A method and system for constructing a high-standard farmland healthy tillage layer
Through multi-dimensional feature portrait and graph convolutional autoencoder technology, combined with domain knowledge graph, the problem of soil quality in high-standard farmland is solved, the construction of healthy tillage layers and the improvement of crop stress resistance is achieved, and the rapid recovery of farmland and Efficient production.
Patent Information
- Application Number
- CN202411207680.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-08-30
AI Technical Summary
The soil quality of high-standard farmland does not meet the standards and the grain productivity is low, making it difficult to build a healthy fertile tillage layer suitable for the growth of crop roots.
By obtaining the soil physical and chemical properties, microbial flora, disease and pest distribution and fertility levels of farmland, we construct multi-dimensional feature portraits, using graph convolution autoencoder and domain knowledge graphs, we determine the adaptive environment and soil improvement methods, and generate healthy tillage construction work. flow, coordinate the water, fertilizer, gas and heat conditions of the soil.
It improves the stress resistance of crops, builds a healthy and fertile soil environment in tillage, quickly restores farmland soil to suitable standards, and accelerates the construction and put into use of high-standard farmland.
Smart Images

Figure CN118735716B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of farmland management technology, and more specifically, to a method and system for constructing a high-standard farmland healthy tillage layer. Background Art
[0002] High-standard farmland refers to arable land that is flat, concentrated and contiguous, has complete facilities, is equipped with agricultural electricity, has fertile soil, good ecology, strong disaster resistance, and is compatible with modern agricultural production and management methods. It is high-yield and stable, and is designated as permanent basic farmland. In the process of farmland construction, due to the need for trenching, turning the soil and digging operations, it is easy to destroy the original ecological construction of the farmland. In order to quickly restore the planting needs of the constructed farmland, the soil of the constructed farmland needs to be improved.
[0003] In view of the current problems of substandard soil quality and low grain productivity in high-standard farmland, it is an important scientific and technological measure to improve the grain productivity of current high-standard farmland by applying appropriate soil improvement technologies, reasonable soil conditioners, precise water and fertilizer control technologies and scientific management and maintenance methods, breaking the surface tension of soil clay particles, forming water-binding semi-micelles on the surface of soil clay particles and organic matter colloids, forming soil aggregates through hydrogen bonds, van der Waals forces, etc., increasing soil porosity, reducing soil bulk density, loosening the soil, breaking soil compaction, deepening the effective tillage layer, expanding the soil's capacity to absorb water and nutrients, coordinating the "water, fertilizer, air, and heat" conditions of the soil tillage layer, improving crop resistance, and building a healthy and fertile tillage layer soil environment for crop growth and development. Therefore, how to cultivate a healthy and fertile tillage layer suitable for crop root growth through organic fertilization, interface regulation, water and fertilizer management and other measures is an urgent problem to be solved. Summary of the invention
[0004] In order to solve the above technical problems, the present invention proposes a method and system for constructing a high-standard healthy tillage layer of farmland, with the aim of conducting rational management based on the current farmland pests and diseases, soil properties and the distribution of beneficial microbial colonies in the soil, constructing a healthy and fertile tillage layer, and ensuring that the farmland soil meets the needs of crop growth.
[0005] The first aspect of the present invention provides a method for constructing a high-standard farmland healthy plough layer, comprising:
[0006] Obtain the soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level of the target farmland, analyze the multidimensional influencing factors of the cultivated soil environment, and construct a multidimensional characteristic portrait of the target farmland;
[0007] Obtain the desired crops of the target farmland, access the relevant domain knowledge graph using the crop category, perform similarity integration in the domain knowledge graph based on the multi-dimensional feature portrait, and determine the suitable growth environment of the desired crops;
[0008] The stacked graph convolutional autoencoder is used to learn the graph structure of the knowledge graph subgraph corresponding to the suitable environment, the multi-dimensional feature data of the target farmland is reconstructed, and the direction of improving the soil quality of the topsoil layer of the target farmland is obtained by using the data distribution deviation;
[0009] Based on the direction of improving the soil quality of the topsoil layer, appropriate soil improvement measures, soil conditioners and management methods are obtained, and a workflow for constructing a healthy topsoil layer of the target farmland is generated.
[0010] In this plan, the multi-dimensional influencing factors of the soil environment of the target farmland are analyzed, specifically:
[0011] Using soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level as search tags, a big data method is used to obtain soil environmental assessment instances corresponding to each dimensional tag, read the characteristic variables involved in the soil environmental assessment instance, and generate a multi-level characteristic system;
[0012] Data collection and processing are performed based on a multi-level feature system, the feature parameters of the feature variables are standardized and normalized, and the feature variables are encoded in sequence to construct an original feature variable set;
[0013] Calculate the Pearson correlation coefficient between each characteristic variable in the original characteristic variable set and the soil environment of the cultivated layer to represent the correlation, and obtain the characteristic variable with the greatest correlation and put it into the optimal characteristic set;
[0014] Introducing the characteristic variables in the original characteristic variable set into the optimal characteristic set one by one, taking the maximum correlation index between the characteristic variables and the soil environment of the cultivated layer and the minimum redundancy index between the characteristic variables as measurement indicators, and introducing the characteristic variables into the optimal characteristic set;
[0015] In addition, in the optimal feature set, the feature variables are measured in terms of the Pearson correlation coefficient and the Gini index, and the measurement results are standardized. The standardized measurement results are mapped into a score vector, and the importance of the feature variables is represented according to the vector length, and a preset number of feature variables are selected as multidimensional influencing factors.
[0016] In this scheme, a multi-dimensional feature portrait of the target farmland is constructed, specifically:
[0017] According to the multidimensional characteristic variables corresponding to the multidimensional influencing factors, parameters are extracted from multiple label data of soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level of the target farmland, and multidimensional characteristic data are constructed after preprocessing;
[0018] Input the multi-dimensional feature data into the CNN network for re-feature extraction, use several filters in the convolution layer to obtain local context features, obtain low-dimensional representations of feature variables, and aggregate to generate feature sets;
[0019] The feature set is subjected to a maximum pooling operation, the maximum values of all features are concatenated, a self-attention layer is introduced to assign different weights to different feature variables, the feature variables of multidimensional influencing factors are weighted and then connected to a fully connected layer, and a multidimensional feature portrait of the target farmland is output.
[0020] In this solution, similarity integration is performed in the domain knowledge graph based on the multi-dimensional feature portrait to determine the suitable environment for the desired crop, specifically:
[0021] Access the knowledge graph of related fields of the expected crop categories of the target farmland, and construct the query intent using the multi-dimensional feature portrait of the target farmland and the crop planting;
[0022] The Euclidean distance is preset as the metric function to represent the similarity. Entity recognition is performed by calculating the similarity between the query intent and the entities in the domain knowledge graph, identifying the key entities as known conditions and determining the positions of the key entities in the domain knowledge graph.
[0023] According to the position of the key entity, all entities within a preset distance from the key entity are searched and marked, and a knowledge graph subgraph containing the key entity, the marked entity and the relationship is obtained. The standard soil environmental parameters corresponding to the feature variables in the multidimensional feature portrait are extracted from the knowledge graph subgraph to determine the suitable growth environment of the desired crop.
[0024] In this solution, a stacked graph convolutional autoencoder is used to learn the graph structure of the knowledge graph subgraph corresponding to the suitable environment, specifically:
[0025] In the knowledge graph subgraph obtained by the query, the weight information involved in the multidimensional feature portrait of the target farmland is used to set the edge weights of the relationship between entities and construct a weight matrix;
[0026] The autoencoder is improved by using graph convolutional neural network to obtain stacked graph convolutional autoencoders. The standard soil environmental parameters corresponding to the suitable environment are used to train each layer of autoencoders in a layer-by-layer unsupervised manner. The graph structure of the knowledge graph subgraph is learned through information propagation of graph convolution.
[0027] The output of the graph convolution is obtained as the hidden layer feature and imported into the next layer for training. The hidden layer feature matrix output by the last layer of graph convolution is combined with the weight matrix and sent to the decoder network for data reconstruction. Back propagation is used to optimize the parameters of the stacked graph convolution autoencoder.
[0028] In this scheme, the multidimensional characteristic data of the target farmland is reconstructed, and the direction of improving the soil quality of the topsoil layer of the target farmland is obtained by using the data distribution deviation, specifically:
[0029] The multi-dimensional feature data of the target farmland is obtained, and input data is generated by using a sliding window segmentation, and a stacked graph convolution autoencoder is imported to obtain potential data distribution, and the input data is reconstructed using the potential data distribution, and the data distribution of the reconstructed output is obtained;
[0030] The distribution deviation between the input data and the reconstructed data under different feature variable categories is calculated to obtain the anomaly score of each feature variable. If the anomaly score is greater than the preset threshold, it is judged as abnormal. The direction of improving the topsoil quality of the target farmland is determined according to the category of the abnormal feature variable.
[0031] In this scheme, appropriate soil improvement means, soil conditioners and management methods are obtained based on the direction of improving the quality of the cultivated layer soil, specifically:
[0032] The direction of improving the quality of the cultivated layer soil is used to locate in the domain knowledge graph, the domain knowledge graph is represented and learned through MetaPath random walk, and the nodes of the domain knowledge graph are pre-trained and embedded into the high-dimensional representation space;
[0033] Based on soil improvement methods, soil conditioners and management methods as target nodes, several meta-paths containing positioning nodes and target nodes for improving the quality of cultivated layer soil are obtained. The meta-paths are represented using the graph attention mechanism to obtain the importance of the meta-paths corresponding to different directions of improving the quality of cultivated layer soil.
[0034] The most important meta-path is selected in the direction of improving soil quality in different cultivated layers to obtain the corresponding target node, and the soil improvement means, soil conditioners and management methods required for the target farmland are determined according to the target node.
[0035] The second aspect of the present invention provides a high-standard farmland healthy plough layer construction system, including: a memory, a processor, a soil multi-dimensional feature acquisition module, a multi-dimensional feature image construction module, a crop suitable environment analysis module, a plough layer soil improvement recommendation module and a healthy plough layer program integration module, the memory and the processor store and execute a high-standard farmland healthy plough layer construction method program;
[0036] The soil multi-source data acquisition module obtains the soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level of the target farmland, and constructs a multi-dimensional feature data sequence after preprocessing;
[0037] The multidimensional feature image construction module analyzes the multidimensional feature variable influencing factors of the target farmland plough layer soil environment and constructs a multidimensional feature image of the target farmland;
[0038] The crop suitable environment analysis module accesses the relevant field knowledge graph based on the desired crops in the target farmland to determine the suitable environment for the desired crops;
[0039] The topsoil improvement recommendation module uses a graph convolutional autoencoder to learn the domain knowledge graph structure, determine the direction of improving the topsoil quality of the target farmland, and obtain appropriate soil improvement means, soil conditioners and management methods;
[0040] The healthy tillage layer program integration module integrates the acquired soil improvement means, soil conditioners and management methods, and generates a target farmland healthy tillage layer construction workflow according to key steps and sequences.
[0041] Compared with the prior art, the present invention has the following beneficial effects:
[0042] The present invention coordinates the "water, fertilizer, air, and heat" conditions of the soil plough layer by acquiring soil improvement technical means, reasonable soil conditioners, and scientific management and maintenance methods that match the target farmland, improves crop stress resistance, and builds a healthy and fertile plough layer soil environment for crop growth and development;
[0043] Generate a workflow for improving farmland topsoil, guide the rapid construction of a healthy topsoil, enable the treated topsoil to recover to the suitable standards for crops in a short period of time, and accelerate the construction and commissioning of high-standard farmland. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] 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.
[0045] Figure 1 A flow chart showing a method for constructing a healthy tillage layer of high-standard farmland is shown;
[0046] Figure 2 A flowchart for learning to represent a knowledge graph subgraph using a stacked graph convolutional autoencoder is shown;
[0047] Figure 3 A flow chart for obtaining suitable soil improvement means, soil conditioners and management methods is shown;
[0048] Figure 4 The present invention shows an architecture diagram of a high-standard farmland healthy tillage layer construction system. DETAILED DESCRIPTION
[0049] 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.
[0050] 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.
[0051] Figure 1 A flow chart of a method for constructing a healthy tillage layer of high-standard farmland is shown.
[0052] like Figure 1 As shown, the present invention provides a method for constructing a high-standard farmland healthy tillage layer, comprising:
[0053] S102, obtain the soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level of the target farmland, analyze the multidimensional influencing factors of the cultivated layer soil environment, and construct a multidimensional characteristic portrait of the target farmland;
[0054] S104, obtaining the desired crops of the target farmland, accessing the relevant domain knowledge graph using the crop categories, integrating the similarities in the domain knowledge graph according to the multi-dimensional feature portraits, and determining the suitable growth environment of the desired crops;
[0055] S106, using a stacked graph convolutional autoencoder to learn the graph structure of the knowledge graph subgraph corresponding to the suitable environment, reconstructing the multi-dimensional feature data of the target farmland, and using the data distribution deviation to obtain the direction of improving the soil quality of the topsoil layer of the target farmland;
[0056] S108, based on the direction of improving the soil quality of the topsoil layer, appropriate soil improvement measures, soil conditioners and management methods are obtained, and a workflow for constructing a healthy topsoil layer of the target farmland is generated.
[0057] It should be noted that, according to the labels such as soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level as search labels, the search engine combined with the big data retrieval method is used to obtain the soil environment assessment instances corresponding to each dimensional label, the characteristic variables involved are read in the soil environment assessment instance, and a multi-level characteristic system is generated, such as soil physical and chemical property measurement indicators, including: effective potassium, soil pH, soil bulk density, effective carbon, soil temperature, etc.; data collection and processing are carried out based on the multi-level characteristic system, the characteristic parameters of the characteristic variables are standardized and normalized, and the characteristic variables are encoded in sequence to construct the original characteristic variable set; the Pearson correlation coefficient between each characteristic variable in the original characteristic variable set and the tillage layer soil environment is calculated to characterize the correlation, and the characteristic variables with the largest correlation are obtained and put into the optimal feature set; the characteristic variables in the original characteristic variable set are introduced one by one into the optimal feature set, and the maximum correlation index between the characteristic variables and the tillage layer soil environment and the minimum redundancy index between the characteristic variables are used as measurement indicators, and the characteristic variables are introduced into the optimal feature set using the mRMR algorithm; the expression of the mRMR algorithm is:
[0058]
[0059] in, Indicates Feature variables Due to the soil environment of the arable layer The Pearson correlation coefficient between Indicates Feature variables With Feature variables The Pearson correlation coefficient between represents the feature variables put into the optimal feature set, represents the number of feature variables in the optimal feature set, Represents the original feature variable set.
[0060] In addition, in the optimal feature set, the feature variables are measured in terms of the Pearson correlation coefficient and the Gini index. The Gini index measures the degree of confusion of each feature variable in the optimal feature set. The smaller the Gini index, the smaller the uncertainty. The measurement results are standardized, and the standardized measurement results are mapped into a score vector. The importance of the feature variables is represented according to the length of the vector, and a preset number of feature variables are selected as multidimensional influencing factors. Feature selection through the Pearson correlation coefficient and the Gini index reduces the volatility of the feature variables, enhances the characterization measurement of the feature variables on the topsoil environment, and removes noise while selecting important feature variables that significantly affect the topsoil environment.
[0061] It should be noted that distributed sampling and testing is carried out on the cultivated layer soil of the target farmland to obtain parameters such as soil compaction analysis, pH value analysis, soil composition analysis, trace element analysis in the soil, soil pest and disease distribution analysis and microbial count to determine the soil physical and chemical properties, soil microbial flora, soil pest and disease distribution and soil fertility level. Parameters are extracted from multiple label data of soil physical and chemical properties, soil microbial flora, soil pest and disease distribution and soil fertility level of the target farmland according to the multidimensional feature variables corresponding to the multidimensional influencing factors, and multidimensional feature data is constructed after preprocessing; the multidimensional feature data is input into the CNN network for re-feature extraction, and several filters are used in the convolution layer to obtain local context features, obtain several feature maps, and aggregate to generate a feature set; the feature set is subjected to the maximum pooling operation to extract the most important information, the maximum values of all features are concatenated, and the output of the convolution layer after the maximum pooling layer is obtained. , import the self-attention layer to assign different weights to different feature variables ,in, are the model parameters corresponding to the query vector, value vector and key vector respectively, The characteristic variables of the multidimensional influencing factors are weighted as scaling factors and then connected to the fully connected layer to output a multidimensional feature portrait of the target farmland.
[0062] It should be noted that the relevant domain knowledge graph of the expected crop categories of the target farmland is accessed, and the query intent is constructed using the multidimensional feature portrait of the target farmland and crop planting; the Euclidean distance is preset as the metric function to characterize the similarity, and entity recognition is performed by calculating the similarity between the query intent and the entities in the domain knowledge graph, identifying the key entities as known conditions, and judging the position of the key entities in the domain knowledge graph; according to the position of the key entity, all entities within the preset distance from the key entity are searched for marking, and the preset distance is preferably set to 1, and the soil environment parameter entity directly connected to the key entity is selected. A knowledge graph subgraph containing key entities, marked entities and relationships is obtained, and the query results are displayed in the form of a knowledge graph subgraph. Through the intuitive and flexible characteristics of the graph database, the relevant knowledge entities are clearly and intuitively displayed. The standard soil environment parameters corresponding to the feature variables in the multidimensional feature portrait are extracted in the knowledge graph subgraph to determine the suitable environment for the expected crops.
[0063] Figure 2 A flowchart showing how to use stacked graph convolutional autoencoders to learn to represent knowledge graph subgraphs.
[0064] According to an embodiment of the present invention, a stacked graph convolutional autoencoder is used to learn the graph structure of the knowledge graph subgraph corresponding to the adaptive environment, specifically:
[0065] S202, in the knowledge graph subgraph obtained by the query, using the weight information involved in the multidimensional feature portrait of the target farmland to set the edge weights of the relationship between entities and construct a weight matrix;
[0066] S204, using a graph convolutional neural network to improve the autoencoder to obtain a stacked graph convolutional autoencoder, using standard soil environmental parameters corresponding to the suitable environment to train each layer of the autoencoder in a layer-by-layer unsupervised manner, and learning the graph structure of the knowledge graph subgraph through information propagation of graph convolution;
[0067] S206, obtaining the output of the graph convolution as the hidden layer feature, and importing it into the next layer for training, combining the hidden layer feature matrix output by the last layer of graph convolution with the weight matrix, transmitting it to the decoder network for data reconstruction, and using back propagation to optimize the parameters of the stacked graph convolution autoencoder.
[0068] It should be noted that the weight matrix is constructed according to the edge weights of the relationship between entity nodes in the knowledge graph subgraph, and the weight matrix contains the graph structure of the knowledge graph subgraph. The multi-dimensional feature data of the target farmland is obtained. The input data is generated by sliding window segmentation and imported into the stacked graph convolution autoencoder. The message propagation between the graph convolution layers is represented as ,in Expressed as and The output of the layer graph convolution, represents the activation function, represents the weight matrix, represents the degree matrix, Indicates Trainable parameters of layer graph convolution. The hidden feature matrix obtained by the graph convolution layer is combined with the weight matrix and imported into the decoder to obtain the potential data distribution, and the potential data distribution is used to reconstruct the input data and obtain the data distribution of the reconstructed output; the distribution deviation of the input data and the reconstructed data under different feature variable categories is calculated to obtain the abnormal score of each feature variable. If the abnormal score is greater than the preset threshold, it is judged as abnormal, and the direction of improving the soil quality of the topsoil layer of the target farmland is determined according to the category of the abnormal feature variable.
[0069] Figure 3 A flow chart for obtaining appropriate soil improvement means, soil conditioners and management methods is shown.
[0070] According to an embodiment of the present invention, appropriate soil improvement means, soil conditioners and management methods are obtained based on the direction of improving the quality of the cultivated layer soil, specifically:
[0071] S302, using the direction of improving the quality of the cultivated layer soil to locate in the domain knowledge graph, performing representation learning on the domain knowledge graph through MetaPath random walk, and embedding the nodes of the domain knowledge graph into a high-dimensional representation space through pre-training;
[0072] S304, using soil improvement means, soil conditioners and management methods as target nodes, obtaining several meta-paths including positioning nodes and target nodes for improving the quality of cultivated soil, using a graph attention mechanism to represent the meta-paths, and obtaining the importance of the meta-paths corresponding to different directions of improving the quality of cultivated soil;
[0073] S306, selecting the most important meta-path in the direction of improving soil quality in different cultivated layers to obtain the corresponding target node, and determining the soil improvement means, soil conditioners and management methods required for the target farmland according to the target node.
[0074] It should be noted that the meta-path is defined as a sequence composed of different entities and relationships in the knowledge graph, which describes the degree of association between the head and tail entities of the sequence, takes the direction of improving the quality of the topsoil of the target farmland as the starting point, presets the soil improvement means, soil conditioners and management methods as the target nodes, and uses the random walk method to perform representation learning on the knowledge graph. According to the meta-path, the different soil improvement means, soil conditioners and management methods corresponding to a certain direction of improving the quality of the topsoil are obtained. A direction of improving the quality of the topsoil may involve several soil improvement means, soil conditioners and management methods, and the management methods include water and fertilizer control technology and farmland management methods. Various methods of improving the topsoil include returning straw to the field, applying organic fertilizer, planting green manure, tillage, mulching, organic fertilization, interface regulation, water and fertilizer management and other measures. They can form soil aggregates through hydrogen bonds, van der Waals forces, etc., increase soil porosity, reduce soil bulk density, loosen the soil, break up soil compaction, deepen the effective topsoil layer, expand the soil's capacity to absorb water and nutrients, and coordinate the "water, fertilizer, air, and heat" conditions of the soil topsoil.
[0075] Graph attention is introduced in each meta-path to characterize the importance of the improvement methods provided by different meta-paths. The graph attention mechanism is introduced in the meta-path to obtain the weights of each entity node on different paths to the head entity node, and the weights of each neighbor entity node are used to perform weighted average on the neighbor entity nodes to obtain the output of the head entity node after graph attention aggregation, that is, to obtain the output of the updated positioning node in the direction of improving the quality of the cultivated layer soil; for the n meta-paths where the head entity node exists, the graph attention mechanism is introduced between the meta-paths, and the graph attention aggregation output of the head entity in the meta-path is used to obtain the weights of different meta-paths to characterize the importance of the i-th meta-path.
[0076] Use data retrieval to obtain historical improvement case data corresponding to the soil improvement means, soil conditioners and management methods required for the target farmland, screen the historical improvement case data that meet the preset standards for integration and extraction of data requirements that can meet the crop growth conditions, and generate adjustment coefficients based on the climate and environmental parameters of the target farmland to correct the data requirements, use the corrected data requirements to obtain the improvement implementation time point and the improvement implementation data to generate the corresponding task workflow, such as the type, dosage and application time of the soil conditioner to be applied, and generate the topsoil improvement task guidance through the task workflow to guide the target farmland to build a healthy topsoil.
[0077] Figure 4 A block diagram of a high-standard farmland healthy tillage layer construction system is shown.
[0078] This embodiment provides a high-standard farmland healthy plough layer construction system 4, including: a memory 41, a processor 42, a soil multi-dimensional feature acquisition module 43, a multi-dimensional feature image construction module 44, a crop suitable environment analysis module 45, a plough layer soil improvement recommendation module 46 and a healthy plough layer program integration module 47, the memory 41 and the processor 42 store and execute a high-standard farmland healthy plough layer construction method program;
[0079] The soil multi-source data acquisition module 43 obtains the soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level of the target farmland, and constructs a multi-dimensional feature data sequence after pre-processing;
[0080] The multi-dimensional feature image building module 44 analyzes the multi-dimensional feature variable influencing factors of the target farmland plough layer soil environment and builds a multi-dimensional feature image of the target farmland;
[0081] The crop suitable environment analysis module 45 accesses the relevant field knowledge graph based on the desired crops of the target farmland to determine the suitable environment for the desired crops;
[0082] The topsoil improvement recommendation module 46 uses the graph convolution autoencoder to learn the domain knowledge graph structure, determine the direction of improving the topsoil quality of the target farmland, and obtain appropriate soil improvement means, soil conditioners and management methods;
[0083] The healthy tillage layer solution integration module 47 integrates the acquired soil improvement means, soil conditioners and management methods, and generates a target farmland healthy tillage layer construction workflow according to key steps and sequences.
[0084] 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.
[0085] 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.
[0086] 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.
[0087] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above method embodiment 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 embodiment; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.
[0088] 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.
[0089] 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 constructing a high-standard farmland healthy plough layer, characterized in that: The following steps are involved: Using soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level as search tags, a big data method is used to obtain soil environmental assessment instances corresponding to each dimensional tag, read the characteristic variables involved in the soil environmental assessment instance, and generate a multi-level characteristic system to analyze the multi-dimensional influencing factors of the soil environment of the target farmland plough layer, and construct a multi-dimensional characteristic portrait of the target farmland; Obtain the desired crops of the target farmland, access the relevant domain knowledge graph using the crop category, perform similarity integration in the domain knowledge graph based on the multi-dimensional feature portrait, and determine the suitable growth environment of the desired crops; The stacked graph convolutional autoencoder is used to learn the graph structure of the knowledge graph subgraph corresponding to the suitable environment, the multi-dimensional feature data of the target farmland is reconstructed, and the direction of improving the soil quality of the topsoil layer of the target farmland is obtained by using the data distribution deviation; Based on the direction of improving the soil quality of the topsoil layer, appropriate soil improvement means, soil conditioners, and management methods are selected, and a workflow for constructing a healthy topsoil layer of the target farmland is generated; Construct a multi-dimensional feature portrait of the target farmland, specifically: According to the multidimensional characteristic variables corresponding to the multidimensional influencing factors, parameters are extracted from multiple label data of soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level of the target farmland, and multidimensional characteristic data are constructed after preprocessing; Input the multi-dimensional feature data into the CNN network for re-feature extraction, use several filters in the convolution layer to obtain local context features, obtain low-dimensional representations of feature variables, and aggregate to generate feature sets; Performing a maximum pooling operation on the feature set, concatenating the maximum values of all features, importing a self-attention layer to assign different weights to different feature variables, weighting the feature variables of multidimensional influencing factors, and connecting them to a fully connected layer to output a multidimensional feature portrait of the target farmland; Based on the multi-dimensional feature portrait, similarity is integrated in the domain knowledge graph to determine the suitable environment for the desired crop, specifically: Access the knowledge graph of related fields of the expected crop categories of the target farmland, and construct the query intent using the multi-dimensional feature portrait of the target farmland and the crop planting; The Euclidean distance is preset as the metric function to represent the similarity. Entity recognition is performed by calculating the similarity between the query intent and the entities in the domain knowledge graph, identifying the key entities as known conditions and determining the positions of the key entities in the domain knowledge graph. According to the position of the key entity, all entities within a preset distance from the key entity are searched and marked, and a knowledge graph subgraph containing the key entity, the marked entity and the relationship is obtained. The standard soil environment parameters corresponding to the feature variables in the multidimensional feature portrait are extracted from the knowledge graph subgraph to determine the suitable growth environment of the desired crop; The stacked graph convolutional autoencoder is used to learn the graph structure of the knowledge graph subgraph corresponding to the suitable environment, specifically: In the knowledge graph subgraph obtained by the query, the weight information involved in the multidimensional feature portrait of the target farmland is used to set the edge weights of the relationship between entities and construct a weight matrix; The autoencoder is improved by using graph convolutional neural network to obtain stacked graph convolutional autoencoders. The standard soil environmental parameters corresponding to the suitable environment are used to train each layer of autoencoders in a layer-by-layer unsupervised manner. The graph structure of the knowledge graph subgraph is learned through information propagation of graph convolution. The output of the graph convolution is obtained as the hidden layer feature and imported into the next layer for training. The hidden layer feature matrix output by the last layer of graph convolution is combined with the weight matrix and transmitted to the decoder network for data reconstruction. The parameters of the stacked graph convolution autoencoder are optimized by back propagation. The multi-dimensional characteristic data of the target farmland are reconstructed, and the direction of improving the soil quality of the topsoil layer of the target farmland is obtained by using the data distribution deviation, specifically: The multi-dimensional feature data of the target farmland is obtained, and input data is generated by using a sliding window segmentation, and a stacked graph convolution autoencoder is imported to obtain potential data distribution, and the input data is reconstructed using the potential data distribution, and the data distribution of the reconstructed output is obtained; Calculate the distribution deviation between input data and reconstructed data under different feature variable categories, obtain the anomaly score of each feature variable, and if the anomaly score is greater than a preset threshold, it is judged as abnormal, and determine the direction of improving the soil quality of the topsoil layer of the target farmland according to the category of the abnormal feature variable; Based on the direction of improving the quality of the cultivated layer soil, appropriate soil improvement means, soil conditioners and management methods are selected, specifically: The direction of improving the quality of the cultivated layer soil is used to locate in the domain knowledge graph, the domain knowledge graph is represented and learned through MetaPath random walk, and the nodes of the domain knowledge graph are pre-trained and embedded into the high-dimensional representation space; Based on soil improvement methods, soil conditioners and management methods as target nodes, several meta-paths containing positioning nodes and target nodes for improving the quality of cultivated layer soil are obtained. The meta-paths are represented using the graph attention mechanism to obtain the importance of the meta-paths corresponding to different directions of improving the quality of cultivated layer soil. The most important meta-path is selected in the direction of improving soil quality in different cultivated layers to obtain the corresponding target node, and the soil improvement means, soil conditioners and management methods required for the target farmland are determined according to the target node.
2. A method for constructing a high-standard farmland healthy plough layer according to claim 1, characterized in that: Based on soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level as search tags, the big data method is used to obtain soil environmental assessment instances corresponding to each dimensional tag, read the characteristic variables involved in the soil environmental assessment instance, and generate a multi-level characteristic system to analyze the multi-dimensional influencing factors of the target farmland plough layer soil environment, specifically: Data collection and processing are performed based on a multi-level feature system, the feature parameters of the feature variables are standardized and normalized, and the feature variables are encoded in sequence to construct an original feature variable set; Calculate the Pearson correlation coefficient between each characteristic variable in the original characteristic variable set and the soil environment of the cultivated layer to represent the correlation, and obtain the characteristic variable with the greatest correlation and put it into the optimal characteristic set; Introducing the characteristic variables in the original characteristic variable set into the optimal characteristic set one by one, taking the maximum correlation index between the characteristic variables and the soil environment of the cultivated layer and the minimum redundancy index between the characteristic variables as measurement indicators, and introducing the characteristic variables into the optimal characteristic set; In addition, in the optimal feature set, the feature variables are measured in terms of the Pearson correlation coefficient and the Gini index, and the measurement results are standardized. The standardized measurement results are mapped into a score vector, and the importance of the feature variables is represented according to the vector length, and a preset number of feature variables are selected as multidimensional influencing factors.
3. A high-standard farmland healthy tillage layer construction system, characterized in that: Implementing the method for constructing a healthy plough layer of high-standard farmland as described in any one of claims 1-2, comprising: a memory, a processor, a soil multi-dimensional feature acquisition module, a multi-dimensional feature portrait construction module, a crop suitable environment analysis module, a plough layer soil improvement recommendation module and a healthy plough layer program integration module, wherein the memory and the processor store and execute a method program for constructing a healthy plough layer of high-standard farmland; The soil multi-source data acquisition module obtains the soil physical and chemical properties, soil microbial flora, soil pest distribution and soil fertility level of the target farmland, and constructs a multi-dimensional feature data sequence after preprocessing; The multidimensional feature image construction module analyzes the multidimensional feature variable influencing factors of the target farmland plough layer soil environment and constructs a multidimensional feature image of the target farmland; The crop suitable environment analysis module accesses the relevant field knowledge graph based on the desired crops in the target farmland to determine the suitable environment for the desired crops; The topsoil improvement recommendation module uses a graph convolutional autoencoder to learn the domain knowledge graph structure, determine the direction of improving the topsoil quality of the target farmland, and obtain appropriate soil improvement means, soil conditioners and management methods; The healthy tillage layer program integration module integrates the acquired soil improvement means, soil conditioners and management methods, and generates a target farmland healthy tillage layer construction workflow according to key steps and sequence.
