Tobacco disease and insect pest pesticide matching method and system

Abstract

The invention relates to the technical field of tobacco information processing, in particular to a tobacco pest and disease damage pesticide matching method and system. The tobacco disease and insect pest pesticide matching method comprises the steps that a first set is generated according to recognized disease and insect pests and candidate pesticides, the first set comprises a plurality of tensor sets, and each tensor set comprises a pesticide purchasing decision tensor and a pesticide mixing decision tensor; inputting each tensor set in the first set into a first calculation device for processing to obtain a target value set corresponding to each tensor set; and recommending a pesticide matching scheme according to the tensor set corresponding to the optimal target value set in the first set. According to the invention, intelligent pesticide matching is realized, the pesticide matching scheme has a higher disease and pest treatment effect, the pesticide matching time is saved, and the cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of tobacco information processing, and in particular to a method and system for matching tobacco pests and pesticides. Background Art

[0002] The use of pesticides for tobacco pests and diseases is an important part of the tobacco planting process. Scientific and reasonable use of pesticides is of great significance to tobacco pest control and tobacco harvest. At present, the guidance on the use of pesticides for tobacco pests and diseases mainly relies on the experience of farmers and the guidance of pharmacists. It is difficult to carry out pest control for large-scale tobacco intelligent planting in a regional scope, and there is a lack of systematic and comprehensive evaluation of the use of pesticides for tobacco pests and diseases. There are many studies on the precision application of pesticides at home and abroad. Precision application of pesticides refers to reducing the use of pesticides as much as possible without reducing the effectiveness of pesticide spraying, thereby reducing pesticide pollution. It mainly focuses on related equipment, systems and platforms, while there are relatively few studies on pesticide matching schemes. Therefore, it is of great significance to further explore pesticide matching schemes. In the prior art, research on tobacco pests and diseases focuses on identifying tobacco pests and diseases and searching for pesticides for treatment. There is a lack of recommendations on scientific and reasonable pesticide matching schemes, including recommendations on the types of purchased pesticides and the amount of pesticides. Summary of the invention

[0003] In order to solve the above technical problems, the embodiments of the present invention disclose a tobacco pest and disease pesticide matching method and system.

[0004] In a first aspect, an embodiment of the present invention provides a method for matching pesticides for tobacco pests and diseases, the method comprising:

[0005] Generate a first set according to the identified pests and diseases and candidate pesticides, the first set including a plurality of tensor sets, each tensor set including a pesticide purchase decision tensor and a pesticide mixing decision tensor;

[0006] Each tensor set in the first set is input into the first computing device for processing, and a target value set corresponding to each tensor set is obtained; the operations performed by the first computing device on the input tensor set include: obtaining a first target value according to the pesticide purchase decision tensor and cost information; obtaining a second target value according to the pesticide purchase decision tensor, the pesticide treatment utility coefficient and the infection area coefficient; obtaining the cost C of the pesticide mixed spraying for treating each pest i :

[0007]

[0008] Among them, C i represents the cost of spraying pesticide mixtures to treat the i-th pest, M is the set of pesticide types, |M| is the pesticide type coefficient, and the pesticide mixture decision tensor d ijmCharacterizes whether the pesticide types j and m used to treat pests i can be mixed, and the pesticide candidate characterization tensor x ij Indicates whether pesticide j is used to treat pest i; sums up the cost of spraying mixed pesticides for all types of pests to obtain the third target value;

[0009] A pesticide matching scheme is recommended based on the tensor set corresponding to the best target value set in the first set.

[0010] Preferably, the pesticide procurement decision tensor includes a first procurement decision tensor and a second procurement decision tensor; the first procurement decision tensor represents the quantity of pesticides purchased from pesticide sellers for treating pests and diseases, and the dimensions of the first procurement decision tensor include: pest and disease type dimension, pesticide type dimension; the second procurement decision tensor represents the quantity of pesticides purchased from other farmers for treating pests and diseases, and the dimensions of the second procurement decision tensor include: pest and disease type dimension, farmer dimension, pesticide type dimension.

[0011] Preferably, generating the first set according to the identified pests and diseases and candidate pesticides comprises:

[0012] For each other farmer, the holding amount of various pesticides used to treat each type of pests and diseases is obtained, and the initial treatment utility of each type of pests and diseases is obtained based on the holding amount of pesticides and the treatment utility coefficient of the pesticides; a correction coefficient is generated based on the cost of mixed spraying of pesticides for each type of pests and diseases, and the initial treatment utility is corrected using the correction coefficient to obtain the corrected treatment utility of each type of pests and diseases;

[0013] The degree of association among other farmers was calculated based on their corrected treatment utility for each pest and disease type;

[0014] Classify the set of other farmers according to the degree of connection between them, and obtain sub-sets of other farmers in different categories;

[0015] Farmers are selected from other subsets of farmers in different categories, and a second purchase decision tensor is generated according to the pesticide holdings of the selected farmers.

[0016] Preferably, before the step of generating the first set according to the identified pests and diseases and candidate pesticides, the step further includes:

[0017] Collect images of tobacco growing areas;

[0018] Identify pests and diseases in images of tobacco growing areas;

[0019] Candidate pesticides for identified pests and diseases.

[0020] Preferably, before the step of recommending a pesticide matching scheme according to the tensor set corresponding to the best target value set in the first set, the step further includes:

[0021] The first set is input into the second computing device for processing: the second computing device performs crossover and mutation operations on the first set with a set probability, and obtains the second set after repair; each tensor set in the second set is input into the first computing device, and a second result set is obtained according to the output of the first computing device; all results in the first result set and the second result set are sorted, and the first set is modified according to the sorting result;

[0022] The step of inputting the first set into the second computing device for processing is repeatedly executed until a termination condition is reached.

[0023] Preferably, the step of inputting each tensor set in the first set into the first computing device for processing and obtaining the first result set according to the output of the first computing device also includes: generating a constraint set; the input of the first computing device also includes the constraint set.

[0024] Preferably, the constraint set includes:

[0025] Constraints between the pesticide mixture decision tensor and the pesticide mixture representation tensor:

[0026]

[0027] Among them, d ijm is the pesticide mixture decision tensor, k jm is the pesticide mixture representation tensor, N is the set of pests and diseases, M is the set of pesticide types, i, j, m are indexes;

[0028] Constraints on the amount of pesticides purchased:

[0029]

[0030] Among them, y ij is the pesticide purchasing decision tensor, g ij The area coefficient of pest i that can be treated by a unit of pesticide j, h i is the infection area coefficient of pest i, a ij is an auxiliary variable;

[0031] Pesticide candidate constraints:

[0032]

[0033] Among them, x ij is the pesticide candidate representation tensor, e ij Indicates whether pesticide j can treat pest i;

[0034] Constraints between the pesticide candidate representation tensor and the pesticide procurement decision tensor:

[0035]

[0036] Among them, y ij is the pesticide purchasing decision tensor;

[0037] Pesticide spraying restrictions:

[0038]

[0039] Where N is the set of pest and disease species.

[0040] In a second aspect, an embodiment of the present invention further provides a tobacco pest and pesticide matching system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a tobacco pest and pesticide matching method when executing the computer program.

[0041] The technical solution provided by the embodiment of the present invention has at least the following advantages compared with the prior art:

[0042] The present invention provides a method and system for matching tobacco pests and diseases with pesticides, which can perform unmanned and autonomous tobacco pest and disease image collection on tobacco in a large area, and obtain a scientific and reasonable tobacco pest and disease pesticide matching scheme through multi-objective optimization reasoning calculation, including procurement structure, pesticide mixed spraying instructions, etc., effectively improving the treatment effect of tobacco pests and diseases, improving the efficiency of pesticide spraying, and reducing costs.

[0043] Specifically, the existing pesticide matching scheme only considers purchasing pesticides from pesticide sellers. Compared with the prior art, the present invention is based on the sharing theory, considers purchasing pesticides from other farmers, and obtains a scientific and reasonable pesticide matching scheme through a scientific and reasonable constraint set and solution method, thereby achieving the rational use of resources, and while improving resource utilization, effectively reducing the cost of the pesticide matching scheme. Further, the present invention considers the cost of mixed pesticide spraying, and obtains the cost of mixed pesticide spraying through an effective and accurate method for calculating the cost of mixed pesticide spraying, thereby reducing the cost of mixed pesticide spraying while ensuring the effectiveness of pest treatment. Furthermore, the present invention obtains the degree of correlation between farmers by obtaining the therapeutic utility vector of the amount of pesticide held by each farmer, thereby improving the quality of the initialization tensor set, thereby improving the efficiency of obtaining a scientific and reasonable pesticide matching scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The figure is an exemplary flow chart of a tobacco pest and disease pesticide matching method according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0045] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0046] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0047] In addition, in order to better illustrate the present disclosure, many specific details are given in the specific implementation below. Those skilled in the art should understand that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present disclosure. An embodiment of the present invention provides a vehicle-mounted image processing method.

[0048] Figure 1 FIG. 2 is a flow chart showing a method for matching tobacco pests and diseases with pesticides according to a first embodiment of the present invention. Figure 1 As shown, the method includes step 101, step 102, and step 103.

[0049] Step 101, generating a first set based on the identified pests and diseases and candidate pesticides, wherein the first set includes a plurality of tensor sets, each of which includes a purchase decision tensor and a pesticide mixing decision tensor.

[0050] Step 102, input each tensor set in the first set into the first computing device for processing, and obtain a target value set corresponding to each tensor set; the operations performed by the first computing device on the input tensor set include: obtaining a first target value based on a procurement decision tensor and cost information; obtaining a second target value based on the procurement decision tensor, a treatment utility coefficient and an infection area coefficient; and calculating the pesticide mixing cost as the third target value based on the pesticide mixing decision tensor, the pesticide type coefficient and the pesticide candidate characterization tensor.

[0051] Step 103: recommend a pesticide matching solution according to the tensor set corresponding to the best target value set in the first set.

[0052] This embodiment takes into account the cost of mixed pesticide spraying, and obtains the cost of mixed pesticide spraying through an effective and accurate method for calculating the cost of mixed pesticide spraying, thereby reducing the cost of mixed pesticide spraying while ensuring the effectiveness of pest treatment.

[0053] On the basis of the first embodiment, optionally, before step 101, the method further includes:

[0054] Step 100a, collecting images of tobacco growing areas.

[0055] Use drones with cameras to automatically collect overhead images of tobacco-growing areas.

[0056] First, the tobacco planting area image is divided into at least one grid. The area that can be covered by the drone during aerial photography is a grid. Then, the drone's cruise route is generated based on the center point of the grid, and an image is taken for each grid cruise. The cruise route can be generated by connecting the center points of adjacent grids in sequence to ensure that each grid is covered. The connection method includes the serpentine antenna connection method.

[0057] Step 100b, using a pre-trained pest and disease recognition neural network to identify pests and diseases in the tobacco planting area image.

[0058] Pests and diseases include diseases and pests, which are a general term for the two. Pre-train a tobacco pest and disease recognition neural network, and perform pest and disease recognition on the collected tobacco planting area images. Specifically, first collect a tobacco planting area image data set, and annotate the collected image data, and the annotations include the types of pests and diseases and whether they are disease-free. One implementation method is that the pest and disease recognition neural network is based on a convolutional neural network structure, including an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer. After the tobacco planting area image used for training is input into the network, the pest and disease prediction results are obtained after the input layer, convolution layer, pooling layer, fully connected layer, and output layer. The network is trained and optimized according to the difference between the prediction result and the annotation result until convergence, and the loss function adopts the cross entropy loss function. One implementation method is to build a tobacco pest and disease recognition neural network based on a residual network. The residual network introduces a residual block and has a deeper level. After step S3, the infected area coefficient is required to reason about the pesticide matching scheme. One implementation method is to count the number of tobacco planting area images infected with a certain pest, and multiply the number by the area covered by the grid to obtain the infected area of ​​the pest.

[0059] Step 100c, using the tobacco pest and disease knowledge graph to identify pesticide candidates for the pests and diseases identified by the neural network.

[0060] Knowledge graphs are used to express entities and relationships in the real world and aggregate a large amount of knowledge, thereby achieving rapid response and reasoning of knowledge. First, crawl data from the tobacco pest data source and perform preprocessing to obtain a plain text corpus. Perform entity recognition based on the plain text corpus, extract knowledge from it, fuse knowledge, and build a knowledge representation model to finally obtain a knowledge graph of tobacco pests and diseases. Commonly used graph databases include Neo4j. The knowledge graph of tobacco pests and diseases includes at least the pest objects, drug properties, and precautions for different pesticides. Drug properties include the treatment area coefficient, which represents the infected area that can be treated by a unit of pesticide. Precautions include: what kind of pesticides cannot be mixed with when used. According to the pests and diseases identified by the neural network, the tobacco pest and disease knowledge graph outputs the corresponding treatment drugs, drug properties, and precautions.

[0061] Next, the tobacco pest inference model can be used to match the pests and diseases identified by the neural network and the candidate pesticides in the knowledge graph, determine the dosage, and infer the drug application parameters. The specific implementation method is as shown in steps 101-103.

[0062] A tobacco pest and disease pesticide matching method according to the second embodiment of the present invention comprises the following steps.

[0063] Step 201, generating a first set based on the identified pests and diseases and candidate pesticides, wherein the first set includes a plurality of tensor sets, each of which includes a first purchase decision tensor (y_s ij ), the second purchasing decision tensor (y_f rij ), the third purchasing decision tensor (y ij ), pesticide mixture decision tensor (d ijm ), the first intermediate tensor (δ_s), the second intermediate tensor (δ_f r ).

[0064] The first set is a set of tensor sets, that is, each element in the first set is a tensor set. Each tensor set also includes: a pesticide candidate representation tensor (x ij ). Each element in the first set is of the form (y_s ij ,y_f rij ,y_ ij ,d ijm ,δ_s,δ_f r ,x ij ).

[0065] In the prior art, when generating a pesticide matching solution, only the case of purchasing pesticides from pesticide sellers is considered. In reality, other tobacco growers may have unused pesticides from the previous time or a large amount of stockpiled pesticides. Pesticides have a certain shelf life. If they are stored for a long time, the efficacy will inevitably be reduced and resources will be wasted. Based on the sharing theory, the present invention considers the way of obtaining pesticides from other tobacco growers, which helps to save social resources and improve social efficiency.

[0066] The first purchasing decision tensor y_s ij Represents the number of pesticide type j purchased from pesticide sellers for treating pest type i. The values ​​of the elements in the first purchase decision tensor can be non-negative, preferably non-negative integers. The dimensions of the first purchase decision tensor include: pest type dimension and pesticide dimension. The second purchase decision tensor y_f rij Represents the number of pesticide types j purchased from other farmers r for treating corresponding pests i. The dimensions of the second purchase decision tensor include: pest type dimension, farmer dimension, and pesticide type dimension. The values ​​of the elements in the second purchase decision tensor can be non-negative, preferably non-negative integers. The third purchase decision tensor y_ ij (i.e., the pesticide purchasing decision tensor) represents the amount of pesticide type j used to treat the corresponding pest i, which is the total purchase amount of each pesticide used to treat each pest. Dimensions include: pesticide type dimension, pest dimension. The values ​​of the elements in the third purchasing decision tensor are non-negative, preferably non-negative integers. The first intermediate tensor represents whether pesticides are purchased from pesticide sellers. The elements in this tensor are binary values, preferably 0 and 1. The second intermediate tensor represents whether pesticides are purchased from farmer r. The elements in this tensor are binary values, preferably 0 and 1. The relationship between the first, second, and third purchasing decision tensors is as follows:

[0067]

[0068] Among them, N is the set of pest and disease types, M is the set of pesticide types, R represents the set of other farmers, and r is the index of the farmer.

[0069] Pesticide candidate representation tensor x ij Characterizes whether pesticide j is used to treat pest i. The elements in this tensor are binary values, preferably 0 and 1. The element values ​​of each tensor set in the generated first set are randomly generated within the corresponding value range. Pesticide mixing decision tensor d ijmCharacterizes whether the pesticide types j and m used to treat pests i can be mixed. The elements in this tensor are binary values, preferably 0 and 1. The dimensions of the pesticide mixing decision tensor include: pest type dimension, first pesticide dimension, and second pesticide dimension. The pesticide mixing decision tensor characterizes whether the pesticides in the first pesticide dimension can be mixed with the pesticides in the second pesticide dimension when applying pesticides to the pest type. The pesticide mixing spraying cost characterizes the lowest mixed spraying cost when treating pests, which is the number of mixed spraying in this embodiment. For example, if five pesticides A, B, C, D, and E are used to treat pests i, and BC can be mixed, the lowest mixed spraying cost is 4, that is, A, D, and E are sprayed separately, and B and C are mixed and sprayed.

[0070] Step 202: input each tensor set in the first set into a first computing device for processing, and obtain a first result set according to the output of the first computing device.

[0071] The operations performed by the first computing device on the input tensor set include: obtaining the first pesticide procurement cost based on the first procurement decision tensor and the cost coefficient, obtaining the first pesticide distribution cost based on the first intermediate tensor and the first distribution fee, obtaining the second pesticide procurement cost based on the second procurement decision tensor, the cost coefficient and the discount coefficient, and obtaining the second pesticide distribution cost based on the second intermediate tensor and the second distribution fee; obtaining the first target value based on the first pesticide procurement cost, the second pesticide procurement cost, the first pesticide distribution cost, and the second pesticide distribution cost; obtaining the second target value based on the third procurement decision tensor, the treatment utility coefficient, the treatment area coefficient, and the infection area coefficient; outputting the first target value and the second target value; calculating the pesticide mixing cost based on the pesticide mixing decision tensor, the pesticide type coefficient, and the pesticide candidate representation tensor, and outputting it as the third target value; outputting the first target value, the second target value, and the third target value.

[0072] The first target value f1 is as follows:

[0073]

[0074] Among them, p j is the price of unit pesticide j, dis_s is the first distribution cost, dis_fr is the second distribution cost, α is the discount coefficient, and R is the set of other farmers.

[0075] The second target value f2 is as follows:

[0076]

[0077] Among them, w ij is the therapeutic efficacy coefficient of pesticide j against pest i, g ijh is the treatment area coefficient of unit pesticide j for pest i, that is, how much area can be treated by unit pesticide. i is the infection area coefficient of pest i.

[0078] According to the precautions given in the knowledge graph, when treating pests and diseases, which pesticides can be mixed and sprayed, and which pesticides cannot be mixed and sprayed. Using pesticides that can be mixed and sprayed can reduce the spraying cost. Therefore, this embodiment uses the third target value to obtain the spraying cost of pesticides. The third target value is the sum of the mixed spraying costs of each pest and disease, and the mixed spraying cost of each pest and disease is as follows:

[0079]

[0080] Among them, |M| is the number of pesticide types, C i represents the mixed spraying cost of pesticides for treating the i-th pest. The mixed spraying costs of pesticides for all pests are summed to obtain the third target value. The purpose of the present invention is to obtain an optimal pesticide matching scheme with a minimum first target value, a minimum third target value, and a maximum second target value.

[0081] Step 203, input the first set into the second computing device for processing, the processing includes: the second computing device performs crossover and mutation operations on the first set with a set probability, and then repairs it to obtain the second set; input each tensor set in the second set into the first computing device, and obtain the second result set according to the output of the first computing device; sort all the results in the first result set and the second result set, and modify the first set according to the sorting result.

[0082] The second computing device is a computing device based on a genetic algorithm. Preferably, a non-dominated sorting genetic algorithm II is selected. For the crossover operation, two elements in the set are randomly selected with a certain probability, and the genetic material between the two elements will be exchanged, so that two offspring different from either parent are obtained, that is, two new elements in the second set. In this way, a crossover operation is completed. Preferably, the probability of crossover is 0.7. For the mutation operation, a mutation point is selected with a certain probability for mutation. ij For example, if the selected x ij If x is 1, the mutation is 0; if x is 0, the mutation is 1. If the mutation point is a continuous variable, a feasible value can be selected. Preferably, the probability of mutation is 0.7. The repair operation mainly depends on the problem and is used to ensure that the algorithm only searches in the feasible space. There is a certain logical relationship between the variables. The logical relationship includes: the first, second, and third purchase decision tensors should satisfy the aforementioned summation relationship, and the amount of pesticide purchased should be the sum of the amount of pesticide purchased from the pesticide sales office and the farmer; when the element x in the pesticide candidate representation tensor ij When it is 0, the related yij ,y_s ij ,y_f rij d ijm should all be 0; ij The number of should not exceed the amount required when pesticide j is used alone to treat pest i, etc. Logical relationships are used to constrain the internal relationships between variables to ensure that the search is always within the feasible space. When the solution is inconsistent with the actual situation and does not meet the logic that the actual situation should meet, the repair operation can be used to repair it. For example, when x 12 = 0, which means that the second drug cannot treat the first disease, so y 12 The corresponding value must be 0, that is, the second drug is not purchased to treat the first disease. By adding repair operations, the algorithm convergence speed can be accelerated and the robustness of the algorithm can be improved.

[0083] The second result set is a result set corresponding to each element in the second set. The result corresponding to each element includes a first target value, a second target value, and a third target value.

[0084] Step 204, repeatedly executing the step of inputting the first set into the second computing device for processing until a termination condition is reached, and recommending a pesticide matching scheme based on the tensor set corresponding to the best result in the first set.

[0085] The termination condition includes the step S203 being executed for a set number of times. After the execution is completed, the Pareto solution set can be obtained. A person skilled in the art can determine without doubt that the first set modified after the previous execution is input into the second computing device in this step. The result corresponding to each element in the first set includes a first target value, a second target value, and a third target value. The first target value, the second target value, and the third target value are weighted and summed, and the summation results are sorted, and the best result is selected according to the sorting. Specifically, the first target value, the second target value, and the third target value are first normalized, and then the normalized results are weighted and summed. A person skilled in the art can adjust the weight of the weighted summation according to the dimension during implementation and the focus of the target. Preferably, the weights are 0.1818, 0.7273, and 0.0909, respectively. It should be noted that the optimization goal is to minimize the first target value and the third target value and maximize the second target value. Therefore, when the second target value is normalized, the opposite process needs to be performed. For example, a common normalization process is: x-min / (max-min), where x is the value to be normalized, and max and min are the maximum and minimum values, respectively. Then, when processing the second target value, the following opposite process needs to be done: 1-[(x-min) / (max-min)]. The one with the lowest sum is the optimal target value set. The pesticide matching plan can be obtained based on the tensor set corresponding to the optimal result. The pesticide matching plan includes: how much of a certain type of pesticide to purchase from pesticide sales offices and farmers for a certain type of pest (i.e., the first procurement decision tensor and the second procurement decision tensor), and how to spray when spraying for a certain type of pest (pesticide mixing decision tensor).

[0086] This embodiment takes into account the cost of mixed pesticide spraying, and obtains the cost of mixed pesticide spraying through an effective and accurate method for calculating the cost of mixed pesticide spraying, thereby reducing the cost of mixed pesticide spraying while ensuring the effectiveness of pest treatment.

[0087] Based on the above embodiments, optionally, generating a first set according to the identified pests and diseases and candidate pesticides includes the following steps.

[0088] Step 200a, for each other farmer, obtain the holding amount of various pesticides used to treat each type of pests and diseases, and obtain the initial treatment utility of each type of pests and diseases based on the pesticide holding amount and the pesticide treatment utility coefficient; generate a correction coefficient based on the cost of mixed pesticide spraying for each type of pests and diseases, and use the correction coefficient to correct the initial treatment utility to obtain the corrected treatment utility of each type of pests and diseases.

[0089] Specifically, the calculation method of initial treatment utility is as follows:

[0090]

[0091] Among them, U ri represents the initial therapeutic utility of the amount of pesticide held by other farmer r for pest type i, M r represents the set of pesticide types held by farmer r, e ij When it is 1, it means that pesticide j can treat pest i, and when it is 0, it cannot treat it. It is a binary variable, w ij is the therapeutic efficacy coefficient of pesticide j against pest i, g ij is the treatment area coefficient of unit pesticide j for pest i, hav_f rj represents the amount of pesticide type j held by farmer r. ij With x ij The difference is that eij is generated based on whether the pesticide given by the knowledge graph can treat pests and diseases, while x ij Indicates whether pesticide type j is used to treat pest type i.

[0092] According to whether the pesticides held by farmer r for treating pest type i can be mixed and sprayed, the cost COST of mixed spraying of pesticides held by farmer r for pest type i is obtained. ri . COST ri It does not involve decision tensors, and can be obtained by whether the pesticides it holds can be mixed and sprayed. ri If five pesticides A, B, C, D, and E are used to treat pest i, and BC can be mixed, the lowest mixed spraying cost is 4, that is, A, D, and E are sprayed separately, and B and C are mixed. If BC can be mixed and DE can be mixed, the mixed spraying cost for the pest type is 3.

[0093] For each pest type i, obtain the farmer mixed spraying cost COST of all other farmers ri , perform reverse normalization, use the reverse normalized pesticide spraying cost as the correction coefficient, multiply it by the initial treatment utility, and you can get the corrected treatment utility of each farmer for each type of pest and disease. A common normalization process: x-min / (max-min), where x is the value to be normalized, max and min are the maximum and minimum values, respectively. Then, the reverse normalization process needs to be: 1-[(x-min) / (max-min)].

[0094] Step 200b, calculating the degree of association between other farmers based on the corrected treatment utility of other farmers for each type of pests and diseases.

[0095] The corrected treatment utilities of all pests and diseases of an other farmer form a vector, called the corrected treatment utility vector. Each other farmer has a corrected treatment utility vector. The similarity between the corrected treatment utility vectors of every two other farmers is calculated as the degree of association. Optionally, cosine similarity can be used as a measure.

[0096] Step 200c, classifying the set of other farmers according to the degree of association between other farmers to obtain sub-sets of other farmers of different categories.

[0097] According to the degree of association between two farmers, all other farmers are classified. Optionally, a clustering algorithm is used for classification. The clustering algorithm includes a density-based clustering method and a k-means-based clustering method. The implementer can choose according to the specific implementation scenario. Optionally, when using the k-means clustering method, k is selected as After the clustering algorithm, all other farmers are divided into multiple other farmer subsets. Optionally, after clustering, outliers can be ignored.

[0098] Step 200d, selecting farmers from other farmer subsets of different categories, and generating a second purchase decision tensor according to the pesticide holdings of the selected farmers.

[0099] There are many ways to select farmers from other farmer subsets of different categories, and it is necessary to be able to cover the identified types of pests and diseases. One implementation method is to select representative farmers from other farmer subsets of different categories. That is, select the core points of different clusters obtained after clustering. Considering the distribution cost, one implementation method is to select other farmers who are closest to the farmers who need to purchase pesticides from other farmer subsets of different categories. Then, initialize the second purchase decision tensor based on the selected other farmers. Optionally, purchase all pesticides related to the types of pests and diseases held by the selected other farmers. The first purchase decision tensor can be randomly initialized within its feasible domain. Thus, the first, second, and third purchase decision tensors can be obtained.

[0100] The embodiment of the present invention obtains the correlation between farmers by obtaining the therapeutic utility vector of the amount of pesticide held by each farmer, thereby improving the quality of the initialization tensor set, thereby improving the efficiency and quality of obtaining a scientific and reasonable pesticide matching solution.

[0101] A third embodiment of the present invention is given below.

[0102] Step 301, generate a first set based on the identified pests and diseases and candidate pesticides, the first set including multiple tensor sets, each tensor set including a first procurement decision tensor, a second procurement decision tensor, a third procurement decision tensor, a pesticide mixing decision tensor, a first intermediate tensor, and a second intermediate tensor.

[0103] Step 302, generating a constraint set; the input of the first computing device also includes a constraint set. The constraint set includes: a constraint that the amount of pesticides purchased from farmers does not exceed the amount of pesticides they hold; a constraint that the elements in the third purchase decision tensor are the sum of the relevant elements in the first purchase decision tensor and the second purchase decision tensor; a constraint that each pest must be treated with at least one pesticide; a constraint that only incompatible pesticides can be mixed for spraying; and a constraint that only used pesticides can be mixed for spraying.

[0104] The quantity of pesticides purchased from farmers must not exceed the quantity of pesticides they hold:

[0105]

[0106] Among them, hav_f rj is the quantity of pesticide j owned by farmer r.

[0107] The elements in the third purchase decision tensor are the sum of the relevant elements in the first purchase decision tensor and the second purchase decision tensor:

[0108]

[0109] Among them, y ij is the third purchasing decision tensor, y_s ij is the first purchase decision tensor, y_f rij is the second purchasing decision tensor.

[0110] Each pest has at least one constraint on pesticide treatment:

[0111]

[0112] Among them, x ij Characterizes whether pesticide j is used to treat pest i.

[0113] The restrictions on the use of pesticides for mixed spraying are:

[0114]

[0115] Among them, k jm It is a pesticide mixture characterization tensor, which is generated according to whether the pesticides can be mixed in the precautions output by the knowledge graph. Characterizes whether pesticide j and pesticide m can be mixed, where the elements are binary variables. Preferably, it is 1 if it cannot be mixed, and 0 if it can be mixed. The difference between the pesticide mixture characterization tensor and the pesticide mixture decision characterization tensor lies in whether it is used for decision-making and whether it contains the pest and disease type dimension. The pesticide mixture characterization tensor is the knowledge obtained based on the knowledge graph, and does not represent whether two pesticides are ultimately used to treat pests and diseases. The pesticide mixture decision characterization tensor characterizes whether two pesticides are ultimately used to treat pests and diseases by spraying.

[0116] Constraints on the amount of pesticides purchased:

[0117]

[0118] Among them, g ij The area coefficient of pest i that can be treated by a unit of pesticide j, h i is the infection area coefficient of pest i, and aij is an auxiliary variable.

[0119] Pesticide candidate constraints:

[0120]

[0121] Among them, e ij When it is 1, it means that pesticide j can treat pest i, and when it is 0, it cannot treat it. It is a binary variable.

[0122] Constraints between the pesticide candidate representation tensor and the first purchase decision tensor:

[0123]

[0124] The above constraints are used to ensure that when one side is 0, the other side is also 0.

[0125] Pesticide spraying restrictions:

[0126]

[0127] The above constraints are used to ensure that the infected area is sprayed with pesticides.

[0128] Step 303, input each tensor set in the first set into the first computing device for processing, and obtain the target value set according to the output of the first computing device. The operations performed by the first computing device on the input tensor set include: obtaining the first pesticide purchase cost according to the first purchase decision tensor and the cost coefficient, obtaining the first pesticide distribution cost according to the first intermediate tensor and the first distribution fee, obtaining the second pesticide purchase cost according to the second purchase decision tensor, the cost coefficient and the discount coefficient, and obtaining the second pesticide distribution cost according to the second intermediate tensor and the second distribution fee; obtaining the first target value according to the first pesticide purchase cost, the second pesticide purchase cost, the first pesticide distribution cost, and the second pesticide distribution cost; obtaining the second target value according to the third purchase decision tensor, the treatment utility coefficient, the treatment area coefficient, and the infection area coefficient; and outputting the first target value and the second target value.

[0129] The first set is based on a mathematical optimization solver, which can use a Gurobi solver. The optimization is performed within the solution range limited by the constraint set, so that the first objective value is minimized and the second objective value is maximized.

[0130] Step 304: recommend a pesticide matching solution based on the tensor set corresponding to the best target value set in the first set.

[0131] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A tobacco pest and disease pesticide matching method, characterized in that: The method specifically includes: Generate a first set according to the identified pests and diseases and candidate pesticides, the first set including a plurality of tensor sets, each tensor set including a pesticide purchase decision tensor and a pesticide mixing decision tensor; Each tensor set in the first set is input into the first computing device for processing, and a target value set corresponding to each tensor set is obtained; the operations performed by the first computing device on the input tensor set include: obtaining a first target value according to the pesticide purchase decision tensor and cost information; obtaining a second target value according to the pesticide purchase decision tensor, the pesticide treatment utility coefficient and the infection area coefficient; obtaining the cost C of the pesticide mixed spraying for treating each pest i : Among them, C i represents the cost of spraying pesticide mixtures to treat the i-th pest, M is the set of pesticide types, |M| is the pesticide type coefficient, i, j, m are indexes, and the pesticide mixture decision tensor d ijm Characterizes whether the pesticide types j and m used to treat pests i can be mixed, and the pesticide candidate characterization tensor x ij Indicates whether pesticide j is used to treat pest i; sums up the cost of spraying mixed pesticides for all types of pests to obtain the third target value; A pesticide matching scheme is recommended based on the tensor set corresponding to the best target value set in the first set.

2. A tobacco pest pesticide matching method as claimed in claim 1, characterized in that: The pesticide procurement decision tensor includes a first procurement decision tensor and a second procurement decision tensor; the first procurement decision tensor represents the quantity of pesticides purchased from pesticide sellers for treating pests and diseases, and the dimensions of the first procurement decision tensor include: pest and disease type dimension, pesticide type dimension; the second procurement decision tensor represents the quantity of pesticides purchased from other farmers for treating pests and diseases, and the dimensions of the second procurement decision tensor include: pest and disease type dimension, farmer dimension, pesticide type dimension.

3. A tobacco pest pesticide matching method as claimed in claim 2, characterized in that: Generating a first set according to the identified pests and diseases and candidate pesticides includes: For each other farmer, the holding amount of various pesticides used to treat each type of pests and diseases is obtained, and the initial treatment utility of each type of pests and diseases is obtained based on the holding amount of pesticides and the treatment utility coefficient of the pesticides; a correction coefficient is generated based on the cost of mixed spraying of pesticides for each type of pests and diseases, and the initial treatment utility is corrected using the correction coefficient to obtain the corrected treatment utility of each type of pests and diseases; The degree of association among other farmers was calculated based on their corrected treatment utility for each pest and disease type; Classify the set of other farmers according to the degree of connection between them, and obtain sub-sets of other farmers in different categories; Farmers are selected from other subsets of farmers in different categories, and a second purchase decision tensor is generated according to the pesticide holdings of the selected farmers.

4. A tobacco pest pesticide matching method as claimed in claim 1, characterized in that: Before the step of generating the first set according to the identified pests and diseases and candidate pesticides, the step further includes: Collect images of tobacco growing areas; Identify pests and diseases in images of tobacco growing areas; Candidate pesticides for identified pests and diseases.

5. A tobacco pest pesticide matching method as claimed in claim 1, characterized in that: Before the step of recommending a pesticide matching solution according to the tensor set corresponding to the best target value set in the first set, the following step is further included: The first set is input into the second computing device for processing: the second computing device performs crossover and mutation operations on the first set with a set probability, and obtains the second set after repair; each tensor set in the second set is input into the first computing device, and a second result set is obtained according to the output of the first computing device; all results in the first result set and the second result set are sorted, and the first set is modified according to the sorting result; The step of inputting the first set into the second computing device for processing is repeatedly executed until a termination condition is reached.

6. A tobacco pest pesticide matching method as claimed in claim 1, characterized in that: The step of inputting each tensor set in the first set into the first computing device for processing and obtaining the first result set according to the output of the first computing device also includes: generating a constraint set; the input of the first computing device also includes the constraint set.

7. A tobacco pest pesticide matching method as claimed in claim 6, characterized in that: The constraint set includes: Constraints between the pesticide mixture decision tensor and the pesticide mixture representation tensor: Among them, d ijm is the pesticide mixture decision tensor, k jm is the pesticide mixture representation tensor, N is the set of identified pests and diseases, M is the set of candidate pesticide types, and i, j, and m are indexes; Constraints on the amount of pesticides purchased: Among them, y ij is the pesticide purchasing decision tensor, g ij The area coefficient of pest i that can be treated by a unit of pesticide j, h i is the infection area coefficient of pest i, a ij is an auxiliary variable; Pesticide candidate constraints: Among them, x ij is the pesticide candidate representation tensor, e ij Indicates whether pesticide j can treat pest i; Constraints between the pesticide candidate representation tensor and the pesticide procurement decision tensor: Among them, x ij is the pesticide candidate representation tensor, y ij is the pesticide purchasing decision tensor; Pesticide spraying restrictions: Among them, y ij is the pesticide purchasing decision tensor, g ij It represents the area coefficient of pest i that can be treated by unit pesticide j, and N is the set of pest types.

8. A tobacco pest and disease pesticide matching system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, a tobacco pest and disease pesticide matching method according to any one of claims 1 to 7 is implemented.

Images