Green control measures recommendation method and system based on litchi pest symptoms

Through deep learning technology, the damage symptoms of lychee pests are extracted and multi-source monitoring data is analyzed. Combined with graph convolution network and neural collaborative filtering, a recommendation model for green prevention and control measures is constructed, which solves the difficulty of lychee growers to obtain personalized green prevention and control measures. The problem has improved the effectiveness and professionalism of litchi pest control, and reduced the use of chemical pesticides.

CN118626733BActive Publication Date: 2024-11-19PLANT PROTECTION RES INST OF GUANGDONG ACADEMY OF AGRI SCI
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411112221.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-19
Estimated Expiration
2044-08-14

AI Technical Summary

Technical Problem

The existing technology is difficult to analyze the occurrence of pests based on the damage symptoms and natural conditions of litchi diseases and pests through deep learning methods, and recommends personalized and professional green prevention and control measures for litchi growers, resulting in poor green prevention and control effects and insufficient professionalism of litchi insects. .

Method used

By obtaining pest data sets of different categories of litchi plants, we extract the characteristics that characterize the symptoms of pests that are harmful to litchi plants, and use multi-source monitoring data to judge pests to analyze the symptoms and categories of pests. We use graph convolution network and neural collaborative filtering to build green prevention The control measures recommendation model calculates the interaction probability between lychee pests and green prevention and control measures, and recommends green prevention and control measures that meet the standards.

Benefits of technology

It has improved the recommended quality and prevention and control effect of green prevention and control measures for litchi pests, enhanced personalization and specialization, reduced the use of chemical pesticides, and ensured the quality of litchi agricultural products and ecological environment safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118626733B_ABST
    Figure CN118626733B_ABST
Patent Text Reader

Abstract

The present invention discloses a green control measure recommendation method and system based on the symptoms of litchi pests, including obtaining a pest data set containing different types of pest data of litchi plants, extracting evaluation indicators of the symptoms of the pests; obtaining multi-source monitoring data of the target litchi planting area according to the evaluation indicators, analyzing the symptoms of the pests, the types of pests and the degree of the pests; obtaining pest portraits of the target litchi planting area to retrieve a green control measure data set, obtaining an interaction graph based on the interaction between litchi pests and green control measures; constructing a green control measure recommendation model, using graph convolutional network learning combined with neural collaborative filtering to calculate the interaction probability between litchi pests and green control measures, and selecting green control measures that meet the standards as recommended outputs. The present invention pushes personalized and specialized green control measures for litchi pests, enhances the control effect of litchi pests, and promotes the green development of the litchi industry.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of litchi pest control, and more specifically, to a method and system for recommending green control measures based on litchi pest damage symptoms. Background Art

[0002] In order to effectively prevent and control litchi pests and diseases, reduce the use of pesticides, and improve the quality of litchi, several green control technologies for litchi pests and diseases, such as biological control and physical control, have been proposed. Since litchi planting areas are often subject to natural conditions and low acceptance by growers, and green control technology is highly professional, agricultural technicians with rich theoretical and practical experience are needed to provide timely and effective guidance on green control of pests and diseases. Therefore, how to use deep learning methods to analyze the occurrence of pests based on the symptoms of litchi pests and diseases and natural conditions, and recommend personalized and professional green control measures for litchi growers, and promote the development of green control technology for litchi pests and diseases is an urgent problem to be solved. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a method and system for recommending green control measures based on the symptoms of litchi pests, so as to solve the problems of poor control effect and lack of professionalism of the existing green control measures for litchi pests implemented by growers themselves, and improve the control effect of green control measures.

[0004] The present invention provides a method for recommending green control measures based on the symptoms of litchi pests, comprising:

[0005] Obtaining an insect pest data set containing data on insect pests of different categories on litchi plants, performing feature selection based on the insect pest data set, extracting features representing symptoms of litchi insect pests, and generating evaluation indicators;

[0006] Acquire multi-source monitoring data of the target litchi planting area according to the evaluation index, perform pest judgment after pre-processing, analyze the pest symptoms of the target litchi planting area if there is pest, and determine the pest category and degree based on the pest symptoms;

[0007] Obtaining a pest portrait of the target litchi planting area through the pest category and the degree of damage, retrieving a green control measures data set based on the pest portrait, and obtaining an interaction graph based on the interaction between litchi pests and green control measures;

[0008] A green control measures recommendation model was constructed, and graph convolutional network learning combined with neural collaborative filtering was used to calculate the interaction probability between litchi pests and green control measures, and green control measures that met the standards were selected as recommended outputs.

[0009] In this solution, a pest dataset containing different types of pest data of litchi plants is obtained, feature selection is performed based on the pest dataset, and features representing the symptoms of litchi pests are extracted to generate evaluation indicators, specifically:

[0010] According to the occurrence of insect pests in litchi planting areas, common litchi insect pest categories are statistically screened, data is retrieved based on the common litchi insect pest categories, and secondary screening is performed in the retrieved data using similarity based on environmental parameters of the litchi planting areas;

[0011] The data obtained from the two screenings are used to generate a retrieval data set according to a preset ratio, the monitoring data used for pest identification and judgment are obtained from the retrieval data set as pest data, and all categories of pest data are aggregated to construct a pest data set;

[0012] Comparing different categories of pest data in the pest data set horizontally, counting the frequency of occurrence of each data type in different categories of pest data subsets, and selecting a preset number of data types as primary features based on the frequency of occurrence;

[0013] The mRMR algorithm is expanded using the Pearson correlation coefficient. Based on the primary selected features, other data types in the pest data set are introduced one by one. The correlation data is obtained as the correlation index for feature selection based on the principle that the correlation between the data type and the pest identification judgment is the largest and the correlation between the data types is the smallest.

[0014] Assume a litchi pest classification model and initialize it, divide the pest dataset into several groups and ensure that each group contains the primary selected features. Train and verify the corresponding number of litchi pest classification models in turn to obtain the average classification accuracy as an importance indicator.

[0015] The improved particle swarm algorithm is used to search for evaluation indicators, randomly generate data type combinations containing preliminary features, calculate the fitness of the data type combination according to the correlation index and the importance index, mutate based on the fitness, and output the optimal data type combination as the evaluation indicator representing the harmful symptoms.

[0016] In this scheme, multi-source monitoring data of the target litchi planting area is obtained according to the evaluation indicators, and pest judgment is performed after pre-processing, specifically:

[0017] Performing plant monitoring on the target litchi planting area according to the data types corresponding to the evaluation indicators, obtaining multi-source monitoring data of the target litchi planting area, and performing dimension reduction, windowing and normalization processing on the multi-source monitoring data;

[0018] Obtaining data critical values ​​of multi-source monitoring data before and after the occurrence of pests in the litchi pest retrieval data set, obtaining a cumulative probability distribution function through the data critical values, and analyzing the probability of pest occurrence according to the cumulative probability distribution functions of different monitoring data;

[0019] A probability matrix is ​​constructed according to the probability of occurrence of pests corresponding to different monitoring data. If the probability matrix is ​​greater than a preset probability threshold matrix, it is determined that there is pest infestation in the target litchi planting area.

[0020] In this scheme, if there is insect pest, the symptoms of insect pest in the target litchi planting area are analyzed, and the type and degree of insect pest are determined based on the symptoms, specifically:

[0021] When there is an insect pest, the symptoms are characterized by multi-source monitoring data, the data distribution of the multi-source monitoring data and the probability distribution of the insect pest are obtained, and the insect pest category is determined by similarity;

[0022] Building a normal growth state model based on multi-source monitoring data under normal growth conditions of litchi plants, using the normal growth state model to obtain an estimated value of multi-source monitoring data of a target litchi planting area, and calculating a residual between the estimated value of the multi-source monitoring data and the actual value of the current multi-source monitoring data;

[0023] By comparing the residual with a preset threshold interval, the degree of damage of the insect pest is determined according to the corresponding level of the threshold interval into which the residual falls.

[0024] In this solution, the green control measures dataset is retrieved based on the pest portrait, and an interaction diagram is obtained based on the interaction between litchi pests and green control measures, specifically:

[0025] Determine a pest portrait according to the pest category and damage degree of the target litchi planting area, and use the pest portrait to obtain similar pest categories whose similarity meets the preset standard;

[0026] Through the search of pest categories and similar pest categories in the pest portrait, we obtain historical prevention and control examples and green prevention and control measures involved in relevant literature, and then construct a green prevention and control measures dataset after secondary screening based on the degree of damage.

[0027] The green control measures dataset is represented graphically, and a pest-green control measures interaction graph is constructed based on whether there is interaction between pests and green control measures.

[0028] In this plan, a green prevention and control measures recommendation model is constructed, specifically:

[0029] A green control measures recommendation model is constructed based on graph convolutional networks and neural collaborative filtering, and the interaction graph is learned to obtain initial features of pests, initial features of collections, and initial features of green control measures;

[0030] Through graph convolution, the interactive data between pests and green control measures, as well as the interactive data between sets and green control measures, are used to alternately learn and update the features of pests, sets, and green control measures.

[0031] The updated pest features are fused using the attention mechanism to generate fused features, and the fused features are added to the updated set features to obtain the final set features;

[0032] Calculate the inner product vector of the final feature of the set and the updated green control measures feature embedding, concatenate the inner product vector, the final feature of the set and the updated green control measures feature, import them into the multi-layer perceptron to obtain the interaction probability of each green control measure, and select the corresponding green control measure as the output according to the interaction probability sorting.

[0033] In this scheme, the attention mechanism is used to fuse the updated pest features to generate fused features, specifically:

[0034] Obtain the green control measure node sets connected to different pest nodes in the interaction graph, determine the deviations between the green control measure node sets corresponding to different pest nodes, and obtain the deviation matrix between all pest nodes and other pest nodes;

[0035] The deviation matrix is ​​fused through a multi-layer perceptron, and the attention weights of all pest nodes in the set are obtained by normalization. The updated pest features are weighted and summed by the attention weights to generate fused features.

[0036] The present invention provides a green control measure recommendation system based on the symptoms of litchi pests, including: an evaluation index acquisition module, a data acquisition module, a pest image evaluation module, a green control measure recommendation module and a data output module;

[0037] The evaluation index acquisition module performs feature selection based on the pest data set of different categories of pest data to extract evaluation indicators that characterize the symptoms of litchi pests;

[0038] The data acquisition module acquires multi-source monitoring data of the target litchi planting area based on the evaluation index and performs data preprocessing;

[0039] The pest image assessment module determines whether there is pest in the target litchi planting area. If there is pest, the pest symptoms in the target litchi planting area are analyzed to determine the pest type and degree of damage and evaluate the pest image.

[0040] The green control measures recommendation module constructs a green control measures recommendation model, uses a graph convolutional network to learn the interaction graph to generate feature representations of pest and green control measures data sets, fuses pest features with set features, uses a multi-layer perceptron to calculate the interaction probability between fused features and green control measures, and selects green control measures that meet the standards;

[0041] The data output module will output green prevention and control measures that meet the standards, and visualize and push them in a preset manner.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention improves the recommendation quality of green control measures for litchi pests, enhances the personalization and specialization of green control measures, improves the control effect of litchi pests, popularizes the importance and advantages of green control to the majority of growers through professional green control measures, and improves the growers' understanding of green control technology;

[0044] By integrating new and optimal green control technology models, green control measures suitable for target litchi planting areas are recommended based on graph convolution and collaborative filtering, which can comprehensively improve the level of green control of litchi pests and diseases, reduce the use of highly toxic and high-residue chemical pesticides, and ensure the quality and safety of litchi agricultural products and the safety of the ecological environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the drawings required for use in the embodiments or exemplary descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained according to the drawings without paying creative work.

[0046] Figure 1 A flowchart showing a method for recommending green control measures based on the symptoms of litchi pests is shown;

[0047] Figure 2 A flow chart showing pest judgment based on multi-source monitoring data is shown;

[0048] Figure 3 A flowchart for constructing a green control measures recommendation model is shown;

[0049] Figure 4 The diagram shows the framework of the green control measures recommendation system based on the symptoms of litchi pests. DETAILED DESCRIPTION

[0050] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0052] Figure 1 Flowcharts of recommended methods for green control measures based on the symptoms of litchi pests are shown respectively.

[0053] like Figure 1 As shown, this embodiment provides a green control measure recommendation method based on the symptoms of litchi pests, including:

[0054] S102, obtaining an insect pest dataset containing data on insect pests of different categories on litchi plants, performing feature selection based on the insect pest dataset, extracting features representing symptoms of litchi insect pests, and generating evaluation indicators;

[0055] S104, obtaining multi-source monitoring data of the target litchi planting area according to the evaluation index, performing pest judgment after pre-processing, and analyzing the pest symptoms of the target litchi planting area if there is pest, and determining the pest category and degree based on the pest symptoms;

[0056] S106, obtaining a pest image of the target litchi planting area according to the pest category and the degree of damage, retrieving a green control measures data set based on the pest image, and obtaining an interaction graph according to the interaction between litchi pests and green control measures;

[0057] S108, construct a green control measures recommendation model, use graph convolutional network learning combined with neural collaborative filtering to calculate the interaction probability between litchi pests and green control measures, and select green control measures that meet the standards as recommended outputs.

[0058] It should be noted that according to the occurrence of pests in litchi planting areas, common litchi pest categories, such as litchi stink bugs, fruit-sucking armyworms and fruit flies, are statistically screened, data retrieval is performed based on the common litchi pest categories, and secondary screening is performed on the retrieved data using similarity through the environmental parameters of the litchi planting areas; the data obtained from the two screenings are used to generate a retrieval data set according to a preset ratio, and usually the proportion of historical pest-related data similar to the environmental parameters of the planting areas is higher, and the monitoring data used for pest identification and judgment is obtained from the retrieval data set as pest data, and all categories of pest data are aggregated to construct a pest data set; in the pest data set, different categories of pest data are horizontally compared, the frequency of occurrence of each data type in different categories of pest data subsets is counted, and a preset number of data types are selected as primary features based on the frequency of occurrence, such as ambient temperature, leaf images of pests and diseases, and meteorological data.

[0059] The mRMR algorithm is expanded using the Pearson correlation coefficient. Based on the primary selected features, other data types in the pest data set are introduced one by one. The correlation data is obtained as the correlation index of feature selection based on the principle that the correlation between data types and pest identification judgment is the largest and the correlation between data types is the smallest. The expanded mRMR algorithm is expressed as: ,in, Indicates the i-th data type Pest identification and judgment The Pearson correlation coefficient between Indicates the i-th data type With the jth data type The Pearson correlation coefficient between Indicates the data type selected into the data type combination, The number of features representing the combination of data types, Represents the insect pest dataset.

[0060] Assuming that the litchi pest classification model is initialized, the function of the classification model is expressed as , its classification accuracy The calculation formula is: in represents the number of training samples, represents the indicator function, represents the training sample set, Indicates a combination of data types. represents the training sample, Indicates the pest category. The pest dataset is divided into several groups and each group is ensured to contain the primary selected features. The corresponding number of litchi pest classification models are trained and verified in turn to obtain the average classification accuracy as an importance indicator.

[0061] Initialize the population parameters of the particle swarm, set the movement range, speed range, maximum number of iterations and learning factor of the parameters, use the improved particle swarm algorithm to search for evaluation indicators, initialize feature encoding for data types, and randomly generate data type combinations containing preliminary features to ensure the diversity of data type combinations, form an objective function according to the formula of the correlation index and the importance index, associate the fitness with the objective function value, calculate the fitness of the data type combination according to the association relationship, obtain the optimal solution of the particle after iterative optimization and the optimal solution of the group, judge whether the particle meets the constraint conditions, if so, introduce mutation to expand the search space, and update the speed and position of the particle; calculate the fitness of the updated particle, compare the obtained fitness with the historical optimal solution, update the historical optimal solution until the iteration end requirements are met, and output the optimal data type combination as an evaluation indicator representing the harmful symptoms.

[0062] Figure 2 A flow chart of pest judgment based on multi-source monitoring data is shown.

[0063] According to an embodiment of the present invention, multi-source monitoring data of the target litchi planting area is obtained according to the evaluation index, and pest judgment is performed after pre-processing, specifically:

[0064] S202, monitoring plants in the target litchi planting area according to the data types corresponding to the evaluation indicators, obtaining multi-source monitoring data of the target litchi planting area, and performing dimension reduction, windowing and normalization processing on the multi-source monitoring data;

[0065] S204, obtaining data critical values ​​of multi-source monitoring data before and after the occurrence of pests in the litchi pest search data set, obtaining a cumulative probability distribution function through the data critical values, and analyzing the probability of pest occurrence according to the cumulative probability distribution functions of different monitoring data;

[0066] S206, constructing a probability matrix according to the pest occurrence probabilities corresponding to different monitoring data, and if the probability matrix is ​​greater than a preset probability threshold matrix, it is determined that pests exist in the target litchi planting area.

[0067] It should be noted that the multi-source monitoring data obtained are pre-processed by data cleaning, dimensionality reduction, windowing and normalization, and the critical values ​​of the multi-dimensional monitoring data before and after the occurrence of litchi pests are obtained. If litchi pests occur in the target litchi planting area, the absolute value of the multi-dimensional monitoring data must be greater than its critical value. For example, when litchi pests occur, the ambient temperature is often greater than the daily value, but litchi pests may not occur when the ambient temperature is greater than the daily value. Assuming that the multi-dimensional monitoring data follows the extreme value I distribution, the cumulative probability distribution function is obtained , predict the probability of litchi pests occurring in the target litchi planting area , represents a hyperparameter, which is the probability of pest occurrence.

[0068] It should be noted that when there is an insect pest, the damage symptoms are characterized by multi-source monitoring data, the data distribution of the multi-source monitoring data and the probability distribution of insect pest occurrence are obtained, and the insect pest category is determined by similarity; based on data reconstruction models such as autoencoder networks and multivariate state estimation, a normal growth state model is constructed according to the multi-source monitoring data under normal growth conditions of litchi plants, and the normal growth state model is used to obtain the estimated value of the multi-source monitoring data of the target litchi planting area, and the residual of the estimated value of the multi-source monitoring data and the actual value of the current multi-source monitoring data is calculated; by comparing the residual with a preset threshold interval, the degree of damage of the insect pest is determined according to the corresponding level of the threshold interval into which the residual falls.

[0069] Figure 3 A flowchart for building a green control measures recommendation model is shown.

[0070] According to an embodiment of the present invention, a green prevention and control measures recommendation model is constructed, specifically:

[0071] S302, building a green control measure recommendation model based on graph convolutional network and neural collaborative filtering, and learning the interaction graph to obtain initial features of pests, initial features of collections, and initial features of green control measures;

[0072] S304, alternately learning and updating the features of pests, collections, and green control measures by using the interactive data between pests and green control measures and the interactive data between collections and green control measures through graph convolution;

[0073] S306, using the attention mechanism to fuse the updated pest features to generate a fused feature, and adding the fused feature to the updated set feature to obtain a final set feature;

[0074] S308, calculate the inner product vector of the final feature of the set and the updated green control measures feature embedding, concatenate the inner product vector, the final feature of the set and the updated green control measures feature, import them into a multi-layer perceptron to obtain the interaction probability of each green control measure, and select the corresponding green control measure as output according to the interaction probability sorting.

[0075] It should be noted that the pest portrait is determined according to the pest category and degree of damage in the target litchi planting area, and the pest category in the pest portrait is used to obtain similar pest categories whose similarity meets the preset standard; the pest category and similar pest category in the pest portrait are used to retrieve the green control measures involved in historical prevention and control examples and related literature, and the green control measures data set is constructed after secondary screening using the degree of damage. The green control measures data set includes pest categories, similar pest categories, and green control measures corresponding to pest categories and similar pest categories. The green control measures data set is represented by a graph, with different pest categories and green control measures as nodes, and the interactions between different pest categories and green control measures as edge structures. A pest-green control measures interaction graph is constructed based on whether there is interaction between pests and green control measures. A graph convolutional network is used to learn the pest-green control measures interaction graph to capture the high-order features of pests and green control measures, and the collection features are strengthened through the attention mechanism. The interaction probability between the collection features and the green control measures is evaluated using a multi-layer perceptron.

[0076] Since the interactive green control measures of the set are determined by the pests and green control measures that constitute the green control measures data set, the neural collaborative filtering framework is selected to eliminate the interference of noise. The green control measures node set connected to different pest nodes in the interaction graph is obtained, and the deviation between the green control measures node set corresponding to different pest nodes is determined to obtain the deviation matrix between all pest nodes and other pest nodes; the deviation between the green control measures node set corresponding to different pest nodes is used to characterize the preference difference of pests in using green control measures. The deviation matrix is ​​fused through a multi-layer perceptron, and the attention weights of all pest nodes in the set are obtained by normalization. The updated pest features are weighted and summed to generate fusion features through the attention weights. The multi-layer perceptron considers the differences in the use of green control measures by pests during fusion, and the pest features are weighted and summed to obtain their common preferences. The final features of the fused set retain the common preferences of the pests and also integrate the preferences of the pests themselves. The fused features are added to the updated set features to obtain the final features of the set, which reduces the noise influence of the fused pest features, and makes the final features of the fused set contain rich information contained in the pest features and the features of the set itself.

[0077] The pest control effect after using green control measures in the target litchi planting area is obtained, and a visual comparison is made between the pest control effect and the historical chemical control effect of similar damage degree. If the control effect deviation is greater than the preset threshold, the green control measures are replaced, and a visual workflow display is performed according to the specific implementation steps of the green control measures to popularize the specific implementation knowledge of green control for litchi growers.

[0078] Figure 4The diagram shows the framework of the green control measures recommendation system based on the symptoms of litchi pests.

[0079] This embodiment provides a green control measure recommendation system 4 based on the symptoms of litchi pests, including: an evaluation index acquisition module 41, a data acquisition module 42, a pest image evaluation module 43, a green control measure recommendation module 44 and a data output module 45;

[0080] The evaluation index acquisition module 41 performs feature selection based on the pest data set of different categories of pest data to extract evaluation indicators that characterize the symptoms of litchi pests;

[0081] The data acquisition module 42 acquires multi-source monitoring data of the target litchi planting area based on the evaluation index and performs data preprocessing;

[0082] The pest image assessment module 43 determines whether there is pest in the target litchi planting area. If there is pest, the pest symptoms in the target litchi planting area are analyzed to determine the pest type and degree of damage and evaluate the pest image.

[0083] The green control measures recommendation module 44 constructs a green control measures recommendation model, uses a graph convolutional network to learn the interaction graph to generate feature representations of pest and green control measures data sets, fuses pest features with set features, uses a multi-layer perceptron to calculate the interaction probability between fused features and green control measures, and selects green control measures that meet the standards;

[0084] The data output module 45 outputs green prevention and control measures that meet the standards, and displays and pushes them visually in a preset manner.

[0085] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0086] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0087] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0088] Those skilled in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0089] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0090] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for recommending green control measures based on the symptoms of litchi pests, characterized in that: The following steps are involved: Obtaining an insect pest data set containing data on insect pests of different categories on litchi plants, performing feature selection based on the insect pest data set, extracting features representing symptoms of litchi insect pests, and generating evaluation indicators; Acquire multi-source monitoring data of the target litchi planting area according to the evaluation index, perform pest judgment after pre-processing, analyze the pest symptoms of the target litchi planting area if there is pest, and determine the pest category and degree based on the pest symptoms; Obtaining a pest portrait of the target litchi planting area through the pest category and the degree of damage, retrieving a green control measures data set based on the pest portrait, and obtaining an interaction graph based on the interaction between litchi pests and green control measures; Construct a green control measures recommendation model, use graph convolutional network learning combined with neural collaborative filtering to calculate the interaction probability between litchi pests and green control measures, and select green control measures that meet the standards as recommended outputs; Obtain an insect pest dataset containing different types of insect pest data of litchi plants, perform feature selection based on the insect pest dataset, extract features that characterize the symptoms of litchi insect pests, and generate evaluation indicators, specifically: According to the occurrence of insect pests in litchi planting areas, common litchi insect pest categories are statistically screened, data is retrieved based on the common litchi insect pest categories, and secondary screening is performed in the retrieved data using similarity based on environmental parameters of the litchi planting areas; The data obtained from the two screenings are used to generate a retrieval data set according to a preset ratio, the monitoring data used for pest identification and judgment are obtained from the retrieval data set as pest data, and all categories of pest data are aggregated to construct a pest data set; Comparing different categories of pest data in the pest data set horizontally, counting the frequency of occurrence of each data type in different categories of pest data subsets, and selecting a preset number of data types as primary features based on the frequency of occurrence; The mRMR algorithm is expanded using the Pearson correlation coefficient. Based on the primary selected features, other data types in the pest data set are introduced one by one. The correlation data is obtained as the correlation index for feature selection based on the principle that the correlation between the data type and the pest identification judgment is the largest and the correlation between the data types is the smallest. Assume a litchi pest classification model and initialize it, divide the pest dataset into several groups and ensure that each group contains the primary selected features. Train and verify the corresponding number of litchi pest classification models in turn to obtain the average classification accuracy as an importance indicator. Using an improved particle swarm algorithm to search for evaluation indicators, randomly generating a data type combination containing preliminary features, calculating the fitness of the data type combination according to the correlation index and the importance index, performing mutation based on the fitness, and outputting the optimal data type combination as an evaluation indicator for characterizing the harmful symptoms; Construct a green prevention and control measures recommendation model, specifically: A green control measures recommendation model is constructed based on graph convolutional networks and neural collaborative filtering, and the interaction graph is learned to obtain initial features of pests, initial features of collections, and initial features of green control measures; Through graph convolution, the interactive data between pests and green control measures, as well as the interactive data between sets and green control measures, are used to alternately learn and update the features of pests, sets, and green control measures. The updated pest features are fused using the attention mechanism to generate fused features, and the fused features are added to the updated set features to obtain the final set features; Calculate the inner product vector of the final feature of the set and the updated green control measures feature embedding, concatenate the inner product vector, the final feature of the set and the updated green control measures feature, import them into the multi-layer perceptron to obtain the interaction probability of each green control measure, and select the corresponding green control measure as the output according to the interaction probability sorting.

2. A green control measure recommendation method based on the symptoms of litchi pests according to claim 1, characterized in that: According to the evaluation index, multi-source monitoring data of the target litchi planting area is obtained, and pest judgment is performed after pre-processing, specifically: Performing plant monitoring on the target litchi planting area according to the data types corresponding to the evaluation indicators, obtaining multi-source monitoring data of the target litchi planting area, and performing dimension reduction, windowing and normalization processing on the multi-source monitoring data; Obtaining data critical values ​​of multi-source monitoring data before and after the occurrence of pests in a retrieval data set of litchi pests, obtaining a cumulative probability distribution function through the data critical values, and analyzing the probability of pest occurrence according to the cumulative probability distribution functions of different monitoring data; A probability matrix is ​​constructed according to the probability of occurrence of pests corresponding to different monitoring data. If the probability matrix is ​​greater than a preset probability threshold matrix, it is determined that there is pest infestation in the target litchi planting area.

3. A green control measure recommendation method based on litchi pest damage symptoms according to claim 1, characterized in that: If there is pest, the pest symptoms in the target litchi planting area are analyzed, and the pest category and degree are determined based on the pest symptoms, specifically: When there is an insect pest, the symptoms are characterized by multi-source monitoring data, the data distribution of the multi-source monitoring data and the probability distribution of the insect pest are obtained, and the insect pest category is determined by similarity; Building a normal growth state model based on multi-source monitoring data under normal growth conditions of litchi plants, using the normal growth state model to obtain an estimated value of multi-source monitoring data of a target litchi planting area, and calculating a residual between the estimated value of the multi-source monitoring data and the actual value of the current multi-source monitoring data; By comparing the residual with a preset threshold interval, the degree of damage of the insect pest is determined according to the corresponding level of the threshold interval into which the residual falls.

4. A green control measure recommendation method based on the symptoms of litchi pests according to claim 1, characterized in that: Based on the pest portrait, the green control measures dataset is retrieved, and an interaction diagram is obtained according to the interaction between litchi pests and green control measures, specifically: Determine a pest portrait according to the pest category and damage degree of the target litchi planting area, and use the pest portrait to obtain similar pest categories whose similarity meets the preset standard; Through the search of pest categories and similar pest categories in the pest portrait, we obtain historical prevention and control examples and green prevention and control measures involved in relevant literature, and then construct a green prevention and control measures dataset after secondary screening based on the degree of damage. The green control measures dataset is represented graphically, and a pest-green control measures interaction graph is constructed based on whether there is interaction between pests and green control measures.

5. A green control measure recommendation method based on litchi pest damage symptoms according to claim 1, characterized in that: The attention mechanism is used to fuse the updated pest features to generate fused features, specifically: Obtain the green control measure node sets connected to different pest nodes in the interaction graph, determine the deviations between the green control measure node sets corresponding to different pest nodes, and obtain the deviation matrix between all pest nodes and other pest nodes; The deviation matrix is ​​fused through a multi-layer perceptron, and the attention weights of all pest nodes in the set are obtained by normalization. The updated pest features are weighted and summed by the attention weights to generate fused features.

6. A green control measures recommendation system based on the symptoms of litchi pests, characterized in that: Implementing the green control measures recommendation method based on the symptoms of litchi pests as described in any one of claims 1 to 5, the system comprises: an evaluation index acquisition module, a data acquisition module, a pest portrait evaluation module, a green control measures recommendation module and a data output module; The evaluation index acquisition module performs feature selection based on the pest data set of different categories of pest data to extract evaluation indicators that characterize the symptoms of litchi pests; The data acquisition module acquires multi-source monitoring data of the target litchi planting area based on the evaluation index and performs data preprocessing; The pest image assessment module determines whether there is pest in the target litchi planting area. If there is pest, the pest symptoms in the target litchi planting area are analyzed to determine the pest type and degree of damage and evaluate the pest image. The green control measures recommendation module constructs a green control measures recommendation model, uses a graph convolutional network to learn an interaction graph to generate feature representations of pest and green control measures data sets, fuses pest features with set features, uses a multi-layer perceptron to calculate the interaction probability between fused features and green control measures, and selects green control measures that meet the standards; The data output module will output green prevention and control measures that meet the standards, and visualize and push them in a preset manner.

Citation Information

Patent Citations

  • Information recommendation method for neural collaborative filtering based on graph convolution

    CN112861017A

  • Litchi insect pest monitoring and early warning method and system based on Internet of Things, and storage medium

    CN114445785A

  • Deep learning-based pest control scheme generation method and system, and storage medium

    CN114707773A

  • Litchi insect pest recognition and resistance detection method and system and storage medium

    CN117151342A

  • Low-carbon treatment scheme recommendation method and system for litchi insect pests

    CN117173400A