A method, system and medium for generating anti-intelligent camouflage simulation test scene

Abstract

This patent relates to a method, system and medium for generating a simulation test scene for anti-intelligent camouflage, which includes: step 1: establishing a target model appearance database and configuring the target model; step 2: building a geographical environment model and configuring the geographical environment; step 3: building a weather model and configuring the meteorological environment; step 4: establishing a camouflage database, configuring the camouflage style, and realizing the generation of a camouflage simulation test scene; step 5: based on the camouflage simulation test scene, rendering the camouflage to the surface of the target model, transforming the target three-dimensional position to the two-dimensional coordinates of the image, and generating a test data set for camouflage effect evaluation. The present invention can realize the simulation of large-scale anti-intelligent camouflage tests and automatically realize the optimization task of arbitrary camouflage data.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method, system and medium for generating an anti-intelligent camouflage simulation test scene. Background Art

[0002] In modern society, artificial intelligence (AI) technology has been widely used, especially in the fields of image recognition, monitoring and autonomous driving. AI detection models have a strong ability to identify and analyze target objects. However, with the advancement of AI technology, how to deal with the efficient recognition of AI detection models has become a new technical challenge. Traditional camouflage technology is mainly designed for human eyes or optical devices, but these methods are often limited in the face of AI detection models. At present, the mainstream camouflage design collects color cards manually in the field scene environment, completes the camouflage scheme planning, and completes the final design through the camouflage effect in the field scene. The field scene test cycle is long and the desired results cannot be obtained in a timely and fast manner. It is difficult to meet the requirements of camouflage by manual methods alone. For this reason, the development of camouflage technology for artificial intelligence detection models has become a research hotspot. Through specially designed textures and patterns, the recognition ability of AI models is interfered, thereby avoiding monitoring or detection, and achieving the effect of hiding or blurring targets in AI systems. This technology shows broad application prospects in the fields of privacy protection and security. In summary, providing a method for generating highly camouflaged camouflage simulation test scenes is of great significance to help researchers evaluate and improve the robustness of AI detection algorithms. Summary of the invention

[0003] The present invention aims to solve the above problems and provide a method, system and medium for generating anti-intelligent camouflage simulation test scenarios to simulate a variety of complex real-world environments for researchers to evaluate and improve the robustness of AI detection algorithms, while providing support for evaluating and improving the effectiveness of camouflage design.

[0004] The technical solution to achieve the purpose of the present invention is:

[0005] A method for generating an anti-intelligent camouflage simulation test scene, comprising:

[0006] Step 1: Establish a target model appearance database and configure the target model;

[0007] Step 2: construct a geographic environment model and configure the geographic environment;

[0008] Step 3: Build a weather model and configure the meteorological environment;

[0009] Step 4, establish a camouflage database, configure camouflage patterns, and realize the generation of camouflage simulation test scenes;

[0010] Step 5: Based on the camouflage simulation test scene, render the camouflage onto the surface of the target model, transform the target's three-dimensional position into the image's two-dimensional coordinates, and generate a test data set for camouflage effect evaluation.

[0011] Furthermore, the establishment of the target model appearance database specifically includes:

[0012] Step 1.1, starting from the equipment model drawings or industrial models, use Maya, 3Dmax or Blender model software to form high, medium and low-level equipment basic models;

[0013] Step 1.2, draw the map and bind the bones in the modeling software, perform animation production, animation state machine configuration, UE material configuration and special effects configuration, and generate the target model appearance database.

[0014] Furthermore, constructing the geographic environment model specifically includes:

[0015] Step 2.1, the collected geographic vector data is subjected to vector cleaning processing under a unified geographic coordinate system;

[0016] Step 2.2, extracting data from the cleaned vector data to extract relevant attribute information from the vector data;

[0017] Step 2.3, modeling based on relevant attribute information in the data;

[0018] Step 2.4, construct terrain for model data based on DEM elevation data and obtain DOM image data;

[0019] Step 2.5, create a terrain map based on the DOM image data to generate a textured full-factor 3D model.

[0020] Furthermore, the weather model calculates the hour angle h, declination δ, solar altitude angle α and solar azimuth angle β according to the longitude Λ, latitude φ and Beijing time t of the simulation area, and simulates the weather, temperature, air pressure, wind speed and day and night conditions of the simulation area.

[0021] Furthermore, the step 5 specifically includes:

[0022] Step 5.1, randomly place the target in the simulation scene;

[0023] Step 5.2, capture the target image from multiple angles through the simulation camera, and detect the distance d between the camera and the target during the capture process. When the distance d is less than the set threshold Avoid when

[0024] Step 5.3, transforming the coordinates of the target position in the simulation world to the coordinates of the target in the image through view transformation, projection transformation, perspective division and viewport transformation;

[0025] Step 5.4, after coordinate transformation, the test data set is obtained. The test data set contains the RGB image of the target. With label ,in, represents the i-th picture; , represents the target category, Represents the location, Represents size.

[0026] Furthermore, the distance d is:

[0027]

[0028] in, is the position of the data acquisition camera in the three-dimensional coordinate system of the simulation scene, It is the position of the target object in the three-dimensional coordinate system in the simulation scene.

[0029] Furthermore, the step 5.3 specifically includes:

[0030] Step 5.3.1, through the view matrix Set the target's position coordinates in the simulated world Convert from world coordinates to camera coordinates ,for:

[0031] ;

[0032] Step 5.3.2, through the projection matrix P, the 3D view coordinates of the object Convert to clip coordinates :

[0033] ;

[0034] Step 5.3.3, set the clipping coordinates Convert to normalized device coordinates by perspective division, that is: ;

[0035] Step 5.3.4, by Normalize:

[0036]

[0037] Step 5.3.5, convert the normalized device coordinates to screen coordinates through the viewport transformation, as follows:

[0038]

[0039]

[0040] in, and are the width and height of the viewport, and is the starting coordinate of the viewport. Through this transformation, the 3D coordinate of the target is successfully mapped to the 2D coordinate on the screen for use in generating image annotations.

[0041] Furthermore, the evaluation of the camouflage effect in step 5 is performed by weighted attack success rate, and the weighted attack success rate is:

[0042]

[0043] in, is the total number of samples in the test data set, is the weight of the i-th sample, representing the importance of the sample, is the indicator function of whether the i-th sample is successfully attacked, which is:

[0044]

[0045] in, is the threshold of target confidence c , is the IoU threshold, IoU is the ratio of the intersection and union of the predicted bounding box and the true label box.

[0046] A system for generating a simulation test scenario for an anti-intelligent camouflage pattern includes a main server, a communication module and a display monitoring station. The main server is used for data processing and is connected to a UPS uninterruptible power supply, which is used to provide power for the platform. The communication module is used to realize information interaction between the main server and the display monitoring station. The display monitoring station includes an operating station and a central dispatching station. The operating station is used for human-computer interaction and configuration of data information required for the test scenario. The central dispatching station is used for displaying the simulation test scenario and data. The main server specifically includes:

[0047] A target model configuration unit, which establishes a target model appearance database for configuring a target model for an anti-intelligent camouflage simulation platform;

[0048] A geographic environment configuration unit, which constructs a geographic environment model for configuring the geographic environment of the anti-intelligent camouflage simulation platform;

[0049] A meteorological environment configuration unit, which constructs a weather model for configuring the meteorological environment for the anti-intelligent camouflage simulation platform;

[0050] The simulation test scenario generation unit establishes a camouflage database, configures camouflage patterns for the anti-intelligent camouflage simulation platform, and realizes the generation of camouflage simulation test scenarios;

[0051] The test data set generation unit renders the camouflage onto the surface of the target model based on the camouflage simulation test scene, transforms the target's three-dimensional position into the image's two-dimensional coordinates, and generates a test data set for camouflage effect evaluation.

[0052] A computer storage medium stores an executable program, and the executable program is executed by a processor to implement the steps of the method for generating a simulation test scene for an anti-intelligent camouflage pattern.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] (1) This patent can automatically generate test scenes with weather characteristics such as cloudy, sunny, rainy, and snowy days, and comprehensively cover the application of various camouflage patterns. The camera captures images of objects in the scene from multiple angles, which improves the efficiency of data collection and greatly enriches the diversity of data sets, making the generated data more extensive and applicable. In addition, it can automatically switch between different virtual scenes and continuously generate data in multiple environments, so that the system does not require human intervention during long-term operation, maintaining the consistency and efficiency of data generation. The present invention ensures that the RGB image, the target position frame, and the true coordinate value are highly consistent. This high-precision annotation provides reliable data support. The camera can automatically avoid collision with objects in the scene during movement, ensuring the continuity and stability of data collection and improving the operating efficiency of the entire system.

[0055] (2) The weighted attack success rate index designed by the present invention can more comprehensively and reasonably evaluate the effectiveness of camouflage and the robustness of the detection model.

[0056] (3) The present invention can effectively use intelligent means to accelerate the design and evaluation of camouflage patterns. This design not only supports multiple staff members to participate in algorithm testing at the same time, but also ensures comprehensive evaluation and optimization of the algorithm by monitoring multi-dimensional output information, thereby improving test efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is the platform hardware architecture diagram of this system;

[0058] Figure 2 This is a car model diagram in the embodiment;

[0059] Figure 3 Construct flowcharts for geographic environments;

[0060] Figure 4Test scene graph for rainy days in the city;

[0061] Figure 5 Test scene map for a sunny day in the jungle;

[0062] Figure 6 Test scene map for desert foggy days;

[0063] Figure 7 This is a schematic diagram of the camouflage pattern database;

[0064] Figure 8 The UV mask map of the camouflaged surface of the target car model;

[0065] Fig. 9 Rendering camouflage to the target car surface;

[0066] Fig.10 It is a schematic diagram of multi-angle image capture;

[0067] Fig.11 A weather distribution ratio diagram of the camouflage simulation test scene data generated in the embodiment;

[0068] Fig.12 is a position scale diagram of the target car relative to the center of the screen in the embodiment;

[0069] Fig.13 is a screen-to-body ratio diagram of the target car in the embodiment;

[0070] Fig.14 This is a picture of the original car and the camouflage color of the car in the embodiment. DETAILED DESCRIPTION

[0071] This embodiment selects to simulate the relevant data of camouflage camouflage against AI test for urban surveillance in a certain area for case analysis. The cameras in this area can autonomously detect vehicles traveling on the road through AI intelligent algorithms, and can not only identify the appearance of the vehicle, but also track the behavior pattern of the vehicle. This embodiment provides a method for generating anti-intelligent camouflage simulation test scenes based on the relevant data of camouflage camouflage against AI test for urban surveillance, which is used for testing AI intelligent algorithms, generating camouflage simulation test scenes and evaluating camouflage effects, and provides effective support for the subsequent improvement of AI intelligent algorithms.

[0072] A system for generating a simulation test scenario for an anti-intelligent camouflage pattern includes a main server, a communication module, and a display monitoring station. The main server is used for data processing and is connected to a UPS uninterruptible power supply, which is used to provide power to the platform. The communication module is used to realize information interaction between the main server and the display monitoring station. The display monitoring station includes an operating station and a central dispatching station. The operating station is used for human-computer interaction and configuration of data information required for the test scenario. The central dispatching station is used to simulate the test scenario and display data. The main server specifically includes:

[0073] A target model configuration unit, which establishes a target model appearance database for configuring a target model for an anti-intelligent camouflage simulation platform;

[0074] A geographic environment configuration unit, which constructs a geographic environment model for configuring the geographic environment of the anti-intelligent camouflage simulation platform;

[0075] A meteorological environment configuration unit, which constructs a weather model for configuring the meteorological environment for the anti-intelligent camouflage simulation platform;

[0076] The simulation test scenario generation unit establishes a camouflage database, configures camouflage patterns for the anti-intelligent camouflage simulation platform, and realizes the generation of camouflage simulation test scenarios;

[0077] The test data set generation unit renders the camouflage onto the surface of the target model based on the camouflage simulation test scene, transforms the target's three-dimensional position into the image's two-dimensional coordinates, and generates a test data set for camouflage effect evaluation.

[0078] Based on the above system, this embodiment provides a method for generating an anti-intelligent camouflage simulation test scene, which specifically includes:

[0079] Step 1, build an anti-intelligent camouflage simulation platform. A demonstration and verification platform containing hardware modules needs to be built as shown in Figure 1. The platform consists of a server cabinet, a UPS uninterruptible power supply, a main server communication module, and a display monitoring console. Through the UPS power supply, the system can ensure the continuity and stability of the power supply, prevent system downtime due to sudden power outages or surges, and ensure that training tasks based on data sets can continue. The computing fiber optic switch equipped with the system can provide 10 Gigabit bandwidth to ensure high-speed communication between nodes and avoid the failure of parallel computing due to insufficient bandwidth. At the same time, the system's storage nodes provide storage support for large-scale data sets for the algorithm to meet high-intensity computing requirements.

[0080] Step 1.1, the main server deployment includes node management server, CPU node server, GPU node server, storage node server, computing fiber switch, management operation node server and UPS uninterruptible power supply. All equipment must be installed in a 42U cabinet and powered by a 220V AC socket. This design ensures efficient integration and stable operation of the system, and provides a reliable hardware foundation for various computing tasks.

[0081] Step 1.2, deploy the communication module to enable the server to have the ability to exchange information with other platforms during joint debugging. Step 1.3, deploy the display monitoring console, which provides the functions of the operation console and the central dispatch console. The operation console is used by operators to perform various operations during the algorithm evaluation process, including parameter setting, simulation scenario selection, sensor information input node configuration, and calling other algorithm nodes. After the algorithm is running, the operation console can also monitor the execution of the software background. The central dispatch console is mainly used to display the output results of the algorithm, visualize the data and algorithm evaluation results. This design can not only help multiple staff members participate in algorithm testing at the same time, but also assist in comprehensive evaluation and optimization of the algorithm by monitoring a variety of output information.

[0082] Step 2: Configure the target model. Create a target model appearance database. The 3D appearance model uses a sophisticated modeling process to restore the shape and visual details of the equipment.

[0083] Step 2.1, first start from the equipment model drawing or industrial model, and use professional modeling software (Maya, 3Dmax or Blender) to form high, medium and low-level basic equipment models.

[0084] Step 2.2, use Substance Designer or Substance Painter to draw the texture and bind the skeleton in the modeling software. Finally, enter Unreal Engine to perform animation production, animation state machine configuration, UE material configuration and special effect configuration, etc., to give users a real visual perception of the equipment, such as Figure 3 shown.

[0085] Step 3: Configure the geographical environment and build simulation scenarios in several expected environments (such as cities, mountains, jungles, etc.). The geographical environment model mainly simulates the natural features of the simulation area, such as topography, landforms, vegetation, water systems, road traffic, etc. Figure 3 As shown in the figure, the system process of geographic environment construction goes through the following steps:

[0086] Step 3.1, vector data cleaning. Vector data such as buildings, roads, vegetation, rivers, terrain, etc. are cleaned in a unified geographic coordinate system.

[0087] Step 3.2, vector data extraction. The cleaned building, road, vegetation, river, and terrain vector data are extracted, and the relevant attribute information in the terrain element vector data, such as elevation value, length, area, etc., is extracted through attribute query and spatial analysis methods.

[0088] Step 3.3, vector data modeling: Model the cleaned building, road, vegetation, river, and terrain vector data.

[0089] Step 3.4, construct terrain. Construct terrain based on DEM elevation data for modeled buildings, roads, vegetation, rivers, and terrain model data.

[0090] Step 3.5, image mapping. Create terrain mapping based on DOM image data for the buildings, roads, vegetation, rivers, and terrain model data with good elevation, and generate a high-precision 3D map model of all elements including buildings, roads, vegetation, rivers, etc. with textures.

[0091] Step 4: Configure the meteorological environment and construct the time period, light, and weather changes related to the simulation scenarios in several expected environments (such as cities, mountains, jungles, etc.) in step 3. Figure 4-Figure 6 The meteorological environment model mainly simulates the weather, temperature, air pressure, wind speed, day and night conditions of the simulation area. By establishing a meteorological environment model, the weather, temperature, air pressure, wind speed, etc. of the simulation scene area are configured to simulate the impact of the meteorological environment on the effect of the AI ​​detection model. The meteorological environment model can set the simulation time and associate it with weather elements such as day, night, dawn, and dusk. The law of weather changes basically conforms to the actual situation of the simulated area; it can also dynamically set meteorological elements such as clouds, rain, snow, fog, wind, and dust, and their level intensity.

[0092] The weather model calculates the hour angle (h), declination (δ), solar altitude angle (α) and solar azimuth angle (β) according to the longitude (Λ), latitude (φ) and Beijing time (t) of the simulation area to ensure that the simulated lighting effect is consistent with the actual one.

[0093] The hour angle (h) is calculated as follows:

[0094]

[0095] The calculation formula of declination (δ) is as follows, where N is the accumulated day, which can be obtained from t.

[0096]

[0097] The calculation formula of the solar altitude angle (α) is as follows:

[0098]

[0099] The calculation formula of solar azimuth (β) is as follows:

[0100]

[0101] For the simulation of weather, the position of the sun can be calculated by the latitude and longitude data of the simulated area, and the position and direction of the light source can be adjusted in the simulated scene, so that the lighting conditions and the time points corresponding to the alternation of day and night in the simulated scene are basically consistent with the actual conditions such as the time zone and season of the real area. The weather model is used to describe weather conditions such as sunny, rainy, snowy, cloudy, foggy, and dusty. The weather effect is visualized and simulated based on the Unreal Engine's atmospheric system and particle special effects system.

[0102] Step 5: Configure the camouflage pattern and establish a camouflage database, such as Figure 7 As shown, select the desired camouflage pattern and apply it to the target model according to its UV mask (such as Figure 8 The target model is cropped and loaded into the test scene synchronously with the target model.

[0103] Step 6: Render camouflage onto the surface of the target model, transform the target's 3D position into the image's 2D coordinates, generate a test data set, and achieve the simultaneous generation of RGB images, object categories, object coordinates, and object sizes. Through precise annotation, the generated data is ensured to have a high degree of accuracy and consistency. The core of this module is to integrate the corresponding relationship between the target object position in the scene and the target object in the image, and provide high-precision annotation data.

[0104] Step 6.1, randomly place targets (cars in this embodiment, such as Figure 2 ). Rendering step 5 camouflage to the target model surface, such as Fig. 9 shown.

[0105] Step 6.2, simulate the camera to capture the target image from multiple angles, such as Fig.10 The five angles shown include front, back, left, right, and top. Randomly generate perspective offsets and position changes to ensure data diversity and richness. At the same time, by detecting the position of the camera and the car in the scene, ensure that the camera does not collide with the car during movement, and capture the target image such as Figure 8 As shown, the consistency and accuracy of the data collection process is guaranteed. The core formula of collision detection is distance detection. When the distance (d) between the camera and the object is less than the set threshold Trigger avoidance.

[0106]

[0107] in is the position of the data acquisition camera in the three-dimensional coordinate system of the simulation scene, It is the position of the target object in the three-dimensional coordinate system in the simulation scene.

[0108] Step 6.3, coordinate transformation. Transform the target position coordinates in the simulation world Transform to the coordinates of the target in the image The specific transformation process includes view transformation, projection transformation, perspective division and viewport transformation.

[0109] The 3D coordinates in the simulation environment Convert from world coordinates to camera coordinates (view coordinates). This conversion is done using the view matrix (V):

[0110]

[0111] Through the projection matrix (P), the 3D view coordinates of the object are Convert to clip coordinates:

[0112]

[0113] Next, the clip coordinates are converted to normalized device coordinates (NDC) using perspective division.

[0114] Need to pass Normalize:

[0115]

[0116] The perspective division process ensures that the coordinates are mapped to the normalized range ([-1, 1]), laying the foundation for subsequent viewport transformations.

[0117] Finally, the normalized device coordinates are converted to screen coordinates through viewport transformation. The specific transformation formula is:

[0118]

[0119]

[0120] in, and are the width and height of the viewport, and is the starting coordinate of the viewport. Through this transformation, the 3D coordinates of the object are successfully mapped to the 2D coordinates on the screen for use in generating image annotations.

[0121] Step 6.4, output the test data set. The test data set contains the RGB image where the target is located With label .in, represents the i-th picture; , represents the target category, Represents the location, Represents size.

[0122] The distribution of the camouflage effect verification dataset is shown in Table 1. It contains a total of 3,032 frames, covering a variety of terrain scenes and weather conditions, to comprehensively verify the concealment effect and adaptability of camouflage in different environments. Terrain scenes: The dataset contains three main terrain scenes, namely city (903 frames), desert (952 frames) and jungle (1,177 frames). It covers urban, desert and jungle scenes at the same time to ensure the applicability of camouflage in diverse terrains. Weather conditions: The dataset contains the frame number distribution of three typical weather conditions, namely sunny days (1,031 frames), rainy days (953 frames) and foggy days (1,048 frames). Such diverse weather conditions verify the robustness of camouflage under different meteorological conditions. It contains weather distribution, such as Fig.11 As shown in the figure; the position ratio of the target object relative to the center of the screen, Fig.12 As shown in the figure; the distribution of the screen ratio of the target is as follows: Fig.13 The generated test data set is shown in Fig.14 shown.

[0123] Table 1. Distribution table of camouflage effect verification dataset

[0124]

[0125] Step 7, perform the test, input the test data set into the AI ​​detector f(x) for testing, evaluate the weighted attack success rate index of the AI ​​detector f(x), and analyze the concealment of the camouflage and the robustness of the detection model. The AI ​​detector is a network model obtained by deep learning method, and the weighted attack success rate formula is as follows:

[0126]

[0127] In the formula, is the total number of samples in the test dataset. is the weight of the i-th sample, representing the importance of the sample, Is the i-th sample successfully attacked? The indicator function sets the threshold of the detector's confidence level c for detecting the target. The threshold of intersection over union (IoU) IoU is the ratio of the intersection over union of the predicted bounding box and the true label box, defined as:

[0128] ,

[0129] in, is the bounding box area predicted by the detector, is the bounding box area of ​​the true label. The intersection of the two is the overlapping area, and the union is the total area covered by the two. Defined as

[0130]

[0131] The greater the difference between the sample confidence and the threshold, the greater the sample weight.

[0132]

[0133] The experimental results show that camouflage effectively interferes with AI target detectors. The system efficiently analyzes the effectiveness of camouflage and the robustness of the detection model by comparing the weighted attack success rate indicators.

[0134] Table 2 Camouflage countermeasure test data table

[0135]

[0136] The present invention can help researchers evaluate and improve the robustness of AI detection algorithms and the effectiveness of camouflage design by building a highly simulated environment, simulating the effects of various camouflage patterns on AI detection models, and designing reasonable evaluation indicators. These test scenarios can not only simulate a variety of complex real-world environments, but also continuously adjust camouflage strategies to ensure that AI detection models can still accurately identify camouflaged targets in various scenarios. Through the test results, researchers can better understand the weaknesses of AI models, optimize camouflage design algorithms, and optimize AI models in a targeted manner, laying the foundation for the development of future AI detection technology.

[0137] This embodiment also provides a computer storage medium, which stores an executable program, and the executable program is executed by a processor to implement the steps of the anti-intelligent camouflage simulation test scene generation method.

[0138] In summary, the invention can automatically generate test scenes with weather characteristics such as cloudy, sunny, rainy, and snowy days, and comprehensively cover the application of various camouflage patterns. The system supports importing camouflage patterns into the target surface, and can perform surface assignment operations on three-dimensional models. At the same time, the system also allows operations such as zooming in, zooming out, adjusting the viewing angle, changing the viewing distance, and rotating the camouflaged target. In addition, the system has the function of automatically recording the location information of the camouflaged target. Based on three-dimensional scenes or images, the system can carry out simulation tests of camouflage effects, and analyze the performance differences before and after camouflage through artificial intelligence target detectors. This invention can not only provide guidance for camouflage design, but also improve the security and reliability of target detection, and has broad application prospects in the fields of privacy protection and security.

[0139] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0140] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, if these modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for generating a simulation test scene for an anti-intelligent camouflage, characterized in that: include: Step 1, establish a target model appearance database and configure the target model; Step 2: construct a geographic environment model and configure the geographic environment; Step 3: Build a weather model and configure the meteorological environment; Step 4, establish a camouflage database, configure camouflage patterns, and realize the generation of camouflage simulation test scenes; Step 5: based on the camouflage simulation test scene, render the camouflage onto the surface of the target model, transform the target's three-dimensional position into the image's two-dimensional coordinates, and generate a test data set for camouflage effect evaluation; The evaluation of the camouflage effect in step 5 is performed by weighted attack success rate, and the weighted attack success rate is: ； in, is the total number of samples in the test data set, is the weight of the i-th sample, representing the importance of the sample, is the indicator function of whether the i-th sample is successfully attacked, which is: ； in, is the threshold of target confidence c, is the IoU threshold, IoU is the ratio of the intersection and union of the predicted bounding box and the true label box.

2. The method for generating an anti-intelligent camouflage simulation test scene according to claim 1, characterized in that: The establishment of the target model appearance database specifically includes: Step 1.1, starting from the equipment model drawings or industrial models, use Maya, 3Dmax or Blender model software to form high, medium and low-level equipment basic models; Step 1.2, draw the map and bind the bones in the modeling software, perform animation production, animation state machine configuration, UE material configuration and special effects configuration, and generate the target model appearance database.

3. The method for generating an anti-intelligent camouflage simulation test scene according to claim 1, characterized in that: Building a geographic environment model specifically includes: Step 2.1, the collected geographic vector data is subjected to vector cleaning processing under unified geographic coordinates; Step 2.2, extracting data from the cleaned vector data to extract relevant attribute information from the vector data; Step 2.3, modeling based on relevant attribute information in the data; Step 2.4, construct terrain for model data based on DEM elevation data and obtain DOM image data; Step 2.5, create a terrain map based on the DOM image data to generate a textured full-factor 3D model.

4. The method for generating an anti-intelligent camouflage simulation test scene according to claim 1, characterized in that: The weather model calculates the hour angle h, declination δ, solar altitude angle α and solar azimuth angle β according to the longitude Λ, latitude φ and Beijing time t of the simulation area, and simulates the weather, temperature, air pressure, wind speed and day and night conditions of the simulation area.

5. The method for generating an anti-intelligent camouflage simulation test scene according to claim 1, characterized in that: The step 5 specifically includes: Step 5.1, randomly place the target in the simulation scene; Step 5.2, capture the target image from multiple angles through the simulation camera, and detect the distance d between the camera and the target during the capture process. When the distance d is less than the set threshold Avoid when Step 5.3, transforming the coordinates of the target position in the simulation world to the coordinates of the target in the image through view transformation, projection transformation, perspective division and viewport transformation; Step 5.4, after coordinate transformation, the test data set is obtained. The test data set contains the RGB image of the target. With label ,in, represents the i-th picture; , represents the target category, Represents the location, Represents size.

6. The method for generating a simulation test scene for anti-intelligent camouflage according to claim 5, characterized in that: The distance d is: ； in, is the position of the data acquisition camera in the three-dimensional coordinate system of the simulation scene, It is the position of the target object in the three-dimensional coordinate system in the simulation scene.

7. The method for generating a simulation test scene for anti-intelligent camouflage according to claim 5, characterized in that: The step 5.3 specifically includes: Step 5.3.1, through the view matrix Set the target's position coordinates in the simulated world Convert from world coordinates to camera coordinates ,for: ； Step 5.3.2, through the projection matrix P, the 3D view coordinates of the object Convert to clip coordinates : ； Step 5.3.3, set the clipping coordinates Convert to normalized device coordinates by perspective division, that is: ; Step 5.3.4, by Normalize: ； Step 5.3.5, convert the normalized device coordinates to screen coordinates through the viewport transformation, as follows: ； ； in, and are the width and height of the viewport, and is the starting coordinate of the viewport. Through this transformation, the 3D coordinate of the target is successfully mapped to the 2D coordinate on the screen for use in generating image annotations.

8. A system for generating a simulation test scene for anti-intelligent camouflage for implementing any of the methods described in claims 1-7, characterized in that: It includes a main server, a communication module and a display monitoring station. The main server is used for data processing and is connected to a UPS uninterruptible power supply, which is used to provide power for the platform; the communication module is used to realize information interaction between the main server and the display monitoring station. The display monitoring station includes an operating console and a central dispatching console. The operating console is used for human-computer interaction and configuration of data information required for the test scenario. The central dispatching console is used to simulate the test scenario and display the data; The main server specifically includes: A target model configuration unit, which establishes a target model appearance database for configuring a target model for an anti-intelligent camouflage simulation platform; A geographic environment configuration unit, which constructs a geographic environment model for configuring the geographic environment of the anti-intelligent camouflage simulation platform; A meteorological environment configuration unit, which constructs a weather model for configuring the meteorological environment for the anti-intelligent camouflage simulation platform; The simulation test scenario generation unit establishes a camouflage database, configures camouflage patterns for the anti-intelligent camouflage simulation platform, and realizes the generation of camouflage simulation test scenarios; The test data set generation unit renders the camouflage onto the surface of the target model based on the camouflage simulation test scene, transforms the target's three-dimensional position into the image's two-dimensional coordinates, and generates a test data set for camouflage effect evaluation.

9. A computer storage medium, characterized in that The computer storage medium stores an executable program, and the executable program is executed by a processor to implement the steps of the anti-intelligent camouflage camouflage simulation test scene generation method described in any one of claims 1-7.

Images