Multi-source data fusion intelligent agricultural knowledge graph construction system and method

Abstract

The invention relates to the technical field of agricultural informatization and artificial intelligence crossing, and discloses a multi-source data fusion intelligent agricultural knowledge graph construction system and method. Comprising a data acquisition module, a data processing and analysis module, a data safety monitoring module, a knowledge graph construction module, a multi-modal feature analysis module and an agricultural large model application module, multi-source agricultural data is acquired through multiple channels of an acquisition tool and a middleware platform, and the acquired data is preprocessed and analyzed by adopting a data processing technology; based on a block chain and an identification analysis technology, security in a data acquisition process is monitored, attribute-based encryption ABE is adopted to complete dynamic desensitization transmission of data, an agricultural knowledge graph database is constructed according to an analysis result, and feature extraction is performed on crop growth states and pest and disease damage conditions to realize crop full-life-cycle management. The agricultural production benefit and the market competitiveness of agricultural products are improved.

Description

Technical Field

[0001] The present invention relates to the cross - technical field of agricultural informatization and artificial intelligence, and more specifically to a system and method for constructing a knowledge graph of smart agriculture with multi - source data fusion. Background Art

[0002] Smart agriculture is an advanced stage of modern agricultural development. It integrates various high - tech means such as information technology, Internet of Things technology, and artificial intelligence, aiming to improve agricultural production efficiency, reduce production costs, optimize resource allocation, and protect the ecological environment. With the growth of the population and the acceleration of urbanization, agriculture faces multiple challenges such as limited resources, environmental pollution, and climate change. Therefore, the development of smart agriculture has become a key way to address these challenges.

[0003] Current agricultural management systems have deficiencies in aspects such as data fusion, knowledge graph construction and update, and multi - modal feature analysis. Existing systems often can only process single or a few types of data sources, making it difficult to achieve effective fusion of multi - source data, resulting in insufficient data utilization; the construction and update of the knowledge graph are not timely and accurate enough to provide the latest knowledge support for agricultural production; in the monitoring of crop growth status and the identification of pests and diseases, there is a lack of multi - modal feature representation and recognition technology, making it difficult to intuitively display the growth of crops and the risk of pests and diseases. Therefore, it is of great significance to develop a smart agriculture full - life - cycle intelligent decision - making system that can integrate multi - source data, construct a dynamic knowledge graph, and achieve multi - modal feature analysis. Summary of the Invention

[0004] In order to overcome the above - mentioned defects of the prior art, the present invention provides a system for constructing a knowledge graph of smart agriculture with multi - source data fusion to solve the problems existing in the above - mentioned background art.

[0005] The present invention provides the following technical solution: A system for constructing a knowledge graph of smart agriculture with multi - source data fusion, comprising: a data acquisition module, a data processing and analysis module, a data security monitoring module, a knowledge graph construction module, a multi - modal feature analysis module, and an agricultural large - model application module;

[0006] The data acquisition module, based on the identification and resolution technology, collects the hardware data of the sensing network through the active identification chip carrier, collects and integrates the agricultural data of each node's heterogeneous management software across departments and enterprises through the industrial software connector, and transmits the data to the data processing and analysis module;

[0007] The data processing and analysis module receives the data transmitted by the data acquisition module and performs pre - processing and analysis on the received data using data processing technology.

[0008] The data security monitoring module monitors the security during the data collection process based on blockchain and identification resolution technologies, and uses attribute-based encryption (ABE) to complete the dynamic desensitization transmission of data;

[0009] The knowledge graph construction module, based on the analysis results of the data preprocessing and analysis by the data processing and analysis module, mines agricultural domain knowledge from the analysis results, constructs an agricultural knowledge graph database, and realizes the dynamic update of knowledge;

[0010] The multi-modal feature analysis module extracts features of the growth status and pest and disease conditions of crops based on the agricultural knowledge graph database constructed by the knowledge graph construction module;

[0011] The agricultural large model application module completes the full life cycle management of crops based on the features of the growth status and pest and disease conditions of crops extracted by the multi-modal feature analysis module.

[0012] Preferably, in the data collection module, the hardware data of the sensing network is collected through an active identification chip carrier. The sensing network includes monitoring probes and temperature sensors. The data collection module interacts with the hardware devices of the sensing network through the active identification chip carrier, collects the hardware data of the sensing network, and converts the collected data into digital signals through sensors; the agricultural data of each node's heterogeneous management software is collected and integrated across departments and enterprises through an industrial software connector. The agricultural data includes: multi-source agricultural data such as meteorology, soil, crop growth, pest and disease, and market collected from multiple channels.

[0013] Preferably, in the data processing and analysis module, data processing technologies are used to preprocess and analyze the received data. The data processing technology is a collaborative technology architecture including multi-source data convergence, alignment and completion, and derivative estimation. The specific content is as follows:

[0014] In the multi-source data convergence stage, a stream-batch integrated processing engine based on Flink is used to construct a data cube, CUBE = (S, T, V), where S is the spatial dimension, T is the time window, and V is the set of observed values;

[0015] Alignment and completion uses an improved ST-MVL algorithm, introducing a spatio-temporal constraint factor λ, and the expression is: where d is the spatial distance and σ is the covariance coefficient;

[0016] Derivative estimation uses a multi-scale feature fusion method, combines wavelet transform and LSTM network, and constructs a feature cross layer to enhance the temporal feature representation ability.

[0017] Preferably, in the data processing and analysis module, data processing technologies are used to preprocess and analyze the received data. The preprocessing process is as follows:

[0018] Step S1: Anomaly detection: Based on the improved LOF algorithm, with the neighborhood radius k = 20 and the outlier threshold θ = 1.5σ;

[0019] Step S2: Standardization: Use RobustScaler, where x represents the original data point, median represents the median of the data, Q3 represents the third quartile of the data, Q1 represents the first quartile of the data, and x' represents the standardized data point;

[0020] Step S3: Feature engineering: Construct the time series difference feature △x t = x t - x t-24h , where x t represents the data corresponding to time point t, x t-24h represents the data corresponding to 24 hours before time point t, and △x t represents the output value of the constructed time series difference feature.

[0021] Preferably, in the data security monitoring module, the security during the data collection process is monitored based on blockchain and identification resolution technology, and the specific content of using attribute-based encryption ABE to complete the dynamic desensitization transmission of data is as follows:

[0022] Trusted collection based on blockchain + identification resolution technology: Embed a lightweight consensus protocol in the active identification chip, and each data packet is attached with a Merkle-Patricia tree verification path to ensure that each piece of data is encrypted and verified during the transfer process through the identification device management platform under the identification resolution system;

[0023] Dynamic desensitization transmission: Use attribute-based encryption ABE, define that the depth of the access policy tree is ≥4 layers, and the key update period <30 minutes;

[0024] Security audit: Deploy an improved BLS signature scheme to achieve non-tamperable traceability of operation logs.

[0025] Preferably, in the knowledge graph construction module, the steps of mining knowledge in the agricultural field through data mining and machine learning technologies, constructing an agricultural knowledge graph library, and realizing the dynamic update of knowledge are as follows:

[0026] Multi-modal entity recognition: Use the BERT-BiLSTM-CRF model for multi-modal entity recognition;

[0027] Relationship extraction: Combine syntactic dependency trees and semantic role annotation;

[0028] Graph fusion: Apply the improved PARIS algorithm, and set the similarity threshold τ = 0.85;

[0029] In the knowledge graph construction module, the dynamic update mechanism for mining knowledge in the agricultural field through data mining and machine learning technologies, constructing an agricultural knowledge graph library, and realizing the dynamic update of knowledge is as follows:

[0030] Event-triggered update: Define the event expression E = (subject, predicate, object, confidence > 0.8);

[0031] Incremental learning: Design an online learning strategy for the graph convolutional network GCN, with an update delay < 3 minutes;

[0032] Conflict resolution: Adopt an inference mechanism based on the description logic ALCQ to support TBox evolution.

[0033] Preferably, in the multimodal feature analysis module, based on the agricultural knowledge graph library constructed by the knowledge graph construction module, the specific content of feature extraction for crop growth status and pest and disease conditions is as follows:

[0034] Cross-modal attention mechanism: Design a vision-text alignment model, construct a joint embedding space, and use an improved CKA method for similarity calculation;

[0035] Three-dimensional point cloud parsing: Develop a plant modeling algorithm based on PointNet++ within the collaborative technology valley language model, with a point cloud registration error < 1.9 mm;

[0036] Multispectral fusion: Construct a band attention network to achieve an optimized combination of 17 feature channels within the spectral range of 400 - 1200 nm, and extract features for crop growth status and pest and disease conditions.

[0037] Preferably, in the agricultural large model application module, based on the knowledge graph and multimodal feature analysis, functions such as crop full life cycle management, intelligent decision-making support, and market prediction are realized.

[0038] A method for constructing an intelligent agricultural knowledge graph with multi-source data fusion includes the following steps:

[0039] Step S01: Collect multi-source agricultural data through multiple channels such as collection tools and middleware platforms, covering multi-dimensional information on meteorology, soil, crop growth, pests and diseases, and the market;

[0040] Step S02: Preprocess and analyze the collected data using multi-source data aggregation, alignment and completion, and derivative estimation technologies;

[0041] Step S03: Monitor the security during the data collection process based on blockchain and identification resolution technologies, and use attribute-based encryption ABE to complete the dynamic desensitization transmission of data;

[0042] Step S04: Based on the analysis results after the data preprocessing and analysis by the data processing and analysis module, extract agricultural domain knowledge from the analysis results, construct an agricultural knowledge graph database, and achieve dynamic update of knowledge;

[0043] Step S05: Based on the constructed agricultural knowledge graph database, extract features of the growth status and pest and disease conditions of crops;

[0044] Step S06: Based on the knowledge graph and multimodal feature analysis, achieve the whole life cycle management, intelligent decision-making support and market prediction of crops.

[0045] Technical effects and advantages of the present invention:

[0046] The present invention integrates multi-source agricultural data through a data acquisition module, a data processing and analysis module, a data security monitoring module, a knowledge graph construction module, a multimodal feature analysis module and an agricultural large model application module, constructs a dynamically updated agricultural knowledge graph, combines multimodal feature analysis with artificial intelligence algorithms, realizes the intelligent management of the whole life cycle of crops, improves the yield and quality of crops, optimizes the utilization efficiency of agricultural resources, and promotes the intelligent upgrading of the agricultural industrial chain;

[0047] Through multi-source data fusion, it provides more comprehensive data support for agricultural production, constructs a dynamically updated agricultural knowledge graph, ensures the timeliness and accuracy of knowledge, provides the latest knowledge support for agricultural production, uses multimodal feature representation and recognition technology to intuitively display the crop growth status and pest and disease risks, provides richer information support for farmers, realizes the intelligent management of the whole life cycle of crops, improves the yield and quality of crops, optimizes the utilization efficiency of agricultural resources, promotes the intelligent upgrading of the agricultural industrial chain, enhances the security and credibility of agricultural data, and guarantees the stable operation of agricultural production. Brief Description of the Drawings

[0048] Figure 1 It is a structural schematic diagram of a multi-source data fusion-based intelligent agricultural knowledge graph construction system.

[0049] Figure 2 It is a flow schematic diagram of a multi-source data fusion-based intelligent agricultural knowledge graph construction method. Detailed Embodiments

[0050] The technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. In addition, the forms of the various structures described in the following embodiments are merely examples, and a multi-source data fusion-based intelligent agricultural knowledge graph construction system and method according to the present invention are not limited to the various structures described in the following embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0051] As Figure 1 shown, the present invention provides a multi-source data fusion-based intelligent agricultural knowledge graph construction system, including: a data acquisition module, a data processing and analysis module, a data security monitoring module, a knowledge graph construction module, a multi-modal feature analysis module, and an agricultural large model application module;

[0052] The data acquisition module, based on the identification and resolution technology, collects the hardware data of the sensing network through the active identification chip carrier, collects and integrates the agricultural data of each node's heterogeneous management software across departments and enterprises through the industrial software connector, and transmits the data to the data processing and analysis module;

[0053] The data processing and analysis module receives the data transmitted by the data acquisition module and performs preprocessing and analysis on the received data using data processing technology;

[0054] The data security monitoring module monitors the security during the data acquisition process based on blockchain and identification and resolution technology, and uses attribute-based encryption ABE to complete the dynamic desensitization transmission of the data;

[0055] The knowledge graph construction module, based on the analysis results after the data processing and analysis module preprocesses and analyzes the data, mines agricultural domain knowledge from the analysis results, constructs an agricultural knowledge graph database, and realizes the dynamic update of knowledge;

[0056] The multi-modal feature analysis module extracts features of the growth status and pest and disease conditions of crops based on the agricultural knowledge graph database constructed by the knowledge graph construction module;

[0057] The agricultural large model application module completes the full life cycle management of crops based on the features of the growth status and pest and disease conditions of crops extracted by the multi-modal feature analysis module.

[0058] In this embodiment, it should be specifically noted that in the data acquisition module, the hardware data of the sensing network is acquired through an active identification chip carrier. The sensing network includes monitoring probes and temperature sensors. The data acquisition module interacts with the hardware devices of the sensing network through the active identification chip carrier, acquires the hardware data of the sensing network, and converts the acquired data into digital signals through sensors; the agricultural data of each node's heterogeneous management software is collected and integrated across departments and enterprises through an industrial software connector. The agricultural data includes: multi-source agricultural data such as meteorology, soil, crop growth, pests and diseases, and the market collected through multiple channels.

[0059] In this embodiment, it should be specifically noted that in the data processing and analysis module, data processing technologies are used to preprocess and analyze the received data. The data processing technology is a collaborative technology architecture including multi-source data aggregation, alignment and completion, and derivative estimation. The specific content is as follows:

[0060] In the multi-source data aggregation stage, a stream-batch integrated processing engine based on Flink is used to construct a data cube, CUBE = (S, T, V), where S is the spatial dimension, T is the time window, and V is the set of observed values;

[0061] Alignment and completion uses an improved ST-MVL algorithm, introducing a spatio-temporal constraint factor λ, and the expression is: where d is the spatial distance and σ is the covariance coefficient;

[0062] Derivative estimation uses a multi-scale feature fusion method, combines wavelet transform and LSTM network, and constructs a feature cross layer to enhance the temporal feature representation ability.

[0063] In this embodiment, it should be specifically noted that in the data processing and analysis module, data processing technologies are used to preprocess and analyze the received data. The preprocessing process is as follows:

[0064] Step S1: Anomaly detection: Based on an improved LOF algorithm, the neighborhood radius k = 20, and the outlier threshold θ = 1.5σ;

[0065] Step S2: Standardization: Use RobustScaler, Where \(x\) represents the original data point, \(median\) represents the median of the data, \(Q3\) represents the third quartile of the data, \(Q1\) represents the first quartile of the data, and \(x'\) represents the standardized data point. The median is the number in the middle position after sorting a set of data from smallest to largest. If the number of data is odd, the median is the middle number; if the number of data is even, the median is the average of the two middle numbers. The third quartile of the data: Among the sorted data, 75% of the data is less than or equal to \(Q3\), and 25% of the data is greater than \(Q3\). The first quartile of the data: Among the sorted data, 25% of the data is less than or equal to \(Q1\), and 75% of the data is greater than \(Q1\). The interquartile range is the difference between \(Q3\) and \(Q1\), which is an index to measure the degree of data dispersion and is less sensitive to outliers.

[0066] Step S3: Feature engineering: Construct the time series difference feature \(\Delta x\) t = \(x\) t - \(x\) t-24h , where \(x\) t represents the data corresponding to time point \(t\), and \(x\) t-24h represents the data corresponding to 24 hours before time point \(t\), and \(\Delta x\) t represents the output value of the constructed time series difference feature.

[0067] In this embodiment, it should be specifically noted that in the data security monitoring module, the security during the data collection process is monitored based on blockchain and identity resolution technologies. The specific content of using attribute-based encryption ABE to complete the dynamic desensitization transmission of data is as follows:

[0068] Trusted collection based on blockchain + identity resolution technology: Embed a lightweight consensus protocol in the active identification chip, and each data packet is attached with a Merkle-Patricia tree verification path to ensure that each piece of data is encrypted and verified during the transfer process through the identity device management platform under the identity resolution system;

[0069] Dynamic desensitization transmission: Use attribute-based encryption ABE, define that the depth of the access policy tree is ≥ 4 layers, and the key update period is < 30 minutes;

[0070] Security audit: Deploy an improved BLS signature scheme to achieve non-tamperable traceability of operation logs and improve the audit efficiency by 40%.

[0071] In this embodiment, it should be specifically noted that in the knowledge graph construction module, the knowledge in the agricultural field is mined through data mining and machine learning technologies, an agricultural knowledge graph library is constructed, and the construction steps for realizing the dynamic update of knowledge are as follows:

[0072] Multi-modal entity recognition: Use the BERT-BiLSTM-CRF model for multi-modal entity recognition;

[0073] Relation extraction: Combining syntactic dependency trees and semantic role labeling;

[0074] Knowledge graph fusion: Applying the improved PARIS algorithm and setting the similarity threshold τ = 0.85;

[0075] In the knowledge graph construction module, the following dynamic update mechanism for mining knowledge in the agricultural field through data mining and machine learning techniques, constructing an agricultural knowledge graph library, and realizing the dynamic update of knowledge is as follows:

[0076] Event-triggered update: Defining the event expression E = (subject, predicate, object, confidence > 0.8);

[0077] Incremental learning: Designing an online learning strategy for the graph convolutional network GCN with an update delay < 3 minutes;

[0078] Conflict resolution: Adopting an inference mechanism based on the description logic ALCQ to support TBox evolution.

[0079] In this embodiment, it should be specifically noted that in the multi-modal feature analysis module, based on the agricultural knowledge graph library constructed by the knowledge graph construction module, the specific content of feature extraction for the growth status and pest and disease conditions of crops is as follows:

[0080] Cross-modal attention mechanism: Designing a vision-text alignment model, constructing a joint embedding space, and using the improved CKA method for similarity calculation;

[0081] 3D point cloud parsing: Developing a plant modeling algorithm based on PointNet++ within the Xietong Keji Guyu large model, with a point cloud registration error < 1.9mm;

[0082] Multi-spectral fusion: Constructing a band attention network to achieve an optimized combination of 17 feature channels within the spectral range of 400 - 1200nm, and extracting features for the growth status and pest and disease conditions of crops.

[0083] In this embodiment, it should be specifically noted that in the agricultural large model application module, based on the knowledge graph and multi-modal feature analysis, the specific content of realizing the functions of crop whole-life cycle management, intelligent decision-making support, and market prediction is as follows:

[0084] Whole-life cycle management:

[0085] Stage recognition: Constructing a hybrid model with a ResNet-Transformer architecture based on the Guyu large model, and dividing it into 6 growth stages: germination stage, seedling stage, vegetative growth stage, flowering stage, fruiting stage, and maturity stage;

[0086] Water and Fertilizer Regulation: Develop a multi-objective optimization algorithm based on NSGA-II, combined with the Guyu large model, with the decision variable dimension reaching 15 dimensions.

[0087] Intelligent Decision Support:

[0088] Planting Planning: The Guyu large model covers a mixed integer programming model, and the decision variables include the resource allocation of 72 time segments.

[0089] Risk Warning: Construct a spatio-temporal graph convolutional network ST-GCN to predict the probability of pest and disease outbreaks 7 days in advance.

[0090] Market Forecast:

[0091] Price Forecast: Construct a combined model of the Guyu large model-LSTM-ARIMA that integrates knowledge graph embedding, with MAPE < 8.1%;

[0092] Yield Estimation: Apply the multi-scale remote sensing analysis algorithm of the Guyu large model, combined with the NDVI index and meteorological data, with an error rate < 4%;

[0093] Demand Forecast: Apply the graph attention network GAT to model the supply chain relationship, with a prediction accuracy of 91.1%.

[0094] As Figure 2 shown, in this embodiment, it should be specifically noted that a method for constructing a smart agriculture knowledge graph with multi-source data fusion includes the following steps:

[0095] Step S01: Collect multi-source agricultural data through multiple channels such as collection tools and middleware platforms, covering multi-dimensional information on meteorology, soil, crop growth, pests and diseases, and the market;

[0096] Step S02: Preprocess and analyze the collected data using multi-source data aggregation, alignment and complementation, and derivative estimation techniques;

[0097] Step S03: Monitor the security of the data collection process based on blockchain and identification resolution technologies, and use attribute-based encryption ABE to complete the dynamic desensitization transmission of the data;

[0098] Step S04: Based on the analysis results after preprocessing and analyzing the data by the data processing and analysis module, mine agricultural domain knowledge from the analysis results, construct an agricultural knowledge graph library, and realize the dynamic update of knowledge;

[0099] Step S05: Extract features of the growth status and pest and disease conditions of crops based on the constructed agricultural knowledge graph library;

[0100] Step S06: Based on knowledge graph and multi-modal feature analysis, realize the whole life cycle management, intelligent decision support, and market prediction of crops.

[0101] In this embodiment, it should be specifically noted that the main difference between this embodiment and the prior art is that this embodiment integrates multi-source agricultural data through a data acquisition module, a data processing and analysis module, a data security monitoring module, a knowledge graph construction module, a multi-modal feature analysis module, and an agricultural large model application module, constructs a dynamically updated agricultural knowledge graph, combines multi-modal feature analysis with artificial intelligence algorithms, realizes intelligent management of the entire life cycle of crops, improves crop yield and quality, optimizes the utilization efficiency of agricultural resources, and promotes the intelligent upgrading of the agricultural industrial chain;

[0102] Through multi-source data fusion, it provides more comprehensive data support for agricultural production, constructs a dynamically updated agricultural knowledge graph, ensures the timeliness and accuracy of knowledge, provides the latest knowledge support for agricultural production, uses multi-modal feature representation and recognition technology to intuitively display the growth status of crops and the risk of pests and diseases, provides richer information support for farmers, realizes intelligent management of the entire life cycle of crops, improves crop yield and quality, optimizes the utilization efficiency of agricultural resources, promotes the intelligent upgrading of the agricultural industrial chain, enhances the security and credibility of agricultural data, and ensures the stable operation of agricultural production.

[0103] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

[0104] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or replacements, which should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A multi-source data fusion smart agriculture knowledge graph construction system, characterized by: include: Data collection module, data processing and analysis module, data security monitoring module, knowledge graph construction module, multimodal feature analysis module and agricultural large model application module; The data collection module, based on the identification resolution technology, collects the hardware data of the perception network through the active identification chip carrier, collects the agricultural data of the heterogeneous management software of each node across departments and enterprises through the industrial software connector, and transmits the data to the data processing and analysis module; The data processing and analysis module receives the data transmitted by the data acquisition module and uses data processing technology to pre-process and analyze the received data; The data security monitoring module monitors the security of the data collection process based on blockchain and identity resolution technology, and uses attribute-based encryption (ABE) to complete dynamic desensitization transmission of data; The knowledge graph construction module mines agricultural knowledge from the analysis results after the data processing and analysis module preprocesses and analyzes the data, constructs an agricultural knowledge graph library, and realizes dynamic update of knowledge; The multimodal feature analysis module extracts features of crop growth status and pest and disease conditions based on the agricultural knowledge graph library constructed by the knowledge graph construction module; The agricultural large model application module completes the full life cycle management of crops based on the crop growth status and pest and disease situation characteristics extracted by the multimodal feature analysis module.

2. According to the multi-source data fusion smart agriculture knowledge graph construction system of claim 1, it is characterized by: In the data acquisition module, the hardware data of the perception network is collected through the active identification chip carrier. The perception network includes a monitoring probe and a temperature sensor. The data acquisition module interacts with the hardware devices of the perception network through the active identification chip carrier to collect the hardware data of the perception network, and converts the collected data into a digital signal through the sensor; Agricultural data from heterogeneous management software at each node is collected and integrated across departments and enterprises through industrial software connectors. The agricultural data includes: multi-channel collection of multi-source agricultural data on meteorology, soil, crop growth, pests and diseases, and the market.

3. According to the multi-source data fusion smart agriculture knowledge graph construction system of claim 1, it is characterized by: In the data processing and analysis module, data processing technology is used to pre-process and analyze the received data. The data processing technology is a collaborative technology architecture including multi-source data aggregation, alignment completion and derivative estimation. The specific contents are as follows: In the multi-source data aggregation stage, a Flink-based stream-batch integrated processing engine is used to construct a data cube, CUBE = (S, T, V), where S is the spatial dimension, T is the time window, and V is the set of observations; Alignment completion uses the improved ST-MVL algorithm and introduces the spatiotemporal constraint factor λ, which is expressed as: Where d is the spatial distance and σ is the covariance coefficient; Derivative estimation adopts a multi-scale feature fusion method, combines wavelet transform and LSTM network, and constructs a feature cross layer to improve the time series feature representation capability.

4. According to claim 3, a multi-source data fusion smart agriculture knowledge graph construction system is characterized by: In the data processing and analysis module, data processing technology is used to preprocess and analyze the received data. The preprocessing process is as follows: Step S1: Anomaly detection: based on the improved LOF algorithm, k = 20 neighborhood radius, outlier threshold θ = 1.5σ; Step S2: Standardization: using RobustScaler, Where x represents the original data point, median represents the median of the data, Q3 represents the third quartile of the data, Q1 represents the first quartile of the data, and x′ represents the standardized data point; Step S3: Feature Engineering: Constructing the time series difference feature △x t =x t -x t-24h , where x t represents the data corresponding to time point t, x t-24h Indicates the data corresponding to 24 hours before time point t, △x t Represents the output value of the constructed temporal difference feature.

5. According to the multi-source data fusion smart agriculture knowledge graph construction system of claim 1, it is characterized by: In the data security monitoring module, the security of the data collection process is monitored based on blockchain and identity resolution technology, and attribute-based encryption ABE is used to complete the dynamic desensitization transmission of data. The specific contents are as follows: Trusted collection based on blockchain + identity resolution technology: A lightweight consensus protocol is embedded in the active identification chip, and each data packet is accompanied by a Merkle-Patricia tree verification path to ensure that each piece of data is encrypted and signed during the flow process through the identity device management platform under the identity resolution system; Dynamic desensitized transmission: Attribute-based encryption (ABE) is used, and the access policy tree depth is defined as ≥ 4 layers, and the key update cycle is < 30 minutes; Security Audit: Deploy an improved BLS signature scheme to achieve tamper-proof traceability of operation logs.

6. The multi-source data fusion smart agriculture knowledge graph construction system according to claim 1 is characterized by: In the knowledge graph construction module, the steps of mining agricultural knowledge through data mining and machine learning technology, building an agricultural knowledge graph library, and realizing dynamic updating of knowledge are as follows: Multimodal entity recognition: The BERT-BiLSTM-CRF model is used for multimodal entity recognition; Relation extraction: combining syntactic dependency trees with semantic role labeling; Graph fusion: Apply the improved PARIS algorithm and set the similarity threshold τ = 0.85; In the knowledge graph construction module, the knowledge in the agricultural field is mined through data mining and machine learning technology, an agricultural knowledge graph library is constructed, and the dynamic update mechanism of the knowledge is realized as follows: Event trigger update: define event expression E = (subject, predicate, object, confidence > 0.8); Incremental learning: Designing an online learning strategy for graph convolutional networks (GCNs) with an update delay of < 3 minutes; Conflict resolution: Adopts the reasoning mechanism based on description logic ALCQ to support TBox evolution.

7. The multi-source data fusion smart agriculture knowledge graph construction system according to claim 1 is characterized by: In the multimodal feature analysis module, the specific content of feature extraction of crop growth status and pest and disease conditions based on the agricultural knowledge graph library constructed by the knowledge graph construction module is as follows: Cross-modal attention mechanism: Design a visual-text alignment model, construct a joint embedding space, and use an improved CKA method for similarity calculation; 3D point cloud analysis: Developed a plant modeling algorithm based on PointNet++ in the collaborative technology Guyu model, with a point cloud registration error of <1.9mm; Multispectral fusion: Build a band attention network to achieve the optimal combination of 17 feature channels in the 400-1200nm spectral range, and extract features of crop growth status and pest and disease conditions.

8. The multi-source data fusion smart agriculture knowledge graph construction system according to claim 1 is characterized by: In the agricultural big model application module, based on knowledge graph and multimodal feature analysis, the full life cycle management of crops, intelligent decision support and market forecasting functions are realized.

9. A method for constructing a smart agricultural knowledge graph by fusion of multi-source data, used for constructing a smart agricultural knowledge graph by fusion of multi-source data according to any one of claims 1 to 8, characterized in that: The following steps are involved: Step S01: Collect multi-source agricultural data through multiple channels through collection tools and middleware platforms, covering multi-dimensional information on weather, soil, crop growth, pests and diseases, and markets; Step S02: pre-processing and analyzing the collected data using multi-source data aggregation, alignment completion, and derivative estimation techniques; Step S03: Monitor the security of data collection process based on blockchain and identity resolution technology, and use attribute-based encryption (ABE) to complete dynamic desensitization transmission of data; Step S04: Based on the analysis results of the data preprocessing and analysis by the data processing and analysis module, agricultural knowledge is mined from the analysis results, an agricultural knowledge graph library is constructed, and dynamic updating of knowledge is achieved; Step S05: Based on the constructed agricultural knowledge graph library, feature extraction is performed on the growth status of crops and the conditions of pests and diseases; Step S06: Based on the knowledge graph and multimodal feature analysis, realize the full life cycle management of crops, intelligent decision support and market forecasting.

Images