A multi-modal large model-based combat micro-scene intelligent modeling method and system

Abstract

The application belongs to the technical field of intelligent scene modeling, and specifically discloses a combat micro-scene intelligent modeling method and system based on a multi-modal large model, which comprises the following steps: obtaining multi-modal original data, extracting a structured feature vector after preprocessing; constructing a multi-modal large model, outputting a tactical intention analysis result and a battle damage interpretation result; establishing a global parameter system and constraint rules; calling a model library to construct an initial combat micro-scene; inputting post-war data into the large model to obtain a battle damage interpretation result and a parameter adjustment instruction, and modifying the structure parameters and appearance parameters of the corresponding buildings in real time to generate an updated combat micro-scene. The system of the application is designed to realize the above method. The application constructs an initial scene based on pre-war data and dynamically updates the scene based on post-war data, so that the real-time collection or manual labeling of changed areas is not required, and the authenticity and timeliness of scene construction are fundamentally guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of intelligent scene modeling technology, and specifically relates to an intelligent modeling method and system for combat micro-scenes based on a multimodal large model. Background Technology

[0002] Micro-scenario modeling is a core foundation of military simulation and exercise, especially in urban combat scenarios, where its accuracy and efficiency directly impact tactical assessment, command decisions, and training effectiveness. Urban combat scenarios are characterized by dense buildings, complex terrain, and intricate entity interactions, creating an urgent need for detailed scenario depiction and rapid construction. Current micro-scenario modeling technologies primarily rely on manual modeling or traditional parametric modeling and scripting methods. Scenario construction is achieved by pre-setting entity models, manually configuring environmental parameters, and writing behavioral logic scripts. A typical application is the drag-and-drop entity deployment and logic flowchart editing method in zero-code scenario building platforms. However, these methods are time-consuming when dealing with urban combat scenarios and fail to meet the timeliness requirements of operations.

[0003] The existing technology has the following significant shortcomings: This traditional approach has significant shortcomings when dealing with urban warfare scenarios: First, it struggles to integrate information from multiple sources, such as images captured by drones and laser scanning data, resulting in incomplete scene detail. For example, it cannot accurately reproduce key information such as building structures and road layouts. Second, it lacks modeling efficiency and flexibility. Urban warfare scenarios contain a large number of detailed entities such as buildings, roads, and underground facilities. Traditional parametric modeling requires professional technicians to manually configure massive amounts of parameters, including building dimensions, materials, and battle damage status. Even with visual drag-and-drop tools, constructing a combat scenario in the core urban area still takes hours or even days. Furthermore, urban warfare involves significant changes in scenarios before and after battle, requiring recompilation and debugging after parameter adjustments. This makes it difficult to quickly adapt to the dynamic needs of pre-battle scenario construction and post-battle damage assessment, failing to meet the timeliness requirements for rapid battlefield decision-making. Third, it is difficult to accurately understand the commander's tactical intentions. For example, abstract tactical requirements such as conducting fire suppression based on a specific building are difficult to translate into precise scenario settings.

[0004] In recent years, multimodal large models have shown great potential in the field of military decision-making due to their powerful cross-modal fusion, semantic understanding and content generation capabilities. However, there is currently no mature solution for applying them to intelligent modeling of urban combat micro-scenes. This cannot solve the core pain points of the traditional modeling technology in the construction of urban combat scenarios, and it is difficult to meet the practical needs of rapid and accurate construction of urban combat scenarios and pre- and post-war damage assessment. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide an intelligent modeling method for combat micro-scenes based on a multimodal large model.

[0006] This invention provides an intelligent modeling method for combat micro-scenes based on a multimodal large model, comprising: Acquire multimodal raw data consisting of pre-war / post-war 3D scene data, geospatial data, equipment parameter data, tactical rule data, and environmental perception data; The multimodal raw data is preprocessed, and the preprocessed data is used to extract features to obtain structured feature vectors; A multimodal large model is constructed, which is used to output tactical intent analysis results and battle damage judgment results based on the input structured feature vector; Establish a full-domain parameter system for operational micro-scenarios, including static basic parameters, dynamic environmental parameters, tactical behavior parameters, and evaluation index parameters, and define the logical relationships and constraint rules between each parameter; Based on the scenario construction requirements, the corresponding structured feature vectors are input into the multimodal large model to obtain the tactical intent parsing results. Combined with the parameter system, the initial parameters for scenario construction that match the tactical intent are automatically generated. Based on the scenario, the initial parameters are called to construct the preset scenario basic model library to build the initial combat micro-scenario; Based on the structured feature vectors corresponding to the post-war multimodal raw data, the multimodal large model is input to obtain the battle damage assessment results and parameter adjustment instructions. According to the parameter adjustment instructions, the structural parameters and appearance parameters of the corresponding buildings in the initial combat micro-scenario are modified in real time to generate the updated combat micro-scenario.

[0007] A further approach is that the pre-war / post-war 3D scene data is acquired by a drone equipped with a visible light imaging device and a lidar scanning device. The pre-war / post-war 3D scene data includes visible light images and lidar point cloud data in both pre-war and post-war time series. The geospatial data includes digital elevation models (DEM) and satellite imagery. Tactical rules data includes doctrinal texts and historical battle examples; Environmental sensing data includes communication signals and meteorological data.

[0008] A further solution involves preprocessing the laser point cloud data, including: using a geometric attribute analysis method to classify the point cloud into ground points, building points, vegetation points, and artificial facility points based on the differences in point cloud attributes; extracting the boundary and structural features of each type of point cloud to form feature vectors that describe the geometric shape of the scene; and preprocessing the visible light image, including brightness and contrast adjustment, to improve the accuracy of subsequent image recognition.

[0009] A further proposed solution involves constructing the full-domain parameter system for the operational micro-scenario using a hierarchical modeling approach. Static basic parameters include terrain type, landform features, equipment type and quantity, defensive fortification configuration, and friendly operational module organization. Dynamic environmental parameters include wind speed, wind direction, visibility, precipitation intensity, electromagnetic interference intensity and frequency band, and light intensity. Tactical behavior parameters include operational module movement trajectory, fire strike timing and intensity, collaborative combat rules, and target identification priority. Evaluation index parameters include target damage probability, mission completion time, total resource consumption, and operational module survival probability. The parameter system also includes semantic models, data types, constraints, and logical relationships between parameters. Correlation analysis algorithms quantify the correlation strength between parameters, constraint satisfaction problem-solving algorithms verify the rationality of parameter combinations, and parameter sensitivity analysis algorithms identify key influencing parameters.

[0010] A further solution is that the scene basic model library contains pre-stored three-dimensional geometric models, including various types of buildings, roads, equipment, and environmental effects; The construction of the initial combat micro-scenario includes: according to the terrain features and building distribution specified by the initial parameters of the scenario construction, retrieving matching model instances from the scenario basic model library, and adjusting the geometric size, spatial position, material properties and initial state of the model instances according to the parameter values ​​to generate the initial combat micro-scenario.

[0011] A further solution is that the generation of the updated combat micro-scenario includes: Building-by-building analysis was performed on postwar visible light images to extract the image regions of each building; damage features in the images were identified through multimodal large models, including wall damage, roof collapse, and structural deformation. Based on the identified combination of damage characteristics and compared with the preset damage level standards, the battle damage level of each building is determined as minor damage, severe damage, or complete collapse; Based on the preset mapping rules, adjustment instructions are generated for the structural integrity parameters, appearance damage parameters, and protection capability parameters of the building. Differential analysis is performed on pre-war and post-war laser point cloud data to generate change area detection results; the change area detection results are compared and verified with the battle damage assessment results, and if there is a significant difference between the two, manual review is triggered.

[0012] A further solution includes building a knowledge base, which is used to store multimodal large model parameters, scene parameter sets, battle damage assessment results, and correction records.

[0013] A second aspect of the present invention provides an intelligent modeling system for combat micro-scenes based on a multimodal large model, comprising: The data acquisition module is used to acquire multimodal raw data consisting of pre-war / post-war 3D scene data, geospatial data, equipment parameter data, tactical rule data, and environmental perception data; The data processing module is used to preprocess the multimodal raw data and extract features from the preprocessed data to obtain structured feature vectors. The multimodal large model module includes a specially trained multimodal large model, which is used to output tactical intent analysis results and battle damage assessment results based on the input structured feature vector; the tactical intent analysis results are used to determine the initial parameters for scenario construction, and the battle damage assessment results include the location of building damage, damage type, battle damage level, and corresponding parameter adjustment instructions; The parameter management module is used to establish and maintain a full-domain parameter system for operational micro-scenarios, including static basic parameters, dynamic environmental parameters, tactical behavior parameters, and evaluation index parameters, and to store the logical relationships and constraint rules between the parameters. The scenario generation module is used to automatically generate initial parameters for scenario construction based on the tactical intent analysis results corresponding to the scenario construction requirements and the parameter system, and to call the preset scenario basic model library to construct the initial combat micro-scenario based on the initial parameters for scenario construction; and to modify the structural parameters and appearance parameters of the corresponding buildings in the initial combat micro-scenario in real time based on the battle damage judgment results and parameter adjustment instructions corresponding to the post-war multimodal raw data, so as to generate the updated combat micro-scenario.

[0014] A further embodiment is that the multimodal large model includes: The tactical intent parsing module is used to receive scenario construction requirements input by commanders in the form of voice, text description or hand-drawn tactical diagrams, and to parse the tactical intents in them through a multimodal large model, and output structured tactical intent parsing results. The automatic battle damage assessment module receives post-battle visible light images, analyzes buildings in the images building by building using a multimodal large model, identifies damage features, determines the battle damage level by comparing them with preset damage level standards, and generates adjustment instructions for building structural integrity parameters, appearance damage parameters, and protection capability parameters according to preset mapping rules.

[0015] A further embodiment of the system includes: The knowledge base is used to store multimodal large model parameters, historical scenario parameter sets, battle damage judgment cases and user correction records. It supports the retrieval of similar cases based on similar tactical requirements or scenario features in subsequent modeling tasks to reuse parameter configurations or model weights. The visualization and interaction module provides a parameter editing interface to support commanders in making real-time fine-tuning of the initial parameters of the generated scenario construction, and provides a display interface for the battle damage assessment results to support manual review and correction. The corrected results are fed back to the multimodal large model inference module and the scenario parameters are updated synchronously.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention utilizes a drone equipped with visible light imaging and lidar scanning equipment to conduct full-coverage data collection of the target combat area before and after combat, obtaining dual-time-series visible light image sequences and lidar point cloud data. The pre-combat 3D scene data comprehensively records the original geographical environment, building morphology, road network, and the original state of man-made facilities, forming a baseline template for scene construction. The post-combat 3D scene data captures real-time changes in the scene after combat operations, including key battle damage information such as physical damage to buildings, road blockages, and destruction of fortifications. This allows for the construction of an initial scene based on pre-combat data and dynamic updates based on post-combat data during the micro-scene modeling process, eliminating the need for manual re-collection or labeling of changed areas, fundamentally ensuring the authenticity and timeliness of scene construction.

[0017] The multimodal large-scale model constructed in this invention, after specialized training, possesses both tactical intent parsing and automatic battle damage assessment capabilities. The tactical intent parsing capability is used to parse tactical requirements input by commanders in the form of voice, text descriptions, or hand-drawn tactical maps into structured tactical intent parsing results, which are directly mapped to the initial parameters for scenario construction. The automatic battle damage assessment capability is used to analyze input post-war visible light imagery building by building, identify damage features, determine the battle damage level against preset damage level standards, and generate adjustment instructions for building structural integrity parameters, appearance damage parameters, and protection capability parameters according to preset mapping rules. During training, the model establishes a mapping relationship between damage features, battle damage levels, and modeling parameter adjustment rules. Furthermore, the model size is optimized and pruned to suit the specific characteristics of combat scenarios, ensuring its rapid operational capability on field deployment equipment.

[0018] This invention employs a hierarchical modeling method to construct a comprehensive parameter system for operational micro-scenarios, covering static basic parameters, dynamic environmental parameters, tactical behavior parameters, and evaluation index parameters. Static basic parameters include terrain type, landform features, equipment type and quantity, defensive fortification configuration, and friendly combat module organization. Dynamic environmental parameters include wind speed, wind direction, visibility, precipitation intensity, electromagnetic interference intensity and frequency band, and light intensity. Tactical behavior parameters include combat module movement trajectory, fire strike timing and intensity, collaborative combat rules, and target identification priority. Evaluation index parameters include target damage probability, mission completion time, total resource consumption, and combat module survival probability. The parameter system also includes semantic models, data types, constraints, and logical relationships between parameters. Correlation analysis algorithms quantify the correlation strength between parameters, constraint satisfaction problem-solving algorithms verify the rationality of parameter combinations, and parameter sensitivity analysis algorithms identify key influencing parameters to determine parameter optimization priorities. The hierarchical construction method ensures that each component of the urban combat scenario has a clear parameterized description. The logical constraints between parameters ensure the rationality of the generated parameter combinations at the level of physical laws and tactical rules, while parameter sensitivity analysis provides a quantitative basis for subsequent parameter optimization.

[0019] This invention, after assessing battle damage using a multimodal large-scale model on post-war visible light imagery, further performs differential analysis on pre- and post-war laser point cloud data to generate change area detection results. These change area detection results are then compared and verified with the battle damage assessment results. If significant differences exist, manual review is triggered. This comparative verification mechanism fully leverages the complementary advantages of the geometric precision of laser point cloud data and the rich texture of visible light imagery. Laser point cloud data can accurately quantify geometric damage features such as structural deformation, volume loss, and displacement changes, while visible light imagery excels at identifying textural damage features such as surface damage, burn marks, and color changes. By cross-validating the analysis results from two independent data sources, misjudgments or omissions caused by factors such as lighting conditions, occlusion, and sensor noise that may occur with a single data source are effectively avoided, significantly improving the reliability and objectivity of battle damage assessment. Attached Figure Description

[0020] The following figures are for illustrative purposes only and are not intended to limit the scope of the invention, wherein: Figure 1 : Schematic diagram of the method flow of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, design methods, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0022] Example 1 like Figure 1 As shown, this embodiment takes the construction of a combat scenario involving the attack and defense of a core urban street and the pre- and post-battle damage assessment as an example to provide a detailed explanation of the intelligent modeling method for micro-scenarios based on a multimodal large model proposed in this invention. The method includes: Multi-source operational data collection: Using drones equipped with visible light imaging and lidar scanning equipment, full-coverage flight paths were conducted over the core area before and after the battle to acquire dual-time-series visible light image sequences and lidar point cloud data, collectively forming pre- and post-battle 3D scene data. Geospatial data of the area was also collected, including a digital elevation model (DEM) and satellite imagery. Equipment parameter data was collected, including performance parameters of drones, armored vehicles, and infantry squads. Tactical rule data was collected, including urban offensive and defensive tactical doctrine texts and historical examples of similar street battles. Environmental perception data was also collected, including real-time communication signal strength and meteorological data such as wind speed, wind direction, visibility, and precipitation intensity provided by weather stations. All of the above data together constitute the multimodal raw data required for this implementation embodiment.

[0023] Data Preprocessing and Feature Extraction: Laser point cloud data is preprocessed using geometric attribute analysis. Based on echo intensity, normal vector variation, and spatial distribution density differences, the point clouds are classified into ground points, building points, vegetation points, and artificial facility points. Boundary and structural features of each type of point cloud are extracted to form geometric feature vectors describing the scene's geometry. Visible light images are preprocessed, including brightness and contrast adjustments, to enhance image quality and improve the accuracy of subsequent image recognition. Tactical rule text undergoes natural language preprocessing to extract key semantic information. A modality-specific feature extractor fuses the features from these data types to generate a unified structured feature vector. This structured feature vector contains multi-dimensional features, including geometric, textural, and semantic information of the scene, providing a standardized data foundation for subsequent input to multimodal large-scale models.

[0024] Multimodal Large-Scale Model Adaptation and Optimization: A multimodal large-scale model is constructed, requiring specialized training. The training process begins by collecting historical battle damage case data, building damage feature data, terrain data, equipment data, tactical rule data, and commander command corpora as training samples. The focus is on training the model's ability to identify building damage features, establishing a mapping relationship between "damage features—battle damage level—modeling parameter adjustment rules." For example, when the model identifies a combination of features such as "large-area wall cracking and partial collapse," it should classify the damage as "severe," and adjust the building model's "structural integrity parameters" and "appearance damage parameters" accordingly. Simultaneously, the model's ability to interpret tactical intentions is trained, enabling it to transform abstract tactical requirements such as "using a building for fire suppression" into structured scenario parameter adjustment intentions. During training, the model size is optimized and pruned, and training content is adjusted in a timely manner by monitoring training effectiveness to avoid recognition bias and ensure the model's ability to run rapidly on edge computing devices. The optimized multimodal large model is deployed in the multimodal large model module. Its core function is to output tactical intent analysis results or battle damage judgment results and parameter adjustment instructions based on the input structured feature vector.

[0025] Operational Scenario Parameter System Establishment: This embodiment employs a hierarchical modeling approach to construct a comprehensive parameter system. The first-level dimension of this system includes static basic parameters, dynamic environmental parameters, tactical behavior parameters, and evaluation index parameters. Static basic parameters are further subdivided into terrain type (e.g., densely built-up urban areas), landform features, equipment type and quantity (e.g., 3 reconnaissance drones, 2 armored vehicles, several infantry squads), defensive fortification configuration (e.g., building walls, roadblocks), and friendly operational module organization. Dynamic environmental parameters include wind speed (3 m / s), wind direction, visibility, precipitation intensity, electromagnetic interference intensity and frequency band, and light intensity. Tactical behavior parameters include operational module movement trajectories (drone grid reconnaissance routes, armored vehicle coordinated advance routes), fire strike timing and intensity (attack initiated after target confirmation), coordinated combat rules (1 drone for primary reconnaissance, 2 drones for coordinated attack), and target identification priority. Evaluation index parameters include target damage probability (not less than 80%), mission completion time (not exceeding 120 seconds), total resource consumption, and operational module survival probability (not less than 85%). Meanwhile, the parameter management module also defines the logical relationships and constraint rules between parameters, such as "the drone's flight altitude is not less than 50 meters and not higher than the building height + 10 meters" and "the drone's flight speed needs to be reduced when the wind speed exceeds 5 m / s". It also quantifies the correlation strength between parameters through correlation analysis algorithms, verifies the rationality of parameter combinations using constraint satisfaction problem-solving algorithms, and identifies key influencing parameters and determines the priority of parameter optimization through parameter sensitivity analysis algorithms.

[0026] Tactical Intent-Driven Parameter Generation: Commanders express their scenario construction needs via voice input through a visual interaction module: "Construct a core urban street attack and defense scenario, and complete pre- and post-battle damage comparison." This voice command is converted into a structured feature vector and input into a multimodal large model module. The tactical intent parsing module in the model parses out core tactical intents such as "urban attack and defense" and "damage comparison," outputting structured tactical intent parsing results. The parameter management module receives these results and, combined with the established parameter system and associated constraint rules, automatically generates an initial parameter set for scenario construction that matches the tactical intent. For example, it sets the scenario type to "core urban street," the deployment location of combat modules, and initial environmental parameters. Commanders can fine-tune the generated initial parameters in real time through the visual interaction module, such as adjusting the drone's flight altitude from 50 meters to 60 meters. The system automatically verifies that this adjustment conforms to the constraint rule of "avoiding building obstruction," generating an optimized parameter set. Based on the optimized parameter set, the scenario generation module retrieves matching model instances from a pre-set scenario basic model library. The model library contains pre-stored 3D geometric models including various types of buildings, roads, equipment, and environmental effects. The scene generation module adjusts the geometric size, spatial position, material properties and initial state of the model instance based on the parameter values, and quickly builds an initial combat micro-scene framework that includes terrain, buildings, roads, combat modules and environmental elements.

[0027] Urban combat scenario generation and visualization: After the combat operation, post-battle visible light images captured by UAVs are input into the system. The automatic damage assessment module in the multimodal large model module analyzes the image building by building, extracts the image area of ​​each building, and identifies damage features in the image through the model, including wall damage, roof collapse, and structural deformation. Based on the identified damage feature combinations and compared with the preset damage level standards, the battle damage level of each building is determined to be minor damage, severe damage, or complete collapse. For example, in this embodiment, building A1 is identified as having "large-area damage to the east wall (damage area approximately 40%) and partial roof collapse (collapse area approximately 30%)", and the battle damage level of this building is determined to be "severe damage". According to the preset mapping rules, parameter adjustment instructions for this building are automatically generated: the "structural integrity parameter" is modified from 100% to 60%, the "appearance damage parameter" is set to "40% damage to the east wall + 30% collapse of the roof", and the "protection capability parameter" is reduced from 80% to 30%. To ensure the accuracy of the interpretation results, the system further performs differential analysis on the pre- and post-battle laser point cloud data to generate change area detection results. These change area detection results are then compared and verified with the battle damage interpretation results, showing a high degree of consistency. After reconnaissance personnel review and confirm the interpretation results through the visual interaction module, the scene generation module modifies the structural and appearance parameters of the corresponding buildings in the initial combat micro-scene in real time according to parameter adjustment instructions, generating an updated battle damage scene. The multimodal large model parameters, scene parameter sets, battle damage interpretation results, and correction records generated in the above process are all stored in a knowledge base for rapid construction and reuse of similar scenes in the future.

[0028] Example 2 This embodiment discloses an intelligent modeling system for combat micro-scenes based on a multimodal large model that implements the method of Embodiment 1. The system includes a data acquisition module, a data processing module, a multimodal large model module, a parameter management module, a scene generation module, a knowledge base, and a visualization interaction module.

[0029] The data acquisition module includes a UAV equipped with visible light imaging equipment and lidar scanning equipment. It is used to conduct full-coverage data collection of the target combat area according to a preset flight path, both before and after the battle, obtaining visible light image sequences and lidar point cloud data in both pre- and post-battle time series. In this embodiment, the data acquisition module also integrates or can connect to other data collection devices to simultaneously acquire geospatial data (such as DEM, satellite imagery), equipment parameter data, tactical rule data (doctrine texts, historical battle examples), and environmental perception data (communication signals, meteorological data). All multimodal raw data collected by the data acquisition module is transmitted to the data processing module via a network.

[0030] The data processing module communicates with the data acquisition module to receive and process multimodal raw data. The data processing module includes a data cleaning unit, a data standardization unit, and a feature extraction unit. The data cleaning unit removes noise and invalid information from various types of raw data; the data standardization unit unifies the format and coordinate system of data from different sources; and the feature extraction unit classifies and extracts structural features from laser point cloud data using geometric attribute analysis methods, enhances visible light images, and converts the processed data into structured feature vectors using a modality-specific feature extractor. The processed structured feature vectors are then transmitted to the multimodal large model module and a knowledge base for storage.

[0031] The multimodal large model module communicates with the data processing module, parameter management module, and scene generation module. In this embodiment, the multimodal large model module is a specially trained and optimized multimodal large model, which internally includes a tactical intent parsing module and a battle damage automatic assessment module. The tactical intent parsing module receives tactical requirements (voice, text, hand-drawn diagrams) from the visualization interaction module, converts them into structured feature vectors, and then parses the tactical intents within them through the model, outputting a structured tactical intent parsing result. This result is sent to the parameter management module to determine the initial parameters for scene construction. The battle damage automatic assessment module receives structured feature vectors corresponding to wartime or post-war visible light images from the data processing module, analyzes the buildings in the images building by building through the model, identifies damage features, and determines the battle damage level by comparing them with preset damage level standards. Finally, it generates parameter adjustment instructions for buildings according to preset mapping rules, and these instructions are sent to the scene generation module for scene updates.

[0032] The parameter management module communicates with the multimodal large model module, the scenario generation module, and the visualization interaction module. The parameter management module is a parameter database used to establish and maintain a comprehensive parameter system covering static basic parameters, dynamic environmental parameters, tactical behavior parameters, and evaluation index parameters for operational micro-scenarios. This module stores the logical relationships and constraint rules between parameters and incorporates algorithms for correlation analysis, constraint satisfaction problem solving, and parameter sensitivity analysis to quantify parameter relationships, verify the rationality of parameter combinations, and identify key influencing parameters. When it receives tactical intent analysis results from the multimodal large model module, the parameter management module automatically generates initial scenario construction parameters that meet the constraints, based on the parameter system, and sends them to the scenario generation module. Simultaneously, it can also receive parameter fine-tuning instructions from the visualization interaction module, perform rationality checks, and update the parameter set.

[0033] The scene generation module communicates with the data processing module, the multimodal large model module, the parameter management module, and the visualization interaction module. The scene generation module has a built-in scene basic model library, pre-stored with 3D geometric models and their material and animation attributes, including various types of buildings, roads, equipment, and environmental effects. In the initial scene construction phase, this module retrieves matching model instances from the model library based on the initial scene construction parameters received from the parameter management module, and adjusts the geometric dimensions, spatial position, material attributes, and initial state of the model instances according to the parameter values ​​to generate the initial combat micro-scene. In the battle damage scene update phase, this module modifies the structural and appearance parameters of corresponding buildings in the scene in real time based on the battle damage assessment results and parameter adjustment instructions received from the multimodal large model module, generating the updated combat micro-scene. The generated scene data is sent to the visualization interaction module for rendering and is simultaneously stored in the knowledge base.

[0034] The knowledge base module is connected to the data processing module, multimodal large model module, parameter management module, and scene generation module, respectively. It centrally stores various types of data generated during system operation, including multimodal large model parameters, historical scene parameter sets, battle damage assessment cases, and user correction records. This module supports semantic-based rapid retrieval. In subsequent modeling tasks, when similar tactical requirements or scene features are detected, similar cases can be quickly retrieved to reuse parameter configurations or model weights, thereby accelerating the construction process of new scenes.

[0035] The visualization and interaction module serves as the system's human-computer interface, communicating bidirectionally with the multimodal large model module, parameter management module, and scene generation module. This module provides a parameter editing interface, allowing commanders to input scene construction requirements via voice, text, or graphics, and to fine-tune the initial parameters of the generated scene in real time. Simultaneously, this module provides a display interface for battle damage assessment results, visually presenting battle damage distribution through heatmaps, dynamic effects, and other methods, and supporting reconnaissance personnel to manually review and correct the automatically generated assessment results. The corrected results are fed back to the multimodal large model module and parameter management module for synchronously updating scene parameters, and are ultimately stored in the knowledge base.

[0036] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for intelligent modeling of operational micro-scenes based on a multimodal large model, characterized in that, include: Acquire multimodal raw data consisting of pre-war / post-war 3D scene data, geospatial data, equipment parameter data, tactical rule data, and environmental perception data; The multimodal raw data is preprocessed, and the preprocessed data is used to extract features to obtain structured feature vectors; A multimodal large model is constructed, which is used to output tactical intent analysis results and battle damage judgment results based on the input structured feature vector; Establish a full-domain parameter system for operational micro-scenarios, including static basic parameters, dynamic environmental parameters, tactical behavior parameters, and evaluation index parameters, and define the logical relationships and constraint rules between each parameter; Based on the scenario construction requirements, the corresponding structured feature vectors are input into the multimodal large model to obtain the tactical intent parsing results. Combined with the parameter system, the initial parameters for scenario construction that match the tactical intent are automatically generated. Based on the scenario, the initial parameters are called to construct the preset scenario basic model library to build the initial combat micro-scenario; Based on the structured feature vectors corresponding to the post-war multimodal raw data, the multimodal large model is input to obtain the battle damage assessment results and parameter adjustment instructions. According to the parameter adjustment instructions, the structural parameters and appearance parameters of the corresponding buildings in the initial combat micro-scenario are modified in real time to generate the updated combat micro-scenario.

2. The intelligent modeling method for combat micro-scenes based on a multimodal large model according to claim 1, characterized in that, The pre-war / post-war 3D scene data was acquired by a drone equipped with a visible light imaging device and a lidar scanning device. The pre-war / post-war 3D scene data includes visible light images and lidar point cloud data in both pre-war and post-war time series. The geospatial data includes digital elevation models (DEM) and satellite imagery. Tactical rules data includes doctrinal texts and historical battle examples; Environmental sensing data includes communication signals and meteorological data.

3. The intelligent modeling method for combat micro-scenes based on a multimodal large model according to claim 2, characterized in that, The preprocessing of the laser point cloud data further includes: using a geometric attribute analysis method to classify the point cloud into ground points, building points, vegetation points, and artificial facility points based on the attribute differences of the point cloud; extracting the boundary features and structural features of each type of point cloud to form a feature vector for describing the geometric shape of the scene; and preprocessing the visible light image including brightness adjustment and contrast adjustment to improve the accuracy of subsequent image recognition.

4. The intelligent modeling method for combat micro-scenes based on a multimodal large model according to claim 3, characterized in that, The operational micro-scenario global parameter system is constructed using a hierarchical modeling method. Static basic parameters include terrain type, landform features, equipment type and quantity, defensive fortification configuration, and friendly operational module organization. Dynamic environmental parameters include wind speed, wind direction, visibility, precipitation intensity, electromagnetic interference intensity and frequency band, and light intensity. Tactical behavior parameters include operational module movement trajectory, fire strike timing and intensity, collaborative combat rules, and target identification priority. Evaluation index parameters include target damage probability, mission completion time, total resource consumption, and operational module survival probability. The parameter system also includes semantic models, data types, constraints, and logical relationships between parameters. Correlation analysis algorithms quantify the correlation strength between parameters, constraint satisfaction problem-solving algorithms verify the rationality of parameter combinations, and parameter sensitivity analysis algorithms identify key influencing parameters.

5. The intelligent modeling method for combat micro-scenes based on a multimodal large model according to claim 4, characterized in that, The scene basic model library contains pre-stored three-dimensional geometric models, including various types of buildings, roads, equipment, and environmental effects; The construction of the initial combat micro-scenario includes: according to the terrain features and building distribution specified by the initial parameters of the scenario construction, retrieving matching model instances from the scenario basic model library, and adjusting the geometric size, spatial position, material properties and initial state of the model instances according to the parameter values ​​to generate the initial combat micro-scenario.

6. The intelligent modeling method for combat micro-scenes based on a multimodal large model according to claim 5, characterized in that, The generated updated combat micro-scenarios include: Building-by-building analysis was performed on postwar visible light images to extract the image regions of each building; damage features in the images were identified through multimodal large models, including wall damage, roof collapse, and structural deformation. Based on the identified combination of damage characteristics and compared with the preset damage level standards, the battle damage level of each building is determined as minor damage, severe damage, or complete collapse; Based on the preset mapping rules, adjustment instructions are generated for the structural integrity parameters, appearance damage parameters, and protection capability parameters of the building. Differential analysis is performed on pre-war and post-war laser point cloud data to generate change area detection results; the change area detection results are compared and verified with the battle damage assessment results, and if there is a significant difference between the two, manual review is triggered.

7. The intelligent modeling method for combat micro-scenes based on a multimodal large model according to claim 6, characterized in that, It also includes building a knowledge base, which is used to store multimodal large model parameters, scene parameter sets, battle damage judgment results and correction records.

8. An intelligent modeling system for combat micro-scenes based on a multimodal large model, characterized in that, include: The data acquisition module is used to acquire multimodal raw data consisting of pre-war / post-war 3D scene data, geospatial data, equipment parameter data, tactical rule data, and environmental perception data; The data processing module is used to preprocess the multimodal raw data and extract features from the preprocessed data to obtain structured feature vectors. The multimodal large model module includes a specially trained multimodal large model, which is used to output tactical intent analysis results and battle damage assessment results based on the input structured feature vector; the tactical intent analysis results are used to determine the initial parameters for scenario construction, and the battle damage assessment results include the location of building damage, damage type, battle damage level, and corresponding parameter adjustment instructions; The parameter management module is used to establish and maintain a full-domain parameter system for operational micro-scenarios, including static basic parameters, dynamic environmental parameters, tactical behavior parameters, and evaluation index parameters, and to store the logical relationships and constraint rules between the parameters. The scenario generation module is used to automatically generate initial parameters for scenario construction based on the tactical intent analysis results corresponding to the scenario construction requirements and the parameter system, and to call the preset scenario basic model library to construct the initial combat micro-scenario based on the initial parameters for scenario construction. Based on the battle damage assessment results and parameter adjustment instructions corresponding to the post-war multimodal raw data, the structural and appearance parameters of the corresponding buildings in the initial combat micro-scenario are modified in real time to generate the updated combat micro-scenario.

9. The intelligent modeling system for combat micro-scenes based on a multimodal large model according to claim 8, characterized in that, The multimodal large model includes: The tactical intent parsing module is used to receive scenario construction requirements input by commanders in the form of voice, text description or hand-drawn tactical diagrams, and to parse the tactical intents in them through a multimodal large model, and output structured tactical intent parsing results. The automatic battle damage assessment module receives post-battle visible light images, analyzes buildings in the images building by building using a multimodal large model, identifies damage features, determines the battle damage level by comparing them with preset damage level standards, and generates adjustment instructions for building structural integrity parameters, appearance damage parameters, and protection capability parameters according to preset mapping rules.

10. The intelligent modeling system for combat micro-scenes based on a multimodal large model according to claim 9, characterized in that, The system also includes: The knowledge base is used to store multimodal large model parameters, historical scenario parameter sets, battle damage judgment cases and user correction records. It supports the retrieval of similar cases based on similar tactical requirements or scenario features in subsequent modeling tasks to reuse parameter configurations or model weights. The visualization and interaction module provides a parameter editing interface to support commanders in making real-time fine-tuning of the initial parameters of the generated scenario construction, and provides a display interface for the battle damage assessment results to support manual review and correction. The corrected results are fed back to the multimodal large model inference module and the scenario parameters are updated synchronously.

Images