Small sample visible light and infrared image simulation method

Abstract

The application provides a small sample visible light and infrared image simulation method, and relates to the technical field of image simulation.The application inputs a joint generation model of a small sample target text description, a prior image sample and physical constraints to generate a target structure expression, a three-dimensional geometric model and a part model;further combines a part-level material attribute library, a heat conduction relationship, solar radiation, atmospheric radiation and sensor imaging parameters to generate infrared radiation distribution and infrared image results matched with the target geometric structure.The application can realize generation from original input to simulation under small sample limited conditions, and the generated results have higher authenticity.

Description

Technical Field

[0001] This application belongs to the field of image simulation technology, specifically relating to a small-sample visible light and infrared image simulation method. Background Technology

[0002] In application scenarios such as target recognition, infrared detection, intelligent aircraft perception, unmanned system training, and simulation testing, it is usually necessary to simultaneously acquire visible light images, infrared images, and their corresponding structural models of the target to support algorithm training, data augmentation, and system evaluation.

[0003] In the field of image simulation technology, if the amount of small sample input information is very limited, only a small number of target images, local appearance information, or even a brief text description can be provided. If a small number of target images, local appearance information, or brief text descriptions are directly input into a generative model, and the overall shape of the target is restored through feature fitting, it is difficult for the model to complete the deep semantic and image analysis of the target within the limited input information. The final generated model only has the appearance shape, and its infrared simulation has low precision and lacks physical radiation characteristics. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this application is to provide a small-sample visible light and infrared image simulation method, which solves the problem of high difficulty in small-sample simulation in existing technologies.

[0005] Firstly, the objective of this invention can be achieved through the following technical solution: a method for simulating small-sample visible light and infrared images, comprising the following steps: Obtain any one or a combination of small sample text description information of the target object, target image samples, or both. The text description information is segmented, entity normalized, and structural semantic encoded to extract target category, geometric shape, components, and surface feature information. The target image samples are then processed for size normalization, viewpoint filtering, background suppression, and key feature enhancement. Establish a unified multimodal input representation for the processed text description information and image samples; Based on the generative model, semantic and image parsing and structural constraint modeling are performed on the multimodal input representation. Combined with the physical thermal radiation simulation algorithm, infrared simulation images and corresponding visible light simulation images are generated.

[0006] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: obtaining a target structure expression result containing target main structure, local component relationships and topological connection information through the semantic and image parsing and structural constraint modeling, and obtaining a target visible light image using an image generative model; Based on the target structure representation results and the 3D model generative model, a target 3D model is generated. Then, the component-level structural model is obtained by segmentation and annotation using the component model generative model. After matching the material library, it is converted into a thermal radiation model. Combined with environmental parameters, multiple types of radiation brightness are calculated to generate infrared simulation images and corresponding visible light simulation images.

[0007] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the generation process of the component-level structural model is as follows: generating a multi-view image representation of the target through an image generative model and the target structure expression result; reconstructing the original three-dimensional geometric model of the target through a three-dimensional model generative model; performing scale correction, topology repair, and smoothing optimization on the original model to eliminate holes, mesh disorder, and size deviation, thereby obtaining an optimized target three-dimensional model; performing global semantic segmentation on the target three-dimensional model through a component model generative model to obtain candidate component regions; aggregating and semantically labeling the candidate regions based on shape features, component connection relationships, and target category priors to generate a component-level structural model containing component numbers, component boundaries, and hierarchical relationships.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: combining the material parameters and external environment parameters of the thermal radiation model, sequentially calculating the equivalent heat transfer coefficient between components of multiple types of radiation brightness, the component's own radiation brightness, the solar reflected radiation brightness, and the sensor received radiation brightness after atmospheric attenuation, to generate a target infrared radiation model and a corresponding infrared image.

[0009] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the thermal radiation model conversion process is as follows: based on the target name, component labels and application scenarios, candidate materials for each component are determined from the large model knowledge base, rule base and preset material database, and the emissivity, reflectivity, thermal conductivity, specific heat capacity and density parameters of the corresponding materials are extracted; after material consistency verification and component relationship constraint verification, a target component-level material attribute library is established; after mapping the material attribute library to the component-level structural model, it is converted into a voxel format or finite element format thermal radiation model.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the formula for calculating the equivalent heat transfer coefficient between the components is: In the formula, This represents the equivalent heat transfer coefficient between the i-th component and the j-th component; This represents the equivalent thermal conductivity of the contact area between two components. This indicates the contact area between the two components; It represents the equivalent heat conduction distance; the heat conduction process is solved based on the equivalent heat transfer coefficient and the initial temperature of the components, and the temperature distribution of each component is updated.

[0011] The formula for calculating the radiance of the component itself is: In the formula, Indicates the radiance of the component itself; Indicates the surface emissivity of the component; Indicates at wavelength Planck radiation function under component temperature T.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the formula for calculating the solar reflected radiance is: In the formula, Indicates the component of solar reflected radiation; Indicates the surface reflectivity of the component; Indicates solar irradiance; Represents the surface normal vector of the component; This represents the vector of the direction of solar incidence.

[0013] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the formula for calculating the atmospheric attenuation-received radiance of the sensor is: In the formula, Indicates the radiance received by the sensor; Indicates distance from transmission ,wavelength Atmospheric transmittance related to atmospheric humidity (h); Indicates the environmental radiation component; It represents the atmospheric path radiation component; based on the sensor received radiance combined with the imaging band and sensor noise parameters, it generates the target infrared radiation distribution and corresponding infrared radiation model and infrared image.

[0014] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the infrared simulation image is generated by performing a fusion process on the target infrared radiation model and the target three-dimensional model as input, specifically: mapping the target infrared radiation model to an infrared texture map on the surface of the target three-dimensional model, or projecting the target infrared radiation model onto the target imaging plane in the form of pixel-by-pixel radiation values ​​to obtain an infrared simulation image that matches the target pose and viewpoint.

[0015] Secondly, in order to achieve the above objectives, the present invention discloses a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein when the processor executes the computer program, it implements the method of the first aspect.

[0016] The beneficial effects of this application are: This technology requires only a small sample of text description, a small number of image samples, or a combination of both to complete the modeling. First, the text and images are preprocessed according to normalization to extract key features such as the target shape and components, and a unified multimodal input representation is constructed. Semantic and image parsing is completed based on the generative model. At the same time, triple constraints of contour consistency, component position, and scale are applied to carry out structural modeling. The output is a complete structural expression that includes the target main structure, component relationships, and topological connections. Based on this expression, the 3D model is automatically reconstructed and optimized, and component segmentation and semantic annotation are completed simultaneously. The entire modeling process is fully automated and does not require a large number of multi-view samples. Attached Figure Description

[0017] 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.

[0018] Figure 1 This is a schematic diagram of the overall process of an embodiment of this application; Figure 2 This is a schematic diagram of the target image generated by the text-to-image generative model in the embodiments of this application; Figure 3 This is a schematic diagram of the target 3D model and texture map generated by the generative model in the embodiments of this application; Figure 4 This is a schematic diagram of the process of segmenting and annotating the target component using the component model generative model in the embodiments of this application; Figure 5 This is a schematic diagram of the small sample target infrared simulation process according to an embodiment of this application; Figure 6 This is a schematic diagram of the final small sample infrared image generated in the embodiments of this application. Detailed Implementation

[0019] 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.

[0020] Example 1: Reference Figures 1-6 As shown, this invention proposes a small-sample visible light and infrared image simulation method based on generative models and physical constraints, comprising the following steps: This technology first performs sample acquisition and normalization preprocessing operations, collecting one or two types of data from a small sample of textual description information and a small number of target image samples. The textual description information includes the target name, category, geometric shape, components, surface features, and application scenario; the target image samples include at least one of visible light images and infrared images.

[0021] For text description information, the first step is to perform text cleaning, word segmentation, entity recognition, and entity normalization to standardize the size units, component names, and attribute descriptions. Then, the data is input into a large model for semantic and image analysis to extract five core information categories: target category, geometric shape, constituent components, surface features, and application scenario. Combining the knowledge base, rule base, and target category priors, semantic completion is performed on missing component information, structural relationships, and size attributes to form a structured text description.

[0022] For the target image sample, size normalization, image quality assessment, viewpoint selection, background suppression and key feature enhancement are performed sequentially. The target region is extracted using the target detection or segmentation model, and key structural information such as target contour, component boundary and texture is preserved. The overall contour, local features and spatial distribution features of the target are extracted to form a structured visual feature representation.

[0023] Finally, text and visual features are uniformly encoded using text encoders and image encoders, respectively, and mapped to the same feature space. A cross-modal attention mechanism is used to align and match the text entity sequence with the image region feature sequence, matching the corresponding local feature block and segmentation mask in the image for each text component entity, establishing a one-to-one mapping relationship between entities and regions. The aligned text feature vector and image feature vector are concatenated along the channel dimension, and a fused feature vector is generated by combining the target category prior feature vector. After consistency verification and removal of feature pairs with matching degree below a set threshold, a unified multimodal input representation containing target category, geometric structure, components, and spatial relationships is generated, providing standardized input data for subsequent target structure generation, 3D model reconstruction, and infrared simulation.

[0024] This technology utilizes generative models to perform semantic and image parsing and multi-view visible light structure constraint modeling. Diffusion generative models and multimodal large models can be used as the base network. First, the model performs semantic expansion on the text description information to generate the initial prior content of the target structure. Then, pre-processed target image samples are imported, and triple correction conditions—contour consistency constraint, component position constraint, and scale constraint—are applied to the initial structure prior. The contour consistency constraint uses the target outer contour mask of the input sample image as a reference, calculating the intersection-union ratio (IUGR) between the target contour of the generated view and the reference contour. The IUGR loss is used as a constraint term to inversely correct the feature parameters of the generative model, ensuring that the deviation between the generated contour and the sample contour is less than 5%. The component position constraint uses the relative spatial positions of each component obtained from text parsing as a reference, calculating the relative deviation of the center coordinates of the corresponding component regions in the generated view. The position loss function corrects the generated component positions, ensuring that the component topology is consistent with the text description. The scale constraint uses the overall target size and key component size parameters described in the text as a reference, converting the pixel scale of the generated view into the actual physical size through focal length conversion. The size loss function constrains the relative error between the generated size and the nominal size, achieving scale normalization correction. After constraint correction, the output is a complete target structure representation, which fully includes the target main structure and multiple angle views. During the multi-view generation process, the same set of underlying structure encoding features is shared, and multi-view reprojection geometric consistency loss is introduced to constrain the three-dimensional coordinates of the same component under different views to coincide with the component area in the corresponding view after projection, thus ensuring the consistency of the target structure under different views.

[0025] Figure 2 The input is "Generate a photo of a Ukrainian unmanned surface vessel at sea. The unmanned surface vessel is 5 meters long and 2 meters wide, with a smooth, streamlined exterior and no antenna." The target image is generated by an image-generative model.

[0026] This technology generates and reconstructs the target's 3D geometric model based on the target structure representation obtained in the previous step: The first step generates multi-view intermediate representation data of the target based on the structure representation; the second step selects a 3D model generative model based on the multi-view intermediate representation to complete the basic reconstruction of the 3D model, such as... Figure 3 As shown; the third step is to uniformly perform three types of post-processing on the generated original three-dimensional geometric model: scale correction, topology repair, and smoothing optimization, to eliminate problems such as model holes, mesh disorder, and size deviation, so that the geometry of the three-dimensional model can meet the basic requirements of subsequent infrared radiation simulation calculations.

[0027] This technology performs component segmentation and structural annotation on the optimized target 3D geometric model. It selects a point cloud-based Transformer-based 3D semantic segmentation model or a mesh-based semantic segmentation network as the generative model for the component models, and performs global semantic segmentation on the complete 3D model. Figure 4 As shown, multiple independent candidate component regions are divided. By combining component shape features, component connection relationships, and prior knowledge of target categories, the scattered candidate regions are aggregated, classified, and semantically labeled. Finally, a standardized component-level structural model is generated. The model fully stores three types of structured information: component number, component boundary contour, and component hierarchical relationship.

[0028] This technology builds a dedicated material attribute library based on the target name, component tags, and spatial structural relationships of the parts, and completes the mapping and binding. First, it retrieves the large model knowledge base, preset rule base, and preset material database to match corresponding candidate materials for each component. The candidate materials are then comprehensively scored based on three dimensions: component functional matching degree, scene adaptability, and physical property rationality. The material with the highest score is selected as the initial material. Next, the corresponding physical parameters for each material are extracted, including surface emissivity, infrared reflectivity, material thermal conductivity, specific heat capacity, and material density. Emissivity and reflectivity are the equivalent values ​​for the corresponding sensor imaging band, the target's typical operating temperature range, and the surface roughness state. The parameters can be calculated based on the material's basic parameters combined with temperature correction coefficients and surface condition correction coefficients. Then, material consistency verification and component association constraint verification are performed. The material conflict judgment rules include: property contradictions where the same component matches both conductive and insulating materials, heat transfer logic contradictions where the difference in thermal conductivity between adjacent contact components exceeds three orders of magnitude, and functional contradictions where moving parts match brittle materials. When a conflict occurs, the candidate material with the lower score is eliminated, and the next best material is selected for re-verification. The conflicting material parameters are then eliminated, and a precise component-level material attribute library is built. Finally, the entire set of material parameters is mapped one-to-one to the corresponding component in the component-level structural model.

[0029] This technology loads and maps the material parameters of the component-level structural model to perform a complete physical thermal radiation simulation algorithm, such as... Figure 5 As shown, the component-level structural model is first converted into a voxel or finite element method (FEM) thermal radiation calculation model. When using the voxel method, the 3D model space is divided into a uniform cubic mesh, with the mesh size set to no more than 1 / 5 of the smallest component feature size. Material and temperature parameters are assigned in units of voxels. When using the FEM method, the surface and interior of the 3D model are divided into tetrahedral meshes, and the conjugate gradient method is selected as the solver. The time step for transient heat transfer calculation is set to 1 second by default, which can be adjusted according to the calculation accuracy requirements. Then, the multi-dimensional radiation is calculated by combining the component material parameters and external environment parameters. When heat is conducted between components, the equivalent heat transfer coefficient between the components is calculated using the following formula: in, Let represent the equivalent heat transfer coefficient between the i-th component and the j-th component. Indicates the equivalent thermal conductivity of the contact area. The contact area between two components is represented by the sum of the mesh areas of the contact surfaces of adjacent components in the component-level structural model. The equivalent heat conduction distance is represented by the average thickness of the two components in the normal direction of their contact surfaces. The heat conduction calculation supports both steady-state and transient modes: in steady-state calculation, the heat balance equation is solved directly based on the equivalent heat conduction; in transient calculation, the heat capacity of the components is calculated by combining the density and specific heat capacity of each component, and then substituted into the unsteady-state heat conduction equation to solve the temperature change process over time.

[0030] The initial temperature of components is taken as the ambient air temperature by default, but the initial temperature of core components can also be preset according to the target operating state. The internal heat source is determined based on the target category and component function, and the corresponding heat power density is assigned to heat-generating components such as the engine and avionics compartment. The external heat transfer boundary includes the convective heat transfer coefficient and the ambient radiation temperature. The convective heat transfer coefficient is calculated from the ambient wind speed using an empirical formula, and the ambient radiation temperature is taken as the ambient air temperature. The heat conduction process is solved based on the equivalent heat transfer coefficient and the initial temperature of the components, and the temperature distribution of each component is updated.

[0031] The self-radiation brightness of a component is calculated using the following formula: in, This indicates the radiant brightness of the component itself. Indicates the surface emissivity of the component. Indicates at wavelength The Planck radiation function under temperature T is calculated; in actual calculations, the Planck function is multiplied by the sensor spectral response function over the sensor's operating band and integrated to obtain the equivalent self-radiance within the band.

[0032] The solar reflected radiation luminance is calculated using the following formula: in, This represents the component of solar reflected radiation. Indicates the reflectivity of the components. Representing solar irradiance, it is calculated based on the target's geographical location, date, and time, taking into account the solar altitude angle and atmospheric mass. It is obtained by combining the standard solar constant with atmospheric attenuation, and is not calculated directly using the Stefan-Boltzmann law. Represents the surface normal vector of the component. This represents the direction vector of solar incidence. Dividing by π in the formula achieves the dimensional conversion from irradiance to radiance, and the Lambert reflection assumption simplifies the bidirectional reflection distribution characteristics.

[0033] The calculations involve determining heat conduction between components, global solar radiation, overall atmospheric radiation, and environmental reflected radiation. Then, atmospheric attenuation is calculated by considering imaging band, imaging distance, atmospheric transmittance, and sensor noise parameters. Atmospheric transmittance, environmental radiation components, and atmospheric path radiation components can be calculated using atmospheric transmission models such as MODTRAN and LOWTRAN, with parameters including observation distance, imaging band, altitude, and atmospheric humidity. Alternatively, these parameters can be obtained by looking up tables using standard atmospheric profiles. The final received radiance of the sensor is calculated using the following formula: in, This indicates the radiance received by the sensor. Indicates distance from transmission ,wavelength Atmospheric transmittance related to humidity h Indicates the environmental radiation component. It represents the atmospheric path radiation component; based on the sensor received radiance combined with the imaging band and sensor noise parameters, the radiance is mapped to the image grayscale value through a preset radiometric calibration linear coefficient, while the sensor's dark current noise, readout noise and random shot noise are superimposed to generate a complete target infrared radiation distribution.

[0034] By integrating all radiation components, the complete infrared radiation distribution field of the target is obtained. Based on the radiation distribution, an infrared radiation model of the target and a corresponding two-dimensional infrared image are generated, such as... Figure 6 As shown.

[0035] The generated infrared radiation model of the target is fused with the original 3D model of the target and a complete set of simulation results is output. Two infrared radiation mapping methods are provided. The first mapping method is to convert the infrared radiation numerical distribution into an infrared texture map exclusive to the surface of the 3D model. The second mapping method is to project the pixel-by-pixel radiance value in the infrared radiation model onto the virtual imaging plane. Both methods can generate infrared images that match any pose and any viewing angle of the target. After fusion, a complete set of simulation data is output, including the target 3D geometric model, component annotation files, component material attribute files, global infrared radiation field data, visible light simulation images, and all results of infrared simulation images.

[0036] 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.

[0037] 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 simulating small-sample visible light and infrared images, characterized in that, Includes the following steps: Obtain any one or a combination of small sample text description information of the target object, target image samples, or both. The text description information is segmented, entity normalized, and structural semantic encoded to extract target category, geometric shape, components, and surface feature information. The target image samples are then processed for size normalization, viewpoint filtering, background suppression, and key feature enhancement. Establish a unified multimodal input representation for the processed text description information and image samples; Based on the generative model, semantic and image parsing and structural constraint modeling are performed on the multimodal input representation. Combined with the physical thermal radiation simulation algorithm, infrared simulation images and corresponding visible light simulation images are generated.

2. The small-sample visible light and infrared image simulation method according to claim 1, characterized in that, Through semantic and image parsing and structural constraint modeling, a target structure expression result containing information on the target main structure, local component relationships and topological connections is obtained, and a visible light image of the target is obtained using an image generative model. Based on the target structure representation results and visible light images, a target 3D model is generated using a 3D model generative model, and a component-level structural model is obtained by segmentation and annotation using a component model generative model. After matching with the material library, it is converted into a thermal radiation model. Combined with environmental parameters, multiple types of radiation brightness are calculated to generate infrared simulation images and corresponding visible light simulation images.

3. The small-sample visible light and infrared image simulation method according to claim 2, characterized in that, The generation process of the component-level structural model is as follows: a multi-view image representation of the target is generated through an image generative model and the target structure expression results; the original three-dimensional geometric model of the target is reconstructed through a three-dimensional model generative model; scale correction, topology repair and smoothing optimization are performed on the original model to eliminate holes, mesh disorder and size deviation, and an optimized target three-dimensional model is obtained; global semantic segmentation of the target three-dimensional model is performed through a component model generative model to obtain candidate component regions; based on shape features, component connection relationships and target category priors, the candidate regions are aggregated and semantically labeled to generate a component-level structural model containing component numbers, component boundaries and hierarchical relationships.

4. The small-sample visible light and infrared image simulation method according to claim 2, characterized in that, The material parameters and external environment parameters of the thermal radiation model are combined to calculate the equivalent heat transfer coefficient between components and the component's own radiation brightness, solar reflected radiation brightness, and sensor received radiation brightness after atmospheric attenuation in turn, thereby generating the target infrared radiation model and the corresponding infrared image.

5. The small-sample visible light and infrared image simulation method according to claim 3, characterized in that, The thermal radiation model conversion process is as follows: Based on the target name, component labels, and application scenario, candidate materials for each component are determined from the large model knowledge base, rule base, and preset material database, and the emissivity, reflectivity, thermal conductivity, specific heat capacity, and density parameters of the corresponding materials are extracted; after material consistency verification and component relationship constraint verification, a target component-level material attribute library is established; after mapping the material attribute library to the component-level structural model, it is converted into a thermal radiation model in voxel format or finite element format.

6. The small-sample visible light and infrared image simulation method according to claim 4, characterized in that, The formula for calculating the equivalent heat transfer coefficient between the components is: In the formula, This represents the equivalent heat transfer coefficient between the i-th component and the j-th component; This represents the equivalent thermal conductivity of the contact area between two components. This indicates the contact area between the two components; The equivalent heat transfer distance is represented; the heat transfer process is solved based on the equivalent heat transfer coefficient and the initial temperature of the components, and the temperature distribution of each component is updated. The formula for calculating the radiance of the component itself is: In the formula, Indicates the radiance of the component itself; Indicates the surface emissivity of the component; Indicates at wavelength Planck radiation function under component temperature T.

7. The small-sample visible light and infrared image simulation method according to claim 4, characterized in that, The formula for calculating the solar reflected radiation brightness is as follows: In the formula, Indicates the component of solar reflected radiation; Indicates the surface reflectivity of the component; Indicates solar irradiance; Represents the surface normal vector of the component; This represents the vector of the direction of solar incidence.

8. The small-sample visible light and infrared image simulation method according to claim 4, characterized in that, The formula for calculating the radiance received by the sensor after atmospheric attenuation is as follows: In the formula, Indicates the radiance received by the sensor; Indicates distance from transmission ,wavelength Atmospheric transmittance related to atmospheric humidity (h); Indicates the environmental radiation component; It represents the atmospheric path radiation component; based on the sensor received radiance combined with the imaging band and sensor noise parameters, it generates the target infrared radiation distribution and corresponding infrared radiation model and infrared image.

9. The small-sample visible light and infrared image simulation method according to claim 1, characterized in that, The infrared simulation image is generated by performing a fusion process on the target infrared radiation model and the target three-dimensional model as input. Specifically, the target infrared radiation model is mapped to an infrared texture map on the surface of the target three-dimensional model, or the target infrared radiation model is projected onto the target imaging plane in the form of pixel-by-pixel radiation values ​​to obtain an infrared simulation image that matches the target's pose and viewing angle.

10. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the computer program, it implements the small sample visible light and infrared image simulation method as described in any one of claims 1 to 9.

Images