Radar confrontation domain knowledge extraction method and system based on BERT model
By using a knowledge extraction method for radar countermeasures based on the BERT model, the problems of incomplete radar countermeasures model generation and insufficient interpretability in existing technologies are solved, and efficient and professional knowledge extraction and decision support are achieved.
Patent Information
- Application Number
- CN202511050824.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-29
- Publication Date
- 2025-11-11
AI Technical Summary
In air defense and missile defense scenarios, existing technologies struggle to generate complete radar countermeasure models, and deep neural networks lack interpretability, failing to meet the demands for highly agile decision-making.
A knowledge extraction method based on the BERT model is adopted for radar countermeasures. By establishing a unified description architecture that combines static and dynamic elements, unstructured data is preprocessed, entity recognition and relation extraction are performed, and knowledge fusion is carried out to form a unified knowledge set.
It improves the interpretability and accuracy of reasoning and decision-making in the field of radar countermeasures, enhances the professionalism and efficiency of knowledge extraction, and meets the needs of highly agile decision-making.
Smart Images
Figure CN120930765A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radar electronic countermeasures technology, and in particular to a method and system for extracting radar countermeasures domain knowledge based on the BERT (Bidirectional Encoder Representations from Transforms) model. Background Technology
[0002] Currently, in air defense and missile defense scenarios, self-defense radar countermeasures face numerous challenges. On the one hand, target reconnaissance and identification data, as well as combat effectiveness data, are difficult to obtain during peacetime, making it difficult for data-driven reinforcement learning and deep learning methods to generate complete models. On the other hand, the reasoning of deep neural network models lacks interpretability and cannot provide reasonable explanations for the current decision-making process and results. Furthermore, the high speed of target maneuverability and short decision-making time place high demands on the agility of the decision-making model.
[0003] Traditional knowledge-driven methods rely on expert systems based on manual modeling. The methods and strategies for target recognition and decision-making reasoning in these expert systems are primarily derived from expert experience through manual modeling, lacking the ability to learn autonomously and evolve rapidly. In recent years, domain task models based on natural language processing have developed rapidly. For example, invention CN 118551840A discloses a knowledge extraction system and method based on a large language model algorithm. This system includes a data collection and cleaning module, an entity and relation extraction module, an entity disambiguation and alignment module, and a knowledge storage module, respectively located in the data collection layer, data filtering layer, knowledge extraction layer, knowledge disambiguation layer, and knowledge storage layer. This allows for knowledge extraction and dynamic evaluation of the actual needs of the external knowledge base during the extraction process, improving extraction efficiency and outputting higher-quality results. However, its extraction of knowledge in the radar countermeasures domain suffers from insufficient professionalism and accuracy, as well as low extraction efficiency.
[0004] Therefore, there is an urgent need to study a knowledge extraction method for the radar countermeasures field. This method aims to address the intelligent reasoning and decision-making needs at each stage of the radar countermeasures process by using pre-trained language models to automatically extract knowledge from massive, open text data. Based on a unified description architecture for the radar countermeasures field, a knowledge system based on a unified logical description system is formed to improve the interpretability and accuracy of reasoning and decision-making in the radar countermeasures process. Summary of the Invention
[0005] The purpose of this invention is to provide a knowledge extraction method and system for the radar countermeasures field based on the BERT model, which is highly professional, accurate, and efficient in extracting and uniformly describing unstructured text data in the radar countermeasures field.
[0006] The technical solution to achieve the purpose of this invention is: a radar countermeasure domain knowledge extraction method based on the BERT model, comprising the following steps:
[0007] Step 1: Establish a unified description framework for radar countermeasure knowledge that combines static and dynamic elements;
[0008] Step 2: Perform unstructured data preprocessing on radar countermeasure text data from different data sources;
[0009] Step 3: Use a sequence labeling method based on a pre-trained language model for entity recognition and relation extraction;
[0010] Step 4: Based on the unified description architecture of radar countermeasure knowledge, knowledge data extracted from different data sources are fused and represented to form a unified description knowledge set for radar countermeasure.
[0011] A radar countermeasures domain knowledge extraction system based on the BERT model is disclosed. This system implements the aforementioned radar countermeasures domain knowledge extraction method based on the BERT model. The system comprises four modules, each with the following functions:
[0012] The first module is used to establish a unified description architecture for radar countermeasure knowledge that combines static and dynamic elements.
[0013] The second module is used for unstructured data preprocessing of radar countermeasure text data from different data sources;
[0014] The third module uses a sequence labeling method based on a pre-trained language model for entity recognition and relation extraction.
[0015] The fourth module, based on the unified description architecture of radar countermeasure knowledge, performs knowledge fusion and knowledge representation on knowledge data extracted from different data sources to form a unified description knowledge set for radar countermeasure.
[0016] A mobile terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the radar countermeasure domain knowledge extraction method based on the BERT model.
[0017] Compared with the prior art, the present invention has the following significant advantages: (1) In response to the intelligent reasoning and decision-making needs of each stage in the radar confrontation process, the invention uses a pre-trained language model to automatically extract knowledge of the radar confrontation field from massive and open text data, and forms a knowledge system based on a unified logical description system based on a unified description architecture of the radar confrontation field, which effectively improves the interpretability and accuracy of reasoning and decision-making in the radar confrontation field; (2) The invention uses the BERT model to perform autonomous learning and rapid evolution of knowledge such as target recognition and decision reasoning, which improves the professionalism and extraction efficiency of knowledge extraction in the radar confrontation field. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating a radar countermeasure domain knowledge extraction method based on the BERT model according to the present invention.
[0019] Figure 2 This is a schematic diagram of the unified description architecture for radar countermeasure knowledge that combines static and dynamic elements in an embodiment of the present invention.
[0020] Figure 3 This is a structural diagram of the radar countermeasure domain knowledge joint extraction model based on BERT in an embodiment of the present invention.
[0021] Figure 4 This is a flowchart illustrating the knowledge fusion and representation process in an embodiment of the present invention. Detailed Implementation
[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.
[0023] This invention provides a radar countermeasures domain knowledge extraction method based on the BERT model, comprising the following steps:
[0024] Step 1: Establish a unified description framework for radar countermeasure knowledge that combines static and dynamic elements;
[0025] Step 2: Perform unstructured data preprocessing on radar countermeasure text data from different data sources;
[0026] Step 3: Use a sequence labeling method based on a pre-trained language model for entity recognition and relation extraction;
[0027] Step 4: Based on the unified description architecture of radar countermeasure knowledge, knowledge data extracted from different data sources are fused and represented to form a unified description knowledge set for radar countermeasure.
[0028] As a specific example, the unified description architecture for radar countermeasure knowledge that combines static and dynamic elements, as described in step 1, is as follows:
[0029] 1) Dynamic knowledge includes state analysis reasoning and jamming decision reasoning knowledge in the radar countermeasure process, which are reasoning models for problems in the field of radar countermeasures, including radar state reasoning models, air defense and anti-missile mission planning models, active and passive jamming decision models, and jamming resource optimization models.
[0030] 2) Static knowledge includes common knowledge and knowledge structure in the field of radar countermeasures, which constitute the knowledge ontology and general knowledge system of the radar countermeasures field. The domain knowledge ontology includes radar object ontology, electronic countermeasures object ontology, game countermeasure concept ontology, and game event ontology. The general knowledge system includes predicate logic system, nondeterministic system, deep network reasoning system, graph search model, and general mathematical model solution method.
[0031] As a specific example, step 2, which involves unstructured data preprocessing of radar countermeasures text data from different data sources, is as follows:
[0032] Step 2.1: Perform noise reduction processing to merge and delete data in the text data that is irrelevant to the radar countermeasures domain, as follows:
[0033] Step 2.1.1: Use an HTML parsing library to extract the text content;
[0034] Step 2.1.2: Use the re library for regular expressions to filter special characters;
[0035] Step 2.1.3: Use the NLTK library to filter stop words;
[0036] Step 2.2: Use a dictionary-based word segmentation algorithm to segment continuous multi-source radar countermeasures text data into sequences of words and phrases, as follows:
[0037] Step 2.2.1: Perform text matching on the preprocessed continuous unstructured text data to form a directed acyclic word graph;
[0038] Step 2.2.2: Use a greedy algorithm to find the shortest path from the starting point to the ending point, forming a sequence of words and phrases.
[0039] As a specific example, step 3 describes the use of a sequence labeling method based on a pre-trained language model for entity recognition and relation extraction, as follows:
[0040] Step 3.1: Convert the preprocessed text sequence data of the multi-source radar countermeasures domain into a BERT-compatible input format;
[0041] Step 3.2: Load the BERT-based pre-trained radar adversarial domain knowledge joint extraction model, including the entity extraction module NER Module, the relation classification module RE Module, and the shared feature extraction module BERT;
[0042] Step 3.3: Input unstructured text data after format conversion to complete entity recognition and relation extraction in the radar countermeasures domain.
[0043] As a specific example, the unstructured text data after input format conversion described in step 3.3 is used to complete entity recognition and relation extraction in the radar countermeasures domain, as follows:
[0044] Step 3.3.1: Encode the input sequence using BERT to obtain the feature sequence, and then pass it through a feedforward neural network and the SoftMax function to obtain the named entity recognition output;
[0045] Step 3.3.2: The named entity recognition module outputs a sequence with the same length as the input sequence and a fixed dimension after argmax processing.
[0046] Step 3.3.3: Pass the original input sequence and the sequence transformed by argmax through the feedforward neural network FFCN layer respectively;
[0047] Step 3.3.4: Using the outputs of the two FFCN layers as input, the relationships between entities are predicted by mapping through a biaffine classifier.
[0048] The equivalent calculation formula for the Biaffine classifier is:
[0049]
[0050] in, Indicates the degree of matching between entities. and This represents the two sets of sequences output after passing through the pre-neural network. b represents the bias term, and W represents the linear weight vector.
[0051] As a specific example, the radar countermeasures knowledge unified description architecture described in step 4 performs knowledge fusion and knowledge representation on knowledge data extracted from different data sources to form a radar countermeasures unified description knowledge set, as follows:
[0052] Step 4.1: Divide the extracted radar countermeasures domain knowledge into blocks;
[0053] Step 4.2: Perform multiple Map-Reduce operations on the multi-block knowledge data after it has been divided to achieve load balancing.
[0054] Step 4.3: For the multiple knowledge data blocks after load balancing, calculate the attribute similarity and entity similarity in sequence, and complete the knowledge representation to form a unified description of the radar countermeasure knowledge set.
[0055] As a specific example, step 4.2 describes performing load balancing on the multi-block knowledge data after partitioning using multiple MapReduce operations, as follows:
[0056] Step 4.2.1: Map all data blocks and generate a set of intermediate key-value pairs;
[0057] Step 4.2.2: Sort and group all intermediate key-value pairs by their keys;
[0058] Step 4.2.3: Use the reduce function to combine the key values of each group and calculate the redundancy of the number of entities in each data block.
[0059] As a specific example, step 4.3 describes the sequential calculation of attribute similarity and entity similarity for multiple blocks of knowledge data after load balancing, and the completion of knowledge representation to form a unified knowledge set describing radar countermeasures, as follows:
[0060] Step 4.3.1: For the two pieces of knowledge data, calculate the similarity of each attribute in turn;
[0061] Step 4.3.2: Combine the similarity of each attribute to form an attribute similarity vector;
[0062] Step 4.3.3: Obtain entity similarity by calculating the cosine similarity of the two attribute similarity vectors;
[0063]
[0064] Where A and B are attribute similarity vectors, and α is entity similarity;
[0065] Step 4.3.4: Calculate the entity similarity between different data blocks in sequence, complete the knowledge representation based on the threshold decision results, and form a unified knowledge set describing radar countermeasures.
[0066] This invention also provides a radar countermeasures domain knowledge extraction system based on the BERT model. This system is used to implement the aforementioned radar countermeasures domain knowledge extraction method based on the BERT model. The system includes a first module to a fourth module, and the functions of each module are as follows:
[0067] The first module is used to establish a unified description architecture for radar countermeasure knowledge that combines static and dynamic elements.
[0068] The second module is used for unstructured data preprocessing of radar countermeasure text data from different data sources;
[0069] The third module uses a sequence labeling method based on a pre-trained language model for entity recognition and relation extraction.
[0070] The fourth module, based on the unified description architecture of radar countermeasure knowledge, performs knowledge fusion and knowledge representation on knowledge data extracted from different data sources to form a unified description knowledge set for radar countermeasure.
[0071] The present invention also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the radar countermeasure domain knowledge extraction method based on the BERT model.
[0072] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0073] Example
[0074] like Figure 1 As shown, this invention provides a radar countermeasures domain knowledge extraction method based on the BERT model, comprising the following steps:
[0075] Step 1: Establish a unified description framework for radar countermeasure knowledge that combines static and dynamic elements, such as... Figure 2 As shown, the details are as follows:
[0076] (1) Dynamic knowledge includes state analysis reasoning and interference decision reasoning knowledge in the radar countermeasure process, which are reasoning models for problems in the field of radar countermeasure, including radar state reasoning model, air defense and anti-missile mission planning model, active and passive interference decision model, and interference resource optimization model.
[0077] (2) Static knowledge includes common knowledge and knowledge structure in the field of radar countermeasures, which are the knowledge ontology and general knowledge system in the field of radar countermeasures. The domain knowledge ontology includes radar object ontology, electronic countermeasure object ontology, game countermeasure concept ontology, and game event ontology. The general knowledge system includes predicate logic system, nondeterministic system, deep network reasoning system, graph search model, and general mathematical model solution method.
[0078] Step 2: Perform unstructured data preprocessing on radar countermeasures text data from different data sources, as follows:
[0079] Step 2.1: Perform noise reduction processing to merge and delete data in the text data that is irrelevant to the radar countermeasures domain. The steps are as follows:
[0080] Step 2.1.1: Use an HTML parsing library to extract the text content;
[0081] Step 2.1.2: Use the re library for regular expressions to filter special characters;
[0082] Step 2.1.3: Use the NLTK library to filter stop words;
[0083] Step 2.2: Use a dictionary-based word segmentation algorithm to segment continuous multi-source radar countermeasures text data into sequences of words and phrases. The steps are as follows:
[0084] Step 2.2.1: Perform text matching on the preprocessed continuous unstructured text data to form a directed acyclic word graph;
[0085] Step 2.2.2: Use a greedy algorithm to find the shortest path from the starting point to the ending point, forming a sequence of words and phrases.
[0086] Step 3: Use a sequence labeling method based on a pre-trained language model for entity recognition and relation extraction, as detailed below:
[0087] Step 3.1: Convert the preprocessed text sequence data of the multi-source radar countermeasures domain into a BERT-compatible input format;
[0088] Step 3.2: Load the BERT-based pre-trained radar adversarial domain knowledge joint extraction model, such as... Figure 3 As shown, it includes the Entity Extraction Module (NER Module), the Relationship Classification Module (RE Module), and the Shared Feature Extraction Module (BERT).
[0089] Step 3.3: Input the unstructured text data after format conversion, and complete entity recognition and relation extraction in the radar countermeasures domain, as detailed below:
[0090] Step 3.3.1: Encode the input sequence using BERT to obtain the feature sequence, and then pass it through a feedforward neural network and the SoftMax function to obtain the named entity recognition output;
[0091] Step 3.3.2: The named entity recognition module outputs a sequence with the same length as the input sequence and a fixed dimension after argmax processing.
[0092] Step 3.3.3: Pass the original input sequence and the sequence transformed by argmax through the feedforward neural network FFCN layer respectively;
[0093] Step 3.3.4: Using the outputs of the two FFCN layers as input, the relationships between entities are predicted by mapping through a biaffine classifier.
[0094] The equivalent calculation formula for the Biaffine classifier is:
[0095]
[0096] in, Indicates the degree of matching between entities. and This represents the two sets of sequences output after passing through the pre-neural network. b represents the bias term, and W represents the linear weight vector.
[0097] Step 4: Based on the unified description architecture of radar countermeasure knowledge, knowledge data extracted from different data sources is fused and represented to form a unified description knowledge set for radar countermeasures, such as... Figure 4 The details are as follows:
[0098] Step 4.1: Divide the extracted radar countermeasures domain knowledge into blocks;
[0099] Step 4.2: For the multi-block knowledge data after partitioning, perform multiple MapReduce operations for load balancing to ensure that the number of entities in all blocks is roughly the same. The steps are as follows:
[0100] Step 4.2.1: Map all data blocks and generate a set of intermediate key-value pairs;
[0101] Step 4.2.2: Sort and group all intermediate key-value pairs by their keys;
[0102] Step 4.2.3: Use the reduce function to combine the key values of each group and calculate the redundancy of the number of entities in each data block;
[0103] Step 4.3: For the multiple knowledge data blocks after load balancing, calculate attribute similarity and entity similarity sequentially, and complete the knowledge representation to form a unified description of the radar countermeasure knowledge set. The steps are as follows:
[0104] Step 4.3.1: For the two pieces of knowledge data, calculate the similarity of each attribute in turn;
[0105] Step 4.3.2: Combine the similarity of each attribute to form an attribute similarity vector;
[0106] Step 4.3.3: Obtain entity similarity by calculating the cosine similarity of the two attribute similarity vectors;
[0107]
[0108] Where A and B are attribute similarity vectors, and α is entity similarity;
[0109] Step 4.3.4: Calculate the entity similarity between different data blocks in sequence, complete the knowledge representation based on the threshold decision results, and form a unified knowledge set describing radar countermeasures.
[0110] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A radar countermeasures domain knowledge extraction method based on the BERT model, characterized in that, Includes the following steps: Step 1: Establish a unified description framework for radar countermeasure knowledge that combines static and dynamic elements; Step 2: Perform unstructured data preprocessing on radar countermeasure text data from different data sources; Step 3: Use a sequence labeling method based on a pre-trained language model for entity recognition and relation extraction; Step 4: Based on the unified description architecture of radar countermeasure knowledge, knowledge data extracted from different data sources are fused and represented to form a unified description knowledge set for radar countermeasure.
2. The radar countermeasure domain knowledge extraction method based on the BERT model according to claim 1, characterized in that, The establishment of a unified description architecture for radar countermeasure knowledge that combines static and dynamic elements, as described in step 1, is as follows: 1) Dynamic knowledge includes state analysis reasoning and jamming decision reasoning knowledge in the radar countermeasure process, which are reasoning models for problems in the field of radar countermeasures, including radar state reasoning models, air defense and anti-missile mission planning models, active and passive jamming decision models, and jamming resource optimization models. 2) Static knowledge includes common knowledge and knowledge structure in the field of radar countermeasures, which constitute the knowledge ontology and general knowledge system of the radar countermeasures field. The domain knowledge ontology includes radar object ontology, electronic countermeasures object ontology, game countermeasure concept ontology, and game event ontology. The general knowledge system includes predicate logic system, nondeterministic system, deep network reasoning system, graph search model, and general mathematical model solution method.
3. The radar countermeasures domain knowledge extraction method based on the BERT model according to claim 1, characterized in that, Step 2 involves unstructured data preprocessing of radar countermeasures text data from different data sources, as detailed below: Step 2.1: Perform noise reduction processing to merge and delete data in the text data that is irrelevant to the radar countermeasures domain, as follows: Step 2.1.1: Use an HTML parsing library to extract the text content; Step 2.1.2: Use the re library for regular expressions to filter special characters; Step 2.1.3: Use the NLTK library to filter stop words; Step 2.2: Use a dictionary-based word segmentation algorithm to segment continuous multi-source radar countermeasures text data into sequences of words and phrases, as follows: Step 2.2.1: Perform text matching on the preprocessed continuous unstructured text data to form a directed acyclic word graph; Step 2.2.2: Use a greedy algorithm to find the shortest path from the starting point to the ending point, forming a sequence of words and phrases.
4. The radar countermeasure domain knowledge extraction method based on the BERT model according to claim 1, characterized in that, Step 3 describes the use of a sequence labeling method based on a pre-trained language model for entity recognition and relation extraction, as follows: Step 3.1: Convert the preprocessed text sequence data of the multi-source radar countermeasures domain into a BERT-compatible input format; Step 3.2: Load the BERT-based pre-trained radar adversarial domain knowledge joint extraction model, including the entity extraction module NER Module, the relation classification module RE Module, and the shared feature extraction module BERT; Step 3.3: Input unstructured text data after format conversion to complete entity recognition and relation extraction in the radar countermeasures domain.
5. The radar countermeasure domain knowledge extraction method based on the BERT model according to claim 4, characterized in that, The unstructured text data after input format conversion described in step 3.3 is used to complete entity recognition and relation extraction in the radar countermeasures domain, as detailed below: Step 3.3.1: Encode the input sequence using BERT to obtain the feature sequence, and then pass it through a feedforward neural network and the SoftMax function to obtain the named entity recognition output; Step 3.3.2: The named entity recognition module outputs a sequence with the same length as the input sequence and a fixed dimension after argmax processing. Step 3.3.3: Pass the original input sequence and the sequence transformed by argmax through the feedforward neural network FFCN layer respectively; Step 3.3.4: Using the outputs of the two FFCN layers as input, the relationships between entities are predicted by mapping through a biaffine classifier. The equivalent calculation formula for the Biaffine classifier is: in, Indicates the degree of matching between entities. and This represents the two sets of sequences output after passing through the pre-neural network. b represents the bias term, and W represents the linear weight vector.
6. The radar countermeasures domain knowledge extraction method based on the BERT model according to claim 1, characterized in that, Step 4 describes a unified knowledge description architecture for radar countermeasures, which integrates and represents knowledge data extracted from different data sources to form a unified knowledge set for radar countermeasures, as detailed below: Step 4.1: Divide the extracted radar countermeasures domain knowledge into blocks; Step 4.2: Perform multiple Map-Reduce operations on the multi-block knowledge data after it has been divided to achieve load balancing. Step 4.3: For the multiple knowledge data blocks after load balancing, calculate the attribute similarity and entity similarity in sequence, and complete the knowledge representation to form a unified description of the radar countermeasure knowledge set.
7. The radar countermeasures domain knowledge extraction method based on the BERT model according to claim 6, characterized in that, Step 4.2 describes the load balancing process performed on the multi-block knowledge data after partitioning using multiple MapReduce operations, as detailed below: Step 4.2.1: Map all data blocks and generate a set of intermediate key-value pairs; Step 4.2.2: Sort and group all intermediate key-value pairs by their keys; Step 4.2.3: Use the reduce function to combine the key values of each group and calculate the redundancy of the number of entities in each data block.
8. The radar countermeasure domain knowledge extraction method based on the BERT model according to claim 6, characterized in that, Step 4.3 describes the sequential calculation of attribute similarity and entity similarity for the multiple knowledge data blocks after load balancing, and the completion of knowledge representation to form a unified knowledge set describing radar countermeasures, as detailed below: Step 4.3.1: For the two pieces of knowledge data, calculate the similarity of each attribute in turn; Step 4.3.2: Combine the similarity of each attribute to form an attribute similarity vector; Step 4.3.3: Obtain entity similarity by calculating the cosine similarity of the two attribute similarity vectors; Where A and B are attribute similarity vectors, and α is entity similarity; Step 4.3.4: Calculate the entity similarity between different data blocks in sequence, complete the knowledge representation based on the threshold decision results, and form a unified knowledge set describing radar countermeasures.
9. A radar countermeasures domain knowledge extraction system based on the BERT model, characterized in that, This system is used to implement the radar countermeasure domain knowledge extraction method based on the BERT model as described in any one of claims 1 to 8. The system includes a first module to a fourth module, and the functions of each module are as follows: The first module is used to establish a unified description architecture for radar countermeasure knowledge that combines static and dynamic elements. The second module is used for unstructured data preprocessing of radar countermeasure text data from different data sources; The third module uses a sequence labeling method based on a pre-trained language model for entity recognition and relation extraction. The fourth module, based on the unified description architecture of radar countermeasure knowledge, performs knowledge fusion and knowledge representation on knowledge data extracted from different data sources to form a unified description knowledge set for radar countermeasure.
10. A mobile terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the radar countermeasure domain knowledge extraction method based on the BERT model as described in any one of claims 1 to 8.