Intelligent monitoring method and system for grassland locusts based on unmanned aerial vehicle remote sensing technology

Abstract

The invention relates to the technical field of intelligent monitoring, in particular to an intelligent monitoring method and system for grassland locusts based on an unmanned aerial vehicle remote sensing technology. The method comprises the following steps: firstly, acquiring grassland multispectral images of different time and regions by using an unmanned aerial vehicle, and acquiring a grassland feature vector group through feature extraction and classification of the multispectral images; secondly, establishing a grassland health prediction model by adopting a random forest algorithm according to the grassland feature vector group, and generating a grassland health score; then, according to the grassland health score data set and the historical locust activity data set, an intelligent locust activity prediction model is established, and the distribution probability, the distribution density, the migration speed and the activity intensity of locusts are obtained; and finally, by analyzing the grassland data set and the interactive feature group, establishing a locust outbreak early warning model, predicting locust outbreak probability, and performing early warning in three risk levels.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring technology, and in particular to an intelligent monitoring method and system for grassland locusts based on unmanned aerial vehicle remote sensing technology. Background Art

[0002] In modern agriculture and ecosystem management, locust monitoring and control is a vital task. The outbreak of locust swarms causes serious damage to grasslands and farmlands, directly affecting agricultural production and ecological balance. Traditional locust monitoring methods usually rely on ground inspections and sampling, which is not only time-consuming and labor-intensive, but also has limited coverage, making it difficult to obtain timely and comprehensive locust distribution information.

[0003] With the advancement of science and technology, the application of UAV remote sensing technology in agricultural and environmental monitoring is becoming more and more widespread. UAVs have the advantages of high efficiency, flexibility and wide coverage. They can carry a variety of sensors to collect data over a large range and with high precision. They are particularly suitable for monitoring tasks in complex terrain and vast areas. Through high-resolution remote sensing images and advanced data analysis algorithms, UAV remote sensing technology can provide accurate locust distribution data, effectively assisting in the early warning and prevention of locust disasters.

[0004] However, there are challenges in effectively identifying and classifying locust targets in remote sensing images, improving the real-time and accuracy of data processing, and comprehensively analyzing multi-source data to provide comprehensive locust monitoring information. To this end, an intelligent monitoring method and system for grassland locusts based on UAV remote sensing technology is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent monitoring method and system for grassland locusts based on unmanned aerial vehicle remote sensing technology. First, a drone is used to obtain multispectral images of grassland at different times and regions, and a grassland feature vector group is obtained by extracting and classifying the features of the multispectral images; secondly, a grassland health prediction model is established based on the grassland feature vector group using a random forest algorithm to generate a grassland health score; then, based on a grassland health score data set and a historical locust activity data set, a locust activity intelligent prediction model is established to obtain the distribution probability, distribution density, migration speed and activity intensity of locusts; finally, a locust outbreak early warning model is established by analyzing the grassland data set and the interactive feature group, the probability of locust outbreak is predicted, and early warning is given according to three risk levels.

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

[0007] An intelligent monitoring method for grassland locusts based on UAV remote sensing technology, comprising:

[0008] Use drones to obtain multispectral images of grassland at different times and in different areas;

[0009] Obtaining a grassland feature vector group and a grassland data set by preprocessing and image analysis of the grassland multispectral image;

[0010] Establishing a grassland health prediction model based on the grassland feature vector group and generating a grassland health score;

[0011] Based on the grassland health score dataset and the historical locust activity dataset, an intelligent prediction model for locust activity was established to obtain locust distribution probability, locust distribution density, locust migration speed and locust activity intensity;

[0012] Based on the locust activity intelligent prediction model, by analyzing the grassland data set, a locust outbreak warning model is established to predict the probability of locust outbreak and to warn of early locust outbreaks; wherein the specific formula for the probability of locust outbreak is:

[0013]

[0014] Where, P(B(t)) is the probability of locust outbreak at time t, E(t) is the intensity of locust activity at time t, H(t) is the grassland health score at time t, P(t) is the precipitation at time t, Y(t) is the grassland stress response at time t, T(t) is the interactive feature of temperature at time t, ξ0 is the initial value of the logistic regression algorithm, ξ1 is the weight of the intensity of locust activity at time t, ξ2 is the weight of the interactive feature of grassland health score and precipitation at time t, and ξ3 is the weight of the interactive feature of grassland stress response and temperature at time t;

[0015] According to the locust outbreak probability, a locust outbreak monitoring and early warning level is obtained.

[0016] Preferably, the preprocessing includes radiation correction, geometric correction and image stitching;

[0017] The radiation correction is used to correct the noise and brightness problems in the radiation value of the grassland multispectral image; the geometric correction is used to eliminate the geometric distortion caused by the grassland multispectral image acquisition process; the image stitching is used to seamlessly connect the grassland multispectral images of different areas to generate a complete image covering a large area;

[0018] The image analysis extracts and classifies features of the grassland multispectral image, and uses the band information of the grassland multispectral image to perform spectral feature analysis to obtain a first grassland feature vector group; the first grassland feature vector group includes an actual grassland health score, grassland coverage ratio, soil moisture content, surface temperature and pest and disease area ratio; the actual grassland health score is obtained through a grassland vegetation index and an enhanced grassland vegetation index; the grassland data set includes grassland stress response data and data on the impact of the environment on locust activity.

[0019] Preferably, the missing values ​​of the vectors in the first grassland feature vector group are filled with mean values ​​and normalized to obtain a standard grassland feature vector group;

[0020] The correlation coefficient between each characteristic variable and the grassland health score is calculated by using the standard grassland characteristic vector group; the specific calculation formula is:

[0021]

[0022] Among them, r j is the correlation coefficient between the jth column feature vector of the grassland feature vector group and the actual grassland health score, x ij is the i-th row data of the j-th column eigenvector, is the mean of the jth column eigenvector, y ij is the i-th row data of the j-th column eigenvector, is the mean of the eigenvectors in the jth column;

[0023] The grassland feature vector group is obtained according to the correlation coefficient, and a grassland health prediction model is established; the grassland health prediction model includes:

[0024] The grassland feature vector group is divided into a training set and a test set in a ratio of 8:2;

[0025] Randomly and recursively extract m Bootstrap sample subsets from the training set, and construct m decision trees accordingly; in each Bootstrap sample subset, randomly select grassland feature vectors for training; based on the decision trees, integrate through the random forest algorithm, use the training set data to train the grassland health prediction model, and establish the relationship between the Bootstrap sample subset parameters and the grassland health score; input the test set into the grassland health prediction model for testing, optimize the grassland health prediction model according to the test results, obtain the optimal grassland health prediction model, and generate a grassland health score;

[0026] The specific formula for the grassland health score is:

[0027]

[0028] Among them, H(t) is the grassland health score at time t, x1(t) is the grassland vegetation index at time t, x2(t) is the enhanced grassland vegetation index at time t, x4(t) is the soil moisture content at time t, x6(t) is the proportion of pest and disease areas at time t, ω1 is the grassland vegetation index weight, ω2 is the enhanced grassland vegetation index weight, ω4 is the soil moisture weight, and ω6 is the pest and disease area proportion weight.

[0029] Preferably, the historical locust activity data set includes the time, location and number of historical locust activities, and the grassland health score data set is the grassland health score corresponding to the time of the historical locust activities; the historical locust activity data set and the grassland health score data set are analyzed by establishing the locust activity intelligent prediction model to obtain the locust distribution probability, locust distribution density, locust migration speed and locust activity intensity;

[0030] The locust activity intelligent prediction model is trained by the historical locust activity data set and the grassland feature vector group, and the specific steps include:

[0031] Data preprocessing: normalize all data to obtain a preprocessed set;

[0032] Dataset division: The preprocessing set is divided into a training set and a test set in a ratio of 7:3;

[0033] Training model: inputting the training set into the locust activity intelligent prediction model until the model training stop condition is met to obtain an initial locust activity intelligent prediction model; the model training stop condition is that the loss function value of the locust activity intelligent prediction model reaches convergence;

[0034] Testing model: inputting the test set into the locust activity intelligent prediction model for testing, optimizing the locust activity intelligent prediction model according to the test results, and obtaining the optimal locust activity intelligent prediction model;

[0035] The locust activity intelligent prediction model includes a preprocessing layer, a feature analysis layer and a parameter prediction layer;

[0036] The preprocessing layer is used to perform preprocessing operations on the historical locust activity data set to obtain a standard data set of historical locust activities; the preprocessing operations include data cleaning, standardization, normalization and missing value processing;

[0037] The feature analysis layer includes an LSTM network structure, which is used to analyze the historical locust activity standard data set to obtain the historical locust activity feature vector and the grassland feature vector;

[0038] The parameter prediction layer includes four output neurons, each of which corresponds to a prediction target; the prediction targets include locust distribution probability, locust distribution density, locust migration speed and locust activity intensity.

[0039] Preferably, the calculation formula for the locust distribution probability is:

[0040] R(t)=β1ln(1+P(t))+β2T(t)+β3W(t)+β4P(t)H(t);

[0041] Among them, R(t) is the locust distribution probability at time t, P(t) is the precipitation distribution at time t, T(t) is the temperature distribution at time t, W(t) is the wind direction distribution at time t, H(t) is the grassland health score distribution at time t, β1 is the precipitation distribution at time t, β2 is the temperature distribution at time t, β3 is the wind direction distribution at time t, and β4 is the grassland health score distribution at time t;

[0042] The calculation formula of the locust distribution density is:

[0043]

[0044] Where D(t) is the locust distribution density within the drone shooting range at time t, A(t) is the area of ​​the drone shooting range at time t, N(t) is the number of locusts within the drone shooting range at time t, x3(t) is the grassland coverage ratio at time t, α1 is the area of ​​the drone shooting range at time t, α2 is the number of locusts within the drone shooting range at time t, and α3 is the weight of the grassland coverage ratio at time t;

[0045] The calculation formula of the locust migration speed is:

[0046]

[0047] Among them, M(t) is the locust migration speed at time t, θ(t) is the wind direction at time t, γ1 is the wind direction distribution at time t, γ2 is the grassland coverage ratio at time t, γ3 is the temperature distribution at time t, and γ4 is the precipitation distribution at time t;

[0048] The calculation formula for the locust activity intensity is:

[0049] E(t)=κ1R(t)+κ2D(t)+κ3M(t);

[0050] Among them, E(t) is the locust activity intensity at time t, κ1 is the locust distribution probability at time t, κ2 is the locust distribution density at time t, and κ3 is the weight of the locust migration speed at time t.

[0051] Preferably, an interactive feature group is established based on the grassland feature vector group and the grassland data set; the interactive feature group includes interactive features of grassland health score and precipitation and grassland stress response and temperature;

[0052] Data analysis is performed on the grassland feature vector group and the grassland data set, and the locust outbreak warning model is constructed based on a logistic regression algorithm; the interactive feature group and the obtained locust activity intensity are input into the locust outbreak warning model to obtain the probability of the locust outbreak, thereby generating the locust outbreak monitoring and warning level.

[0053] Preferably, the locust outbreak monitoring and early warning levels are divided into three levels, namely low risk level, medium risk level and high risk level;

[0054] Set the locust outbreak monitoring and early warning level threshold set TH = {p1, p2}. When the probability of the locust outbreak is less than p1, the corresponding locust outbreak monitoring and early warning is the low risk level; when the probability of the locust outbreak is in the interval [p1, p2), the corresponding locust outbreak monitoring and early warning is the medium risk level; when the probability of the locust outbreak is not less than p2, the corresponding locust outbreak monitoring and early warning is the high risk level; wherein the probability of the locust outbreak ranges from [0, 1], and p1 and p2 are 0.3 and 0.6, respectively.

[0055] An intelligent monitoring system for grassland locusts based on UAV remote sensing technology, including:

[0056] An image acquisition unit, used for acquiring multispectral images of grassland at different times and in different areas by using a drone;

[0057] An image processing unit, used for obtaining a grassland feature vector group and a grassland data set by preprocessing and image analysis of the grassland multispectral image;

[0058] A grassland scoring unit, used to establish a grassland health prediction model according to the grassland feature vector group and generate a grassland health score;

[0059] A model building unit is used to build an intelligent prediction model for locust activity based on the grassland health score data set and the historical locust activity data set to obtain locust distribution probability, locust distribution density, locust migration speed and locust activity intensity;

[0060] A locust outbreak prediction and warning unit is used to establish a locust outbreak warning model based on the locust activity intelligent prediction model and by analyzing the grassland data set, so as to predict the probability of locust outbreak and give early warning of locust outbreak;

[0061] According to the locust outbreak probability, a locust outbreak monitoring and early warning level is obtained.

[0062] Preferably, the locust outbreak prediction and early warning unit establishes an interactive feature group based on the grassland feature vector group and the grassland data set; the interactive feature group includes interactive features of grassland health score and precipitation, and grassland stress response and temperature;

[0063] Data analysis is performed on the grassland feature vector group and the grassland data set, and the locust outbreak warning model is constructed based on a logistic regression algorithm; the interactive feature group and the obtained locust activity intensity are input into the locust outbreak warning model to obtain the probability of the locust outbreak; wherein the specific formula for the probability of locust outbreak is:

[0064]

[0065] Among them, P(B(t)) is the probability of locust outbreak at time t, E(t) is the intensity of locust activity at time t, H(t) is the grassland health score at time t, P(t) is the precipitation at time t, Y(t) is the grassland stress response at time t, T(t) is the interactive feature of temperature at time t, ξ0 is the initial value of the logistic regression algorithm, ξ1 is the weight of the locust activity intensity at time t, ξ2 is the weight of the interactive feature of grassland health score and precipitation at time t, and ξ3 is the weight of the interactive feature of grassland stress response and temperature at time t.

[0066] Preferably, the locust outbreak prediction and early warning unit warns of an early locust outbreak through the probability of the locust outbreak; the locust outbreak monitoring and early warning levels are divided into three levels, namely, low risk level, medium risk level and high risk level;

[0067] Set the locust outbreak monitoring and early warning level threshold set TH = {p1, p2}. When the probability of the locust outbreak is less than p1, the corresponding locust outbreak monitoring and early warning is the low risk level; when the probability of the locust outbreak is in the interval [p1, p2), the corresponding locust outbreak monitoring and early warning is the medium risk level; when the probability of the locust outbreak is not less than p2, the corresponding locust outbreak monitoring and early warning is the high risk level; wherein the probability of the locust outbreak ranges from [0, 1], and p1 and p2 are 0.3 and 0.6, respectively.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1. The present invention obtains grassland health status, grassland coverage, soil moisture content, temperature distribution, pest and disease information and grassland stress response through multispectral images, and establishes a grassland health prediction model; the model generates a grassland health score by analyzing important parameters that affect grassland health, thereby improving the accuracy of grassland health prediction in real time and providing key basic data for intelligent prediction of locust activities.

[0070] 2. The present invention proposes an intelligent prediction model for locust activity; the model determines the distribution probability, distribution density and migration speed of locusts in the monitoring area by comprehensively analyzing time, precipitation, temperature, wind direction and speed, grassland health score, grassland coverage ratio and detection area; at the same time, through the comprehensive analysis of the distribution probability, distribution density and migration speed of locusts, the intensity of locust activity can be accurately calculated.

[0071] 3. The present invention constructs a locust outbreak warning model by real-time monitoring of grassland stress response and environmental condition changes, combining grassland health score with precipitation, grassland stress response with temperature and interactive characteristics. The model predicts the probability of locust outbreak by using a logistic regression algorithm, and provides early warning according to the probability of outbreak, which is divided into three levels: low risk, medium risk and high risk. The model can give early warning of locust outbreaks, thus improving the efficiency and effectiveness of locust disaster prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A schematic flow chart of an intelligent monitoring method for grassland locusts based on unmanned aerial vehicle remote sensing technology provided in an embodiment of the present invention;

[0073] Figure 2 A schematic diagram of the structure of an intelligent monitoring system for grassland locusts based on unmanned aerial vehicle remote sensing technology provided by an embodiment of the present invention;

[0074] Figure 3 A schematic diagram of the structure of an intelligent prediction model for locust activity provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] The present invention provides an intelligent monitoring method for grassland locusts based on UAV remote sensing technology. This method is applied to an intelligent monitoring system for grassland locusts based on UAV remote sensing technology. The flowchart of the specific method and system is shown in FIG. Figure 1 and Figure 2 .

[0077] Embodiment 1

[0078] This embodiment collects multispectral images of grassland at different times and in different areas of grassland A monitoring area, including green light band, red light band, near infrared band, red edge band, short wave infrared band and thermal infrared band; among which the green light band is used for chlorophyll content and grassland health assessment, the red light band is used for grassland density and health monitoring, the near infrared band is used for grassland biomass and health assessment and locust infestation identification, the red edge band is used for grassland stress state and health level analysis, the short wave infrared band is used for soil and grassland moisture content assessment and the thermal infrared band is used for grassland surface temperature monitoring and locust activity identification.

[0079] As an embodiment of the present invention, refer to Figure 1 S10 in the figure is applied to an image acquisition unit of an intelligent monitoring system for grassland locusts based on UAV remote sensing technology, and is used to use a UAV to acquire multispectral images of grassland at different times and in different areas.

[0080] Reference Figure 1 S20 in the figure is applied to an image processing unit of an intelligent monitoring system for grassland locusts based on unmanned aerial vehicle remote sensing technology, and is used to obtain a grassland feature vector group and a grassland data set by preprocessing and image analyzing the grassland multispectral image.

[0081] Further, the preprocessing includes radiation correction, geometric correction and image stitching;

[0082] The radiation correction is used to correct the noise and brightness problems in the radiation value of the grassland multispectral image; the geometric correction is used to eliminate the geometric distortion caused by the grassland multispectral image acquisition process; the image stitching is used to seamlessly connect the grassland multispectral images of different areas to generate a complete image covering a large area;

[0083] The image analysis extracts and classifies features of the grassland multispectral image, and uses the band information of the grassland multispectral image to perform spectral feature analysis to obtain a first grassland feature vector group; the first grassland feature vector group includes an actual grassland health score, grassland coverage ratio, soil moisture content, surface temperature and pest and disease area ratio; the actual grassland health score is obtained through a grassland vegetation index and an enhanced grassland vegetation index; the grassland data set includes grassland stress response data and data on the impact of the environment on locust activity.

[0084] In this embodiment, through preprocessing steps such as radiation correction, geometric correction and image stitching, noise, brightness problems and geometric distortion in the image are eliminated, ensuring the high quality and accuracy of the image data. The spectral characteristics of each band are used for feature extraction and classification to generate a grassland feature vector group including grassland vegetation index, enhanced grassland vegetation index, grassland coverage ratio, soil moisture content, surface temperature and pest and disease area ratio, providing a reliable data basis for subsequent grassland health scoring and locust activity prediction. These data provide important basic information for further intelligent prediction of locust activity.

[0085] Reference Figure 1 S30 in the figure is applied to a grassland scoring unit of an intelligent monitoring system for grassland locusts based on drone remote sensing technology, and is used to establish a grassland health prediction model based on the grassland feature vector group to generate a grassland health score.

[0086] Further, the missing values ​​of the vectors in the first grassland feature vector group are filled with mean values ​​and normalized to obtain a standard grassland feature vector group;

[0087] The correlation coefficient between each characteristic variable and the grassland health score is calculated by using the standard grassland characteristic vector group; the specific calculation formula is:

[0088]

[0089] Among them, r j is the correlation coefficient between the jth column feature vector of the grassland feature vector group and the actual grassland health score, x ij is the i-th row data of the j-th column eigenvector, is the mean of the jth column eigenvector, y ij is the i-th row data of the j-th column eigenvector, is the mean of the eigenvectors in the jth column;

[0090] In this embodiment, by calculating the Pearson correlation coefficient between each characteristic variable and the grassland health score, the characteristic variables that have the greatest impact on the grassland health prediction can be preliminarily screened out. These characteristic variables with high correlation can be used as the main input in the subsequent feature selection and model training process to improve the accuracy of the prediction model.

[0091] The grassland feature vector group is obtained according to the correlation coefficient, and a grassland health prediction model is established; the grassland health prediction model includes:

[0092] The grassland feature vector group is divided into a training set and a test set in a ratio of 8:2;

[0093] Randomly and recursively extract m Bootstrap sample subsets from the training set, and construct m decision trees accordingly; in each Bootstrap sample subset, randomly select grassland feature vectors for training; based on the decision trees, integrate through the random forest algorithm, use the training set data to train the grassland health prediction model, and establish the relationship between the Bootstrap sample subset parameters and the grassland health score; input the test set into the grassland health prediction model for testing, optimize the grassland health prediction model according to the test results, obtain the optimal grassland health prediction model, and generate a grassland health score;

[0094] The specific formula for the grassland health score is:

[0095]

[0096] Among them, H(t) is the grassland health score at time t, x1(t) is the grassland vegetation index at time t, x2(t) is the enhanced grassland vegetation index at time t, x4(t) is the soil moisture content at time t, x6(t) is the proportion of pest and disease areas at time t, ω1 is the grassland vegetation index weight, ω2 is the enhanced grassland vegetation index weight, ω4 is the soil moisture weight, and ω6 is the pest and disease area proportion weight.

[0097] In this embodiment, the model is trained using the training set data by using the random forest algorithm to establish the relationship between the characteristic parameters and the grassland health score. After the model training is completed, the performance of the model is evaluated using the validation set, the model parameters are adjusted according to the evaluation results, and the model is retrained using the optimized parameters. Finally, the trained random forest model is applied to the newly collected grassland data to generate grassland health score evaluation data.

[0098] Reference Figure 1 S40 and S30 are used in the model building unit of the intelligent monitoring system for grassland locusts based on UAV remote sensing technology. They are used to establish an intelligent prediction model of locust activity based on the grassland health score data set and the historical locust activity data set, and to obtain the locust distribution probability, locust distribution density, locust migration speed and locust activity intensity.

[0099] Furthermore, the historical locust activity data set includes the time, location and number of historical locust activities, and the grassland health score data set is the grassland health score corresponding to the time of the historical locust activities; by establishing the locust activity intelligent prediction model, the historical locust activity data set and the grassland health score data set are analyzed to obtain the locust distribution probability, locust distribution density, locust migration speed and locust activity intensity;

[0100] The locust activity intelligent prediction model is trained by the historical locust activity data set and the grassland feature vector group, and the specific steps include:

[0101] Data preprocessing: normalize all data to obtain a preprocessed set;

[0102] Dataset division: The preprocessing set is divided into a training set and a test set in a ratio of 7:3;

[0103] Training model: inputting the training set into the locust activity intelligent prediction model until the model training stop condition is met to obtain an initial locust activity intelligent prediction model; the model training stop condition is that the loss function value of the locust activity intelligent prediction model reaches convergence;

[0104] Testing model: inputting the test set into the locust activity intelligent prediction model for testing, optimizing the locust activity intelligent prediction model according to the test results, and obtaining the optimal locust activity intelligent prediction model;

[0105] The locust activity intelligent prediction model includes a preprocessing layer, a feature analysis layer and a parameter prediction layer;

[0106] The preprocessing layer is used to perform preprocessing operations on the historical locust activity data set to obtain a standard data set of historical locust activities; the preprocessing operations include data cleaning, standardization, normalization and missing value processing;

[0107] The feature analysis layer includes an LSTM network structure, which is used to analyze the historical locust activity standard data set to obtain the historical locust activity feature vector and the grassland feature vector;

[0108] The parameter prediction layer includes four output neurons, each of which corresponds to a prediction target; the prediction targets include locust distribution probability, locust distribution density, locust migration speed and locust activity intensity, see Figure 3 .

[0109] In this embodiment, by analyzing the historical locust activity data set and the grassland health score data set, an intelligent prediction model for locust activity is established to accurately predict the locust distribution probability, distribution density, migration speed and activity intensity. This model can not only predict each key parameter individually, but also provide comprehensive locust activity prediction information. The present invention combines deep learning with multi-objective prediction technology to provide advanced technical support and scientific basis for grassland management and locust disaster prevention and control.

[0110] Furthermore, the calculation formula of the locust distribution probability is:

[0111] R(t)=β1ln(1+P(t))+β2T(t)+β3W(t)+β4P(t)H(t);

[0112] Among them, R(t) is the locust distribution probability at time t, P(t) is the precipitation distribution at time t, T(t) is the temperature distribution at time t, W(t) is the wind direction distribution at time t, H(t) is the grassland health score distribution at time t, β1 is the precipitation distribution at time t, β2 is the temperature distribution at time t, β3 is the wind direction distribution at time t, and β4 is the grassland health score distribution at time t;

[0113] The calculation formula of the locust distribution density is:

[0114]

[0115] Where D(t) is the locust distribution density within the drone shooting range at time t, A(t) is the area of ​​the drone shooting range at time t, N(t) is the number of locusts within the drone shooting range at time t, x3(t) is the grassland coverage ratio at time t, α1 is the area of ​​the drone shooting range at time t, α2 is the number of locusts within the drone shooting range at time t, and α3 is the weight of the grassland coverage ratio at time t;

[0116] The calculation formula of the locust migration speed is:

[0117]

[0118] Among them, M(t) is the locust migration speed at time t, θ(t) is the wind direction at time t, γ1 is the wind direction distribution at time t, γ2 is the grassland coverage ratio at time t, γ3 is the temperature distribution at time t, and γ4 is the precipitation distribution at time t;

[0119] The calculation formula for the locust activity intensity is:

[0120] E(t)=κ1R(t)+κ2D(t)+κ3M(t);

[0121] Among them, E(t) is the locust activity intensity at time t, κ1 is the locust distribution probability at time t, κ2 is the locust distribution density at time t, and κ3 is the weight of the locust migration speed at time t.

[0122] In this embodiment, by integrating multi-dimensional data such as precipitation, temperature, wind speed, wind direction and grassland health, the monitoring system can predict the geographical distribution and movement trend of locusts. The proposed locust monitoring method combines multiple environmental variables and grassland health indicators, and calculates the distribution probability, distribution density, migration speed and activity intensity of locusts through accurate mathematical models, so that prevention and control measures are more targeted and timely. In addition, by calculating the distribution density and migration speed of locusts, high-risk areas can be identified, and resource allocation and emergency response strategies can be optimized.

[0123] Reference Figure 1 S50 in the invention is applied to a locust outbreak prediction and early warning unit of an intelligent monitoring system for grassland locusts based on unmanned aerial vehicle remote sensing technology, and is used to establish a locust outbreak early warning model based on the locust activity intelligent prediction model, analyze the grassland data set, predict the probability of locust outbreak, and issue an early warning of locust outbreak.

[0124] Further, an interactive feature group is established based on the grassland feature vector group and the grassland data set; the interactive feature group includes interactive features of grassland health score and precipitation and grassland stress response and temperature;

[0125] Data analysis is performed on the grassland feature vector group and the grassland data set, and the locust outbreak warning model is constructed based on a logistic regression algorithm; the interactive feature group and the acquired locust activity intensity are input into the locust outbreak warning model to predict the probability of the locust outbreak.

[0126] Among them, the specific formula for the probability of locust outbreak is:

[0127]

[0128] Among them, P(B(t)) is the probability of locust outbreak at time t, E(t) is the intensity of locust activity at time t, H(t) is the grassland health score at time t, P(t) is the precipitation at time t, Y(t) is the grassland stress response at time t, T(t) is the interactive feature of temperature at time t, ξ0 is the initial value of the logistic regression algorithm, ξ1 is the weight of the locust activity intensity at time t, ξ2 is the weight of the interactive feature of grassland health score and precipitation at time t, and ξ3 is the weight of the interactive feature of grassland stress response and temperature at time t.

[0129] In this embodiment, the locust activity intensity and the interactive features are combined and input into the early warning model, so that the system can comprehensively consider the current locust activity status and environmental conditions to predict the probability of locust outbreaks. This method can provide timely early warnings at the early stage of locust activity, help relevant departments and farmers take early prevention and control measures, and effectively reduce agricultural losses and ecological impacts caused by locusts.

[0130] Furthermore, the locust outbreak monitoring and early warning levels are divided into three levels, namely low risk level, medium risk level and high risk level;

[0131] Set the locust outbreak monitoring and early warning level threshold set TH = {p1, p2}. When the probability of the locust outbreak is less than p1, the corresponding locust outbreak monitoring and early warning is the low risk level; when the probability of the locust outbreak is in the interval [p1, p2), the corresponding locust outbreak monitoring and early warning is the medium risk level; when the probability of the locust outbreak is not less than p2, the corresponding locust outbreak monitoring and early warning is the high risk level; wherein the probability of the locust outbreak ranges from [0, 1], and p1 and p2 are 0.3 and 0.6 respectively. Table 1 is an example of locust outbreak probability data;

[0132] Table 1A Probability of locust outbreak in monitoring area

[0133] Intensity of locust activity Grassland Health Score Precipitation Grassland stress response temperature Probability of locust outbreak 0.31 0.82 0.53 0.3 0.4 0.35 0.69 0.46 0.80 0.5 0.6 0.77 0.25 0.84 0.47 0.4 0.5 0.40 0.75 0.32 0.91 0.6 0.7 0.92

[0134] The locust early warning system proposed in this embodiment distinguishes the risk level of locust outbreaks by setting clear probability thresholds, including three levels of low risk, medium risk and high risk. The hierarchical early warning mechanism not only quantifies the risk of locust outbreaks, but also allows relevant management departments to take corresponding prevention and control measures according to different risk levels. In addition, this system effectively improves the ability of early intervention by real-time monitoring of the interactive data of environmental variables and locust activities, which is of vital importance for preventing the expansion of locust scale and minimizing agricultural losses and ecological damage.

[0135] In this embodiment, firstly, through multispectral image acquisition technology, the present invention can obtain high-precision image data of grassland at different times and regions; through multi-band monitoring, rich data support is provided for comprehensive analysis of grassland ecosystems, and the monitoring accuracy of grassland health status is significantly improved. Secondly, pre-processing steps such as radiation correction, geometric correction and image stitching are adopted to eliminate noise, uneven brightness and geometric distortion problems in the image, ensure the high quality of data, and provide a solid data foundation for the construction of grassland health scoring and locust activity prediction model. In addition, by combining historical locust activity data with grassland health scoring data, deep learning and random forest algorithms are used to establish an intelligent locust activity prediction model. The model can accurately predict the distribution probability, density, migration speed and activity intensity of locusts, providing a scientific basis for the prevention and control of locust disasters. Finally, through the locust outbreak warning model, based on the locust activity intensity and grassland data set, the probability of locust outbreak can be effectively predicted and early warning can be carried out. The early warning system adopts a grading mechanism to divide the risk of locust outbreak into three risk levels: low risk, medium risk and high risk, helping managers to quickly identify high-risk areas and take prevention and control measures in advance. This early warning function has greatly improved the initiative and effectiveness of preventing and controlling locust disasters and reduced losses in agricultural production.

[0136] Embodiment 2

[0137] As an embodiment of the present invention, refer to Figure 1 S10 in the embodiment is applied to an image acquisition unit of an intelligent monitoring system for grassland locusts based on drone remote sensing technology, and is used to acquire multispectral images of grassland at different times and in different areas using drones. In this embodiment, multispectral images of grassland at different times and in different areas of grassland B monitoring area are acquired.

[0138] Reference Figure 1 S20 in the figure is applied to an image processing unit of an intelligent monitoring system for grassland locusts based on unmanned aerial vehicle remote sensing technology, and is used to obtain a grassland feature vector group and a grassland data set by preprocessing and image analyzing the grassland multispectral image.

[0139] Further, the preprocessing includes radiation correction, geometric correction and image stitching;

[0140] The radiation correction is used to correct the noise and brightness problems in the radiation value of the grassland multispectral image; the geometric correction is used to eliminate the geometric distortion caused by the grassland multispectral image acquisition process; the image stitching is used to seamlessly connect the grassland multispectral images of different areas to generate a complete image covering a large area;

[0141] The image analysis extracts and classifies features of the grassland multispectral image, and uses the band information of the grassland multispectral image to perform spectral feature analysis to obtain a first grassland feature vector group; the first grassland feature vector group includes an actual grassland health score, grassland coverage ratio, soil moisture content, surface temperature and pest and disease area ratio; the actual grassland health score is obtained through a grassland vegetation index and an enhanced grassland vegetation index; the grassland data set includes grassland stress response data and data on the impact of the environment on locust activity.

[0142] Reference Figure 1 S30 in the figure is applied to a grassland scoring unit of an intelligent monitoring system for grassland locusts based on drone remote sensing technology, and is used to establish a grassland health prediction model based on the grassland feature vector group to generate a grassland health score.

[0143] Further, the missing values ​​of the vectors in the first grassland feature vector group are filled with mean values ​​and normalized to obtain a standard grassland feature vector group;

[0144] The correlation coefficient between each characteristic variable and the grassland health score is calculated by using the standard grassland characteristic vector group; the specific calculation formula is:

[0145]

[0146] Among them, r j is the correlation coefficient between the jth column feature vector of the grassland feature vector group and the actual grassland health score, x ij is the i-th row data of the j-th column eigenvector, is the mean of the jth column eigenvector, y ij is the i-th row data of the j-th column eigenvector, is the mean of the eigenvectors in the jth column;

[0147] The grassland feature vector group is obtained according to the correlation coefficient, and a grassland health prediction model is established; the grassland health prediction model includes:

[0148] The grassland feature vector group is divided into a training set and a test set in a ratio of 8:2;

[0149] Randomly and recursively extract m Bootstrap sample subsets from the training set, and construct m decision trees accordingly; in each Bootstrap sample subset, randomly select grassland feature vectors for training; based on the decision trees, integrate through the random forest algorithm, use the training set data to train the grassland health prediction model, and establish the relationship between the Bootstrap sample subset parameters and the grassland health score; input the test set into the grassland health prediction model for testing, optimize the grassland health prediction model according to the test results, obtain the optimal grassland health prediction model, and generate a grassland health score;

[0150] The specific formula for the grassland health score is:

[0151]

[0152] Among them, H(t) is the grassland health score at time t, x1(t) is the grassland vegetation index at time t, x2(t) is the enhanced grassland vegetation index at time t, x4(t) is the soil moisture content at time t, x6(t) is the proportion of pest and disease areas at time t, ω1 is the grassland vegetation index weight, ω2 is the enhanced grassland vegetation index weight, ω4 is the soil moisture weight, and ω6 is the pest and disease area proportion weight.

[0153] Reference Figure 1 S40 and S30 are used in the model building unit of the intelligent monitoring system for grassland locusts based on UAV remote sensing technology. They are used to establish an intelligent prediction model of locust activity based on the grassland health score data set and the historical locust activity data set, and to obtain the locust distribution probability, locust distribution density, locust migration speed and locust activity intensity.

[0154] Furthermore, the historical locust activity data set includes the time, location and number of historical locust activities, and the grassland health score data set is the grassland health score corresponding to the time of the historical locust activities; by establishing the locust activity intelligent prediction model, the historical locust activity data set and the grassland health score data set are analyzed to obtain the locust distribution probability, locust distribution density, locust migration speed and locust activity intensity;

[0155] The locust activity intelligent prediction model is trained by the historical locust activity data set and the grassland feature vector group, and the specific steps include:

[0156] Data preprocessing: normalize all data to obtain a preprocessed set;

[0157] Dataset division: The preprocessing set is divided into a training set and a test set in a ratio of 7:3;

[0158] Training model: inputting the training set into the locust activity intelligent prediction model until the model training stop condition is met to obtain an initial locust activity intelligent prediction model; the model training stop condition is that the loss function value of the locust activity intelligent prediction model reaches convergence;

[0159] Testing model: inputting the test set into the locust activity intelligent prediction model for testing, optimizing the locust activity intelligent prediction model according to the test results, and obtaining the optimal locust activity intelligent prediction model;

[0160] The locust activity intelligent prediction model includes a preprocessing layer, a feature analysis layer and a parameter prediction layer;

[0161] The preprocessing layer is used to perform preprocessing operations on the historical locust activity data set to obtain a standard data set of historical locust activities; the preprocessing operations include data cleaning, standardization, normalization and missing value processing;

[0162] The feature analysis layer includes an LSTM network structure, which is used to analyze the historical locust activity standard data set to obtain the historical locust activity feature vector and the grassland feature vector;

[0163] The parameter prediction layer includes four output neurons, each of which corresponds to a prediction target; the prediction targets include locust distribution probability, locust distribution density, locust migration speed and locust activity intensity.

[0164] Furthermore, the calculation formula of the locust distribution probability is:

[0165] R(t)=β1ln(1+P(t))+β2T(t)+β3W(t)+β4P(t)H(t);

[0166] Among them, R(t) is the locust distribution probability at time t, P(t) is the precipitation distribution at time t, T(t) is the temperature distribution at time t, W(t) is the wind direction distribution at time t, H(t) is the grassland health score distribution at time t, β1 is the precipitation distribution at time t, β2 is the temperature distribution at time t, β3 is the wind direction distribution at time t, and β4 is the grassland health score distribution at time t;

[0167] The calculation formula of the locust distribution density is:

[0168]

[0169] Where D(t) is the locust distribution density within the drone shooting range at time t, A(t) is the area of ​​the drone shooting range at time t, N(t) is the number of locusts within the drone shooting range at time t, x3(t) is the grassland coverage ratio at time t, α1 is the area of ​​the drone shooting range at time t, α2 is the number of locusts within the drone shooting range at time t, and α3 is the weight of the grassland coverage ratio at time t;

[0170] The calculation formula of the locust migration speed is:

[0171]

[0172] Among them, M(t) is the locust migration speed at time t, θ(t) is the wind direction at time t, γ1 is the wind direction distribution at time t, γ2 is the grassland coverage ratio at time t, γ3 is the temperature distribution at time t, and γ4 is the precipitation distribution at time t;

[0173] The calculation formula for the locust activity intensity is:

[0174] E(t)=κ1R(t)+κ2D(t)+κ3M(t);

[0175] Among them, E(t) is the locust activity intensity at time t, κ1 is the locust distribution probability at time t, κ2 is the locust distribution density at time t, and κ3 is the weight of the locust migration speed at time t.

[0176] Reference Figure 1 S50 in the invention is applied to a locust outbreak prediction and early warning unit of an intelligent monitoring system for grassland locusts based on unmanned aerial vehicle remote sensing technology, and is used to establish a locust outbreak early warning model based on the locust activity intelligent prediction model, analyze the grassland data set, predict the probability of locust outbreak, and issue an early warning of locust outbreak.

[0177] Among them, the specific formula for the probability of locust outbreak is:

[0178]

[0179] Where P(B(t)) is the probability of locust outbreak at time t, E(t) is the intensity of locust activity at time t, H(t)P(t) is the interactive characteristic between grassland health score and precipitation at time t, Y(t)T(t) is the interactive characteristic between grassland stress response and temperature at time t, ξ0 is the initial value of the logistic regression algorithm, ξ1, ξ2 and ξ3 are the weights of the intensity of locust activity at time t, the interactive characteristic between grassland health score and precipitation at time t, and the interactive characteristic between grassland stress response and temperature at time t, respectively.

[0180] Further, an interactive feature group is established based on the grassland feature vector group and the grassland data set; the interactive feature group includes interactive features of grassland health score and precipitation and grassland stress response and temperature;

[0181] Data analysis is performed on the grassland feature vector group and the grassland data set, and the locust outbreak warning model is constructed based on a logistic regression algorithm; the interactive feature group and the acquired locust activity intensity are input into the locust outbreak warning model to predict the probability of the locust outbreak.

[0182] Furthermore, the locust outbreak monitoring and early warning levels are divided into three levels, namely low risk level, medium risk level and high risk level;

[0183] Table 2B Probability of locust outbreak in monitoring area

[0184] Intensity of locust activity Grassland Health Score Precipitation Grassland stress response temperature Probability of locust outbreak 0.16 0.87 0.30 0.3 0.2 0.27 0.59 0.45 0.72 0.4 0.5 0.71 0.37 0.79 0.53 0.3 0.4 0.42 0.69 0.21 0.75 0.5 0.6 0.83

[0185] Set the locust outbreak monitoring and early warning level threshold set TH = {p1, p2}. When the probability of the locust outbreak is less than p1, the corresponding locust outbreak monitoring and early warning is the low risk level; when the probability of the locust outbreak is in the interval [p1, p2), the corresponding locust outbreak monitoring and early warning is the medium risk level; when the probability of the locust outbreak is not less than p2, the corresponding locust outbreak monitoring and early warning is the high risk level; wherein the probability of the locust outbreak ranges from [0, 1], and p1 and p2 are 0.3 and 0.6 respectively. Table 2 is a data example based on the formula for the probability of locust outbreak.

[0186] In this embodiment, first, the present invention realizes the efficient collection and processing of grassland multispectral images through the application of UAV remote sensing technology. Information in different bands, such as green light, red light, near infrared, red edge, short-wave infrared and thermal infrared bands, is fully utilized for grassland health assessment, locust infestation identification and environmental stress response analysis. This multi-band, multi-temporal and multi-scale monitoring capability greatly improves the accuracy and timeliness of grassland management. Secondly, the present invention proposes radiation correction, geometric correction and image splicing preprocessing technology for grassland images to ensure the high quality and consistency of image data. This provides a reliable basis for subsequent feature extraction and classification, and greatly improves the accuracy of grassland health scoring and locust activity prediction model. In addition, by introducing the random forest algorithm, the present invention can establish a robust grassland health prediction model and realize intelligent prediction of locust activity in combination with historical data. The model can not only accurately predict the distribution probability, density and migration speed of locusts, but also classify and assess the risk of locust outbreaks through a real-time locust outbreak warning model, providing a scientific basis for the implementation of prevention and control measures. The intelligent monitoring system of the present invention significantly improves the early warning capability and response efficiency to locust disasters through the combination of multi-source data fusion and deep learning technology, reduces agricultural losses, and promotes the sustainable management of grassland ecosystems.

[0187] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent monitoring method for grassland locusts based on unmanned aerial vehicle remote sensing technology, characterized in that: include: Use drones to obtain multispectral images of grassland at different times and in different areas; Obtaining a grassland feature vector group and a grassland data set by preprocessing and image analysis of the grassland multispectral image; According to the grassland feature vector group, a grassland health prediction model is established to generate a grassland health score; Based on the grassland health score dataset and the historical locust activity dataset, an intelligent prediction model for locust activity was established to obtain locust distribution probability, locust distribution density, locust migration speed and locust activity intensity; Based on the locust activity intelligent prediction model, by analyzing the grassland data set, a locust outbreak warning model is established to predict the probability of locust outbreak and to warn of early locust outbreaks; wherein the specific formula for the probability of locust outbreak is: Where, P(B(t)) is the probability of locust outbreak at time t, E(t) is the intensity of locust activity at time t, H(t) is the grassland health score at time t, P(t) is the precipitation at time t, Y(t) is the grassland stress response at time t, T(t) is the interactive feature of temperature at time t, ξ0 is the initial value of the logistic regression algorithm, ξ1 is the weight of the intensity of locust activity at time t, ξ2 is the weight of the interactive feature of grassland health score and precipitation at time t, and ξ3 is the weight of the interactive feature of grassland stress response and temperature at time t; According to the locust outbreak probability, a locust outbreak monitoring and early warning level is obtained.

2. The intelligent monitoring method for grassland locusts based on unmanned aerial vehicle remote sensing technology according to claim 1 is characterized in that: The preprocessing includes radiation correction, geometric correction and image stitching; The radiation correction is used to correct the noise and brightness problems in the radiation value of the grassland multispectral image; the geometric correction is used to eliminate the geometric distortion caused by the grassland multispectral image acquisition process; the image stitching is used to seamlessly connect the grassland multispectral images of different areas to generate a complete image covering a large area; The image analysis extracts and classifies features of the grassland multispectral image, and uses the band information of the grassland multispectral image to perform spectral feature analysis to obtain a first grassland feature vector group; the first grassland feature vector group includes an actual grassland health score, grassland coverage ratio, soil moisture content, surface temperature and pest and disease area ratio; the actual grassland health score is obtained through a grassland vegetation index and an enhanced grassland vegetation index; the grassland data set includes grassland stress response data and data on the impact of the environment on locust activity.

3. The intelligent monitoring method for grassland locusts based on unmanned aerial vehicle remote sensing technology according to claim 1 is characterized in that: Performing mean filling on the missing values ​​of the vectors in the first grassland feature vector group and performing normalization processing to obtain a standard grassland feature vector group; The correlation coefficient between each characteristic variable and the grassland health score is calculated by using the standard grassland characteristic vector group; the specific calculation formula is: Among them, r j is the correlation coefficient between the jth column feature vector of the grassland feature vector group and the actual grassland health score, x ij is the i-th row data of the j-th column eigenvector, is the mean of the jth column eigenvector, y ij is the i-th row data of the j-th column eigenvector, is the mean of the eigenvectors in the jth column; The grassland feature vector group is obtained according to the correlation coefficient, and a grassland health prediction model is established; the grassland health prediction model includes: The grassland feature vector group is divided into a training set and a test set in a ratio of 8:2; Randomly and recursively extract m Bootstrap sample subsets from the training set, and construct m decision trees accordingly; in each Bootstrap sample subset, randomly select grassland feature vectors for training; based on the decision trees, integrate through the random forest algorithm, use the training set data to train the grassland health prediction model, and establish the relationship between the Bootstrap sample subset parameters and the grassland health score; input the test set into the grassland health prediction model for testing, optimize the grassland health prediction model according to the test results, obtain the optimal grassland health prediction model, and generate a grassland health score; The specific formula for the grassland health score is: Among them, H(t) is the grassland health score at time t, x1(t) is the grassland vegetation index at time t, x2(t) is the enhanced grassland vegetation index at time t, x4(t) is the soil moisture content at time t, x6(t) is the proportion of pest and disease areas at time t, ω1 is the grassland vegetation index weight, ω2 is the enhanced grassland vegetation index weight, ω4 is the soil moisture weight, and ω6 is the pest and disease area proportion weight.

4. The intelligent monitoring method for grassland locusts based on unmanned aerial vehicle remote sensing technology according to claim 1 is characterized in that: The historical locust activity data set includes the time, location and number of historical locust activities, and the grassland health score data set is the grassland health score corresponding to the time of the historical locust activities; the historical locust activity data set and the grassland health score data set are analyzed by establishing the locust activity intelligent prediction model to obtain the locust distribution probability, locust distribution density, locust migration speed and locust activity intensity; The locust activity intelligent prediction model is trained by using the historical locust activity data set and the grassland feature vector group, and the specific steps include: Data preprocessing: normalize all data to obtain a preprocessed set; Dataset division: The preprocessing set is divided into a training set and a test set in a ratio of 7:3; Training model: inputting the training set into the locust activity intelligent prediction model until the model training stop condition is met to obtain an initial locust activity intelligent prediction model; the model training stop condition is that the loss function value of the locust activity intelligent prediction model reaches convergence; Testing model: inputting the test set into the locust activity intelligent prediction model for testing, optimizing the locust activity intelligent prediction model according to the test results, and obtaining the optimal locust activity intelligent prediction model; The locust activity intelligent prediction model includes a preprocessing layer, a feature analysis layer and a parameter prediction layer; The preprocessing layer is used to perform preprocessing operations on the historical locust activity data set to obtain a standard data set of historical locust activities; the preprocessing operations include data cleaning, standardization, normalization and missing value processing; The feature analysis layer includes an LSTM network structure, which is used to analyze the historical locust activity standard data set to obtain the historical locust activity feature vector and the grassland feature vector; The parameter prediction layer includes four output neurons, each of which corresponds to a prediction target; the prediction targets include locust distribution probability, locust distribution density, locust migration speed and locust activity intensity.

5. The intelligent monitoring method for grassland locusts based on unmanned aerial vehicle remote sensing technology according to claim 4 is characterized in that: The calculation formula of the locust distribution probability is: R(t)=β1ln(1+P(t))+β2T(t)+β3W(t)+β4P(t)H(t); Among them, R(t) is the locust distribution probability at time t, P(t) is the precipitation distribution at time t, T(t) is the temperature distribution at time t, W(t) is the wind direction distribution at time t, H(t) is the grassland health score distribution at time t, β1 is the precipitation distribution at time t, β2 is the temperature distribution at time t, β3 is the wind direction distribution at time t, and β4 is the grassland health score distribution at time t; The calculation formula of the locust distribution density is: Where D(t) is the locust distribution density within the drone shooting range at time t, A(t) is the area of ​​the drone shooting range at time t, N(t) is the number of locusts within the drone shooting range at time t, x3(t) is the grassland coverage ratio at time t, α1 is the area of ​​the drone shooting range at time t, α2 is the number of locusts within the drone shooting range at time t, and α3 is the weight of the grassland coverage ratio at time t; The calculation formula of the locust migration speed is: Among them, M(t) is the locust migration speed at time t, θ(t) is the wind direction at time t, γ1 is the wind direction distribution at time t, γ2 is the grassland coverage ratio at time t, γ3 is the temperature distribution at time t, and γ4 is the precipitation distribution at time t; The calculation formula for the locust activity intensity is: E(t)=κ1R(t)+κ2D(t)+κ3M(t); Among them, E(t) is the locust activity intensity at time t, κ1 is the locust distribution probability at time t, κ2 is the locust distribution density at time t, and κ3 is the weight of the locust migration speed at time t.

6. The intelligent monitoring method for grassland locusts based on unmanned aerial vehicle remote sensing technology according to claim 1 is characterized in that: An interactive feature group is established according to the grassland feature vector group and the grassland data set; the interactive feature group includes interactive features of grassland health score and precipitation as well as grassland stress response and temperature; Data analysis is performed on the grassland feature vector group and the grassland data set, and the locust outbreak warning model is constructed based on a logistic regression algorithm; the interactive feature group and the obtained locust activity intensity are input into the locust outbreak warning model to obtain the probability of the locust outbreak, thereby generating the locust outbreak monitoring and warning level.

7. The intelligent monitoring method for grassland locusts based on unmanned aerial vehicle remote sensing technology according to claim 6 is characterized in that: The locust outbreak monitoring and early warning levels are divided into three levels, namely low risk level, medium risk level and high risk level; Set a locust outbreak monitoring and early warning level threshold set TH = {p1, p2}, when the probability of the locust outbreak is less than p1, the corresponding locust outbreak monitoring and early warning is the low risk level; When the probability of the locust outbreak is in the interval [p1, p2), the corresponding locust outbreak monitoring and early warning is the medium risk level; When the probability of the locust outbreak is not less than p2, the corresponding locust outbreak monitoring and early warning is the high risk level; wherein the probability of the locust outbreak ranges from [0,1], and p1 and p2 are 0.3 and 0.6 respectively.

8. An intelligent monitoring system for grassland locusts based on drone remote sensing technology, characterized in that: include: An image acquisition unit, used for acquiring multispectral images of grassland at different times and in different areas by using a drone; An image processing unit, used for obtaining a grassland feature vector group and a grassland data set by preprocessing and image analysis of the grassland multispectral image; A grassland scoring unit, used to establish a grassland health prediction model according to the grassland feature vector group and generate a grassland health score; A model building unit is used to build an intelligent prediction model for locust activity based on the grassland health score data set and the historical locust activity data set to obtain locust distribution probability, locust distribution density, locust migration speed and locust activity intensity; A locust outbreak prediction and warning unit is used to establish a locust outbreak warning model based on the locust activity intelligent prediction model and by analyzing the grassland data set, so as to predict the probability of locust outbreak and give early warning of locust outbreak; According to the locust outbreak probability, a locust outbreak monitoring and early warning level is obtained.

9. The intelligent monitoring system for grassland locusts based on unmanned aerial vehicle remote sensing technology according to claim 8 is characterized in that: The locust outbreak prediction and early warning unit establishes an interactive feature group based on the grassland feature vector group and the grassland data set; the interactive feature group includes interactive features of grassland health score and precipitation as well as grassland stress response and temperature; Data analysis is performed on the grassland feature vector group and the grassland data set, and the locust outbreak warning model is constructed based on a logistic regression algorithm; the interactive feature group and the obtained locust activity intensity are input into the locust outbreak warning model to obtain the probability of the locust outbreak; wherein the specific formula for the probability of locust outbreak is: Among them, P(B(t)) is the probability of locust outbreak at time t, E(t) is the intensity of locust activity at time t, H(t) is the grassland health score at time t, P(t) is the precipitation at time t, Y(t) is the grassland stress response at time t, T(t) is the interactive feature of temperature at time t, ξ0 is the initial value of the logistic regression algorithm, ξ1 is the weight of the locust activity intensity at time t, ξ2 is the weight of the interactive feature of grassland health score and precipitation at time t, and ξ3 is the weight of the interactive feature of grassland stress response and temperature at time t.

10. The intelligent monitoring system for grassland locusts based on drone remote sensing technology according to claim 8, characterized in that: The locust outbreak prediction and early warning unit warns of an early locust outbreak according to the probability of the locust outbreak; the locust outbreak monitoring and early warning levels are divided into three levels, namely, low risk level, medium risk level and high risk level; Set a locust outbreak monitoring and early warning level threshold set TH = {p1, p2}, when the probability of the locust outbreak is less than p1, the corresponding locust outbreak monitoring and early warning is the low risk level; When the probability of the locust outbreak is in the interval [p1, p2), the corresponding locust outbreak monitoring and early warning is the medium risk level; When the probability of the locust outbreak is not less than p2, the corresponding locust outbreak monitoring and early warning is the high risk level; wherein the probability of the locust outbreak ranges from [0,1], and p1 and p2 are 0.3 and 0.6 respectively.

Images