of the overall method flow of the present invention. Detailed implementation manners

[0058] To make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with specific embodiments.

[0059] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meanings understood by those with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar terms used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. The terms such as "connected" or "linked" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left", "right" are only used to represent relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0060] Embodiment:

[0061] Please refer to Figure 1 , the present invention provides a technical solution:

[0062] A real-time control method for precise pesticide application of a plant protection unmanned aerial vehicle, the specific steps include:

[0063] S1, evenly divide the farmland to be sprayed with pesticides into several sub-regions, use a plant protection unmanned aerial vehicle to collect the plant spectral images of each sub-region, extract the reflectance data of the key bands from the spectral images, and at the same time obtain the average leaf thickness data and average photosynthetic rate of the sub-regions;

[0064] In this embodiment, the key band refers to the band in the spectral image that can reflect the chlorophyll content, normalized difference vegetation index, plant water content and leaf area index;

[0065] Calculate the plant chlorophyll content of the sub-region according to the reflection wave data, and the formula for calculating the chlorophyll content is as follows:

[0066] ;

[0067] Among them, is the chlorophyll content, is the reflectance of the 660nm band, is the reflectance of the 450nm band, is a constant used to reflect the influence of chlorophyll content on the reflectance ratio, and is obtained by referring to relevant ecological and plant physiology research literature Reference value; This formula uses the form of the reflectance ratio, which can effectively eliminate the influence of external factors such as different lighting conditions and sensor differences on the reflectance. Moreover, the ratio can better reflect the relative chlorophyll content rather than the absolute value, thus improving the stability and consistency of the data.

[0068] Calculate the plant water content based on the reflected wave data. The formula for calculating the plant water content is as follows:

[0069] ;

[0070] Where, is the plant water content, is the reflectance at the 970nm band, is the reflectance at the 660nm band, is a constant used to reflect the influence of the plant water content on the reflectance ratio, and its reference value is obtained by referring to relevant ecological and plant physiology research literature; Reference value; This formula uses the form of the reflectance ratio, which can effectively eliminate the influence of external factors such as different lighting conditions and sensor differences on the reflectance. Moreover, the ratio can better reflect the relative plant water content rather than the absolute value, thus improving the stability and consistency of the data.

[0071] Calculate the leaf area index based on the reflected wave. The formula is as follows:

[0072] ;

[0073] Where, is the leaf area index, is the light attenuation index, obtained from experimental data, is the reflectance of the area without leaves at the 660nm band, is the reflectance of the area with leaves at the 660nm band; This formula uses the natural logarithm to represent the ratio of the reflectances because the attenuation of light is exponential. According to the Beer-Lambert law, the relationship between the intensity of light and distance can be described by an exponential function. Therefore, taking the logarithm can linearize it when calculating the leaf area index, which is convenient for calculation and understanding; and reflects the change in reflectance between the leafy and non-leafy states. The ratio of the reflectances can eliminate the influence of some external factors, making the calculation results more stable and reliable.

[0074] Calculate the normalized difference vegetation index based on the reflected wave. The formula is as follows:

[0075] ;

[0076] Where, is the Normalized Difference Vegetation Index, is the near-infrared reflectance, is the red light reflectance;

[0077] The specific logic for collecting the average photosynthetic rate is as follows: Three plants of the same mass are collected as samples in the sub-region, placed in a closed chamber, the same light conditions are applied, the change in carbon dioxide is recorded over a certain period of time, and the photosynthetic rate is calculated. The average value of the photosynthetic rates of the three samples is used as the average photosynthetic rate, denoted as ;

[0078] Among them, the formula for calculating the photosynthetic rate is:

[0079] ;

[0080] Among them, is the photosynthetic rate, is the change in the concentration of in the chamber during the measurement period, is the measurement period, is the effective photosynthetic area of the leaf;

[0081] The specific logic for collecting the average leaf thickness data is as follows: The leaf thickness data of the plants in the sub-region are collected multiple times, and the average value is taken as the average leaf thickness data of the sub-region, denoted as .

[0082] The advantage of Step 1 is that by evenly dividing the farmland to be sprayed into several sub-regions, it is possible to achieve detailed monitoring and evaluation of the growth conditions of crops in different regions. This regional management method can effectively identify the distribution of pests and diseases compared with the traditional overall spraying method, avoiding the waste of pesticides and unnecessary environmental impacts, and thus realizing a more precise spraying strategy.

[0083] Compared with the prior art, the beneficial effect of Step 1 is that high-resolution plant spectral images are collected using a plant protection drone, and the reflectance data of key bands are extracted. Combining the acquisition of the average leaf thickness and photosynthetic rate, a rich set of biophysical parameter data is formed. This data-driven approach makes the subsequent pest and disease assessment and spraying decision-making more scientific and precise, and can significantly improve the efficiency of crop health management. In the present invention, adopting Step 1 not only provides basic data support for subsequent data analysis and model construction, but also lays a solid foundation for the implementation of the overall scheme. Through targeted spraying management, it is possible to reduce the amount of pesticides used and environmental impacts while ensuring crop health and increasing yields, promoting the sustainable development of agriculture.

[0084] S2. Calculate the chlorophyll content, normalized difference vegetation index (NDVI), plant water content, and leaf area index (LAI) for each sub-region based on the reflected wave data. Calculate the interaction term between the leaf average thickness data and the leaf area index to generate the leaf structure density index. Generate the plant growth health index based on the chlorophyll content and the normalized difference vegetation index, and generate the water-light synergy evaluation index based on the plant water content and the average photosynthetic rate.

[0085] In this embodiment, the interaction term is calculated between the dimensionless leaf average thickness data and the leaf area index to generate the leaf structure density index. The formula is as follows:

[0086] ;

[0087] In the formula, is the leaf structure density index, which is used to characterize the comprehensive effect of the relative thickness and quantity of plant leaves per unit area. is the leaf average thickness data. is the leaf area index. The formula uses a logarithmic function, which can effectively handle non-linear relationships, and the logarithmic transformation can reduce the influence of outliers.

[0088] Thicker leaves usually mean that the plant can better accumulate water and nutrients and is in a relatively healthy state. A high is also related to the good health and strong competitiveness of the plant because it can use light and other resources more effectively. Therefore, , and are positively correlated.

[0089] Generate the plant growth health index based on the dimensionless chlorophyll content and the normalized difference vegetation index. The formula is as follows:

[0090] ;

[0091] Among them, is the plant growth health index. is the chlorophyll content. is the normalized difference vegetation index. , are the weight coefficients of the chlorophyll content and the normalized difference vegetation index, respectively. , and satisfy ; This is because the chlorophyll content is a direct indicator of plant photosynthesis and directly reflects the photosynthetic ability of the plant. The higher the chlorophyll content, usually the better the growth condition of the plant, so a higher weight value is given to it. is the vegetation index obtained by remote sensing technology, which reflects the vegetation coverage and growth condition. Although It can provide useful information about vegetation health, but it is calculated based on multiple factors, so its indication is not as clear as directly measuring the chlorophyll content.

[0092] Using the calculation in the form of cube root makes the plant growth health index smoother and avoids the over-amplification of large values on the results. The higher the chlorophyll content, generally the better the growth condition of the plant; a high value indicates good plant growth condition; therefore, 、 and show a positive correlation.

[0093] According to the plant water content and average photosynthetic rate after dimensionless processing, a water-light synergy evaluation index is generated, and the formula is as follows:

[0094] ;

[0095] where, is the water-light synergy evaluation index, is the average photosynthetic rate, is the photosynthetic rate reference threshold, is the plant water content, is the water content reference threshold. This formula sums up the two indicators of photosynthetic rate and water content, reflecting the synergistic effect of these two factors in plant growth. Photosynthetic rate and water content are two key factors in plant growth, and their interaction will directly affect the health and growth of plants. Therefore, it is reasonable to combine them for evaluation. Squaring the synthesized result means that the influence on the synergy evaluation index shows a non-linear relationship. A small deviation may lead to a large change in the value, while a large deviation leads to a relative decrease in the value. This non-linear processing can more sensitively reflect the changes in plant status, especially the small changes when approaching the ideal state.

[0096] When increases, it means that the deviation between the average photosynthetic rate and the photosynthetic rate reference threshold increases. Too high or too low photosynthetic rate will have a negative impact on plant growth. Therefore, will decrease as increases; similarly, when increases, will also decrease; that is to say, 、 and show a negative correlation.

[0097] The advantage of Step 2 is that through in-depth analysis of the reflected wave data, it is possible to accurately calculate the plant growth indicators of each sub-region, such as chlorophyll content, normalized difference vegetation index, plant water content, and leaf area index. This refined index calculation makes the assessment of crop health more comprehensive and scientific, can better reflect the actual growth status of plants, and thus provides a solid data basis for subsequent pest and disease assessment and pesticide application decision-making.

[0098] Compared with the prior art, the beneficial effect of Step 2 is that it combines the calculation of multiple biophysical parameters to form a comprehensive evaluation system of plant growth health index and water-light synergy evaluation index. This systematic evaluation method can more effectively identify potential pest and disease risks, enhance the timeliness and accuracy of pesticide application compared with the single index relied on by traditional methods, and reduce the risk of losses caused by pests and diseases to crops. In the present invention, adopting Step 2 not only improves the accuracy and scientificity of the overall pesticide application decision-making, but also provides important support for the generation of subsequent regional pest and disease assessment indicators. Through a comprehensive evaluation of the plant health status, targeted pesticide application management can be achieved, ensuring the rational use of resources, and thus promoting the sustainable development of agriculture.

[0099] S3. Combine the plant growth health index, water-light synergy evaluation index, and leaf structure density index to generate a regional pest and disease assessment index;

[0100] In this embodiment, the plant growth health index, water-light synergy evaluation index, and leaf structure density index are combined to generate an initial regional pest and disease assessment index, and the formula is as follows:

[0101] ;

[0102] Among them, is the regional pest and disease assessment index, is the natural constant, is the leaf structure density index, is the plant growth health index, is the water-light synergy evaluation index, , and are the weight coefficients of the leaf structure density index, plant growth health index, and water-light synergy evaluation index respectively, , and satisfy . This is because is directly related to the overall health status of the plant. Healthy plants are usually more resistant to pests and diseases. Therefore, when assessing the risks of diseases and pests, should be given a higher weight to reflect its core role in pest and disease resistance; while the leaf structure density index It reflects the structural characteristics of plant leaves and affects photosynthetic efficiency, water transpiration, etc. However, although the structural characteristics of leaves are important, their influence is relatively small in overall plant health and pest and disease resistance, so a relatively low weight value is assigned to them.

[0103] This formula makes full use of the rapid growth, positive value characteristics of the exponential function and the advantages of linear combination, which helps to comprehensively evaluate the relationship between plant health status and pest and disease risks. This form not only ensures the interpretability of the model but also enhances its application effect in actual management. When plants are affected by pests and diseases, it may lead to smaller or thinner leaves. Therefore, a high AS value means that the plant is in good growth condition. Similarly, when or increases, it means that the plant is currently in a relatively healthy growth state and the degree of danger from pests and diseases decreases. Therefore, 、 、 and show a positive correlation.

[0104] The advantage of step 3 is that by combining the plant growth health index, the water-light synergy evaluation index and the leaf structure density index, it can generate a regional pest and disease evaluation index. This comprehensive evaluation method makes the judgment of crop health status more comprehensive and accurate. Compared with the traditional method that solely relies on a single index, it can better reflect the actual situation of pests and diseases, thus providing a scientific basis for subsequent pesticide application decisions.

[0105] Compared with the prior art, the beneficial effect embodied in step 3 is that through multi-dimensional comprehensive analysis, it reduces the judgment errors caused by insufficient or one-sided information, and improves the accuracy and reliability of pest and disease evaluation. This method makes the pesticide application strategy more precise, can effectively reduce unnecessary pesticide use, reduce the impact on the environment, and at the same time improve the yield and quality of crops. In the present invention, adopting step 3 not only enhances the scientificity and effectiveness of the overall pesticide application decision-making but also provides important support for the precise management of subsequent pesticide application. By real-time evaluating the crop health status, the pesticide application plan can be adjusted in a timely manner to ensure the efficient use of resources, thus promoting the sustainable development of agricultural production.

[0106] S4. Collect a number of historical farmland fertilization data and regional pest and disease evaluation indexes, map the regional pest and disease evaluation index in each sample to the corresponding farmland fertilization data one by one to generate a sample data set. The farmland fertilization data includes the amount of pesticide applied and the spraying method. Based on the long short-term memory network (LSTM) model, establish a neural network prediction model. Use the regional pest and disease evaluation index in the sample data set as the input and the corresponding farmland fertilization data in the sample data set as the label to train the neural network prediction model to obtain a pesticide application prediction model;

[0107] In this embodiment, a neural network prediction model is established based on the long short-term memory network (LSTM) model, and an activation function and an optimization algorithm are selected. Among them, the Tanh function is selected as the activation function, and Adam is selected as the optimization algorithm for the LSTM model; the formula of the Tanh function is:

[0108] ;

[0109] In the formula, represents the Tanh function, and the independent variable represents the weighted sum of the inputs of the neuron, that is, the result of the weighted sum of the inputs received by the neuron from the previous layer; at the same time, the hyperparameters of the LSTM model are set. The hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the batch processing quantity, and the number of neurons in the hidden layer; among them, the number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the batch processing quantity is set to 256, and the number of neurons in the hidden layer is 32;

[0110] The regional pest and disease assessment index of the sub-region is input into the trained pesticide application prediction model, and the model outputs the predicted value of the farmland fertilization data of the sub-region to be applied with pesticides. The predicted value of the farmland fertilization data includes the amount of pesticides applied and the fertilization method .

[0111] The advantage of step 4 is that by constructing a long short-term memory network model, accurate pesticide application prediction based on historical farmland fertilization data and crop health status can be realized. This process correlates historical data with the current regional pest and disease assessment index, thereby generating high-accuracy predictions of the amount of pesticides applied and the spraying method, providing strong data support for pesticide application decision-making.

[0112] Compared with the prior art, the beneficial effect of step 4 is that it realizes the intelligence and precision of plant protection pesticide application by using a deep learning model. In traditional methods, pesticide application decisions often rely on experience or simple rules, while using the LSTM model can effectively process time series data, capture potential trends and changes, significantly improve the scientificity and adaptability of pesticide application, and reduce errors caused by human factors. In the present invention, adopting step 4 not only improves the accuracy of pesticide application prediction, but also lays a foundation for the automation and intelligence of the entire pesticide application process. By obtaining the regional pest and disease assessment index in real time and making dynamic adjustments in combination with historical data, it can better cope with the changing agricultural environment, ensure the rational use of resources and the healthy growth of crops, and thus promote the sustainable development of agricultural production.

[0113] S5. Input the regional pest and disease assessment index of the sub-region into the trained pesticide application prediction model. The model outputs the predicted value of the farmland fertilization data for the sub-region to be sprayed. Obtain the environmental parameters of the farmland to be sprayed, calculate the environmental correction factor based on the obtained environmental parameters, correct the amount of pesticide application in the predicted value of the farmland fertilization data according to the environmental correction factor to obtain the accurate value of the amount of pesticide application. According to the accurate value of the amount of pesticide application and the fertilization method in the predicted value of the farmland fertilization data, operate the drone to complete the pesticide application for each sub-region in the farmland to be sprayed. The environmental parameters include the average temperature and the average wind speed.

[0114] In this embodiment, the specific logic for obtaining the environmental parameters of the farmland to be sprayed is as follows: Collect the environmental temperature and wind speed data of the farmland to be sprayed multiple times. Take the average value of the environmental temperature as the average temperature of the farmland, denoted as , and take the average value of the wind speed as the average wind speed of the farmland, denoted as . The environmental temperature refers to the atmospheric temperature at a vertical height of 1.5 - 2 m from the ground;

[0115] After dimensionless processing of the obtained environmental parameters, calculate the environmental correction factor according to the following formula:

[0116] ;

[0117] Among them, is the environmental correction factor, is the average temperature, is the average wind speed. Squaring can emphasize the non-linear characteristics of the combined environmental impact of temperature and wind speed. The squaring operation is usually used to represent a certain degree of amplification effect, which is reasonable in ecology because environmental factors may have a non-linear impact on plant growth.

[0118] Correct the amount of pesticide application in the predicted value of the farmland fertilization data according to the environmental correction factor to obtain the accurate value of the amount of pesticide application. The specific formula is as follows:

[0119] ;

[0120] Among them, is the accurate value of the amount of pesticide application, is the amount of pesticide application in the predicted value of the farmland fertilization data, e is the natural constant, is the environmental correction factor, is the set weight value, and . By setting multiple experimental areas, calculate and apply the amount of pesticide application based on different weight values in different experimental areas, and select the weight value corresponding to the area with the best pest control effect as Value. Using the form of an exponential function in this formula can effectively capture the non-linear effects of environmental impacts. Due to the properties of the exponential function, when increases, the correction factor will increase in a rapidly growing manner, thus more significantly reflecting the impact of environmental factors on the amount of fertilizer application.

[0121] When the average wind speed or average temperature increases, it will lead to an increase in the kinetic energy of drug molecules, enhancing the volatility of the drug, thereby affecting the efficacy of the drug. Therefore, under the condition that other conditions remain unchanged, when increases, it is necessary to increase the amount of drug application to maintain the drug effect, that is, is positively correlated with

[0122] According to the exact value of the amount of drug application and the fertilization method in the predicted value of farmland fertilization data , operate the drone to complete the drug application to each sub-region in the farmland to be sprayed.

[0123] The advantage of step S5 is that by inputting the regional pest and disease assessment index into the trained drug application prediction model, the farmland fertilization data of the sub-regions to be sprayed can be accurately predicted, including the amount of drug application and the fertilization method. This method not only improves the accuracy of drug application, but also effectively reduces the cost of drug application, reduces resource waste, and at the same time ensures that the crops obtain the optimal nutrient supply under specific environmental conditions. Through data-driven prediction, farmland managers can make more scientific decisions based on real-time environmental and crop conditions.

[0124] Compared with the existing technology, step S5 can more comprehensively consider the dynamic changes of the farmland environment and the occurrence law of pests and diseases by introducing a deep learning model to process complex time series data. Compared with traditional experience-based drug application methods, the model can update and adjust the drug application strategy in real time, significantly improving the adaptability and efficacy of drug application. At the same time, based on the correction mechanism of the environmental correction factor, the amount of drug application is more in line with the actual needs, thereby improving the crop yield and quality. In this solution, step S5 effectively integrates the regional pest and disease assessment index and environmental parameters to form a scientific drug application decision support system, promoting the development of precision agriculture. This step not only optimizes the drug application process, but also improves the scientificity and adaptability of fertilization, making farmland management more intelligent and data-driven. The introduction of this method can significantly improve the growth health and yield of crops, and promote the development of agricultural production towards high efficiency and sustainability.

[0125] The above formulas are all dimensionless and take their numerical calculations. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formula are set by technicians in this field according to the actual situation. ​

[0126] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0127] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0128] As mentioned above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. A real-time control method for precise pesticide application of a plant protection unmanned aerial vehicle, characterized in that, The specific steps include: S1. Uniformly divide the farmland to be sprayed with pesticides into several sub - regions. Use a plant protection UAV to collect the plant spectral images of each sub - region, extract the reflectance data of the key bands from the spectral images, and at the same time obtain the average leaf thickness data and average photosynthetic rate of the sub - regions. S2. Calculate the chlorophyll content, normalized difference vegetation index (NDVI), plant water content, and leaf area index of each sub - region according to the reflection wave data, and perform an interaction calculation on the average leaf thickness data and the leaf area index to generate a leaf structure density index. Generate a plant growth health index according to the chlorophyll content and the normalized difference vegetation index, and generate a water - light synergy evaluation index according to the plant water content and the average photosynthetic rate. S3. Combine the plant growth health index, the water - light synergy evaluation index, and the leaf structure density index to generate a regional pest and disease assessment index. S4. Collect a number of historical farmland fertilization data and the regional pest and disease assessment index. Map the regional pest and disease assessment index in each sample to the corresponding farmland fertilization data one by one to generate a sample data set. The farmland fertilization data includes the amount of pesticides applied and the spraying method. Establish a neural network prediction model based on the long short - term memory network (LSTM) model. Use the regional pest and disease assessment index in the sample data set as the input and the corresponding farmland fertilization data in the sample data set as the label to train the neural network prediction model to obtain a pesticide application prediction model. S5. Input the regional pest and disease assessment index of the sub - region into the trained pesticide application prediction model. The model outputs the predicted value of the farmland fertilization data for the sub - region to be sprayed with pesticides. Obtain the environmental parameters of the farmland to be sprayed with pesticides, calculate the environmental correction factor based on the obtained environmental parameters, correct the amount of pesticides applied in the predicted value of the farmland fertilization data according to the environmental correction factor to obtain the accurate value of the amount of pesticides applied. According to the accurate value of the amount of pesticides applied and the fertilization method in the predicted value of the farmland fertilization data, operate the UAV to complete the pesticide application for each sub - region in the farmland to be sprayed with pesticides. The environmental parameters include the average temperature and the average wind speed. After dimensionless processing of the obtained environmental parameters, calculate the environmental correction factor. The formula is as follows: ENV = (T + W) 2 Among them, ENV is the environmental correction factor, T is the average temperature, and W is the average wind speed. Correct the amount of pesticides applied X in the predicted value of the farmland fertilization data according to the environmental correction factor to obtain the accurate value of the amount of pesticides applied. The specific formula is as follows: X′ = X * (1 + e μ*ENV ) Among them, X′ is the accurate value of the amount of pesticides applied, X is the amount of pesticides applied in the predicted value of the farmland fertilization data, e is the natural constant, ENV is the environmental correction factor, μ is the set weight value, and μ > 0.

2. The real-time control method for precise pesticide application of a plant protection UAV according to claim 1, characterized in that: The key band refers to the band in the spectral image that can reflect the chlorophyll content, normalized difference vegetation index, plant water content, and leaf area index. Calculate the plant chlorophyll content of the sub - region according to the reflection wave data. The formula for calculating the chlorophyll content is as follows: Among them, CHL is the chlorophyll content, and R red is the reflectance at the 660 nm band, and R blue is the reflectance at the 450 nm band, and k chl is a constant used to reflect the influence of chlorophyll content on the reflectance ratio; Calculate the plant water content according to the reflection wave data. The formula for calculating the plant water content is as follows: Among them, WC is the plant water content, R N is the reflectance at the 970 nm band, R red is the reflectance at the 660 nm band, k wc is a constant used to reflect the influence of plant water content on the reflectance ratio; Calculate the leaf area index according to the reflection wave. The formula is as follows: Among them, LAI is the leaf area index, k0 is the light attenuation index obtained according to experimental data, R0 is the reflectance of the area of plants without leaves at the 660 nm band, and R i is the reflectance of the area of plants with leaves at the 660 nm band; Calculate the normalized difference vegetation index according to the reflection wave. The formula is as follows: Among them, NDVI is the Normalized Difference Vegetation Index, NIR is the near-infrared reflectance, and R is the red-light reflectance; The specific logic for collecting the average photosynthetic rate is as follows: Three plants of the same mass are collected as samples in the sub-region, placed in a closed gas chamber, the same light conditions are applied, the change in carbon dioxide is recorded over a certain period of time, and the photosynthetic rate is calculated. The average value of the photosynthetic rates of the three samples is used as the average photosynthetic rate, denoted as PHO; Among them, the formula for calculating the photosynthetic rate is: where PHO′ is the photosynthetic rate, is the change in the CO2 concentration in the gas chamber during the measurement period, Δt is the measurement period, and A leaf is the effective photosynthetic area of the leaf; The specific logic for collecting the average leaf thickness data is as follows: The leaf thickness data of the plants in the sub-region are collected multiple times, and the average value is taken as the average leaf thickness data of the sub-region, denoted as B.

3. A real-time control method for precise pesticide application of a plant protection unmanned aerial vehicle according to claim 1, characterized in that: The interaction term of the dimensionless processed average leaf thickness data and the leaf area index is calculated to generate the leaf structure density index. The formula is as follows: AS = ln(B * LAI + 1) In the formula, AS is the leaf structure density index, B is the average leaf thickness data, and LAI is the leaf area index; The plant growth health index is generated based on the dimensionless processed chlorophyll content and the normalized difference vegetation index. The formula is as follows: Among them, BS is the plant growth health index, CHL is the chlorophyll content, NDVI is the normalized difference vegetation index, k2 and k3 are the weight coefficients of the chlorophyll content and the normalized difference vegetation index respectively, k2 > k3 > 0, and k2 + k3 = 1; The water-light synergy evaluation index is generated based on the dimensionless processed plant water content and the average photosynthetic rate. The formula is as follows: Among them, CS is the synergistic evaluation index of water and light, PHO is the average photosynthetic rate, and PHO ref is the reference threshold of photosynthetic rate, WC is the plant water content, and WC ref is the reference threshold of water content.

4. A real-time control method for precise pesticide application of a plant protection unmanned aerial vehicle according to claim 3, characterized in that: The plant growth health index, the water-light synergy evaluation index, and the leaf structure density index are combined to generate the initial regional pest and disease assessment index. The formula is: Among them, QS is the regional pest and disease assessment index, e is the natural constant, AS is the leaf structure density index, BS is the plant growth health index, CS is the water-light synergy evaluation index, ω1, ω2, and ω3 are the weight coefficients of the leaf structure density index, the plant growth health index, and the water-light synergy evaluation index respectively, ω2 > ω3 > ω1 > 0, and ω1 + ω2 + ω3 = 1.

5. A real-time control method for precise pesticide application of a plant protection unmanned aerial vehicle according to claim 1, characterized in that: A neural network prediction model is established based on the Long Short-Term Memory Network (LSTM) model, and an activation function and an optimization algorithm are selected. Among them, the Tanh function is selected as the activation function, and Adam is selected as the optimization algorithm for the LSTM model; The formula of the Tanh function is: In the formula, f(r) represents the Tanh function, and the independent variable r represents the weighted sum of the inputs of the neuron, that is, the result after the inputs received by the neuron from the previous layer are weighted and summed; At the same time, the hyperparameters of the LSTM model are set. The hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the batch processing quantity, and the number of neurons in the hidden layer; Among them, the number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the batch processing quantity is set to 256, and the number of neurons in the hidden layer is 32; Input the regional pest and disease assessment index of the sub-region into the trained pesticide application prediction model, and the model outputs the predicted values of the farmland fertilization data for the sub-regions to be sprayed with pesticides. The predicted values of the farmland fertilization data include the amount of pesticide application X and the fertilization method Y.

6. A real-time control method for precise pesticide application of a plant protection unmanned aerial vehicle according to claim 1, characterized in that: Obtain the environmental parameters of the farmland to be sprayed with pesticides. The specific logic is as follows: collect the environmental temperature and wind speed data of the farmland to be sprayed with pesticides multiple times, take the average value of the environmental temperature as the average temperature of the farmland, denoted as T, and take the average value of the wind speed as the average wind speed of the farmland, denoted as W. The environmental temperature refers to the atmospheric temperature at a vertical height of 1.5 - 2 m from the ground.

7. A real-time control method for precise pesticide application of a plant protection unmanned aerial vehicle according to claim 6, characterized in that: According to the exact value of the amount of pesticide application X′ and the fertilization method Y in the predicted values of the farmland fertilization data, operate the unmanned aerial vehicle to complete the pesticide application for each sub-region in the farmland to be sprayed with pesticides.

Images