A pepper pathology reasoning method based on graph computing

Abstract

The application discloses a kind of pepper pathology inference methods based on graph computing, including the construction of static pepper pathology knowledge graph;The construction of time sequence pepper pathology knowledge graph;The construction of time sequence pepper pathology inference model;The application of time sequence pepper pathology inference model and so on four steps.The construction of static pepper pathology knowledge graph will all nodes be divided into pepper disease node and the characteristic node that pepper disease shows, and then pepper disease node is connected with the characteristic node that pepper disease shows;The construction of time sequence pepper pathology knowledge graph describes the evolution process of characteristic in the process that pepper is diseased in the mode of time sequence knowledge graph;The construction of time sequence pepper pathology knowledge graph inference model adopts four graph convolutional neural networks, also includes VAE module and XL-VAE module;The application of time sequence pepper pathology inference model infers the disease of this pepper;The application can be applied to pepper pathology inference, with good accuracy and practical application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, and in particular to a pepper pathology reasoning method based on graph calculation. BACKGROUND

[0002] A knowledge graph represents structured semantic information with nodes, and forms a network graph through the relationship between nodes. Due to different knowledge coverage focuses, the knowledge graph is divided into general domain knowledge graph and vertical domain knowledge graph. The general domain knowledge graph faces open fields such as common sense knowledge, relies on a large amount of data generated by the Internet and knowledge courses, extracts entities and their relationships existing therein, and establishes a general domain knowledge graph. The vertical domain knowledge graph mainly faces specific fields such as finance, medicine, and education, and relies on internal industry data. The vertical domain knowledge graph fully considers the significance of a specific industry, application scenarios, and users. The vertical domain knowledge graph meets the industry demand by processing entity attributes and data patterns. Whether it is a general domain knowledge graph or a vertical domain knowledge graph, a knowledge graph for pepper disease research still needs to be studied. The present application constructs a static pepper disease knowledge graph and a time-series pepper disease knowledge graph, converts pepper disease description data from unstructured data to structured data, and lays a data foundation for pepper disease reasoning.

[0003] A graph convolution network has strong advantages for non-Euclidean data widely existing in the real world. It can process massive social network data, analyze the relationship between people by equating a social network to a graph, model chemical molecular structures to assist new drug research, accurately construct user interaction patterns to improve the accuracy and efficiency of a recommendation system, and has a wide range of applications in language processing and complex word-sentence relationships. Thanks to this inspiration, the static pepper pathology knowledge graph previously constructed by the present application equates pepper diseases and their performance characteristics to the relationship between nodes in a graph, and equates disease reasoning to a node classification task in a graph. The graph convolution network can efficiently process such tasks.

[0004] Knowledge graph reasoning is a key task in natural language processing. When a knowledge graph is upgraded to a time-series knowledge graph, there are relatively few related research tasks, and the related research tasks focus on reasoning about the past and cannot predict future facts. The present application makes the following considerations: the growth process of a pepper plant is a dynamic process, and different diseases have different performance characteristics during the disease process. Based on this theoretical basis, a time-series pepper pathology knowledge graph is constructed, and a reasoning method based on the time-series pepper pathology knowledge graph is explored to provide an effective solution for pepper disease reasoning. SUMMARY

[0005] The application aims to construct a pepper pathology knowledge graph and utilize the constructed pepper pathology knowledge graph to perform pepper disease reasoning, and provides a pepper pathology reasoning method based on graph calculation.

[0006] The application aims to achieve the following technical solutions:

[0007] A pepper pathology reasoning method based on graph calculation comprises the following steps:

[0008] Step S1, construction of a static pepper pathology knowledge graph;

[0009] Step S2, construction of a time-series pepper pathology knowledge graph;

[0010] Step S3, construction of a time-series pepper pathology reasoning model;

[0011] Step S4, application of the time-series pepper pathology reasoning model.

[0012] Specifically, the construction of the static pepper pathology knowledge graph divides all nodes into pepper disease nodes and characteristic nodes of pepper diseases, and then connects the pepper disease nodes with the characteristic nodes of pepper diseases, i.e., there are edges.

[0013] Specifically, the construction of the time-series pepper pathology knowledge graph comprises the following sub-steps:

[0014] Step S21, construction of a time-series knowledge graph for the dynamic growth evolution process of each pepper disease by taking day 1, day 3, day 5 and day 7 as the observation time granularity of the time-series pepper pathology knowledge graph;

[0015] Step S22, division of nodes in the time-series pepper pathology knowledge graph into two categories, i.e., nodes of each part of a pepper and nodes of the performance of each part when suffering from a disease.

[0016] Specifically, the construction of the time-series pepper pathology reasoning model adopts four independent graph convolutional neural networks corresponding to day 1, day 3, day 5 and day 7 in the time-series pepper pathology knowledge graph.

[0017] The input of the graph convolutional neural network takes the time-series pepper pathology knowledge graph at the corresponding time point as the input, and a VAE module and an XL-VAE module are arranged between adjacent time points.

[0018] Specifically, the XL-VAE module takes a feature matrix H output by the graph convolutional neural network as the input.

[0019] The node feature vector, the edge feature vector and the event feature vector are outputs.

[0020] Wherein, the node feature vector and the edge feature vector are taken as the input of the graph convolutional neural network at the next moment, and the event feature vector is the reasoning result of the pepper disease at the current moment.

[0021] The XL-VAE module further comprises an encoder, which is supervised trained by the known pepper disease feature state at the current moment, so that the XL-VAE can more accurately represent the discrete sequence of the feature matrix.

[0022] Specifically, the VAE module updates the weight matrix W shared by all graph convolutional neural networks, including the following sub-steps:

[0023] Step S31, taking the weight matrix W of the previous moment GCN as the input of the VAE, and obtaining the weight matrix W` through VAE sampling;

[0024] Step S32, taking the weight matrix W` as the weight matrix of the next moment GCN, and obtaining the weight matrix W`` through operation of the next moment GCN;

[0025] Step S33, using the weight matrix W`` for reverse training of the VAE.

[0026] Specifically, the application of the time sequence pepper pathology reasoning model comprises the following sub-steps:

[0027] S41. Obtain the disease feature performance of the pepper plant from the diseased pepper picture;

[0028] S42. Refine the feature performance obtained in S41 into an entity node by using a static pepper pathology knowledge graph, and convert the finally obtained pepper pathology knowledge graph into an input format required by the time sequence pepper pathology reasoning model;

[0029] S43. Input the converted pepper pathology knowledge graph in S42 into the time sequence pepper pathology reasoning model to reason the pepper pathology.

[0030] Specifically, the feature performance includes performance characteristics of roots, stem bases, leaves and the like.

[0031] The beneficial effects of the present application are:

[0032] The present application provides a pepper pathology reasoning method based on graph calculation, which converts pepper disease description unstructured data into structured data, and provides a pepper pathology reasoning method based on graph calculation, filling the blank of pepper disease reasoning method based on graph calculation. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 A flowchart of the pepper pathology reasoning method based on graph calculation of the present application;

[0034] Figure 2 A structure diagram of a time sequence pepper pathology reasoning model;

[0035] Figure 3 A node schematic diagram of the static pepper pathology knowledge graph. DETAILED DESCRIPTION

[0036] It should be understood that the specific embodiments described herein are merely illustrative of the present application and do not limit the present application.

[0037] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application. In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on that a person of ordinary skill in the art can realize, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, and is not within the protection scope of the present application.

[0038] As shown in Figure 1 A pepper pathology reasoning method based on graph calculation, which comprises construction of a static pepper pathology knowledge graph, construction of a time sequence pepper pathology knowledge graph, construction of a time sequence pepper pathology reasoning model, and application of the time sequence pepper pathology reasoning model.

[0039] The construction of the static pepper pathology knowledge graph, as shown in Figure 3 The steps include:

[0040] The nodes in the static pepper pathology knowledge graph are divided into two categories, one category is a pepper disease node, and the other category is a characteristic node exhibited by the pepper disease;

[0041] There is only one relationship in the static pepper pathology knowledge graph, which is "exhibited", and it is specifically understood as that a certain type of pepper disease is exhibited by certain characteristics, at this time, the pepper disease node is connected with certain specific characteristic nodes, that is, there is an edge;

[0042] The construction of the time sequence pepper pathology knowledge graph, the steps include:

[0043] Take the 1st day, the 3rd day, the 5th day, and the 7th day as the time granularity of the time sequence pepper pathology knowledge graph, and construct a time sequence knowledge graph for the dynamic growth and evolution process of each pepper disease. In this process, the description for constructing the time sequence pepper pathology knowledge graph may not be clear enough, at which time the static pepper pathology knowledge graph constructed previously can be used to more accurately obtain instance information. Taking pepper blight as an example, part of the data is: (pepper blight, root | first day, brown), (pepper blight, stem base | first day, light brown), (pepper blight, stem base | third day, black brown)…;

[0044] There are two types of nodes in the time sequence pepper pathology knowledge graph, which are the parts of the pepper and the performance of each part when suffering from a certain disease. There is only one type of relationship in the time sequence pepper pathology knowledge graph, which is performance, such as the root performance being brown.

[0045] The construction of the time sequence pepper pathology knowledge graph reasoning model is shown in Figure 2 The steps include:

[0046] The task of the time sequence pepper pathology knowledge graph reasoning model is graph classification, that is, judging the final disease of the pepper plant according to the characteristics shown by the part of the pepper disease at a certain time node;

[0047] As shown in Figure 2 The time sequence pepper pathology knowledge graph takes the 1st day, the 3rd day, the 5th day, and the 7th day as the time granularity of the observation characteristics, so the time sequence pepper pathology knowledge graph reasoning model needs 4 independent graph convolutional neural networks (GCN). The input of each GCN will take the time sequence pepper pathology knowledge graph at the corresponding time point as the input. In addition, there is a VAE module and an XL-VAE module between adjacent time points. The XL-VAE takes the feature matrix H output by the GCN as the input. The node feature vector, the edge feature vector, and the event feature vector are taken as the output, wherein the node feature vector and the edge feature vector are taken as the input of the GCN at the next moment, and the event feature vector is the reasoning result of the pepper disease at the current moment. The encoder of the XL-VAE can be supervised and trained using the known characteristics of the pepper disease at the current moment, so that the XL-VAE can more accurately represent the discrete sequence of the feature matrix. Each time GCN uses a shared weight matrix W, which is updated as the time node moves. The update method is to use the VAE module, take the weight matrix W of the GCN at the previous moment as the input of the VAE, sample the weight matrix W` through the VAE, and take the weight matrix W` as the weight matrix of the GCN at the next moment. The weight matrix W`` obtained by the GCN at the next moment is used for training the VAE in reverse;

[0048] The application of the time sequence pepper pathology reasoning model includes the following steps:

[0049] Obtain the disease characteristics of the pepper plant from the pictures of the diseased pepper, including the performance characteristics of the roots, stem bases, leaves and other parts;

[0050] The characteristics obtained in S1 are more accurately refined into entity nodes by using the previously constructed static pepper pathology knowledge graph, and the finally obtained pepper pathology knowledge graph is converted into the input format required by the time sequence pepper pathology reasoning model;

[0051] The characteristics obtained in S2 from the pictures of the diseased pepper are input into the time sequence pepper pathology reasoning model to reason the pepper pathology.

[0052] The above only describes the preferred embodiments of the present application, and it should be understood that the present application is not limited to the forms disclosed herein, and should not be considered as excluding other embodiments, but various other combinations, modifications and environments can be made within the scope of the concepts described herein, and can be modified by the above-mentioned teaching or related technical or knowledge. The modifications and changes made by the person skilled in the art without departing from the spirit and scope of the present application shall be within the protection scope of the claims of the present application.

Claims

1. A pepper pathology reasoning method based on graph computation, characterized in that, Comprise the following steps: Step S1, construction of static pepper pathology knowledge graph; Step S2, construction of time sequence pepper pathology knowledge graph; Step S3, construction of time sequence pepper pathology reasoning model; Step S4, application of time sequence pepper pathology reasoning model; The construction of the time sequence pepper pathology knowledge graph comprises the following substeps: Step S21, taking the 1st day, the 3rd day, the 5th day, and the 7th day as the time sequence pepper pathology knowledge graph observation time granularity, the dynamic growth evolution process of each pepper disease is constructed into a time sequence knowledge graph; Step S22, the nodes in the time sequence pepper pathology knowledge graph are divided into two categories: pepper part nodes and performance nodes of each part when suffering from a disease; The construction of the time sequence pepper pathology reasoning model adopts four independent graph convolutional neural networks corresponding to the 1st day, the 3rd day, the 5th day, and the 7th day in the time sequence pepper pathology knowledge graph; The input of the graph convolutional neural network takes the time sequence pepper pathology knowledge graph at the corresponding time point as the input, and VAE modules and XL-VAE modules are set between adjacent time points; The XL-VAE module takes the graph convolutional neural network output feature matrix H as the input; The node feature vector, the edge feature vector, and the event feature vector are outputs; Among them, the node feature vector and the edge feature vector are the inputs of the next moment graph convolutional neural network, and the event feature vector is the current moment pepper disease reasoning result; The XL-VAE module further comprises an encoder, which is supervised and trained with the known pepper disease feature state at the current moment, so that the XL-VAE can more accurately represent the discrete sequence of the feature matrix; The VAE module updates the weight matrix W shared by all graph convolutional neural networks, comprising the following substeps: Step S31, taking the weight matrix W of the previous moment GCN as the input of the VAE, and obtaining the weight matrix W` through VAE sampling; Step S32, taking the weight matrix W` as the weight matrix of the next moment GCN, and obtaining the weight matrix W`` through the operation of the next moment GCN; Step S33, using the weight matrix W`` for back training VAE.

2. The pepper pathology reasoning method based on graph computation according to claim 1, wherein, The construction of the static pepper pathology knowledge graph divides all nodes into pepper disease nodes and characteristic nodes of pepper diseases, and then connects the pepper disease nodes and the characteristic nodes of pepper diseases, that is, there is an edge.

3. The pepper pathology reasoning method based on graph computation according to claim 1, wherein, The application of the time sequence pepper pathology reasoning model comprises the following substeps: S41. Obtain the characteristic performance of the pepper plant from the diseased pepper picture; S42. Extract the characteristic performance obtained in S41 into entity nodes with the static pepper pathology knowledge graph, and convert the finally obtained pepper pathology knowledge graph into the input format required by the time sequence pepper pathology reasoning model; S43. Input the converted pepper pathology knowledge graph in S42 into the time sequence pepper pathology reasoning model to reason the pepper pathology.

4. The pepper pathology reasoning method based on graph computation according to claim 3, wherein, The characteristic performance includes the performance characteristics of the root, stem base, and leaf parts.

Images