Agricultural pest risk assessment method and system

Abstract

The invention relates to the technical field of agricultural disease and insect pest monitoring, prevention and control, in particular to an agricultural disease and insect pest risk assessment method and system.The method comprises the steps that farmland multispectral data are collected through a multispectral sensor, and a farmland environment monitoring network is constructed; acquiring pest and disease feature data by adopting a pest and disease feature extraction algorithm, analyzing a relationship between environmental factors and pests by utilizing an environmental factor association model, and evaluating a pest and disease risk level based on a risk probability model; and generating an optimal prevention and treatment strategy set by using an optimization decision algorithm, and updating the risk assessment model through a parameter adaptive adjustment program. The system comprises a data acquisition and networking module, a pest and disease feature extraction module, an environmental factor correlation analysis module, a risk assessment module, a control strategy generation module and a model optimization module. Accurate assessment and dynamic prevention and control of agricultural pest and disease risks are achieved, pest and disease loss can be effectively reduced, the prevention and control cost is reduced, pesticide residues are reduced, and the intelligent level and economic benefits of agricultural production are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural pest monitoring and prevention, and in particular to an agricultural pest risk assessment method and system. Background Art

[0002] In agricultural production, the outbreak of pests and diseases often poses a serious threat to crop yield and quality. Traditional pest and disease monitoring and prevention methods have many limitations and cannot meet the needs of modern agriculture for efficient, precise and sustainable development.

[0003] From the monitoring perspective, the traditional manual field inspection method not only consumes a lot of manpower, material resources and time, and is extremely inefficient, but is also limited by people's subjective judgment and experience differences, making it difficult to achieve early and accurate identification of pests and diseases and simultaneous monitoring of large areas. In some large-scale planting areas, manual inspections are difficult to cover every corner, resulting in untimely detection of pests and diseases, which may cause large-scale losses once an outbreak occurs. For example, in some large-scale orchards, manual inspections of fruit tree pests and diseases require a lot of manpower, and it is easy to miss some fruit trees in the early stages of the disease. When pests and diseases spread, the cost of governance increases greatly, and the yield and quality of the fruit will also be seriously affected.

[0004] In terms of technical means, previous pest and disease monitoring technologies were relatively simple and lacked comprehensive consideration of multiple environmental factors. Relying solely on a single physical or chemical detection method, it is impossible to fully obtain relevant information on the occurrence of pests and diseases. For example, traditional pest and disease monitoring may only focus on the morphological characteristics of pests and diseases, while ignoring the impact of environmental factors such as temperature, humidity, and light intensity on the occurrence and development of pests and diseases. In fact, these environmental factors often have complex relationships with the occurrence of pests and diseases. Changes in temperature may affect the reproduction rate of pests and diseases, and the level of humidity will affect the transmission path and range of pathogens. Light intensity may affect the disease resistance of crops themselves.

[0005] In terms of prevention and control strategy formulation, most of the existing prevention and control plans are based on experience, lacking scientific risk assessment and accurate decision-making basis. In actual operation, farmers may blindly increase the use of pesticides, which not only increases production costs, but also causes serious pollution to the ecological environment such as soil, water sources and air, destroying the ecological balance, and also causes excessive pesticide residues in agricultural products, endangering human health. For example, in the process of vegetable planting, excessive use of pesticides to prevent and control pests and diseases will leave a large amount of pesticide residues on vegetables, which may have adverse effects on consumers' health after consumption. Moreover, a single prevention and control method can easily make pests and diseases resistant, further increasing the difficulty of prevention and control.

[0006] With the rapid development of emerging technologies such as the Internet of Things, big data, and artificial intelligence, new opportunities have been brought to the monitoring and prevention of agricultural pests and diseases. How to apply these advanced technologies to the agricultural field to achieve accurate monitoring, scientific evaluation, and effective prevention and control of pests and diseases has become an important issue that agricultural researchers are currently working on. Based on this background, the present invention proposes a new agricultural pest and disease risk assessment method and system to make up for the shortcomings of traditional technologies and promote the modernization of agriculture. Summary of the invention

[0007] The purpose of the present invention is to provide a method and system for agricultural pest risk assessment to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: a method for risk assessment of agricultural pests and diseases, the method comprising:

[0009] Step S1: Use a multispectral sensor to collect remote sensing data of the target farmland area to obtain multispectral data of the farmland; and deploy network nodes of the multispectral sensor through the Internet of Things communication protocol to build a farmland environment monitoring network;

[0010] Step S2: using a pest and disease feature extraction algorithm to perform biological feature analysis on the farmland multispectral data to obtain pest and disease feature data; wherein the function formula of the pest and disease feature extraction algorithm is as follows:

[0011]

[0012] In the formula, is the pest and disease characteristic data, For farmland multispectral data, is the convolution kernel weight matrix of the i-th layer, is the bias term, σ is the activation function, is the feature weight coefficient;

[0013] Step S3: using an environmental factor association model to perform a multivariate regression analysis on the pest characteristic data to obtain an environmental factor association matrix; the environmental factor association model includes a multidimensional nonlinear regression equation of temperature, humidity, and light intensity;

[0014] Step S4: Performing dynamic risk assessment processing on the environmental factor correlation matrix based on a risk probability model to obtain pest risk level data; the risk probability model uses a Bayesian network to calculate conditional probability distribution;

[0015] Step S5: performing control strategy generation processing on the pest risk level data through an optimization decision algorithm to obtain an optimal control strategy set; the optimization decision algorithm includes a particle swarm optimization objective function under constraint conditions;

[0016] Step S6: using a parameter adaptive adjustment program to perform feedback learning processing on the optimal control strategy set, updating the parameter weights of the risk probability model, and obtaining a dynamically optimized risk assessment model.

[0017] Preferably, step S1 comprises the following steps:

[0018] Step S11: collecting visible light, near infrared and thermal infrared band data of the target farmland through a multispectral sensor to generate an original spectral data set;

[0019] Step S12: using a data normalization algorithm to normalize the original spectral data set to eliminate dimensional differences;

[0020] Step S13: using an outlier detection algorithm to remove outliers from the normalized spectral data to obtain pre-processed farmland multispectral data.

[0021] Preferably, step S2 comprises the following steps:

[0022] Step S21: extracting features from the farmland multispectral data through a convolutional neural network to generate a primary feature map;

[0023] Step S22: using a spatial pyramid pooling layer to perform multi-scale feature fusion on the primary feature map to obtain an enhanced feature vector;

[0024] Step S23: weight the enhanced feature vector based on the attention mechanism, and output the pest and disease feature data.

[0025] Preferably, the functional expression of the environmental factor association model in step S3 is:

[0026]

[0027] In the formula, is the probability of pests and diseases occurring, is the temperature variable, is the humidity variable, is the light intensity variable, is the intercept term, , , is the regression coefficient.

[0028] Preferably, step S4 comprises the following steps:

[0029] Step S41: constructing a Bayesian network structure and defining the causal relationship between environmental factor nodes and pest risk nodes;

[0030] Step S42: using the Markov chain Monte Carlo method to perform parameter learning on the Bayesian network and calculate the posterior probability distribution;

[0031] Step S43: Output the pest risk level data according to the posterior probability distribution, and divide it into low, medium and high risk intervals according to the preset threshold value.

[0032] Preferably, the expression of the particle swarm optimization objective function in step S5 is:

[0033]

[0034] In the formula, is the cost of the mth control measure, is the decision variable, It is a penalty item for pesticide residues. is the penalty coefficient, and are the total number of control measures and pesticide types, respectively.

[0035] Preferably, step S6 comprises the following steps:

[0036] Step S61: collecting farmland response data after the prevention and control strategy is implemented in real time through the online learning module;

[0037] Step S62: back-propagating and updating the parameters of the risk probability model using a gradient descent algorithm;

[0038] Step S63: Use a sliding window mechanism to perform weight decay on historical data to ensure the dynamic adaptability of the model.

[0039] Preferably, the convolutional neural network in step S21 includes 5 convolutional layers, 3 pooling layers and 2 fully connected layers, the convolution kernel size is 3×3, and the step size is 2.

[0040] Preferably, in step S42, a Gibbs sampling algorithm is used to approximate the posterior probability, the number of iterations is not less than 1000 times, and the convergence threshold is 0.001.

[0041] Preferably, the present invention also includes an agricultural pest risk assessment system, comprising:

[0042] Data collection and networking module: Use multispectral sensors to collect remote sensing data of the target farmland area to obtain multispectral data of the farmland, and deploy network nodes of the multispectral sensors through the Internet of Things communication protocol to build a farmland environment monitoring network;

[0043] Pest and disease feature extraction module: using pest and disease feature extraction algorithm to perform biological feature analysis and processing on the farmland multispectral data to obtain pest and disease feature data;

[0044] Environmental factor association analysis module: using the environmental factor association model to perform multivariate regression analysis on the pest characteristic data to obtain an environmental factor association matrix, wherein the environmental factor association model includes a multidimensional nonlinear regression equation of temperature, humidity, and light intensity;

[0045] Risk assessment module: based on the risk probability model, the environmental factor correlation matrix is ​​dynamically assessed to obtain the pest risk level data. The risk probability model uses a Bayesian network to calculate the conditional probability distribution;

[0046] Prevention and control strategy generation module: performing prevention and control strategy generation processing on the pest risk level data through an optimization decision algorithm to obtain an optimal prevention and control strategy set, wherein the optimization decision algorithm includes a particle swarm optimization objective function under constraint conditions;

[0047] Model optimization module: using the parameter adaptive adjustment program to perform feedback learning processing on the optimal prevention and control strategy set, updating the parameter weights of the risk probability model, and obtaining a dynamically optimized risk assessment model.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention uses a multispectral sensor to collect remote sensing data of the target farmland area and builds a farmland environment monitoring network, which can obtain multispectral data of the farmland and environmental information such as temperature, humidity, and light intensity in real time and comprehensively. Through the pest feature extraction algorithm, the multispectral data is deeply analyzed, and the pest feature data can be accurately extracted to achieve early identification and monitoring of pests and diseases. For example, when the pests and diseases just show signs, the system can monitor the subtle changes in the spectral data, so as to issue an early warning in time, buy precious prevention and control time for farmers, and avoid losses caused by large-scale outbreaks of pests and diseases. Compared with traditional manual inspections, the accuracy and timeliness of monitoring are greatly improved, and the degree of harm caused by pests and diseases is effectively reduced.

[0050] The environmental factor association model takes into account a variety of environmental factors such as temperature, humidity, and light intensity. The environmental factor association matrix is ​​obtained through multivariate regression analysis, which can more accurately reveal the complex relationship between environmental factors and the occurrence of pests and diseases. The risk probability model uses a Bayesian network to calculate the conditional probability distribution and dynamically evaluate the risk of pests and diseases. The resulting pest and disease risk level data is more scientific and reasonable. This enables farmers and agricultural managers to formulate targeted prevention and control strategies based on accurate risk assessment results, avoid blind prevention and control, and improve prevention and control effects. For example, in a high temperature and high humidity environment, the system can accurately assess the high risk of certain pests and diseases, and promptly remind farmers to take corresponding preventive measures to effectively reduce the probability of pests and diseases.

[0051] The particle swarm optimization objective function in the optimization decision algorithm takes into account the cost of prevention and control measures and the pesticide residue penalty term, and generates the optimal prevention and control strategy set by processing the pest risk level data. This can not only reduce the cost of prevention and control while ensuring the prevention and control effect, but also reduce the amount of pesticides used and the impact of pesticide residues on the environment and agricultural product quality. For example, when selecting prevention and control measures, the system will comprehensively consider the costs and effects of various methods such as physical control, biological control and chemical control, and recommend the most economical and effective prevention and control plan for farmers, which not only ensures the healthy growth of crops, but also achieves the sustainable development of agriculture. The parameter adaptive adjustment program collects farmland response data after the implementation of the prevention and control strategy in real time through the online learning module, uses the gradient descent algorithm to back-propagate and update the parameters of the risk probability model, and uses the sliding window mechanism to weight the historical data to ensure that the model can be dynamically optimized as the farmland environment and the occurrence of pests and diseases change. In this way, the system can always maintain the accuracy of pest and disease risk assessment and the effectiveness of prevention and control strategies, and adapt to the pest and disease control needs of different farmland environments and different growth stages. For example, in different seasons and under different planting varieties, the system can timely adjust the evaluation model and prevention and control strategies according to actual conditions, thereby improving the level of intelligence in agricultural production.

[0052] The agricultural pest risk assessment method and system of the present invention realizes accurate monitoring, scientific assessment and intelligent prevention of pests and diseases, effectively reduces the damage of pests and diseases to crops, and improves the yield and quality of crops. At the same time, by optimizing the prevention and control strategy, the prevention and control cost is reduced, the pesticide residue is reduced, and the market competitiveness of agricultural products is improved, thereby bringing significant economic benefits to farmers and agricultural enterprises. In addition, the automation and intelligent operation of the system greatly reduces labor input, improves agricultural production efficiency, and promotes the development of agricultural modernization. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a working principle diagram of the agricultural pest risk assessment method of the present invention;

[0054] Figure 2 This is a working principle diagram of the environmental factor association model;

[0055] Figure 3 Diagram of how the objective function calculation for particle swarm optimization works. DETAILED DESCRIPTION

[0056] 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.

[0057] See also Figure 1-3 The present invention provides a technical solution: a method for assessing agricultural pest risk, the method comprising:

[0058] Step S1: Use multispectral sensors to collect remote sensing data of the target farmland area to obtain multispectral data of the farmland; and deploy network nodes of the multispectral sensors through the Internet of Things communication protocol to build a farmland environment monitoring network. Multispectral sensors can collect information on the target farmland in different spectral bands, which contains rich information on the growth status of crops and potential pests and diseases. Connecting multiple multispectral sensors into a network through the Internet of Things communication protocol can realize real-time transmission and sharing of data, which is convenient for subsequent processing.

[0059] Step S2: Use the pest feature extraction algorithm to perform biological feature analysis on the farmland multispectral data to obtain pest feature data. The algorithm extracts key features related to pests and diseases from a large amount of farmland multispectral data through a specific formula operation to provide data support for subsequent analysis. The formula is:

[0060] In the formula, is the pest and disease characteristic data, For farmland multispectral data, For the Layer convolution kernel weight matrix, is the bias term, is the activation function, is the feature weight coefficient.

[0061] Step S3: Perform multivariate regression analysis on the pest characteristic data using the environmental factor association model to obtain an environmental factor association matrix. The environmental factor association model takes into account multiple environmental factors such as temperature, humidity, and light intensity, and analyzes the relationship between these factors and the pest characteristic data through a multidimensional nonlinear regression equation to determine the degree of association between each environmental factor and the occurrence of pests.

[0062] Step S4: Based on the risk probability model, the environmental factor correlation matrix is ​​dynamically evaluated to obtain pest risk level data. The risk probability model uses a Bayesian network to calculate the conditional probability distribution, calculates the risk probability of pests and diseases under different circumstances according to the environmental factor correlation matrix, and divides the corresponding risk levels.

[0063] Step S5: Generate control strategies for the pest risk level data through an optimization decision algorithm to obtain an optimal control strategy set. The optimization decision algorithm includes a particle swarm optimization objective function under constraints, and finds the best combination of control strategies based on factors such as control costs and pesticide residues.

[0064] Step S6: Use the parameter adaptive adjustment program to perform feedback learning processing on the optimal control strategy set, update the parameter weights of the risk probability model, and obtain a dynamically optimized risk assessment model. By monitoring the farmland after the implementation of the control strategy, obtaining relevant data feedback, and adjusting the parameters of the risk probability model, the model can more accurately assess the risk of pests and diseases.

[0065] The present invention will be further described below in conjunction with Examples 1 to 5:

[0066] Embodiment 1:

[0067] This embodiment mainly preprocesses the collected raw spectral data, eliminates dimensional differences in the data and removes outliers, improves data quality, and provides a reliable data basis for subsequent analysis.

[0068] In step S1, it is further refined as follows:

[0069] Step S11: Collect visible light, near infrared and thermal infrared band data of the target farmland through a multispectral sensor to generate an original spectral data set. The multispectral sensor has multiple detection channels, each corresponding to a different spectral band. During the acquisition process, ensure that the sensor is in normal working condition, scan the farmland at preset time intervals, and obtain comprehensive and continuous spectral data. For example, in a farmland with an area of ​​100 mu, data is collected every 1 hour for 24 hours to obtain spectral information at different time periods.

[0070] Step S12: Use a data normalization algorithm to normalize the original spectral data set to eliminate dimensional differences. The data normalization algorithm uses a common normalization formula, such as ,in is the original data, and are the minimum and maximum values ​​in the original data set, respectively. In this way, data of different dimensions are uniformly mapped to the [0, 1] interval, making data of different bands comparable. For example, for a certain eigenvalue in the visible light band, its original range is [100, 500]. After normalization, the eigenvalue is mapped to the [0, 1] interval, which is convenient for subsequent data analysis and model calculation.

[0071] Step S13: Use an outlier detection algorithm to remove outliers from the normalized spectral data to obtain preprocessed farmland multispectral data. The outlier detection algorithm can use the 3σ principle based on statistics, that is, data points that exceed the range of the mean plus or minus 3 times the standard deviation are considered outliers. For the data of each spectral band, calculate its mean and standard deviation, and mark and remove the data points that exceed the range. For example, in the near-infrared band data, the calculated mean is 0.5 and the standard deviation is 0.1. If a data point is 0.9, which exceeds If the data point is out of the range, it will be regarded as an outlier and removed. After this series of preprocessing steps, the obtained farmland multispectral data is more accurate and reliable, providing high-quality data support for subsequent pest and disease feature extraction.

[0072] Embodiment 2:

[0073] This embodiment describes in detail the specific process of pest and disease feature extraction. Through convolutional neural networks, spatial pyramid pooling layers and attention mechanisms, pest and disease features can be more effectively extracted from farmland multispectral data, thereby improving the accuracy and effectiveness of feature extraction.

[0074] In step S2, the specific implementation process is as follows:

[0075] Step S21: Extract features from the farmland multispectral data through a convolutional neural network to generate a primary feature map. In this embodiment, the convolutional neural network includes 5 convolutional layers, 3 pooling layers and 2 fully connected layers, the convolution kernel size is 3×3, and the step size is 2. In the convolutional layer, the 3×3 convolution kernel slides on the farmland multispectral data with a step size of 2, performs a convolution operation on the data, and extracts local features from the data. For example, for a multispectral image of size 256×256, after being processed by the first convolutional layer, due to the setting of the convolution kernel size and step size, the image size will become 127×127, and a series of primary image features, such as edges, textures and other information, are extracted. After layer-by-layer extraction of 5 convolutional layers, rich primary features are obtained, and these features are combined into a primary feature map.

[0076] Step S22: Use the spatial pyramid pooling layer to perform multi-scale feature fusion on the primary feature map to obtain an enhanced feature vector. The spatial pyramid pooling layer divides the primary feature map into regions of different scales, such as 1×1, 2×2, and 4×4 regions, performs a pooling operation on each region, and fuses features of different scales. In this way, image features at different scales can be obtained and feature information is enriched. For example, pooling operations are performed in a 1×1 region to retain the most representative features of the region; pooling operations are performed in a 2×2 region to integrate the feature information of the four small regions in the region. These pooling results of different scales are spliced ​​together to form an enhanced feature vector, which contains more comprehensive image feature information.

[0077] Step S23: Based on the attention mechanism, the enhanced feature vector is weighted and the feature data of pests and diseases is output. The attention mechanism performs weighted processing on the feature vector by calculating the importance weight of each element in the enhanced feature vector. For elements with high correlation with the features of pests and diseases, a larger weight is assigned, while for elements with low correlation, a smaller weight is assigned. For example, after calculation, it is found that some elements in the enhanced feature vector are closely related to specific symptoms of pests and diseases (such as leaf color changes, texture abnormalities, etc.), then these elements are assigned higher weights, so that the pest and disease feature data finally output can more prominently reflect the characteristic information of pests and diseases. In this way, the accuracy and pertinence of pest and disease feature extraction are improved.

[0078] Embodiment 3:

[0079] This embodiment clarifies the specific functional expression of the environmental factor association model, and how to use the model to perform multivariate regression analysis to determine the relationship between environmental factors and the probability of occurrence of pests and diseases, providing a scientific basis for subsequent risk assessment.

[0080] In step S3, the functional expression of the environmental factor association model is:

[0081] In the formula, is the probability of pests and diseases occurring, is the temperature variable, is the humidity variable, is the light intensity variable, is the intercept term, , , is the regression coefficient.

[0082] In practical applications, a large amount of farmland environmental data is first collected, including temperature, humidity, light intensity data in different time periods, and corresponding pest and disease occurrence data. For example, in multiple farmlands in a certain area, environmental data and pest and disease occurrence data within a growing season are continuously monitored, and a total of 100 sets of valid data are collected. Then, these data are substituted into the environmental factor association model, and multivariate regression analysis is performed using statistical analysis software (such as SPSS, R language, etc.). During the analysis process, the error between the model's predicted value and the actual observed value is minimized by adjusting the regression coefficient. For example, after multiple iterative calculations, it was determined that , , (Assuming there are two temperature variables), , Through this model, we can accurately analyze the influence of environmental factors such as temperature, humidity, and light intensity on the probability of occurrence of pests and diseases. For example, when the temperature rises, according to The changing trend and corresponding regression coefficient , we can judge the change of the probability of pests and diseases; the square term of humidity Reflects the nonlinear effect of humidity on the probability of pests and diseases; light intensity variable pass The functional form of more accurately reflects the effect of light intensity on the probability of pests and diseases in different ranges. In this way, the environmental factor correlation matrix is ​​obtained, which provides important data support for subsequent risk assessment.

[0083] Embodiment 4:

[0084] This embodiment describes in detail the risk assessment process based on the Bayesian network, including network structure construction, parameter learning, and risk level classification, so that the risk assessment results are more scientific and accurate, and provide a reliable decision-making basis for pest control.

[0085] In step S4, the specific implementation steps are as follows:

[0086] Step S41: Construct a Bayesian network structure and define the causal relationship between the environmental factor node and the pest risk node. Based on professional knowledge and practical experience in the agricultural field, determine the environmental factors (such as temperature, humidity, and light intensity) as the parent node and the pest risk as the child node. For example, in the Bayesian network structure of a certain pest, different values ​​of the temperature node (such as low temperature, normal temperature, and high temperature) will affect the probability distribution of the pest risk node, and the same is true for the humidity and light intensity nodes. By establishing these causal relationships, a complete Bayesian network structure is constructed, which intuitively shows the relationship between environmental factors and pest risks.

[0087] Step S42: The Markov chain Monte Carlo method is used to learn the parameters of the Bayesian network and calculate the posterior probability distribution. In this embodiment, the Gibbs sampling algorithm is used to approximate the posterior probability, the number of iterations is not less than 1000 times, and the convergence threshold is 0.001. First, the parameters in the Bayesian network are initialized according to the collected environmental factor data and the occurrence data of pests and diseases. Then, multiple iterations of sampling are performed in the Bayesian network by the Gibbs sampling algorithm. In each iteration, the state of each node is updated according to the current state of other nodes, and the posterior probability distribution is gradually approached. For example, at the 100th iteration, the parameters in the network gradually tend to be stable, but have not yet reached the convergence threshold; continue to iterate, when the number of iterations reaches more than 1000 times, the change in the network parameters is less than the convergence threshold of 0.001, and the posterior probability distribution is considered to have converged. In this way, the posterior probability distribution of the occurrence of pests and diseases under different environmental factor combinations is accurately calculated.

[0088] Step S43: Output the pest and disease risk level data according to the posterior probability distribution, and divide the low, medium and high risk intervals according to the preset threshold. Set different risk thresholds according to actual needs and experience. For example, the case where the posterior probability is less than 0.3 is divided into a low risk interval, the case between 0.3-0.7 is divided into a medium risk interval, and the case greater than 0.7 is divided into a high risk interval. According to the calculated posterior probability distribution, determine the pest and disease risk level corresponding to each data point. For example, if the posterior probability of a data point is 0.8, it is determined that the farmland is in a high risk state and prevention and control measures need to be taken in time. In this way, the quantitative assessment and grading of pest and disease risks are realized, providing clear guidance for the subsequent formulation of prevention and control strategies.

[0089] Embodiment 5:

[0090] This embodiment describes how to generate the optimal control strategy set according to the pest risk level, and optimize the risk assessment model through feedback learning, so that the entire pest risk assessment and control system is more perfect and can adapt to different farmland environments and pest conditions.

[0091] In step S5, the expression of the particle swarm optimization objective function is:

[0092] In the formula, For the The cost of the control measures, is the decision variable, It is a penalty item for pesticide residues. is the penalty coefficient, and are the total number of control measures and pesticide types, respectively.

[0093] In practical applications, we first determine the costs of various control measures, such as the cost of using pesticides, the cost of manual control, etc. At the same time, we consider the residues of different pesticides and determine the pesticide residue penalty. For example, the cost of a certain pesticide is Yuan / mu, and the corresponding pesticide residue penalty item is (Determined according to the toxicity and residue standards of pesticides). Then, the particle swarm optimization algorithm is used to solve the objective function and obtain the optimal decision variables under certain constraints (such as control effect requirements, resource limitations, etc.). Combination means determining which control measures to adopt and the types and dosages of pesticides to use, thereby generating an optimal set of control strategies.

[0094] In step S6, the specific implementation process is as follows:

[0095] Step S61: Collect farmland response data after the implementation of the prevention and control strategy in real time through the online learning module. After the implementation of the prevention and control strategy, various sensors installed in the farmland (such as multispectral sensors, soil sensors, etc.) are used to monitor the changes in the farmland in real time, including the growth status of crops, changes in the occurrence of pests and diseases, changes in the soil environment, and other data. For example, after pesticide prevention and control, the spectral changes of crop leaves are monitored by multispectral sensors to determine whether pests and diseases are effectively controlled; at the same time, soil sensors are used to monitor the pesticide residues in the soil.

[0096] Step S62: back-propagate and update the parameters of the risk probability model using the gradient descent algorithm. The collected farmland response data is used as feedback and substituted into the risk probability model. The gradient of the model parameters is calculated by the gradient descent algorithm, and the parameters are adjusted according to the direction of the gradient, so that the prediction results of the model are closer to the actual observed data. For example, at a certain moment, there is a deviation between the risk level of pests and diseases predicted by the model and the actual observed occurrence of pests and diseases. The gradient descent algorithm is used to calculate the parameter values ​​that need to be adjusted, such as adjusting certain conditional probability values ​​in the Bayesian network, so as to update the risk probability model.

[0097] Step S63: Use a sliding window mechanism to attenuate the weight of historical data to ensure the dynamic adaptability of the model. As time goes by, new farmland data is constantly generated, and the timeliness of historical data gradually decreases. Using a sliding window mechanism, a fixed-size window is set, and the data within the window is given a higher weight, while the weight of the data outside the window gradually decays. For example, the window size is set to 10 days, and a higher weight is given to the data within the last 10 days for model updating; for data older than 10 days, the weight gradually decreases, so that the model can pay more attention to recent data changes, adapt to the dynamic changes in farmland environment and pests and diseases, and maintain the accuracy and effectiveness of the model.

[0098] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0099] 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. A method for assessing agricultural pest risk, characterized in that: The following steps are involved: Step S1: Use a multispectral sensor to collect remote sensing data of the target farmland area to obtain multispectral data of the farmland; and deploy network nodes of the multispectral sensor through the Internet of Things communication protocol to build a farmland environment monitoring network; Step S2: using a pest and disease feature extraction algorithm to perform biological feature analysis on the farmland multispectral data to obtain pest and disease feature data; wherein the function formula of the pest and disease feature extraction algorithm is as follows: In the formula, is the pest and disease characteristic data, For farmland multispectral data, is the convolution kernel weight matrix of the i-th layer, is the bias term, σ is the activation function, is the feature weight coefficient; Step S3: using an environmental factor association model to perform a multivariate regression analysis on the pest characteristic data to obtain an environmental factor association matrix; the environmental factor association model includes a multidimensional nonlinear regression equation of temperature, humidity, and light intensity; Step S4: Performing dynamic risk assessment processing on the environmental factor correlation matrix based on a risk probability model to obtain pest risk level data; the risk probability model uses a Bayesian network to calculate conditional probability distribution; Step S5: performing control strategy generation processing on the pest risk level data through an optimization decision algorithm to obtain an optimal control strategy set; the optimization decision algorithm includes a particle swarm optimization objective function under constraint conditions; Step S6: using a parameter adaptive adjustment program to perform feedback learning processing on the optimal control strategy set, updating the parameter weights of the risk probability model, and obtaining a dynamically optimized risk assessment model.

2. The agricultural pest risk assessment method according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: collecting visible light, near infrared and thermal infrared band data of the target farmland through a multispectral sensor to generate an original spectral data set; Step S12: using a data normalization algorithm to normalize the original spectral data set to eliminate dimensional differences; Step S13: using an outlier detection algorithm to remove outliers from the normalized spectral data to obtain pre-processed farmland multispectral data.

3. The agricultural pest risk assessment method according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: extracting features from the farmland multispectral data through a convolutional neural network to generate a primary feature map; Step S22: using a spatial pyramid pooling layer to perform multi-scale feature fusion on the primary feature map to obtain an enhanced feature vector; Step S23: weight the enhanced feature vector based on the attention mechanism, and output the pest and disease feature data.

4. The agricultural pest risk assessment method according to claim 1, characterized in that: The functional expression of the environmental factor association model in step S3 is: In the formula, is the probability of pests and diseases occurring, is the temperature variable, is the humidity variable, is the light intensity variable, is the intercept term, , , is the regression coefficient.

5. The agricultural pest risk assessment method according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: constructing a Bayesian network structure to define the causal relationship between environmental factor nodes and pest risk nodes; Step S42: using the Markov chain Monte Carlo method to perform parameter learning on the Bayesian network and calculate the posterior probability distribution; Step S43: Output the pest risk level data according to the posterior probability distribution, and divide it into low, medium and high risk intervals according to the preset threshold value.

6. The agricultural pest risk assessment method according to claim 1, characterized in that: The expression of the particle swarm optimization objective function in step S5 is: In the formula, is the cost of the mth control measure, is the decision variable, It is a penalty item for pesticide residues. is the penalty coefficient, and are the total number of control measures and pesticide types, respectively.

7. The agricultural pest risk assessment method according to claim 1, characterized in that: Step S6 includes the following steps: Step S61: collecting farmland response data after the prevention and control strategy is implemented in real time through the online learning module; Step S62: back-propagating and updating the parameters of the risk probability model using a gradient descent algorithm; Step S63: Use a sliding window mechanism to perform weight decay on historical data to ensure the dynamic adaptability of the model.

8. The agricultural pest risk assessment method according to claim 3, characterized in that: The convolutional neural network in step S21 includes 5 convolutional layers, 3 pooling layers and 2 fully connected layers, the convolution kernel size is 3×3, and the step size is 2.

9. The agricultural pest risk assessment method according to claim 5, characterized in that: In step S42, the Gibbs sampling algorithm is used to approximate the posterior probability, the number of iterations is not less than 1000 times, and the convergence threshold is 0.

001.

10. An agricultural pest risk assessment system, characterized in that: include: Data collection and networking module: Use multispectral sensors to collect remote sensing data of the target farmland area to obtain multispectral data of the farmland, and deploy network nodes of the multispectral sensors through the Internet of Things communication protocol to build a farmland environment monitoring network; Pest and disease feature extraction module: using pest and disease feature extraction algorithm to perform biological feature analysis and processing on the farmland multispectral data to obtain pest and disease feature data; Environmental factor association analysis module: using the environmental factor association model to perform multivariate regression analysis on the pest characteristic data to obtain an environmental factor association matrix, wherein the environmental factor association model includes a multidimensional nonlinear regression equation of temperature, humidity, and light intensity; Risk assessment module: based on the risk probability model, the environmental factor correlation matrix is ​​dynamically assessed to obtain the pest risk level data. The risk probability model uses a Bayesian network to calculate the conditional probability distribution; Prevention and control strategy generation module: performing prevention and control strategy generation processing on the pest risk level data through an optimization decision algorithm to obtain an optimal prevention and control strategy set, wherein the optimization decision algorithm includes a particle swarm optimization objective function under constraint conditions; Model optimization module: using the parameter adaptive adjustment program to perform feedback learning processing on the optimal prevention and control strategy set, updating the parameter weights of the risk probability model, and obtaining a dynamically optimized risk assessment model.

Images