A battlefield environment intelligence generation method and system based on natural language understanding, an electronic device, and a storage medium

Abstract

The application provides a battlefield environment intelligence generation method and system based on natural language understanding, an electronic device and a storage medium, relates to the technical field of data processing, and extracts enemy force deployment information and communication characteristic parameters from historical combat data and reconnaissance data, constructs an association model to generate a behavior mode library, extracts communication protocol characteristic parameters from battlefield electromagnetic data, matches the behavior mode library to form a communication characteristic mapping relationship, generates associated data in combination with force deployment rules and the communication characteristic mapping relationship, inputs a strategy deduction framework to construct a strategy space of enemy multi-level decision nodes, calculates node interaction behaviors in the strategy space topology structure, generates a decision tree, and then reconstructs the semantic of the decision tree by using battlefield environment feedback data, converts the semantic into a tactical logic chain by using a natural language understanding model, and generates an intelligence report, so that intelligent conversion from multi-source battlefield data to understandable tactical intelligence can be realized.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, system, electronic device, and storage medium for generating battlefield environment intelligence based on natural language understanding. Background Technology

[0002] In multilingual battlefield command environments, core requirements such as multilingual voice command recognition, accurate tactical intent parsing, and cross-lingual intelligence fusion generation need to be addressed. Key challenges include multilingual voice feature separation under complex electromagnetic interference, semantic disambiguation of battlefield terminology and dialectal slang, and the ability to correlate and reason about multi-source heterogeneous intelligence, in order to achieve end-to-end conversion from raw voice to structured tactical intelligence.

[0003] Current mainstream solutions employ a multimodal fusion-based speech-text joint analysis system: the mixed speech stream is separated by a microphone array and noise reduction algorithm, and the speaker's identity is identified by combining voiceprint features; the speech is converted into text using a pre-trained multilingual speech recognition model, and entity annotation is performed by embedding a battlefield terminology knowledge graph; the speech spectral features and text semantic vectors are fused using an attention mechanism to generate instruction parsing results with confidence scores; finally, an intelligence summary is output by linking the rule engine to the battlefield environment database.

[0004] However, this scheme has significant drawbacks. The language-specific speech recognition model has a high error rate in recognizing proper nouns for battlefield terminology, leading to deviations in subsequent semantic parsing. Texture features exhibit poor stability in high-noise battlefield environments, causing misidentification of speakers. The static update mechanism of the knowledge graph struggles to adapt to temporary changes in enemy tactical codenames, resulting in the failure of entity annotation. The rule engine relies on pre-set tactical logic templates and cannot autonomously infer implicit relationships between cross-language commands, resulting in insufficient completeness of intelligence generation. Summary of the Invention

[0005] The purpose of this application is to provide a method, system, electronic device, and storage medium for generating battlefield environment intelligence based on natural language understanding, in order to solve the problems of unclear voice commands and unclear tactical intentions in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a method for generating battlefield environment intelligence based on natural language understanding, comprising:

[0007] Enemy troop deployment information and communication characteristic parameters are extracted from historical combat data and reconnaissance data. A correlation model is constructed based on the multi-dimensional correlation elements between the enemy troop deployment information and the communication characteristic parameters, so as to generate a behavior pattern library through the correlation model.

[0008] Feature parameters reflecting enemy communication protocols are extracted from battlefield electromagnetic data collected based on radio sensor networks, and the evolution patterns of communication features in the behavior pattern library are associated and matched with the feature parameters to form a communication feature mapping relationship.

[0009] Based on the troop deployment patterns in the behavior pattern library and the mapping relationship between the communication features, related data is generated, and the related data is input into the strategy deduction framework to construct a strategy space containing multi-level decision-making nodes of the enemy.

[0010] Based on the topology of the strategy space, the interaction behavior of the enemy's multi-level decision nodes is calculated to generate a decision tree containing the path associated with combat intentions.

[0011] The decision tree is semantically reconstructed based on the acquired battlefield environment feedback data, and the semantic reconstruction result is transformed into a tactical logic chain using a natural language understanding model, and an intelligence report corresponding to the tactical logic chain is generated.

[0012] Optionally, enemy troop deployment information and communication characteristic parameters are extracted from the acquired historical combat data and reconnaissance data. A correlation model is constructed based on the multi-dimensional correlation elements between the enemy troop deployment information and the communication characteristic parameters to generate a behavior pattern library through the correlation model, including:

[0013] The records in historical combat data and reconnaissance data are divided into multiple data segments according to a preset time window, and enemy units are extracted from the multiple data segments as enemy troop deployment information;

[0014] The spectrum characteristics of enemy communication signals recorded in the multiple data segments are scanned to extract target parameters, and the statistical distribution value of the target parameters within the preset time window is calculated to form communication feature parameters.

[0015] Establish multi-dimensional correlation elements between the enemy troop deployment information and the communication feature parameters, use a dynamic connection algorithm to map the multi-dimensional correlation elements layer by layer to generate mapping relationships, and iteratively adjust the weight parameters of the mapping relationships at each layer to achieve a dynamic balance in the correlation strength of each layer.

[0016] The mapping relationships under the dynamic equilibrium state are combined into an association model, and the enemy unit movement patterns, communication equipment operation rules, and signal strength change trends output by the association model are stored in chronological order as a behavior pattern library.

[0017] Optionally, based on the mapping relationship between troop deployment patterns in the behavioral pattern library and the communication features, associated data is generated, and the associated data is input into a strategy deduction framework to construct a strategy space containing multi-level enemy decision nodes, including:

[0018] The movement direction and dwell time of enemy units within a continuous time window are extracted from the behavior pattern library to form a position sequence. The signal strength change value and frequency band switching interval of the enemy units within the continuous time window are extracted from the communication feature mapping relationship to form a signal sequence. The position sequence is paired with the records in the signal sequence to generate associated data containing a combination of position and signal tags.

[0019] The decision level to which the enemy unit belongs is determined based on the hierarchical identifier of the enemy unit, and a hierarchical weight value is assigned to each level. The position and signal combination mark of each enemy unit in the associated data are input into the strategy inference framework according to the hierarchical weight value, so that the node of each decision level in the strategy inference framework corresponds to the enemy unit within the range of the hierarchical weight value.

[0020] In the strategy deduction framework, corresponding parameters are loaded for each node at each decision level. The influence coverage of each node at each decision level is dynamically adjusted by comparing the number of interactions between the parameters and the nodes at each decision level. At the same time, cross-level nodes are constrained to be unidirectionally connected according to the level weight value, thereby forming a strategy space with hierarchical connections and varying coverage.

[0021] Optionally, the decision tree is semantically reconstructed based on the acquired battlefield environment feedback data, and the semantic reconstruction result is transformed into a tactical logic chain using a natural language understanding model, generating an intelligence report corresponding to the tactical logic chain, including:

[0022] The interaction behavior records of enemy units are extracted from the decision tree and matched with battlefield environment feedback data to obtain the movement trajectory, signal change characteristics and communication operation timing of the enemy units.

[0023] The movement trajectory is compared with preset geographical coordinates to generate spatial semantic segments. At the same time, temporal semantic segments are generated based on the signal change characteristics and the communication operation timing. The spatial semantic segments and temporal semantic segments are classified and combined according to the enemy unit to form behavioral semantic blocks.

[0024] The behavior semantic block is analyzed using a natural language understanding model, and the terrain constraints and communication rules in the tactical rule base are matched to generate terrain association chains and protocol association chains. The terrain association chains and protocol association chains are merged, duplicate actions are removed, key action sequences are retained, and a tactical logic chain is formed.

[0025] The time interval and direction change of adjacent actions in the tactical logic chain are detected. If the time interval is less than a set threshold, they are merged into a compound action and combined with the maneuver pattern analysis to generate an intelligence report.

[0026] Optionally, a multi-dimensional correlation element is established between the enemy troop deployment information and the communication feature parameters. A dynamic connection algorithm is used to map the multi-dimensional correlation element layer by layer to generate a mapping relationship. The weight parameters of the mapping relationship at each layer are iteratively adjusted to achieve a dynamic balance in the correlation strength of each layer. This includes:

[0027] The unit identifiers in the enemy troop deployment information are paired with the signal statistics in the communication characteristic parameters in chronological order to form a unit-signal sequence;

[0028] The spatial position change, signal intensity change, and unit type features in the unit and signal sequence are extracted as multidimensional correlation elements.

[0029] The multidimensional related elements are input into the multi-layer mapping structure of the dynamic connection algorithm and processed. In the first layer of the multi-layer mapping structure, a preliminary mapping relationship is generated by combining the spatial position change and the signal strength change. In the second layer of the multi-layer mapping structure, the preliminary mapping relationship is corrected based on the unit type characteristics to generate an intermediate mapping relationship. In the third layer of the multi-layer mapping structure, the intermediate mapping relationship is optimized by combining the historical behavior data of the enemy unit to generate the final mapping relationship.

[0030] Initialize the weight parameters of each layer, calculate the deviation between the predicted result of the final mapping relationship and the actual data, and adjust the weight parameters of each layer in reverse until the prediction deviation is less than the set threshold, so that the mapping relationship of each layer reaches a dynamic equilibrium state.

[0031] Optionally, feature parameters reflecting enemy communication protocols are extracted from battlefield electromagnetic data collected based on radio sensor networks, and the evolution patterns of communication features in the behavior pattern library are correlated and matched with the feature parameters to form a communication feature mapping relationship, including:

[0032] Continuous scanning of battlefield electromagnetic data collected by radio sensor networks is performed to extract recurring fixed-interval pulses from the signal waveforms emitted by enemy communication equipment as protocol time references, and the signal segments between two adjacent protocol time references are divided into communication protocol units.

[0033] Within the protocol unit, the number of frequency jumps and amplitude abrupt changes are detected to form a first feature group, and the frequency band energy distribution is statistically analyzed to form a second feature group. The first feature group is compared with the historical jump patterns in the behavior pattern library, and the amplitude abrupt change point deviation is compensated to generate compensated protocol unit features. The second feature group is compared with historical frequency band usage records, and the operational phase label is matched as the phase label of the protocol unit.

[0034] The protocol unit features and the stage labels are combined to form the feature parameters of the communication protocol. The feature parameters are superimposed with the communication signal strength change curves under the same combat stage identifier in the behavior pattern library. The signal strength difference between the two at the same time scale is calculated. When the difference is less than a set tolerance, a mapping relationship is established between the feature parameters and the corresponding communication feature evolution law in the behavior pattern library.

[0035] Optionally, based on the topology of the strategy space, the interactive behaviors of the enemy's multi-level decision nodes are calculated to generate a decision tree containing the associated paths of combat intentions, including:

[0036] Based on the connection weights and coverage of nodes at each level in the strategy space, the command hierarchy relationship between enemy units is determined. The final mapping relationship of the multi-layer mapping structure is classified according to the enemy unit identifier. The spatial position change and signal strength change value are superimposed on the command hierarchy relationship to generate the unit movement pattern.

[0037] Based on the command hierarchy, the extreme points of the signal strength change values ​​are extracted, and the time interval and strength difference between adjacent extreme points are statistically analyzed to generate the operating rules of communication equipment that reflect the command hierarchy.

[0038] Based on the command hierarchy and the angular change rate of the unit movement pattern, calculate the average intensity of the operational rules at each level and generate dynamic weighting coefficients.

[0039] The unit movement patterns, operational rules, and dynamic weight coefficients are bound to the enemy unit identifiers and timestamps. By analyzing the interaction behavior between nodes at each level, an operational intent association path reflecting the transmission of command intentions is constructed. Based on the changes in the interaction frequency and intensity of each node in the operational intent association path, a decision tree containing the enemy command hierarchy and the operational intent association path is generated.

[0040] Secondly, this application provides a battlefield environment intelligence generation system based on natural language understanding, comprising:

[0041] The acquisition module is used to extract enemy troop deployment information and communication characteristic parameters from the acquired historical combat data and reconnaissance data, and to construct an association model based on the multi-dimensional correlation elements between the enemy troop deployment information and the communication characteristic parameters, so as to generate a behavior pattern library through the association model;

[0042] The matching module is used to extract feature parameters reflecting the enemy's communication protocol from battlefield electromagnetic data collected based on radio sensor networks, and to associate and match the evolution law of communication features in the behavior pattern library with the feature parameters to form a communication feature mapping relationship.

[0043] The construction module is used to generate associated data based on the troop deployment rules of the behavior pattern library and the mapping relationship of the communication features, and input the associated data into the strategy inference framework to construct a strategy space containing multi-level decision nodes of the enemy.

[0044] The calculation module is used to calculate the interaction behavior of the enemy's multi-level decision nodes based on the topology of the strategy space, so as to generate a decision tree containing the path of combat intent association.

[0045] The transformation module is used to semantically reconstruct the decision tree based on the acquired battlefield environment feedback data, and to use a natural language understanding model to transform the semantic reconstruction result into a tactical logic chain, and generate an intelligence report corresponding to the tactical logic chain.

[0046] Thirdly, this application provides an electronic device, comprising:

[0047] Memory, used to store computer programs;

[0048] A processor, configured to implement the steps of the method as described in the first aspect above when executing the computer program.

[0049] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the method described in the first aspect above.

[0050] This application extracts enemy troop deployment information and communication characteristic parameters from historical combat and reconnaissance data, and constructs a multi-dimensional correlation model. Its technical effect lies in the dynamic correlation modeling of enemy combat units and communication behavior, breaking through the limitations of traditional intelligence analysis that separates troop deployment and communication characteristics. This forms a comprehensive behavioral pattern library reflecting the enemy's tactical coordination patterns. Furthermore, by matching the communication protocol characteristics extracted from battlefield electromagnetic data with the behavioral pattern library, a communication characteristic mapping relationship is established. Its technical effect is the precise correspondence between electromagnetic signal characteristics and enemy tactical behavior, providing a reliable signal dimension basis for inferring enemy operational intentions. The technical effect of constructing a strategy space by combining the deployment patterns of forces with the mapping relationship of communication characteristics is that it establishes a topological model of the enemy command system containing multi-level decision nodes, reveals the coordination mechanism between different levels of combat units, and generates a decision tree based on the interaction behavior of the topological computing nodes in the strategy space. The technical effect of this method is that it builds an interpretable reasoning framework that reflects the evolution path of the enemy's combat intentions, supports multi-level deduction of tactical intentions, and reconstructs the semantics of the decision tree and generates a tactical logic chain through battlefield environment feedback data. The technical effect of this method is that it realizes the analysis of enemy intentions and the automated generation of intelligence products in a dynamic battlefield environment, significantly improving the timeliness and accuracy of battlefield situation awareness.

[0051] Furthermore, by dividing the time window to extract enemy troop deployment information, scanning the spectral characteristics of communication signals to form feature parameters, and establishing a multi-dimensional correlation model between troop deployment and communication characteristics, the technical effect is that it breaks through the limitations of static feature extraction in traditional intelligence analysis. Based on a dynamic connection algorithm, it achieves hierarchical correlation modeling of multi-dimensional combat features, and establishes a stable and reliable behavioral pattern library through iterative optimization of weight parameters. Through the time-series storage of enemy unit movement patterns, communication operation rules, and signal change trends, it forms a complete knowledge system reflecting the dynamic evolution of enemy tactical behavior, providing high-fidelity behavioral benchmark data for subsequent strategy deduction, and significantly improving the accuracy of enemy combat pattern identification and the reliability of tactical intent prediction. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 A flowchart illustrating a method for generating battlefield environment intelligence based on natural language understanding, as provided in this application, is shown.

[0054] Figure 2 A scene diagram illustrating a method for generating battlefield environment intelligence based on natural language understanding, as provided in this application, is shown.

[0055] Figure 3 A schematic diagram of the structure of a battlefield environment intelligence generation system based on natural language understanding provided in this application is shown. Detailed Implementation

[0056] Researchers have found that existing battlefield intelligence analysis methods rely on single data sources or static rule bases, making it difficult to dynamically correlate enemy troop deployments and communication behavior characteristics. Furthermore, they lack semantic fusion capabilities for multi-source heterogeneous data, resulting in delayed and inaccurate tactical intent identification. Therefore, this paper proposes a battlefield environment intelligence generation method based on natural language understanding. This method can analyze enemy operational intentions and generate structured intelligence through multi-dimensional data correlation modeling and dynamic strategy deduction. The technical solution of this application is applicable to intelligent decision support in complex combat scenarios such as multi-source intelligence fusion and battlefield situational awareness.

[0057] The entire R&D process embodies the technological synergy of data association modeling and dynamic semantic understanding, aiming to overcome the shortcomings of existing solutions, such as fragmented intelligence elements, static intent reasoning, and low semantic conversion efficiency. Through spatiotemporal correlation analysis of historical combat data and reconnaissance data, it overcomes the limitations of traditional force deployment identification; by combining the mapping and matching of electromagnetic signal characteristics and behavioral pattern libraries, it constructs a correlation reasoning mechanism between communication behavior and tactical intent, achieving multi-dimensional modeling of enemy operational patterns; and based on dynamic optimization of the strategy space and decision tree, it solves the problem of the disconnect between intelligence generation and battlefield feedback in traditional methods. This method significantly improves the timeliness and tactical value of battlefield intelligence through multi-level data fusion and natural language understanding technology.

[0058] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0059] The core of this application is to provide a method for generating battlefield environment intelligence based on natural language understanding. A flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0060] 101. Extract enemy troop deployment information and communication characteristic parameters from the acquired historical combat data and reconnaissance data, and construct a correlation model based on the multi-dimensional correlation elements between the enemy troop deployment information and the communication characteristic parameters, so as to generate a behavior pattern library through the correlation model;

[0061] In some embodiments, step 101 includes:

[0062] 1011. Divide the records in historical combat data and reconnaissance data into multiple data segments according to a preset time window, and extract enemy units from the multiple data segments as enemy troop deployment information;

[0063] 1012. Scan the spectral characteristics of enemy communication signals recorded in the multiple data segments, extract target parameters, and calculate the statistical distribution value of the target parameters within the preset time window to form communication feature parameters;

[0064] 1013. Establish multi-dimensional correlation elements between the enemy troop deployment information and the communication feature parameters, use a dynamic connection algorithm to map the multi-dimensional correlation elements layer by layer to generate mapping relationships, and adjust the weight parameters of the mapping relationships at each layer iteratively to make the correlation strength of each layer reach a dynamic equilibrium state.

[0065] Specifically, step 1013 includes: pairing unit identifiers in enemy troop deployment information with signal statistics in communication feature parameters in chronological order to form unit-signal sequences; extracting spatial position changes, signal strength changes, and unit type features from the unit-signal sequences as multidimensional association elements; inputting the multidimensional association elements into a multi-layer mapping structure of a dynamic connection algorithm and processing them; in the first layer of the multi-layer mapping structure, combining spatial position changes and signal strength changes to generate a preliminary mapping relationship; in the second layer of the multi-layer mapping structure, correcting the preliminary mapping relationship based on unit type features to generate an intermediate mapping relationship; in the third layer of the multi-layer mapping structure, optimizing the intermediate mapping relationship by combining historical behavior data of enemy units to generate a final mapping relationship; initializing the weight parameters of each layer, calculating the deviation between the predicted result and the actual data of the final mapping relationship, and adjusting the weight parameters of each layer in reverse until the prediction deviation is less than a set threshold, so that the mapping relationships of each layer reach a dynamic equilibrium state.

[0066] In step 1013, the multidimensional correlation element refers to the set of correlation dimensions composed of changes in the spatial location of enemy units, changes in signal strength, and unit type characteristics. The mapping relationship refers to the correlation rules generated through layer-by-layer correction in the multi-layer mapping structure of the dynamic connection algorithm. The dynamic equilibrium state refers to the stable convergence state achieved by the multi-layer mapping structure through iterative optimization of weights; the weight parameter directly reflects the degree of influence of the feature on the model through its numerical value.

[0067] 1014. Combine the mapping relationships under the dynamic equilibrium state into an association model, and store the enemy unit movement patterns, communication equipment operation rules, and signal strength change trends output by the association model in chronological order as a behavior pattern library.

[0068] In steps 1011-1014, enemy troop deployment information refers to the enemy combat unit identifiers, location coordinates, and type information extracted from historical combat data and reconnaissance data, including unit numbers, equipment types, and dynamic position changes. Communication characteristic parameters refer to the statistical distribution values ​​extracted after spectral scanning of enemy communication signals, including mean signal strength, frequency band fluctuation variance, and peak frequency band occupancy. Multidimensional correlation elements refer to the set of correlation dimensions composed of changes in the spatial location of enemy units, changes in signal strength, and unit type characteristics. Dynamic connection algorithm refers to a machine learning method that uses a multi-layer neural network structure to map multidimensional correlation elements layer by layer and adjusts weight parameters through backpropagation. Behavioral pattern library refers to a database storing enemy unit movement patterns, communication equipment operation rules, and signal strength change trends.

[0069] In this embodiment of the application, firstly, the historical combat data and reconnaissance data are divided into fixed time windows in step 1011, for example, every 10 minutes is a window. The unique identifier and coordinates of the enemy unit are extracted by the entity recognition algorithm to form the enemy troop deployment information.

[0070] Secondly, in step 1012, the communication signal in each data segment is scanned across the entire frequency band. The signal strength peak and frequency hopping interval are extracted using fast Fourier transform. The mean and variance are calculated using the sliding window statistical method to generate communication characteristic parameters with timestamps.

[0071] Next, in step 1013, the unit identifiers and communication feature parameters in the enemy troop deployment information are aligned with timestamps to construct a sequence of association between units and signals. From this sequence, spatial displacement quantities such as coordinate differences, signal strength change gradients such as the rate of change per minute, and unit type codes such as armored unit type code 01 are extracted to form a multidimensional association vector group. Subsequently, the multidimensional vector is input into a three-layer network of the dynamic connection algorithm: the first layer establishes a preliminary mapping relationship between spatial displacement and signal gradient through a linear regression model, for example, for every 1 kilometer increase in displacement, the signal variance increases by 0.3; the second layer introduces a random forest model to adjust the mapping weights according to the unit type, for example, armored units have a weight of 0.8 while communication units have a weight of 1.2; the third layer combines historical behavior data and optimizes the weight parameters through a dynamic time warping algorithm to ensure that the movement pattern is consistent with historical behavior.

[0072] In one specific embodiment of step 1013, the unit identifiers in the enemy troop deployment information are aligned and paired with the signal statistics in the communication characteristic parameters according to time sequence to form a unit and signal sequence with timestamps. For example, the identifier of armored unit A-203 is bound to its signal strength average of 75 dB and frequency band variance of 0.5 within the time window of 09:00 to 09:10 to generate an associated sequence containing spatiotemporal attributes.

[0073] Multidimensional correlation elements are extracted from the unit and signal sequences, including spatial location changes, signal strength changes, and unit type characteristics. The displacement of enemy units within a preset time window is calculated using coordinate differences; for example, if armored battalion B moves 3.2 kilometers in 10 minutes, this is considered a change in spatial location. Simultaneously, the gradient rate of change in signal strength is statistically analyzed; for example, if the signal strength of communication node C increases by 2 decibels per minute, this is considered a signal strength change. Furthermore, the equipment type code of the enemy unit is analyzed; for example, mechanized infantry type code 02 and electronic warfare unit type code 05 are used as unit type characteristics.

[0074] The multidimensional correlation elements are input into the three-layer mapping structure of the dynamic connection algorithm for processing. In the first layer mapping, a linear regression model is used to establish preliminary correlation rules between spatial displacement and signal strength. For example, for every 1 kilometer increase in displacement, the signal variance increases by 0.3, generating a preliminary mapping relationship such as a positive correlation between armored vehicle movement distance and communication frequency band fluctuations. In the second layer mapping, the weights are adjusted based on unit type characteristics. For example, a random forest model is used to assign a weight of 0.8 to armored vehicles and 1.2 to communication vehicles, adjusting the preliminary rule so that the weight of the impact of armored vehicle movement distance on frequency band fluctuations is reduced by 20%. In the third layer mapping, the weight parameters are optimized by combining historical behavior data. For example, a dynamic time warping algorithm is used to match the current movement trajectory with historical attack patterns, generating a final mapping relationship with an 85% similarity between the armored vehicle movement direction and historical attack paths.

[0075] After initializing the weight parameters for each layer, the prediction bias is iteratively calculated using the backpropagation algorithm. The predicted signal strength output from the mapping relationship is compared with the actual reconnaissance data. For example, if the mean square error between a predicted value of 78 dB and the actual value of 75 dB is 3%, and the error exceeds a set threshold such as 5%, the linear regression coefficients and random forest node splitting parameters are adjusted layer by layer in reverse until the error converges to within 2%. When the weight parameters of each layer stabilize the prediction bias within the set range, the multi-layer mapping relationship is considered to have reached a dynamic equilibrium state, for example, the signal strength prediction error fluctuation is less than ±0.5 dB.

[0076] Finally, in step 1014, the network weights are iteratively adjusted using the backpropagation algorithm until the prediction error converges to within 5%. The weight matrix of the three-layer network in the dynamic connection algorithm, the random forest classification rules, and the dynamic time warping path constraints are coupled with the probabilistic graphical model through tensor decomposition to form an association model. The model parsing engine performs spatiotemporal synchronization processing on the output enemy unit movement patterns, communication device operation rules, and signal strength change trends. An event sorting algorithm based on Lamport logical clocks is used to achieve millisecond-level time axis alignment. Finally, a three-layer index structure containing timestamps, behavior types, and feature vectors is generated and persistently stored in the behavior pattern library using differential encoding technology.

[0077] Here is a specific example:

[0078] In border defense combat systems, the construction of an enemy behavior pattern database begins with spatiotemporal data fusion. The system first segments historical radar trajectories and electronic reconnaissance data into fixed 10-minute time windows, and then extracts enemy combat unit information through entity recognition algorithms. For example, the unique identifier B-203 of Armored Battalion B was identified between 09:00 and 09:10, and its coordinate data showed that the unit had moved 3.2 kilometers eastward and 1.8 kilometers northward.

[0079] Simultaneously, the system performs a full-band scan of communication signals within the same time window. After extracting spectral features using Fast Fourier Transform, the measured signal strength is 78 dB, the frequency band variance is 0.6, and the standard deviation of the frequency hopping interval is 0.3 seconds. These parameters are precisely linked to the movement data of Armored Battalion B through timestamps, forming a spatiotemporal correlation sequence between the unit and the signal.

[0080] Based on this correlation sequence, the system initiates a three-layer dynamic correlation modeling process. In the first layer of mapping, a linear regression model establishes a quantitative rule for spatial displacement and signal fluctuation: for every 1 kilometer the armored battalion moves, the variance of the communication signal increases by 0.28 units. When a displacement of 3.2 kilometers is input, the model predicts that the signal variance should increase to 0.94, but the actual measured value is 0.89, resulting in an initial error of 5.3%. To correct the error, a unit type feature is introduced in the second layer of mapping. The random forest model adjusts the weight coefficient to 0.82 based on the armored unit type code 01. The corrected predicted value drops to 0.74, and the error narrows to 3.4%. At this point, the system triggers the third-layer optimization mechanism. The dynamic time warping algorithm compares the historical database and finds that the current movement trajectory has a similarity of 83% with a typical pincer attack, thus strengthening the weight factor to 1.12.

[0081] After three rounds of backpropagation iterations, the model weight matrix was gradually optimized from the initial 0.5, 0.3, and 0.2 to 0.62, 0.25, and 0.13, and the final prediction error converged to 1.8%. To ensure the temporal consistency of multi-source data, the system adopts the Lamport logical clock alignment mechanism to control the timestamp deviation between radar data and communication signals within 1.2 milliseconds.

[0082] The final generated behavior pattern entries contain a three-layer index structure: the timestamp layer uses wavelet time-frequency analysis to generate a composite index to accurately locate the signal fluctuation cycle; the behavior type layer constructs decision tree labels based on the edge distribution of the probability graph to distinguish tactical types such as maneuver and communication suppression; and the feature vector layer achieves efficient similarity retrieval through 128-bit locality-sensitive hashing driven by kernel tensors.

[0083] 102. Extract feature parameters reflecting the enemy's communication protocol from battlefield electromagnetic data collected based on radio sensor networks, and associate and match the evolution law of communication features in the behavior pattern library with the feature parameters to form a communication feature mapping relationship.

[0084] In some embodiments, step 102 includes:

[0085] 1021. Continuously scan the battlefield electromagnetic data collected by the radio sensor network, extract the recurring fixed-interval pulses in the signal waveforms emitted by the enemy's communication equipment as the protocol time reference, and divide the signal segment between two adjacent protocol time references into protocol units.

[0086] 1022. Within the protocol unit, the number of frequency jumps and amplitude mutation points are detected to form a first feature group, and the frequency band energy distribution is statistically analyzed to form a second feature group. The first feature group is compared with the historical jump patterns in the behavior pattern library, and the amplitude mutation point deviation is compensated to generate compensated protocol unit features. The second feature group is compared with historical frequency band usage records, and the operational phase label is matched as the phase label of the protocol unit.

[0087] 1023. Combine the protocol unit features with the stage label to form the feature parameters of the communication protocol. Superimpose the feature parameters with the communication signal strength change curves under the same combat stage identifier in the behavior pattern library. Calculate the signal strength difference between the two at the same time scale. When the difference is less than a set tolerance, establish a mapping relationship between the feature parameters and the corresponding communication feature evolution law in the behavior pattern library.

[0088] In steps 1021-1023, the protocol time reference refers to the periodically occurring fixed-interval pulses in the enemy communication signal, used to divide protocol units. A protocol unit refers to a complete communication signal segment between two adjacent protocol time references. The first feature group refers to the number of frequency jumps and signal amplitude abrupt changes detected within the protocol unit. The second feature group refers to the statistical values ​​of the frequency band energy distribution within the protocol unit. The stage label refers to the classification identifier generated after matching the frequency band energy distribution with historical combat stages. The protocol unit feature refers to the set of multi-dimensional parameters extracted from the enemy communication signal, used to characterize the core characteristics of the protocol unit. The signal strength difference refers to the degree of matching between the feature parameters and historical patterns. The communication feature evolution law refers to the dynamic change pattern of the enemy protocol summarized from long-term battlefield data.

[0089] In this embodiment, firstly, the battlefield electromagnetic data collected by the radio sensor network is continuously scanned in step 1021. An energy detection algorithm is used to identify periodically occurring fixed-interval pulses in the enemy's communication signals as a protocol time reference. For example, a synchronization pulse occurring every 1.2 seconds is used to regularly segment complete signal segments between adjacent pulses, forming protocol units.

[0090] Secondly, time-frequency analysis is performed within the protocol unit in step 1022. Short-time Fourier transform is used to detect the number of frequency jumps, such as 8 times per second, forming the first feature group. An amplitude abrupt change, such as a sudden drop in signal strength of 20 dB, is detected using a cumulative sum algorithm, and the location and difference of the abrupt change are recorded. Simultaneously, wavelet packet decomposition is used to calculate the energy proportion of each sub-band, for example, the 2.6 GHz band accounts for 70%, forming the second feature group. Next, the first feature group is compared with historical jump patterns in the behavior pattern library. If the number of jumps deviates from the historical average, such as 8 times higher than the average of 6 times, a linear compensation algorithm is used to correct the amplitude abrupt change deviation, such as adjusting 20 dB to 18 dB, ensuring consistency between the protocol features and historical behavior patterns. Simultaneously, the second feature group is matched with historical frequency band usage records, such as a high proportion of 2.6 GHz corresponding to armored force assembly orders, to add an operational phase label to the protocol unit, forming a contextual association.

[0091] Finally, in step 1023, the compensated protocol unit features are merged with the stage label to form the feature parameters of the communication protocol. These feature parameters are then superimposed with the signal strength change curves of the same stage in the behavior pattern library, and a dynamic time-bending algorithm is used to calculate the signal strength difference. If the signal strength difference is less than the multi-level tolerance threshold derived from historical data, a confidence-weighted mapping decision mechanism is triggered: first, the tactical semantic consistency between the stage label and the frequency band energy distribution is verified; then, a triplet mapping relationship of <feature parameters, evolution law, confidence> is generated through the association reasoning engine of the pattern library; finally, this mapping relationship is written into the knowledge graph node of the behavior pattern library using incremental learning, and the corresponding tactical intent recognition interface is activated. For example, when the mean square error of the 2.6GHz band feature parameters is consistently lower than the 5dB threshold of the armored force assembly stage for three consecutive protocol units, the system automatically generates a strong correlation mapping between this feature pattern and the armored force communication protocol upgrade version V2.3, updates the correlation confidence to 0.92, and links the electronic warfare system to preload the corresponding frequency band interference plan.

[0092] Here is a specific example:

[0093] In a real-world case study of battlefield electromagnetic countermeasures systems, an electronic reconnaissance system in a certain area captured enemy tactical radio signals through a radio sensor network. The system first detected periodic synchronization pulses in the UHF band, characterized by a narrow pulse sequence repeating every 0.8 seconds, with a pulse width of 2 milliseconds and a time error of less than 0.05 milliseconds. This characteristic was verified by a pattern library to be the synchronization header structure of a certain type of frequency-hopping radio. Based on this, the signal segment lasting 0.8 seconds between adjacent pulses was segmented into protocol units. Subsequently, the protocol units were analyzed from multiple dimensions: the time-frequency analysis module detected 12 frequency jumps per second, covering the range of 1.2 GHz to 1.8 GHz, and simultaneously captured a modulation feature with an amplitude drop of 15 dB at the 0.45-second position; the frequency domain decomposition module used wavelet packet technology to divide the signal into 32 sub-bands, finding that the 1.5 GHz sub-band accounted for 82% of the energy. Based on historical combat data, the frequency band distribution characteristics match the historical records of the "field command vehicle cluster communication" phase by more than 90%. However, the amplitude mutation value deviates from the 13 dB reference value recorded in the pattern library due to channel attenuation. This triggers an adaptive threshold algorithm to correct the current value to 13 dB, with the error controlled within the ±2 dB tolerance, ensuring the consistency of the characteristic parameters with the historical pattern library.

[0094] After calibration, the system fuses the corrected frequency jump count, amplitude abrupt change points, and 1.5GHz energy proportion characteristics into a multi-dimensional vector, and performs dynamic time-warping matching with the communication template labeled "Command Vehicle Coordination Protocol V5" in the behavior pattern library. The algorithm elastically aligns the three consecutive signal segments of the current protocol unit (with mean square errors of 3.2 dB, 2.9 dB, and 3.1 dB, respectively) with the historical template by constraining the stretching and compression ratio of the time axis, with all error values ​​below the preset 4 dB threshold. At this point, the confidence assessment module is activated. After verifying the strong correlation between the 1.5GHz band energy distribution and the command vehicle's maneuver tactics, it generates a triplet mapping relationship of <12 jumps / 13dB correction / 1.5GHz -82%, Command Vehicle V5 Protocol, 0.95 confidence>, which is then synchronously updated to the command link nodes in the tactical knowledge graph. The system then triggers the electronic countermeasures unit to load an intelligent jamming strategy: based on the jump pattern, it predicts that the communication frequency of the next time slot is 1.65GHz, generates a comb-shaped blocking signal covering 1.45-1.65GHz, and simultaneously injects a deception pulse synchronized with the enemy's protocol time slot, forming a dual attack of spectrum suppression and protocol confusion, and finally achieving a closed-loop tactical response from accurate identification of protocol characteristics to dynamic electromagnetic countermeasures.

[0095] 103. Generate associated data based on the troop deployment patterns in the behavior pattern library and the mapping relationship between the communication features, and input the associated data into the strategy deduction framework to construct a strategy space containing multi-level decision nodes of the enemy;

[0096] In some embodiments, step 103 includes:

[0097] 1031. Extract the movement direction and dwell time of enemy units within a continuous time window from the behavior pattern library to form a position sequence. Extract the signal strength change value and frequency band switching interval of the enemy units within the continuous time window from the communication feature mapping relationship to form a signal sequence. Pair the position sequence with the records in the signal sequence to generate associated data containing a combination of position and signal markers.

[0098] 1032. Determine the decision level to which the enemy unit belongs based on the hierarchical identifier, and assign a hierarchical weight value to each level. Input the position and signal combination mark of each enemy unit in the associated data into the strategy deduction framework according to the hierarchical weight value, so that the node of each decision level in the strategy deduction framework corresponds to the enemy unit within the range of the hierarchical weight value.

[0099] 1033. In the strategy deduction framework, corresponding parameters are loaded for each node of each decision level. The influence coverage of each node of each decision level is dynamically adjusted by comparing the number of interactions between the parameters and the nodes of each decision level. At the same time, cross-level nodes are constrained to be unidirectionally connected according to the level weight value, thereby forming a strategy space with hierarchical connections and coverage changes.

[0100] In steps 1031-1033, the associated data refers to the labeled data generated by pairing enemy unit location sequences with communication signal sequences. The hierarchical weight value refers to the decision influence coefficient assigned according to the enemy unit level, such as division or battalion. The strategy space refers to the network topology containing multi-level decision nodes and their connections. The communication feature mapping relationship refers to the association rules between enemy communication behavior patterns and tactical intentions; its core is to establish a mapping model from communication features to tactical intentions by analyzing the matching degree between communication signal features and historical behavior patterns. The strategy inference framework is a dynamic network model simulating multi-level enemy decision-making behavior; its core is to construct a dynamically adjustable strategy space network through hierarchical node connections and weight allocation.

[0101] In this embodiment, firstly, step 1031 extracts the movement trajectory data of enemy units within a continuous time window from the behavior pattern library, such as a coordinate sequence within 24 hours. Missing coordinate points are then filled in using a cubic spline interpolation algorithm to generate a time-stamped position sequence. For example, the movement trajectory of Armored Battalion A updates its latitude and longitude records every 5 minutes. Simultaneously, communication signal characteristics within the corresponding time window are extracted, including signal strength range (e.g., a difference of 30 dB between the maximum and minimum values) and frequency band switching frequency (e.g., switching 3 times per hour), forming a signal sequence. The position sequence and signal sequence are precisely paired according to the timestamp to generate associated data containing a combination of position and signal markers. For example, the coordinates of unit D-105 corresponding to timestamp 09:00:00 are 118.5 degrees longitude, 32.8 degrees latitude, with a signal strength of 75 dB and 2 frequency band switching times.

[0102] Secondly, in step 1032, hierarchical weight values ​​are assigned based on the hierarchical identifiers of enemy units, such as the prefix "D" representing division-level units. For example, the weight for a division-level unit is set to 0.9, and the weight for a battalion-level unit is set to 0.5. The position and signal combination markers of each enemy unit in the associated data are input into the node generation module of the strategy deduction framework according to the hierarchical weight values. The initial coverage radius of a division-level node is set to the hierarchical weight value multiplied by 10 kilometers, such as 0.9 multiplied by 10 equals 9 kilometers, and the radius of a battalion-level node is set to 4.5 kilometers. A force-directed algorithm is used to dynamically arrange the node positions, linking physical distance with signal strength. For example, nodes with high-strength signals attract each other, forming a preliminary strategy space network topology.

[0103] Next, in step 1033, parameters are loaded for each node at the decision-making level, such as the command issuance frequency and signal strength threshold. By statistically analyzing interaction events between nodes (e.g., a division node sends commands to a battalion node 5 times per hour), the coverage radius is dynamically adjusted using exponential smoothing; for example, the radius expands by 5% for every 10 additional interactions. Simultaneously, cross-level nodes are constrained to unidirectional connections based on weighted values; for example, a battalion node can only receive commands from a division node, ensuring the hierarchical nature of the command chain. Finally, a policy space containing dynamic coverage and hierarchical connection relationships is generated, providing structured data support for subsequent intent inference.

[0104] Here is a specific example:

[0105] In a live-fire simulation of a battlefield intelligent decision-making system, an electronic warfare center integrated multi-source data to construct the enemy's strategy space. The system first extracted 72 consecutive hours of movement trajectory data from a behavior pattern database. It then used cubic spline interpolation to complete the latitude and longitude coordinates sampled every 5 minutes, generating a second-accurate timestamp location sequence. For example, the coordinates for 09:00:00 were (118.5 degrees East, 32.8 degrees North). Simultaneously, it correlated the 30 dB difference in communication signal strength within the same time window with the frequency of 3.2 frequency band switching per hour, forming a spatiotemporally coupled associated dataset. Based on the hierarchical identifier of the D-105 armored division, a decision weight of 0.9 was assigned, while its subordinate Y-307 armored battalion was assigned a weight of 0.5. This established an initial coverage radius of 9 km for division-level nodes and 4.5 km for battalion-level nodes. Using a force-directed algorithm, the high-intensity signal source of 85 dB from the division-level command vehicle was used as the gravitational core, attracting surrounding battalion-level nodes to form an initial command cluster topology.

[0106] Based on the evolution of the initial command cluster topology, after the system loads core parameters such as the command issuance frequency threshold of 3 times per hour and the communication strength threshold of 70 dB for the armored division node, it immediately triggers a dynamic monitoring mechanism: when the interaction frequency between the division-level node and the Y-307 battalion-level node is detected to jump from the baseline of 3 times per hour to 5 times per hour, the exponential smooth expansion algorithm of the coverage radius is immediately activated, extending the radius of the division-level node from 9 kilometers to 10.8 kilometers, simultaneously constraining cross-level connection rules, and cutting off possible lateral interference links between battalion-level nodes. At this time, the newly emerging electronic warfare battalion Y-409 unit connects to the division-level node with an interaction frequency of 4 times per hour. The system matches its signal characteristic sequence through a dynamic time warp algorithm, identifies the electronic warfare relay function undertaken by the unit, shortens its frequency band switching interval to 2.5 minutes, and compresses the signal strength fluctuation range to ±5 dB, showing a significant difference from the communication characteristics of the armored battalion Y-307. Based on this, the system expanded the radius of the Y-409 node from 4.5 km to 6.2 km, and reconstructed the command link in the strategy space with the division-level node D-105 as the hub: after receiving the division-level jamming command, the electronic warfare battalion Y-409 forwards the encrypted control signal to the armored battalion Y-307, forming a three-level cascaded control architecture. The final generated strategy space network contains 12 division-level core nodes and 47 battalion-level execution nodes. All 213 unidirectional command links between nodes are loaded with interaction frequency, signal strength threshold, and response delay parameters, realizing a complete simulation closed loop from static topology construction to dynamic command chain adaptation.

[0107] 104. Based on the topology of the strategy space, calculate the interaction behavior of the enemy's multi-level decision nodes to generate a decision tree containing the path of combat intent association.

[0108] In some embodiments, step 104 includes:

[0109] 1041. Based on the connection weights and coverage of nodes at each level in the strategy space, determine the command hierarchy relationship between enemy units, classify the final mapping relationship of the multi-layer mapping structure according to the enemy unit identifier, and calculate the superposition result of the spatial position change and signal strength change value in combination with the command hierarchy relationship to generate the unit movement pattern.

[0110] 1042. Based on the command hierarchy, extract the extreme points of the signal strength change value, and count the time interval and strength difference between adjacent extreme points to generate the operation pattern of the communication equipment that reflects the command hierarchy;

[0111] 1043. Based on the command hierarchy and the angle change rate of the unit movement mode, calculate the average intensity of the operational rules at each level and generate dynamic weighting coefficients;

[0112] 1044. Bind the unit movement pattern, operation rules and dynamic weight coefficients to the enemy unit identifier and timestamp, construct the combat intent association path that reflects the transmission of command intentions by analyzing the interaction behavior between nodes at each level, and generate a decision tree that includes the enemy command level and the combat intent association path based on the interaction frequency and intensity changes of each node in the combat intent association path.

[0113] In steps 1041-1044, the command hierarchy refers to the hierarchical command link determined by the connection weights of nodes in the strategy space. The unit movement pattern refers to the regularized movement trajectory generated through position sequences. Operational patterns refer to the communication equipment usage characteristics extracted from signal sequences. The dynamic weight coefficient refers to the node influence parameter calculated based on interaction frequency and signal strength. The enemy unit identifier refers to the code or feature set used to uniquely distinguish enemy combat entities; its core function is to support the association and hierarchical analysis of multi-source data. The combat intent association path refers to the dynamic behavioral chain reflecting the logic of enemy command intent transmission; its construction is based on the strategy space topology and node interaction behavior.

[0114] In this embodiment, firstly, step 1041 determines the command hierarchy between enemy units based on the connection weights and coverage of nodes at each level within the strategy space. For example, when the connection weight between a division-level node and a battalion-level node is 0.8, by analyzing the consistency between the unit coordinate sequence and the command direction, it is calculated that 80% of the movement directions of battalion-level units match the division-level commands, thereby establishing a command link between upper and lower levels.

[0115] Secondly, in step 1042, polynomial fitting is performed on the movement trajectory of enemy units to extract regularized parameters with an average speed of 30 kilometers per hour and a standard deviation of turning angle of ±15 degrees, thereby generating unit movement patterns. Simultaneously, the encryption start time at 09:00 daily and the frequency band switching cycle every 15 minutes are extracted from the communication signal sequence to form operational patterns of communication equipment reflecting the command level.

[0116] Next, based on the extreme points of signal strength changes in step 1043, the time interval and strength difference between adjacent extreme points are statistically analyzed. For example, the interval between events where the signal suddenly increases by 20 dB is 2 hours. Combining the command hierarchy, dynamic weight coefficients are calculated. The division-level weight is 1.62, calculated by multiplying the base weight of 0.9 by the total contribution rate of subordinate units' signals (80%). The battalion-level weight is adjusted to 0.5 based on the cosine similarity between the movement direction and the superior's instructions. Subsequently, the unit movement patterns, operational rules, and dynamic weights are bound by timestamps. The A* algorithm is used to search for command paths in the decision tree, such as the instruction transmission path from the division level to the battalion level and then to the company level, and path weights are assigned based on an interaction frequency of 5 times per hour.

[0117] Finally, step 1044 integrates the interactive behaviors and path weights of nodes at each level to construct a decision tree reflecting the enemy's command intentions. For example, at 09:00, the signal of the division-level node D-105 suddenly increases by 20 decibels, triggering the "command issuance" flag; 15 minutes later, the subordinate armored battalion Y-307 moves 3 kilometers northeast and accelerates its communication frequency switching to once every 10 minutes, while the electronic warfare battalion Y-409 simultaneously starts 1.5GHz jamming. The system assigns a weight of 1.6 to the main path "command to maneuver" and a weight of 0.7 to the branch path "command to jamming, then to encrypted communication," and generates a three-layer decision tree based on confidence levels (0.9 for division level, 0.85 for armored battalion level, and 0.8 for electronic warfare): the root node is the signal surge event, the first-level branches are jamming and maneuver responses, and the second-level nodes are associated with the action parameters of each battalion level, forming a complete mapping chain from signal to tactics.

[0118] Here is a specific example:

[0119] In the practical simulation of the battlefield intelligent decision-making system, tactical intent modeling is achieved through hierarchical analysis of enemy command behavior. First, based on the connection weight of 0.8 between the division-level node D-105 and its subordinate battalion-level units in the strategy space and the coverage parameters, the system analyzes the coordinate sequence of armored battalion Y-307 and finds that 83% of its movement direction within 6 hours matches the division-level instructions, generating unit movement patterns, including a maneuver pattern with an average speed of 32 kilometers per hour and a standard deviation of ±14 degrees for the turning angle. At the same time, combined with the fixed operation pattern of encrypted communication starting at 09:00 every day and frequency band switching every 12 minutes, a spatiotemporal behavioral feature database is formed. Subsequently, extreme points of communication signal strength were extracted. After the signal suddenly increased by 22 decibels at 09:00, the subordinate electronic warfare battalion Y-409 responded within 10 minutes by initiating jamming on the 1.5GHz band, with the signal strength gradient reaching 8 decibels per minute. Meanwhile, the armored battalion Y-307 moved 3.2 kilometers east longitude and north latitude at 118.7 degrees east longitude and 32.9 degrees north latitude 15 minutes later and accelerated the frequency band switching to once every 10 minutes. This generated the operational pattern of the command level: the interval between signal surge events is 2 hours, and the standard deviation of the response delay of subordinate units is ±3 minutes.

[0120] Next, based on the cosine similarity of the maneuver direction (0.87) and the signal contribution rate (85%), the weight coefficients were dynamically calculated. The weight of the division-level node was increased from the base value of 0.9 to 1.6, the weight of the electronic warfare battalion was corrected to 0.75, and the weight of the armored battalion decreased to 0.6 due to frequency band switching stability. Finally, the unit movement patterns, operational rules, and weight coefficients were bound by timestamps to construct the operational intent association path: the main path describes the division-level command triggering the armored battalion's maneuver, with a weight of 1.6; the branch path describes the division-level command driving the electronic warfare battalion to jam and activate encrypted communication, with a weight of 0.75. Based on the confidence labels of division-level (0.92), armored battalion (0.86), and electronic warfare (0.78), a three-layer decision tree was generated: the root node is the signal surge event, the first-level branches describe the jamming initiation and maneuver response, and the second-level nodes are refined to the battalion-level coordinate offset of 3.2 kilometers and the frequency band switching standard deviation ±4 minutes, forming an interpretable command link from electromagnetic characteristics to tactical actions.

[0121] 105. Based on the acquired battlefield environment feedback data, the decision tree is semantically reconstructed, and the semantic reconstruction result is transformed into a tactical logic chain using a natural language understanding model, and an intelligence report corresponding to the tactical logic chain is generated.

[0122] In some embodiments, step 105 includes:

[0123] 1051. Extract the interaction behavior records of enemy units from the decision tree, match them with battlefield environment feedback data, and obtain the movement trajectory, signal change characteristics and communication operation timing of the enemy units.

[0124] 1052. Compare the movement trajectory with preset geographical coordinates to generate spatial semantic segments, and generate temporal semantic segments based on the signal change characteristics and the communication operation timing. Classify and combine the spatial semantic segments and temporal semantic segments according to the enemy unit to form behavioral semantic blocks.

[0125] 1053. Analyze the behavioral semantic blocks using a natural language understanding model, match terrain constraints and communication rules in the tactical rule base, generate terrain association chains and protocol association chains, merge the terrain association chains and protocol association chains, remove duplicate actions, retain key action sequences, and form a tactical logic chain.

[0126] 1054. Detect the time interval and direction change of adjacent actions in the tactical logic chain. If the time interval is less than a set threshold, merge them into a compound action and combine it with the maneuver mode analysis to generate an intelligence report.

[0127] In steps 1051-1054, spatial semantic segments refer to the natural language descriptions of enemy unit movement trajectories mapped to geographic coordinates. Temporal semantic segments refer to temporal logical descriptions generated based on signal changes. Tactical logic chains refer to key action sequences that integrate terrain constraints and communication rules. Communication operation timing refers to the precise correspondence between enemy unit signal characteristics and the timeline in battlefield communication. Behavioral semantic blocks refer to semantic units that fuse spatiotemporal and tactical characteristics. Protocol association chains refer to the mapping logic between communication rules and tactical actions. Terrain association chains refer to the causal logic between geographic constraints and maneuver behavior. Intelligence reports are visual decision support documents for the tactical logic chains.

[0128] In this embodiment, firstly, step 1051 extracts the interaction behavior records of enemy units from the decision tree and performs multi-dimensional matching with battlefield environment feedback data. For example, the movement trajectory data of armored battalion Y-307 is extracted, including a coordinate sequence of longitude from 118.5 degrees to 118.7 degrees and latitude from 32.8 degrees to 32.9 degrees. This data is overlaid and analyzed with a digital elevation topographic map to identify the path characteristics of the unit's maneuver to the eastern highlands: a cumulative elevation gain of 65 meters over 3 consecutive hours and an average slope of 12 degrees. Simultaneously, signal change characteristics within the corresponding time window are extracted, including a reduction in the communication band switching interval from the usual 15 minutes to 5 minutes, a signal strength range of 35 dB (peak 80 dB to trough 45 dB), and precise communication operation timing is obtained, such as starting the encryption protocol at 09:05:00 and switching to the 1.5 GHz band at 09:10:00.

[0129] Secondly, step 1052 spatially maps the movement trajectory to preset geographical coordinates to generate spatial semantic fragments. For example, at 09:00:00, armored battalion Y-307 is located at 118.5 degrees east longitude and 32.8 degrees north latitude, and at 09:30:00, it moves to 118.6 degrees east longitude and 32.85 degrees north latitude, forming a geographical description of "moving 3.2 kilometers northeast". At the same time, temporal semantic fragments are generated based on signal change characteristics: at 09:05:00, the signal strength suddenly increases by 20 decibels, triggering encryption; at 09:10:00, the frequency band switches to 1.5 GHz and is maintained for 5 minutes. These are categorized and combined according to enemy unit identifiers to form behavioral semantic blocks. The behavioral semantic block of armored battalion Y-307 includes the spatial semantics of "moving northeast towards the mountains" and the temporal semantics of "high-frequency encrypted communication".

[0130] Next, in step 1053, the natural language understanding model is used to parse the behavioral semantic blocks and match them with terrain constraints and communication rules in the tactical rule base. The terrain association chain generation logic is as follows: when the slope of the eastern highland path is greater than 10 degrees, it is necessary to detour through the valley, which maps the actual maneuver path of the Y-307 armored battalion to the detour rule with a matching degree of 87%. The protocol association chain is based on the temporal relationship between encryption initiation and frequency band switching, and is associated with the rule in the electronic countermeasures manual that "encryption should be initiated within 5 minutes after the assault order is issued". After merging the two association chains, duplicate actions are removed, such as removing two consecutive encryption checks, and the key action sequence is retained: 09:05 encryption initiation, 09:10 frequency band switching, and then 09:30 detour through the valley, forming a tactical logic chain.

[0131] Finally, step 1054 detects the time interval and directional changes of adjacent actions in the tactical logic chain. When the interval between encryption activation (09:05) and frequency band switching (09:10) is less than the set threshold of 10 minutes (5 minutes), they are merged into a "rapid communication mode switching" composite action. Combining the serpentine trajectory characteristics of the maneuver path (directional angle change rate ±12 degrees / minute), an intelligence report is generated that includes the tactical intent inferred as preparation for a high-altitude assault, the predicted action path as a breakthrough towards coordinates 118.7 / 32.9 via the valley, and a risk warning that the 1.5GHz band may be equipped with anti-radar detection modules.

[0132] Here is a specific example:

[0133] In a complex battlefield environment with multi-source heterogeneous data fusion, a situational awareness system detected abnormal movements of an enemy armored formation. By analyzing the dynamic branches of the combat decision tree, it first correlated satellite remote sensing data with electromagnetic spectrum monitoring logs, identifying the Y-307 armored battalion's serpentine maneuver path between 118.5° and 118.7° east longitude from 09:00 to 09:30. At this time, the digital elevation model showed that the unit was continuously climbing between 32.8 meters and 97.3 meters above sea level, with the average gradient of the path exceeding the 12-degree maneuver limit of conventional armored forces. Simultaneously, at 09:05:23, its communication system was intercepted to suddenly emit an 80-decibel high-intensity signal pulse, triggering the AES-256 encryption protocol and completing a rapid switch to the 1.5GHz tactical frequency band within 5 minutes and 37 seconds. Based on the spatial mapping module of the geofencing engine, the unit's 3.2-kilometer displacement trajectory is transformed into a spatiotemporal semantic fragment of "penetration into the northeast valley highlands", and coupled with the temporal features of "pulse-type encrypted communication window" to construct a composite behavioral semantic block containing 12-dimensional battlefield feature vectors.

[0134] Subsequently, the tactical inference engine parses the semantic units of the behavioral semantic blocks using the NLU model, activates the "steep slope detour to avoid fire" clause in the terrain constraint rule base, and deduces that 87% of the unit's maneuver trajectory conforms to the standard tactical profile of a mountain assault team. At the same time, it associates the "frequency band switching - encryption activation" timing constraint in the electronic warfare protocol base, identifying a strong correlation between the communication behavior and clause EC-217 of the NATO Electronic Countermeasures Manual. After multimodal feature fusion, the encryption protocol activation and frequency band switching are merged and marked as a "T+5 minute emergency communication reconstruction" composite event. Combined with the evasion feature of ±12 degrees per minute in the maneuver direction angle, a level 3 threat intelligence is finally generated: it is determined that the enemy will launch a three-dimensional surprise attack on the 118.7 / 32.9 coordinates during the T+60 minute window, and the probability of successful electronic warfare suppression is greatly increased. It is recommended to immediately activate the 1.5GHz band directional jamming array and deploy an unmanned reconnaissance cluster with anti-slope detection capability in the valley area.

[0135] The following is a complete example for steps 101-105, such as Figure 2 As shown:

[0136] During the air defense penetration mission codenamed "Iron Shield," the Red Force's electronic countermeasures battalion detected a sudden change in the Blue Force's radar signal parameters, and the system immediately initiated a multi-level analysis process. The system first retrieved historical deployment data from the Blue Force's 7th Armored Division, processing the electromagnetic spectrum records from last year's Snow Leopard exercise into 15-minute windows. Using an improved Fast-RCNN algorithm, the system identified the radio locations of 32 T-90A tanks, achieving a positioning accuracy of ±0.02 degrees north latitude (48.7 degrees). Simultaneously, L-band signal strength data from 1.4 to 1.6 GHz was extracted during this period, with a standard deviation of 18 dB. A dynamic correlation model between tank movement direction and communication frequency offset was established using a three-layer GRU neural network, where an azimuth rate of change of 2.3 degrees per minute corresponds to a communication frequency offset of 0.4 MHz per kilometer. After 72 hours of model training, the prediction error stabilized within the ±0.12 MHz range.

[0137] During the operation, a new frequency-hopping signal from the Blue Force's S-400 radar site was intercepted, detecting a regular characteristic with a pulse width of 2.3 microseconds ± 5%. A multiphase filter bank separated 17 protocol units, revealing a periodic energy surge every 5.8 seconds at the characteristic frequency of 593 MHz, with a peak power reaching 42 dBmW. Dynamic time warping was performed on this signal characteristic against the frequency band occupancy curves of the air defense alert phase in the behavior pattern database. After compensating for a 3.2% time base offset, the matching accuracy improved to 89.7%, confirming it as a characteristic signal of the 54K6E command vehicle's fire control link.

[0138] The system inputs the identified 12 air defense nodes into the strategy simulation framework, with division-level nodes weighted at 0.92 and battalion-level nodes at 0.65. When the X-band signal strength of air defense node No. 3 at 48.712 degrees North latitude increases sharply by 28 dB between 08:15 and 08:30, the hierarchical constraint algorithm is automatically activated, cutting off the lateral communication paths between battalion-level nodes and forcing the formation of a tree-like command topology. Using the Levenberg-Marquardt optimization algorithm, the node's influence radius is recalculated, and the coverage area of ​​node No. 3 expands from the baseline of 5.7 km to 7.2 km, revealing its key tactical role as the regional air defense command center.

[0139] When constructing a decision tree based on node interaction data, the system captured a 128-bit AES-CBC encrypted command sent by node 3 to seven end nodes at 08:22. A hardware-accelerated Meet-in-the-Middle attack cracked the azimuth parameter within 18 seconds, obtaining precise pointing data of 182 degrees ± 0.5 degrees. Combined with 0.3-meter resolution visible light images transmitted by the UAV, it was determined that the blue force was constructing a fan-shaped identification zone with a radius of 18 kilometers. The system generated a survival probability matrix for three penetration paths, with the low-altitude sea-skimming path increasing the penetration success rate from 68% to 83% at an altitude of 30 meters.

[0140] In the final stage, satellite infrared heat source density data and sonar buoy signal characteristics are integrated. When the active sonar pulse interval of the blue force destroyer is detected to be compressed from 15 seconds to 9 seconds, the semantic reconstruction engine automatically generates a tactical logic chain. The system analyzes the correlation between the enemy ship's speed increasing to 28 knots and the active sonar scanning cycle shortening by 40%, determining that the enemy has entered an anti-submarine warfare preparation state. The system predicts the battlefield situation for the next 30 minutes using a two-way LSTM model, outputs AES-256-GCM encrypted intelligence messages, marks high-probability encounters with submarine-launched missile interception warnings, and recommends a precise frequency hopping timing scheme of ±0.3 seconds every 17 seconds for the electronic countermeasures pod.

[0141] Figure 3 This application provides a schematic diagram of a specific implementation of a battlefield environment intelligence generation system based on natural language understanding, as illustrated in the embodiments of this application. Figure 3 The system may include:

[0142] The acquisition module 31 is used to extract enemy troop deployment information and communication characteristic parameters from the acquired historical combat data and reconnaissance data, and to construct an association model based on the multi-dimensional association elements between the enemy troop deployment information and the communication characteristic parameters, so as to generate a behavior pattern library through the association model;

[0143] The matching module 32 extracts feature parameters reflecting the enemy's communication protocol from the battlefield electromagnetic data collected based on the radio sensor network, and associates and matches the evolution law of communication features in the behavior pattern library with the feature parameters to form a communication feature mapping relationship.

[0144] The construction module 33 is used to generate associated data based on the troop deployment rules of the behavior pattern library and the mapping relationship of the communication features, and input the associated data into the strategy inference framework to construct a strategy space containing multi-level decision nodes of the enemy.

[0145] The calculation module 34 is used to calculate the interaction behavior of the enemy's multi-level decision nodes based on the topology of the strategy space, so as to generate a decision tree containing the path of combat intent association.

[0146] The conversion module 35 is used to semantically reconstruct the decision tree based on the acquired battlefield environment feedback data, and to convert the semantic reconstruction result into a tactical logic chain using a natural language understanding model, and to generate an intelligence report corresponding to the tactical logic chain.

[0147] The battlefield environment intelligence generation system based on natural language understanding in this application is used to implement the aforementioned battlefield environment intelligence generation method based on natural language understanding. Therefore, the specific implementation of the battlefield environment intelligence generation system based on natural language understanding can be found in the embodiment section of the battlefield environment intelligence generation method based on natural language understanding above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0148] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described methods for generating battlefield environment intelligence based on natural language understanding.

[0149] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described methods for generating battlefield environment intelligence based on natural language understanding.

[0150] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0151] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the method for generating battlefield environment intelligence based on natural language understanding.

[0152] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0153] The foregoing has provided a detailed description of a method, system, electronic device, and storage medium for generating battlefield environment intelligence based on natural language understanding, as provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for generating battlefield environment intelligence based on natural language understanding, characterized in that, The method comprises the following steps: extracting enemy force deployment information and communication characteristic parameters from acquired historical combat data and reconnaissance data, and constructing a correlation model according to multi-dimensional correlation elements between the enemy force deployment information and the communication characteristic parameters, to generate a behavior pattern library through the correlation model; extracting characteristic parameters reflecting enemy communication protocols from battlefield electromagnetic data collected based on a wireless sensor network, and correlating and matching the communication characteristic evolution law in the behavior pattern library with the characteristic parameters to form a communication characteristic mapping relationship; generating correlation data according to the force deployment law of the behavior pattern library and the communication characteristic mapping relationship, and inputting the correlation data into a strategy deduction framework to construct a strategy space containing enemy multi-level decision nodes; based on the topological structure of the strategy space, calculating the interactive behavior of the enemy multi-level decision nodes to generate a decision tree containing combat intent correlation paths; according to acquired battlefield environment feedback data, performing semantic reconstruction on the decision tree, and using a natural language understanding model to convert the semantic reconstruction result into a tactical logic chain, and generating an intelligence report corresponding to the tactical logic chain; wherein, according to the acquired battlefield environment feedback data, the semantic reconstruction is performed on the decision tree, and the natural language understanding model is used to convert the semantic reconstruction result into a tactical logic chain, and an intelligence report corresponding to the tactical logic chain is generated, comprising: extracting the interactive behavior records of enemy units from the decision tree, matching with the battlefield environment feedback data, acquiring the moving track, signal change characteristic and communication operation time sequence of the enemy units; comparing the moving track with the preset geographic coordinates to generate a space semantic segment, and generating a time sequence semantic segment according to the signal change characteristic and the communication operation time sequence, classifying and combining the space semantic segment and the time sequence semantic segment according to the enemy units to form a behavior semantic block; using a natural language understanding model to analyze the behavior semantic block, matching the terrain constraints and communication rules in a tactical rule library to generate a terrain association chain and a protocol association chain, merging the terrain association chain and the protocol association chain, removing repeated actions and retaining key action sequences to form a tactical logic chain; detecting the time interval and direction change of adjacent actions in the tactical logic chain, if the time interval is less than a set threshold, merging as a composite action, and generating an intelligence report in combination with a maneuver mode analysis.

2. The method of claim 1, wherein, extracting enemy force deployment information and communication characteristic parameters from acquired historical combat data and reconnaissance data, and constructing a correlation model according to multi-dimensional correlation elements between the enemy force deployment information and the communication characteristic parameters, to generate a behavior pattern library through the correlation model, comprising: dividing the records in the historical combat data and reconnaissance data into multiple data segments according to a preset time window, and extracting enemy units in the multiple data segments as enemy force deployment information; scanning the enemy communication signal spectrum characteristic recorded in the multiple data segments, extracting target parameters, and calculating the statistical distribution value of the target parameters in the preset time window to form communication characteristic parameters; The multi-dimensional association elements between the enemy force deployment information and the communication characteristic parameters are established, a dynamic connection algorithm is used to perform layer-by-layer mapping on the multi-dimensional association elements, mapping relationships are generated, and the weight parameters of the mapping relationships of each layer are iteratively adjusted to make the association strength of each layer reach a dynamic balance state; The mapping relationships in the dynamic balance state are combined into an association model, and the enemy unit movement mode, communication equipment operation rules, and signal strength change trend output by the association model are stored in time sequence as a behavior mode library.

3. The method of claim 1, wherein, According to the force deployment rules of the behavior mode library and the communication characteristic mapping relationship, association data is generated, and the association data is input into a strategy deduction framework to build a strategy space containing enemy multi-level decision nodes, including: The moving direction and stay duration of the enemy unit in a continuous time window are extracted from the behavior mode library to form a position sequence, the signal strength change value and frequency band switching interval of the enemy unit in the continuous time window are extracted from the communication characteristic mapping relationship to form a signal sequence, the position sequence and the signal sequence are paired to generate association data containing position and signal combination markers; According to the level identifier of the enemy unit, the decision level to which the enemy unit belongs is determined, and a level weight value is assigned to each level. The position and signal combination markers of each enemy unit in the association data are input into the strategy deduction framework according to the level weight value, so that the nodes of each decision level in the strategy deduction framework correspond to the enemy units in the range of the level weight value. In the strategy deduction framework, corresponding parameters are loaded for the nodes of each decision level, the influence coverage range of each decision level node is dynamically adjusted by comparing the parameters and the interaction times between each decision level node, and the cross-level nodes are constrained to be unidirectionally connected according to the level weight value, thereby forming a strategy space with hierarchical connection and coverage range change.

4. The method of claim 2, wherein, The multi-dimensional association elements between the enemy force deployment information and the communication characteristic parameters are established, a dynamic connection algorithm is used to perform layer-by-layer mapping on the multi-dimensional association elements, mapping relationships are generated, and the weight parameters of the mapping relationships of each layer are iteratively adjusted to make the association strength of each layer reach a dynamic balance state, including: The unit identifiers in the enemy force deployment information and the signal statistics in the communication characteristic parameters are paired in time sequence to form a unit and signal sequence; The spatial position change, signal strength change value, and unit type characteristics in the unit and signal sequence are extracted as multi-dimensional association elements; The multi-dimensional association elements are input into and processed in a multi-layer mapping structure of the dynamic connection algorithm. In the first layer of the multi-layer mapping structure, the spatial position change and the signal strength change value are combined to generate a preliminary mapping relationship. In the second layer of the multi-layer mapping structure, the preliminary mapping relationship is modified based on the unit type characteristics to generate an intermediate mapping relationship. In the third layer of the multi-layer mapping structure, the intermediate mapping relationship is optimized based on the historical behavior data of the enemy unit to generate a final mapping relationship; The weight parameters of each layer are initialized, the deviation between the predicted result of the last mapping relationship and the actual data is calculated, the weight parameters of each layer are adjusted reversely until the prediction deviation is less than a set threshold, and the mapping relationship of each layer reaches a dynamic balance state.

5. The method of claim 1, wherein, From the battlefield electromagnetic data collected based on the wireless sensor network, the characteristic parameters reflecting the enemy communication protocol are extracted, and the communication characteristic evolution law in the behavior mode library is matched with the characteristic parameters to form a communication characteristic mapping relationship, including: The battlefield electromagnetic data collected by the wireless sensor network is continuously scanned, and the fixed interval pulses repeatedly appearing in the signal waveform emitted by the enemy communication equipment are extracted as the protocol time reference. The signal segment between two adjacent protocol time references is divided into a protocol unit; In the protocol unit, the number of frequency jumps and the amplitude mutation points are detected to form a first feature group, the energy distribution of the frequency band is counted to form a second feature group, the first feature group is compared with the historical jump law in the behavior mode library, the amplitude mutation point deviation is compensated to generate a compensated protocol unit feature, the second feature group is matched with the historical frequency band usage record, and the combat stage label is matched as the stage label of the protocol unit; The protocol unit feature and the stage label are merged to constitute the characteristic parameters of the communication protocol, the characteristic parameters are superimposed with the communication signal intensity change curve in the behavior mode library under the same combat stage identifier, the signal intensity difference value of both on the same time scale is calculated, and when the difference value is less than a set tolerance, the mapping relationship between the characteristic parameters and the corresponding communication characteristic evolution law in the behavior mode library is established.

6. The method of claim 1, wherein, Based on the topology structure of the strategy space, the interaction behavior of the enemy multi-level decision nodes is calculated to generate a decision tree containing a combat intention associated path, including: According to the connection weight and coverage range of each level node in the strategy space, the command hierarchical relationship between the enemy units is determined, the last mapping relationship of the multi-layer mapping structure is classified according to the enemy unit identifier, the superposition result of the spatial position change and the signal intensity change value is calculated in combination with the command hierarchical relationship, and the unit movement mode is generated; Based on the command hierarchical relationship, the extreme points of the signal intensity change value are extracted, and the time interval and intensity difference of adjacent extreme points are counted to generate the operation law of the communication equipment reflecting the command level; According to the command hierarchical relationship and the angle change rate of the unit movement mode, the average intensity of the operation law of each level is calculated to generate a dynamic weight coefficient; The unit movement mode, operation law and dynamic weight coefficient are bound according to the enemy unit identifier and time stamp, the interaction behavior between nodes at each level is analyzed, the combat intention associated path reflecting the transmission of command intention is constructed, and according to the interaction frequency and intensity change of each node in the combat intention associated path, a decision tree containing the enemy command level and the combat intention associated path is generated.

7. A system for generating battlefield environment intelligence based on natural language understanding, for performing the method for generating battlefield environment intelligence based on natural language understanding according to any one of claims 1-6, characterized in that, Including: An acquisition module is configured to extract enemy force deployment information and communication characteristic parameters from acquired historical combat data and reconnaissance data, and construct an association model according to multi-dimensional association elements between the enemy force deployment information and the communication characteristic parameters, so as to generate a behavior pattern library through the association model; A matching module is configured to extract characteristic parameters reflecting enemy communication protocols from battlefield electromagnetic data collected based on a wireless sensor network, and associate and match communication characteristic evolution rules in the behavior pattern library with the characteristic parameters, so as to form a communication characteristic mapping relationship; A construction module is configured to generate associated data according to force deployment rules of the behavior pattern library and the communication characteristic mapping relationship, and input the associated data into a strategy deduction framework, so as to construct a strategy space containing enemy multi-level decision nodes; A calculation module is configured to calculate interaction behaviors of the enemy multi-level decision nodes based on a topological structure of the strategy space, so as to generate a decision tree containing combat intent association paths; A conversion module is configured to perform semantic reconstruction on the decision tree according to acquired battlefield environment feedback data, and convert a semantic reconstruction result into a tactical logic chain by using a natural language understanding model, and generate an intelligence report corresponding to the tactical logic chain.

8. An electronic device, comprising: Comprise: A memory is configured to store a computer program; A processor is configured to implement steps of the method for generating battlefield environment intelligence based on natural language understanding according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program can implement the method for generating battlefield environment intelligence based on natural language understanding according to any one of claims 1 to 6 when executed by the processor.

Images