A method for constructing and acquiring data of an infrared simulation scene.
By parsing scene tasks using a large language model to generate control commands, the Unreal Engine is driven to construct a 3D infrared scene and perform thermophysical calculations. This solves the problems of slow construction speed and deviation in infrared simulation scene construction in existing technologies, and realizes fast and efficient construction of multi-parameter infrared simulation scene and data acquisition, thereby improving the detection accuracy of intelligent models.
Patent Information
- Application Number
- CN202611247816.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-18
- Publication Date
- 2026-11-10
AI Technical Summary
Existing infrared simulation scene construction methods suffer from slow speed and susceptibility to errors when adapting to the generation of multiple types and dimensions of scenes.
The system utilizes a large language model to perform semantic parsing of the natural language description of the scene task, generates control commands that conform to the large model context protocol, drives the Unreal Engine to construct a 3D dynamic infrared scene, and calculates the surface temperature and radiance of the object by combining thermophysical parameters. Infrared data is acquired from multiple perspectives and in multiple environments using AirSim.
It enables the rapid batch construction of differentiated, multi-parameter infrared simulation scenes, improving the efficiency of automated scene construction and the realism of simulation scenes, which helps to improve the detection accuracy and generalization performance of intelligent models.
Smart Images

Figure CN123069672A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of infrared imaging simulation technology, specifically to the construction of infrared simulation scenarios and data acquisition methods. Background Technology
[0002] In the perception and recognition applications of intelligent aircraft, intelligent vehicles, robots, robot dogs and other intelligent agents, it is usually necessary to simulate a large amount of visible light and infrared image data covering multiple environments, multiple perspectives and multiple weather conditions for the training and testing of target detection, target tracking, semantic segmentation and multimodal fusion recognition models.
[0003] Infrared scene acquisition requires the pre-establishment of a simulation scene, within which data is acquired. The speed and realism of infrared scene acquisition are highly dependent on the constructed simulation scene. Current infrared scene construction typically involves first semantically parsing the scene requirements in natural language form to extract a list of scene elements and environmental parameter data. Then, the extracted text data is mapped item by item to the corresponding functional interfaces of the 3D simulation platform, driving the scene construction process. However, large language models belong to the field of natural language processing, with output primarily consisting of natural language semantics and unstructured text logic. 3D simulation engines, on the other hand, belong to the field of computer graphics and modeling simulation, with functional interfaces primarily based on the engine's native parameterized instructions and node-based calling rules. These two are independent of each other, therefore, they can typically only adapt to parameter mappings for fixed scene types and fixed parameter dimensions. If the scene requirements contain complex semantics that cannot be directly broken down into individual interface parameters, semantic information loss and correspondence deviations are likely to occur. This makes it difficult to adapt to the batch generation requirements of multi-type, multi-dimensional infrared simulation scenes. Summary of the Invention
[0004] In view of the shortcomings of the prior art, the purpose of this application is to provide a method for constructing infrared simulation scenes and acquiring data, which solves the problems of slow modeling speed and easy deviation in the infrared simulation of the prior art.
[0005] Firstly, the objective of this invention can be achieved through the following technical solution: a method for constructing and acquiring data of an infrared simulation scene, specifically including the following steps: We use a large language model to perform semantic parsing on the natural language description of scene tasks, and extract scene entities, spatial relationships and constraints. Based on the parsing results, a data structure conforming to the Large Model Context Protocol (MCP) is generated, along with corresponding control commands. The control commands include at least one of scene construction commands, object control commands, material configuration commands, environmental state control commands, and acquisition task commands. The system sends commands to the Unreal Engine to obtain a 3D dynamic infrared scene. A complete infrared simulation scene is obtained by fusing a three-dimensional dynamic infrared scene with a three-dimensional infrared model of the target object; Within the complete infrared simulation scenario, infrared data acquisition from multiple perspectives and environmental parameters is completed using AirSim.
[0006] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the generation process of the Unreal Engine modeling instructions is as follows: the control command drives the Unreal Engine to automatically call various scene resources and complete the scene element deployment according to the scene object deployment priority; wherein the scene object deployment priority scoring formula is: In the formula, Assign priority scores to scene objects. As the semantic importance weight of the object, For task relevance weights, The distance weight between the object and the target region is used. , , Let be the weight coefficient, and satisfy... .
[0007] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the construction method of the three-dimensional dynamic infrared scene is: issuing instructions to the Unreal Engine through control commands to obtain a three-dimensional visible light simulation scene containing terrain, buildings, vegetation, road scene elements, and corresponding material properties; Based on the material properties of each object in the 3D visible light simulation scene, and combined with the environmental thermophysical parameters, the surface temperature of each object and the infrared radiation brightness of the scene are calculated in sequence to generate a 3D dynamic infrared scene with a continuous 3D infrared radiation field. The construction of the 3D visible light simulation scene involves establishing basic terrain in Unreal Engine; loading models of buildings, vegetation, roads, water bodies, obstacles, and background interference objects based on the scene semantic description; adjusting the spatial position, scale, material texture binding, and lighting configuration of the loaded scene objects; and configuring roughness, metallicity, diffuse texture, and normal texture parameters for different scene objects to complete the construction of the 3D visible light simulation scene.
[0008] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the formula for calculating the surface temperature of each object is: In the formula, For the surface temperature of the scene objects, The ambient reference temperature This represents the temperature rise component caused by solar radiation. The temperature change component caused by heat conduction with surrounding objects. This represents the temperature change component caused by convective heat transfer.
[0009] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the formula for calculating the infrared radiance of the scene is: In the formula, The infrared radiation brightness of the scene object. For the surface emissivity of the object, For the surface temperature of the object Planck radiation function under the given conditions For the surface reflectivity of the object, This represents the background radiation component.
[0010] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the construction of the complete infrared simulation scene is as follows: acquiring the three-dimensional geometric model of the target object and the corresponding infrared radiation model; automatically placing the target object in a preset position or candidate area in the three-dimensional dynamic infrared scene according to the spatial constraints between the target object and the scene, wherein the spatial constraints include at least one of constraints based on traversable areas, constraints based on visible areas, and constraints based on the distribution of task targets; adjusting the pose, height, orientation, and occlusion relationship of the target object to obtain the complete infrared simulation scene of the fused target.
[0011] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the Airsim multi-view, multi-environment parameter infrared data acquisition, preferably performed by a UAV, specifically: configuring the aircraft motion path, camera attitude, sensor parameters, and dynamic environmental disturbance parameters, and performing multi-view data acquisition in the fused complete infrared simulation scene; the aircraft motion path is set to at least one of a circling flight path and a straight flight path, the circling flight path is used to perform multi-altitude, multi-angle data acquisition around the target, and the straight flight path is used to perform through-sampling along the front or side of the target; during flight, the weather state, visibility, haze intensity, background interference position, and local temperature parameters are dynamically adjusted according to preset time intervals or waypoint intervals; the camera orientation, focal length, imaging mode, and sampling frequency are controlled using Airsim, and the sensor noise model and imaging post-processing parameters are configured to simulate the imaging effect of the actual infrared sensor.
[0012] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: during the multi-view data acquisition process, judging the target visibility effect through camera target visibility scoring; when the score is lower than a preset threshold, automatically adjusting the camera orientation or aircraft attitude to ensure that the target area appears within the effective field of view of the image; wherein the camera target visibility scoring formula is: In the formula, Rate the camera's target visibility. The score is given based on the area of the target in the image. The score is awarded for the target center being located in the center region of the effective field of view. Score the clarity of the target. , , This is the weighting coefficient; the preset threshold value ranges from 0.55 to 0.85.
[0013] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: performing synchronous output during multi-view data acquisition: synchronously saving RGB visible light image, infrared grayscale image and target semantic segmentation map for each sample; synchronously recording camera pose, flight trajectory, environmental parameters, temperature parameters and target annotation information corresponding to the image; and integrating the image and the accompanying parameter information to output a multimodal dataset for training and testing.
[0014] Secondly, in order to achieve the above objectives, the present invention discloses an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor performs the method of the first aspect.
[0015] The beneficial effects of this application are: This invention can uniformly encapsulate the scene entities, spatial relationships, and constraints extracted after parsing scene requirements using a large language model into standardized structured data, and simultaneously generate corresponding classification control instructions to achieve direct transmission of complete information. It can synchronize the real-time status of scene construction and dynamically iterate and adjust the generated instructions. The parameter mapping logic has universal adaptability and can be compatible with various scene types and multi-dimensional environments and target parameters. It completely retains all original semantic information and can quickly build differentiated, multi-parameter infrared simulation scenes in batches and complete automatic data collection, improving the efficiency of automated scene construction and the realism of simulation scenes, which helps to improve the detection accuracy and generalization performance of intelligent models. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in 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, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a schematic diagram of the overall process of an embodiment of this application; Figure 2 This is a schematic diagram illustrating the use of the Large Model Pyramid (MCP) to control the Unreal Engine to generate scenes in an embodiment of this application. Figure 3 These are schematic diagrams of two data acquisition routes (circumferential / straight line) according to embodiments of this application; Figure 4 These are schematic diagrams illustrating simulation results under different environmental parameters in embodiments of this application; Figure 5 This is a schematic diagram of the output of visible light image, infrared image and target segmentation map according to an embodiment of this application. Detailed Implementation
[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0019] Example 1: Reference Figures 1-5 As shown, this invention proposes an embodiment applied to Unreal Engine infrared scene simulation equipped with the Airsim data acquisition plugin, used to generate a multimodal dataset required for training a UAV target recognition model. The specific implementation process is as follows: The system acquires four categories of full input parameters: scenario task, target object, environmental thermophysics, and flight simulation. The specific parameter values are as follows: Scene task parameters: The natural language input task description is "Generate an island and sea scene, including island and reef topography, rocky reef coastline, island vegetation, simple island and reef buildings, and a sunny summer midday marine environment. Complete multi-view infrared acquisition with conventional submarines as the main target and other ships as interference targets." Extract core information such as scene type, terrain features, building layout, vegetation distribution, coastline structure, weather conditions, target type, and acquisition task requirements.
[0020] Target parameters: The target type is a conventionally powered submarine, with geometric dimensions of 76m in length, 8m in width, and 10m in height above the waterline. The material type is a metal hull with a non-metallic anechoic tile coating. The number of targets is 2, and the initial location is preset to the near-shore navigation area around the island.
[0021] Environmental thermophysical parameters: Weather type: sunny; Visibility: 15 km; Relative humidity: 75%; Wind speed: 5 m / s; Solar altitude angle: 65°; Day / night time: noon; Sea surface background temperature: 26°C. =29℃, atmospheric transmittance 0.82, ambient reference temperature 29 =29℃, solar irradiance 1000W / The initial surface temperature of the sea was 27°C. =29℃.
[0022] Flight simulation configuration parameters: flight speed 8m / s, flight altitude coverage from 50m to 200m, number of waypoints 16, sampling interval 2s, camera field of view 60°.
[0023] The system utilizes a large language model to perform semantic parsing on the natural language description of the scene task, extracting scene entities, spatial relationships, and constraints. Based on the parsing results, it generates data structures and control commands conforming to the Large Model Context Protocol (MCP). These control commands include five categories: scene construction commands, object control commands, material configuration commands, environment state control commands, and data acquisition task commands. These control commands drive Unreal Engine to automatically invoke terrain, building, vegetation, sea surface water, and lighting resources, completing the scene element deployment according to the scene object deployment priority. The scene object deployment priority scoring formula is as follows: In the formula, Assign priority scores to scene objects. As the semantic importance weight of the object, For task relevance weights, The distance between the object and the target region is used as the weight; in this embodiment, the weight coefficient is taken as... , , ,satisfy Based on the scoring ranking, elements with high mission relevance, such as island and reef coastlines and sea areas, are deployed first, followed by distant background vegetation and peripheral obstacles.
[0024] In Unreal Engine, a bedrock island terrain and surrounding open sea surface are generated as the base scene. Based on the scene semantic description, models such as simple island and reef buildings, coastal shrubs, rocky reef coastlines, sea surface water, and distant islands in the background are loaded. Spatial position calibration, scale normalization adjustment, material texture binding, and noon parallel lighting configuration are performed on all loaded scene objects. Material parameters are configured for different scene objects: roughness 0.1 and metallicity 0.0 for sea surface water, roughness 0.7 and metallicity 0.05 for concrete island and reef buildings, and roughness 0.9 and metallicity 0.02 for rocky reef surfaces. Diffuse texture and normal texture are matched at the same time to finally generate a complete 3D visible light simulation scene.
[0025] Based on the material properties of each object in the 3D visible light simulation scene, the surface temperature of each object is calculated using a thermophysical calculation model. Then, the infrared radiation brightness of the scene is calculated to generate a continuous 3D infrared radiation field, thus constructing a 3D dynamic infrared scene. The surface temperature of each object is obtained by superimposing three types of temperature components onto the ambient reference temperature. The calculation formula is as follows: In the formula, For the surface temperature of the scene objects, The ambient reference temperature; The temperature rise component caused by solar radiation is calculated from solar irradiance, material absorptivity, and sunshine duration. The temperature change component caused by heat conduction with surrounding objects is calculated based on the contact area and thermal conductivity of adjacent objects. The temperature change component caused by convective heat transfer is calculated from wind speed, ambient temperature difference, and convective heat transfer coefficient. In this embodiment, the surface temperature of the sea surface after solar radiation at noon is approximately 29°C. =29℃, the surface temperature of concrete island and reef building walls is approximately 37℃. =29℃, the surface temperature of the rocky reef is approximately 39℃. =29℃.
[0026] Based on the surface temperature of each object, and combining the material emissivity, Planck radiation function, surface reflectivity, and background radiation components, the infrared radiance of the scene is calculated using the following formula: In the formula, The infrared radiation brightness of the scene object. For the surface emissivity of the object, For the surface temperature of the object Planck radiation function under the given conditions For the surface reflectivity of the object, This represents the background radiation component.
[0027] The emissivity of the material is set in different ranges according to its type: the emissivity of sea surface water is 0.96 (between 0.75 and 0.98), the emissivity of concrete island and reef building surface is 0.85 (between 0.60 and 0.95), and the emissivity of conventional submarine metal hull surface is 0.30 (between 0.20 and 0.75).
[0028] Table 1 shows some dialogue examples generated by Unreal Engine using the Large Model Pyramid (MCP) to control the scene: like Figure 2 As shown, this is generated through the dialogue in Table 1.
[0029] By traversing the surface mesh of all objects in the entire scene, calculating the infrared radiation brightness mesh by mesh, generating a continuous three-dimensional infrared radiation field, and finally constructing a three-dimensional dynamic infrared scene.
[0030] Obtain the 3D geometric model and corresponding infrared radiation model of the target conventional-powered submarine and the jamming target, and embed the target into a 3D dynamic infrared scene by combining three types of spatial constraints: Navigable area constraints: Conventional-powered submarines may only be placed in navigable waters such as open sea areas and waterways around islands, and are prohibited from being placed in reef or shoal areas; Visibility area constraints: The placement of the target must ensure that the main observation direction is not completely obstructed and retain at least 120° of visibility; Mission target distribution constraints: The two targets are respectively deployed in the near-shore waterway on the east side of the island and in the open sea area on the south side, which is consistent with the distribution pattern of submarine activities in the waters surrounding the island.
[0031] After the target was placed, the navigation attitude, draft, and bow orientation of the conventional submarine were adjusted. The spatial obstruction relationship between the conventional submarine and the sea surface and surrounding islands and reefs was corrected to match the target's infrared radiation characteristics with the ambient temperature of the scene. This completed the target-scene fusion simulation, resulting in a complete infrared simulation scene of the fused target. (See...) Figure 2 .
[0032] Configuring aircraft parameter acquisition based on Airsim plugin: Flight path: Both an orbital flight path and a straight flight path are configured. The orbital path, centered on the target, performs 360° surround data acquisition at altitudes of 50m, 100m, and 200m to generate multi-altitude, multi-angle samples. See [link / details]. Figure 3 A straight path is set up on the side of the channel on the east side of the coastal island to complete continuous sampling and generate motion view samples.
[0033] Dynamic environmental disturbances: Parameters are dynamically adjusted every 10 seconds during flight, including visibility fluctuations, sea fog intensity changes, movement of background interfering vessels, and local temperature ±2°C. =29℃ perturbation, expanding the environmental diversity of the sample, see Figure 4 .
[0034] Sensor configuration: Using Airsim to control camera orientation, focal length, imaging mode and 2Hz sampling frequency, configure the Gaussian noise model of infrared sensor, non-uniformity correction residual and grayscale mapping post-processing parameters to simulate the imaging effect of actual long-wave infrared sensor.
[0035] During the data acquisition process, the camera target visibility score is calculated in real time to assess the target visibility effect. When the score falls below a preset threshold of 0.7, the camera's pitch and yaw angles are automatically adjusted to correct the aircraft's attitude, ensuring that the target area always appears within the effective field of view of the image. The camera target visibility score formula is as follows: In the formula, Rate the camera's target visibility. The score is given based on the area of the target in the image. The score is awarded for the target center being located in the center region of the effective field of view. The target sharpness score is used; the weighting coefficient in this embodiment is set to... , , .
[0036] Each sampling process involves synchronous output: simultaneously saving the RGB visible light image, the 16-bit infrared grayscale image, and the target semantic segmentation map; simultaneously recording the camera pose, flight trajectory coordinates, ambient temperature and humidity, atmospheric visibility, and target 2D bounding box annotation information corresponding to each image; and integrating and packaging all images and their associated parameters according to the acquisition sequence number to output a multimodal dataset that can be directly used for target detection and segmentation model training and testing. (See...) Figure 5 .
[0037] In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0038] The foregoing has shown and described the basic principles, main features, and advantages of this application. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this application. Various changes and modifications can be made to this application without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of the claims of this application.
Claims
1. A method for constructing and acquiring data of an infrared simulation scene, characterized in that, Specifically, the following steps are included: We use a large language model to perform semantic parsing on the natural language description of scene tasks, and extract scene entities, spatial relationships and constraints. Based on the parsing results, a data structure conforming to the Large Model Context Protocol (MCP) is generated, along with corresponding control commands. The control commands include at least one of scene construction commands, object control commands, material configuration commands, environmental state control commands, and acquisition task commands. The system sends commands to the Unreal Engine to obtain a 3D dynamic infrared scene. A complete infrared simulation scene is obtained by fusing a three-dimensional dynamic infrared scene with a three-dimensional infrared model of the target object. Within the complete infrared simulation scenario, multi-view infrared data acquisition is performed based on AirSim.
2. The method for constructing and acquiring data of an infrared simulation scene according to claim 1, characterized in that, The generation process of the Unreal Engine modeling instructions is as follows: the control commands drive the Unreal Engine to automatically call various scene resources and complete the scene element deployment according to the scene object deployment priority; The scoring formula for scene object deployment priority is as follows: In the formula, Assign priority scores to scene objects. As the semantic importance weight of the object, For task relevance weights, The distance weight between the object and the target region is used. , , Let be the weight coefficient, and satisfy... .
3. The method for constructing and acquiring data of an infrared simulation scene according to claim 1, characterized in that, The construction method of the three-dimensional dynamic infrared scene is as follows: by issuing instructions to the Unreal Engine through control commands, a three-dimensional visible light simulation scene containing terrain, buildings, vegetation, road scene elements, and corresponding material properties is obtained; Based on the material properties of each object in the 3D visible light simulation scene, and combined with the environmental thermophysical parameters, the surface temperature of each object and the infrared radiation brightness of the scene are calculated in sequence to generate a 3D dynamic infrared scene with a continuous 3D infrared radiation field. The construction of the 3D visible light simulation scene involves establishing basic terrain in Unreal Engine; loading models of buildings, vegetation, roads, water bodies, obstacles, and background interference objects based on the scene semantic description; adjusting the spatial position, scale, material texture binding, and lighting configuration of the loaded scene objects; and configuring roughness, metallicity, diffuse texture, and normal texture parameters for different scene objects to complete the construction of the 3D visible light simulation scene.
4. The method for constructing and acquiring data of an infrared simulation scene according to claim 3, characterized in that, The formula for calculating the surface temperature of each object is as follows: In the formula, For the surface temperature of the scene objects, The ambient reference temperature This represents the temperature rise component caused by solar radiation. The temperature change component caused by heat conduction with surrounding objects. This represents the temperature change component caused by convective heat transfer.
5. The method for constructing and acquiring data of an infrared simulation scene according to claim 3, characterized in that, The formula for calculating the infrared radiance of the scene is: In the formula, The infrared radiation brightness of the scene object. For the surface emissivity of the object, For the surface temperature of the object Planck radiation function under the given conditions The surface reflectivity of the object. This represents the background radiation component.
6. The method for constructing and acquiring data of an infrared simulation scene according to claim 1, characterized in that, The complete infrared simulation scene is constructed by: obtaining the three-dimensional geometric model and corresponding infrared radiation model of the target object; Based on the spatial constraints of the target object and the scene, the target object is automatically placed in a preset position or candidate area in the three-dimensional dynamic infrared scene. The spatial constraints include at least one of constraints based on the passable area, constraints based on the visible area, and constraints based on the distribution of the task target. The pose, height, orientation, and occlusion relationship of the target object are adjusted to obtain a complete infrared simulation scene of the fused target.
7. The method for constructing and acquiring data of an infrared simulation scene according to claim 1, characterized in that, The Airsim multi-view infrared data acquisition preferably uses a drone for acquisition. Specifically, it involves configuring the aircraft's motion path, camera attitude, sensor parameters, and dynamic environmental disturbance parameters, and performing multi-view data acquisition in the fused complete infrared simulation scene. The aircraft's motion path is set to at least one of an orbital flight path and a straight flight path. The orbital flight path is used to collect data around the target at multiple altitudes and angles, while the straight flight path is used to perform through-sampling along the front or side of the target. During flight, weather conditions, visibility, haze intensity, background interference positions, and local temperature parameters are dynamically adjusted according to preset time intervals or waypoint intervals. Airsim is used to control the camera orientation, focal length, imaging mode, and sampling frequency, and sensor noise models and imaging post-processing parameters are configured to simulate the imaging effect of an actual infrared sensor.
8. The method for constructing and acquiring data of an infrared simulation scene according to claim 7, characterized in that, During the multi-view data acquisition process, the target visibility effect is judged by the camera target visibility score. When the score is lower than a preset threshold, the camera orientation or aircraft attitude is automatically adjusted to ensure that the target area appears within the effective field of view of the image. The camera target visibility score formula is as follows: In the formula, Rate the camera's target visibility. The score is given based on the area of the target in the image. The score is awarded for the target center being located in the center of the effective field of view. Score the clarity of the target. , , These are the weighting coefficients; The preset threshold value is between 0.55 and 0.
85.
9. The method for constructing and acquiring data of an infrared simulation scene according to claim 8, characterized in that, During the multi-view data acquisition process, synchronous output is performed: RGB visible light image, infrared grayscale image and target semantic segmentation map are saved synchronously for each sample; camera pose, flight trajectory, environmental parameters, temperature parameters and target annotation information corresponding to the image are recorded synchronously; the image and supporting parameter information are integrated and output as a multimodal dataset for training and testing.
10. An electronic device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for constructing and acquiring data of infrared simulation scenes as described in any one of claims 1 to 9.
