Attribute feature driven scene-to-image cross-domain mapping sample set construction method

Abstract

The invention discloses an attribute feature driven scene-to-image cross-domain mapping sample set construction method, and belongs to the technical field of depth model compression, and the method comprises the steps: carrying out the augmentation of a basic sample set through employing an image feature transformation method, and obtaining a material library used for sample synthesis; the basic sample set comprises a simulation target material and a real shooting environment material; performing spatial position registration on an actual target and an actual environment in an actual application scene to obtain position information of the actual target in an environment coordinate system; selecting a corresponding simulation target material and a real shooting environment material from a material library based on an actual application scene; based on the position information of an actual target in an environment coordinate system, performing threshold segmentation on a selected simulation target material and fusing the selected simulation target material with a real shot environment material to obtain a high-confidence synthetic sample, so as to solve the problems of obvious segmentation of pixel blocks in a synthetic sample region, sample distortion and high reliability of the existing sample synthesis method. And therefore, a high-confidence sample set required by intelligent algorithm training cannot be effectively synthesized.

Description

Technical Field

[0001] The present invention relates to the technical field of deep model compression, and in particular to a method for constructing a scene-to-image cross-domain mapping sample set driven by attribute features. Background Art

[0002] At present, the main sources of sample sets required for intelligent perception algorithm training include real-life acquisition, simulation modeling, and sample synthesis. Limited by the diversity and complexity of application scenarios, it is costly to obtain large-scale real-life samples. A large number of simulation samples can be generated by simulation modeling, but there are usually obvious cross-domain differences between simulation samples and real-life samples, and it is urgent to eliminate the cross-domain influence of samples through sample synthesis and other means. The current mainstream method usually uses sample synthesis based on generative adversarial networks. By learning the target and environmental features of simulated samples and real-life samples, and designing adversarial loss functions to constrain the sample generation process, samples with similar distribution to the original data but different features are generated, and a certain degree of sample confidence improvement is achieved in the form of style transfer. However, the training process of the synthesis method based on generative adversarial networks has poor convergence and it is difficult to avoid the problem of mode collapse. In addition, the synthesized samples have obvious regional pixel block splitting and sample distortion problems, making it difficult to synthesize high-confidence sample sets required for intelligent algorithm training. Summary of the invention

[0003] The purpose of the present invention is to provide a method for constructing a scene-to-image cross-domain mapping sample set driven by attribute features, so as to solve the problem that the existing sample synthesis methods have obvious pixel block fragmentation and sample distortion in the synthesized sample area, resulting in the inability to effectively synthesize the high-confidence sample set required for intelligent algorithm training.

[0004] To achieve the above object, the present invention adopts the following technical solution: On the one hand, this specification provides a method for constructing a scene-to-image cross-domain mapping sample set driven by attribute features, comprising: Step 102, using an image feature transformation method to augment the basic sample set to obtain a material library for sample synthesis; the basic sample set includes simulation target materials and real-shot environment materials; Step 104, performing spatial position registration on the actual target and the actual environment in the actual application scene to obtain position information of the actual target in the environment coordinate system; Step 106, selecting corresponding simulation target materials and real-shot environment materials from a material library based on the actual application scenario; Step 108 , based on the position information of the actual target in the environmental coordinate system, threshold segmentation is performed on the selected simulated target material and fused with the real-shot environmental material to obtain a high-confidence synthetic sample.

[0005] On the other hand, the present specification provides an attribute feature driven scene-to-image cross-domain mapping sample set construction device, comprising: A sample augmentation module, used to augment the basic sample set by using an image feature transformation method to obtain a material library for sample synthesis; the basic sample set includes simulated target materials and real-shot environment materials; The spatial registration module is used to perform spatial registration of the actual target and the actual environment in the actual application scenario to obtain the position information of the actual target in the environmental coordinate system; The actual material acquisition module is used to select the corresponding simulation target material and real-shot environment material from the material library based on the actual application scenario; The material fusion module is used to perform threshold segmentation on the selected simulated target material based on the position information of the actual target in the environmental coordinate system and fuse it with the real-shot environmental material to obtain a high-confidence synthetic sample.

[0006] Based on the above technical solution, this specification can achieve the following technical effects: This method constructs a synthetic feature library based on simulation modeling data and real-shot data, and performs sample augmentation based on image features to improve the diversity of the feature library. Then, the rationality of the spatial position relationship between the target and the environment in the synthetic samples is improved through spatial position registration of the target and the environment. The target threshold segmentation and background adaptive fusion technology are used for sample synthesis, which effectively solves the problems of obvious regional pixel block segmentation and sample authenticity, and improves the confidence of the synthetic samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 The figure is a flow chart of a method for constructing a scene-to-image cross-domain mapping sample set driven by attribute features in one embodiment of the present invention.

[0008] Figure 2 The schematic diagram is a structural diagram of an attribute feature driven scene-to-image cross-domain mapping sample set construction device according to an embodiment of the present invention.

[0009] Figure 3 The figure is a schematic diagram of an electronic device according to the present invention. DETAILED DESCRIPTION

[0010] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present invention will become more apparent from the following description and claims. It should be noted that the drawings are all in very simplified form and are not in precise proportions, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.

[0011] It should be noted that, in order to clearly explain the content of the present invention, the present invention specifically cites multiple embodiments to further illustrate different implementations of the present invention, wherein the multiple embodiments are enumerated rather than exhaustive. In addition, for the sake of brevity of explanation, the contents mentioned in the previous embodiments are often omitted in the subsequent embodiments. Therefore, the contents not mentioned in the subsequent embodiments can refer to the previous embodiments accordingly. Example 1

[0012] Please refer to Figure 1 , Figure 1 The following is a method for constructing a scene-to-image cross-domain mapping sample set driven by attribute features provided in this embodiment. In this embodiment, the method includes: Step 102, using an image feature transformation method to augment the basic sample set to obtain a material library for sample synthesis; the basic sample set includes simulation target materials and real-shot environment materials; In this embodiment, before step 102, the following steps are also included: Step 1011, using a simulation modeling method, establishing a target model and generating a number of simulation target materials; the simulation target materials include a target type, a target scale, and a target direction; Step 1012, using a real-shot acquisition method to obtain real-shot environment materials; the real-shot environment materials include scene information, time of day information, lighting information, and weather information; In this embodiment, one implementation of step 102 is: Step 202, analyzing the mapping relationship between scene conditions and image features on the basic sample set to obtain augmented image features; the augmented image features are image features that are strongly coupled with the intelligent detection and recognition algorithm; Step 204: perform image preprocessing and image feature transformation on the basic sample set based on the augmented image features to obtain a material library for sample synthesis.

[0013] Step 104, performing spatial position registration on the actual target and the actual environment in the actual application scene to obtain position information of the actual target in the environment coordinate system; In this embodiment, one implementation of step 104 is: Step 302, based on the actual application scenario, determine the actual target and the actual environment; Step 304, extracting features from the actual target and the actual environment to obtain key features; Step 306, performing feature point registration on the actual target and the actual environment based on the key features, and obtaining a transformation matrix from the target coordinate system to the environment coordinate system; In this embodiment, one implementation of step 306 is: Step 3061, based on the position information of the actual target in the target coordinate system and the rotation angle of the actual target relative to the actual environment, obtain a rotation matrix for converting from the target coordinate system to the environment coordinate system; Step 3062: Based on the translation direction, scaling ratio, and rotation matrix, obtain a transformation matrix for converting from the target coordinate system to the environment coordinate system.

[0014] Step 308: based on the transformation matrix, the coordinates of the actual target are transformed into the environment coordinate system to obtain the position information of the actual target in the environment coordinate system after spatial registration.

[0015] In this embodiment, the key features include a rotation angle, a translation vector, and a scaling ratio of the actual target relative to the actual environment.

[0016] Step 106, selecting corresponding simulation target materials and real-shot environment materials from a material library based on the actual application scenario; Step 108 , based on the position information of the actual target in the environmental coordinate system, threshold segmentation is performed on the selected simulated target material and fused with the real-shot environmental material to obtain a high-confidence synthetic sample.

[0017] In this embodiment, one implementation of step 108 is: Step 402, based on the position information of the actual target in the environment coordinate system, calculate the gradient of the local area in the image of the simulated target material and determine the segmentation threshold; Step 404, segmenting the image of the simulated target material into a target part and a background part based on the segmentation threshold; Step 406, determining the edge of the segmented simulation target material based on the Sobel detection operator and performing edge smoothing optimization to obtain a smoothed simulation target material; In this embodiment, one implementation of step 406 is: Step 4061, using the Sobel detection operator to perform convolution on the simulation target material along the X-axis and the Y-axis respectively, to determine the image edge of the simulation target material; Step 4062, performing edge smoothing optimization on the image edge of the simulation target material through Gaussian kernel filtering to obtain a smoothed simulation target material.

[0018] Step 408, synthesizing the smoothed simulation target material with the real-shot environment material to generate a simulation-real-shot synthesized sample; Step 410: Perform contrast adjustment on the simulated real shot synthetic sample to obtain a high confidence synthetic sample.

[0019] In this embodiment, one implementation of step 410 is: Step 4101, performing grayscale histogram statistics on the simulated real shot synthetic samples; Step 4102: Based on the grayscale histogram, the contrast of the simulated real-shot synthetic sample is adjusted by using a histogram equalization method to obtain a high-confidence synthetic sample.

[0020] Specifically, a method for constructing a scene-to-image cross-domain mapping sample set driven by attribute features includes the following steps: The first step is to construct a basic sample set of simulation targets and real-shot environments, establish a material library required for sample synthesis, and construct a basic sample set consisting of a small number of real samples and a large number of simulation samples. By analyzing the mapping relationship between scene conditions and image features, the image feature transformation method is used to expand the basic sample set, improve sample diversity, and form a material library required for sample synthesis.

[0021] By adopting simulation modeling to establish the target model, a large number of simulation samples of different types, scales and perspectives are generated.

[0022] Through real-life shooting, we obtain a small number of real samples of different scenes, different times of day, different lighting, different weather conditions, and other conditions to build a basic sample set.

[0023] Based on a small number of real samples and a large number of simulated samples, the relationship between scene conditions and image feature mapping is analyzed, and different features are used to conduct intelligent detection and recognition test analysis. By comparing the degree of influence of different features on the accuracy of the detection and recognition algorithm, image features that are strongly coupled with the performance of the intelligent detection and recognition algorithm are found as the feature basis for sample augmentation. Through preprocessing such as image rotation, scaling, cropping, denoising, and image feature transformation operations such as geometric transformation, color transformation, and texture transformation, the basic sample set is augmented and the sample diversity is improved.

[0024] The augmented target materials and environmental materials are standardized and uniformly stored to form a material library required for sample synthesis.

[0025] The second step is to align the spatial position of the target and the environment in the sample By extracting the key features of the target material and environmental information, the feature points of the target and environmental information are aligned, and the transformation matrix between the target and the environment is calculated. The target is transformed into the environmental information coordinate system to achieve spatial position alignment.

[0026] Combined with the actual application scenario, the target type and environmental information are determined, and the key features of the target material and environmental information are extracted respectively.

[0027] Based on the extracted features, the feature points of the target and the environment information are registered, the corresponding feature point pairs are found, and the transformation matrix between the target and the environment is calculated.

[0028] Apply the change matrix to transform the target into the environmental information coordinate system to achieve spatial position alignment, clarify the position, size, and direction of the target in the environment, and ensure that the relative position of the target and the environment is relevant and reasonable.

[0029] Assume the position of the target in its own coordinate system is , the rotation angle relative to the environment coordinate is , the rotation matrix from the target system to the environment coordinate system is R,

[0030] If the translation vector is , the scaling factor is , then the change matrix T of the target transformation to the environment information coordinate system is: ; Coordinates after spatial position registration The calculation is as follows: ; The target position converted to the environment coordinate is , .

[0031] The third step is to perform adaptive synthesis of the target environment through threshold segmentation and background fusion technology The edge distribution of the target material is adjusted through threshold segmentation, and the background is fused according to the adaptive adjustment of the target and environmental materials to generate simulated real-shot synthetic samples, effectively reducing the problem of large edge gradient changes in the synthetic samples.

[0032] For the image-level sample synthesis of the target and the environment, according to the scene information such as the target type, lighting, weather, and environment contained in the expected synthetic sample, the target simulation material and the real-shot environment material related to the scene information are selected from the synthetic material library; Using the relative position relationship between the target and the environment in the sample determined by the spatial position registration in step 2, the threshold is set by calculating the gradient of the local area in the target material image (usually a 3×3 or 5×5 pixel area is selected as the local area), and the target part and the background part in the target material are segmented according to the threshold to reduce the interference of background information in the target material; The edge is determined by calculating the Sobel detection operator (3×3 pixel size), and the edge distribution of the target material is adjusted through Gaussian kernel filtering to perform fusion boundary smoothing optimization. Then the target and environmental materials are synthesized to generate simulated real-life synthetic samples, and the grayscale histogram of the synthetic samples is calculated. The contrast is adjusted through histogram equalization to make the grayscale value distribution more uniform, effectively reducing the problem of large gradient changes at the edges of synthetic samples.

[0033] For example, for the target material I, the edge features are calculated using the Sobel operator.

[0034] For the pixel points in the target material image , the Sobel operator performs 3×3 convolution along the X-axis and Y-axis respectively, and its gradient is: , in,

[0035] When the 3×3 convolution traverses the entire image, the edge of the image can be obtained.

[0036] After determining the edge of the image, the edge distribution of the target material is adjusted through Gaussian kernel filtering, and a 3×3 Gaussian kernel is used for average rate filtering. The Gaussian smoothing filter formula is:

[0037] After Gaussian smoothing filtering, the smooth optimization of the fusion boundary can be achieved.

[0038] After smoothing and optimizing the target material, the target and environment materials are synthesized to generate simulated real-life synthetic samples. The grayscale histogram of the synthetic samples is calculated, and the contrast is adjusted using histogram equalization to make the grayscale value distribution of the synthetic samples more uniform and reduce the edge distortion problem of the synthetic image.

[0039] This method proposes a sample augmentation method based on image feature domain. A synthetic feature library is constructed based on simulation modeling data and real-shot data. The scene domain is mapped to the image feature domain through the coupling relationship analysis between scene conditions and image features. Image feature transformation is used for sample augmentation, realizing the transformation from a qualitative augmentation method based on scene description to a quantitative controllable augmentation method based on feature domain, which effectively improves sample efficiency and sample diversity.

[0040] This method proposes a target environment adaptive synthesis method based on threshold segmentation and edge gradient adjustment. Based on the sample synthesis element library, spatial position registration is used to determine the relative position relationship between the target and the environment, which improves the rationality of the spatial position of the synthetic sample. By performing threshold segmentation and edge gradient adjustment on the target and environmental materials, the adaptive synthesis of the target and the environment is achieved, which effectively reduces the regional deformation and sample distortion problems of the synthetic samples.

[0041] In summary, this method constructs a feature library based on simulation modeling data and real-shot data, constructs a basic sample set of simulation targets and real-shot environments, and establishes a material library required for sample synthesis. According to the actual application scenario, the target style, environmental information, and the relative spatial position of the target and the environment in the required samples are analyzed, and the target foreground threshold segmentation is performed and adaptively fused with the background to generate high-confidence synthetic samples. This solves the problem of synthetic sample distortion in existing methods, and achieves the beneficial effects of improving the confidence of synthetic samples and reducing the construction of large-scale training sets. It has outstanding substantive characteristics and significant progress. Example 2

[0042] Please refer to Figure 2 , Figure 2 The following is a device for constructing a scene-to-image cross-domain mapping sample set driven by attribute features provided in this embodiment. In this embodiment, the device includes: A sample augmentation module, used to augment the basic sample set by using an image feature transformation method to obtain a material library for sample synthesis; the basic sample set includes simulated target materials and real-shot environment materials; The spatial registration module is used to perform spatial registration of the actual target and the actual environment in the actual application scenario to obtain the position information of the actual target in the environmental coordinate system; The actual material acquisition module is used to select the corresponding simulation target material and real-shot environment material from the material library based on the actual application scenario; The material fusion module is used to perform threshold segmentation on the selected simulated target material based on the position information of the actual target in the environmental coordinate system and fuse it with the real-shot environmental material to obtain a high-confidence synthetic sample.

[0043] Optionally, also include: A simulation target material acquisition module is used to establish a target model and generate a number of simulation target materials using a simulation modeling method; the simulation target materials include a target type, a target scale, and a target direction; The real-shot environment material acquisition module is used to acquire the real-shot environment material by adopting the real-shot acquisition method; the real-shot environment material includes scene information, time of day information, lighting information and weather information.

[0044] Optional, sample augmentation modules include: An image feature mapping unit, used to analyze the mapping relationship between scene conditions and image features of a basic sample set to obtain augmented image features; the augmented image features are image features that are strongly coupled with an intelligent detection and recognition algorithm; The image augmentation unit is used to perform image preprocessing and image feature transformation on the basic sample set based on the augmented image features to obtain a material library for sample synthesis.

[0045] Optional spatial registration modules include: An actual scenario determination unit, used to determine an actual target and an actual environment based on an actual application scenario; A feature extraction unit is used to extract features of the actual target and the actual environment to obtain key features; A transformation matrix calculation unit, used to perform feature point registration of the actual target and the actual environment based on the key features, and obtain a transformation matrix for converting from the target coordinate system to the environment coordinate system; The spatial registration unit is used to transform the coordinates of the actual target into the environment coordinate system based on the transformation matrix, and obtain the position information of the actual target in the environment coordinate system after spatial registration.

[0046] Optionally, the key features include a rotation angle, a translation vector, and a scaling ratio of the actual target relative to the actual environment.

[0047] Optionally, the transformation matrix calculation unit includes: A rotation matrix calculation subunit, used to obtain a rotation matrix for converting from a target coordinate system to an environment coordinate system based on position information of the actual target in the target coordinate system and a rotation angle of the actual target relative to the actual environment; The coordinate system conversion subunit is used to obtain a transformation matrix from a target coordinate system to an environment coordinate system based on a translation vector, a scaling factor, and a rotation matrix.

[0048] Optional, material fusion modules include: A segmentation threshold determination unit, used to calculate the gradient of a local area in the image of the simulated target material and determine the segmentation threshold based on the position information of the actual target in the environmental coordinate system; The target and background segmentation unit is used to segment the image of the simulated target material into a target part and a background part based on a segmentation threshold; An edge smoothing optimization unit, used for determining the edge of the segmented simulation target material based on a Sobel detection operator and performing edge smoothing optimization to obtain a smoothed simulation target material; A material fusion unit is used to synthesize the smoothed simulation target material with the real-shot environment material to generate a simulation-real-shot synthesis sample; The contrast adjustment unit is used to adjust the contrast of the simulated real-shot synthetic sample to obtain a high-confidence synthetic sample.

[0049] Optionally, the edge smoothing optimization unit includes: A convolution subunit, used to perform convolution on the simulation target material along the X-axis and the Y-axis respectively using a Sobel detection operator to determine the image edge of the simulation target material; The filtering subunit is used to perform edge smoothing optimization on the image edge of the simulation target material through Gaussian kernel filtering to obtain the smoothed simulation target material.

[0050] Optionally, the contrast adjustment unit comprises: A histogram statistics subunit, used to perform grayscale histogram statistics on the simulated real-shot synthetic samples; The contrast adjustment subunit is used to adjust the contrast of the simulated real-shot synthetic sample based on the grayscale histogram and using the histogram equalization method to obtain a high-confidence synthetic sample.

[0051] Based on this, the device constructs an element library based on simulation modeling data and real-shot data, constructs a basic sample set of simulation targets and real-shot environments, and establishes a material library required for sample synthesis. According to the actual application scenario, the target style, environmental information, and the relative spatial position of the target and the environment in the required samples are analyzed, and the target foreground threshold segmentation and adaptive fusion with the background are performed to generate high-confidence synthetic samples. This solves the problem of synthetic sample distortion in existing methods, and achieves the beneficial effects of improving the confidence of synthetic samples and reducing the construction of large-scale training sets. It has outstanding substantive characteristics and significant progress. Example 3

[0052] Please refer to Figure 3 , this embodiment provides an electronic device, which includes a processor, an internal bus, a network interface, a memory and a non-volatile memory, and of course may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a method for constructing a scene-to-image cross-domain mapping sample set driven by attribute features at the logical level. Of course, in addition to software implementations, this specification does not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0053] The network interface, processor and memory can be connected to each other through a bus system. The above bus can be divided into an address bus, a data bus, a control bus, etc.

[0054] The memory is used to store programs. Specifically, the program may include program code, and the program code includes computer operation instructions. The memory may include a read-only memory and a random access memory, and provides instructions and data to the processor.

[0055] The processor is used to execute the program stored in the above memory, and specifically perform: Step 102, using an image feature transformation method to augment the basic sample set to obtain a material library for sample synthesis; the basic sample set includes simulation target materials and real-shot environment materials; Step 104, performing spatial position registration on the actual target and the actual environment in the actual application scene to obtain position information of the actual target in the environment coordinate system; Step 106, selecting corresponding simulation target materials and real-shot environment materials from a material library based on the actual application scenario; Step 108 , based on the position information of the actual target in the environmental coordinate system, threshold segmentation is performed on the selected simulated target material and fused with the real-shot environmental material to obtain a high-confidence synthetic sample.

[0056] The processor may be an integrated circuit chip with signal processing capability. In the implementation process, each step of the above method may be completed by an integrated logic circuit of the processor hardware or by instructions in the form of software.

[0057] Based on the same invention, the embodiment of this specification also provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple application programs, the electronic device executes Figure 1 The corresponding embodiment provides a method for constructing a scene-to-image cross-domain mapping sample set driven by attribute features.

[0058] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-readable storage media containing computer-usable program code.

[0059] In addition, for the above-mentioned specific implementation of the system, since it is basically similar to the method implementation, the description is relatively simple, and the relevant parts can be referred to the partial description of the method implementation. Moreover, it should be noted that in each module of the system of the present application, the components therein are logically divided according to the functions to be implemented, but the present application is not limited thereto, and the components can be re-divided or combined as needed.

[0060] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other. Each embodiment focuses on the differences between it and other embodiments.

[0061] The above is a description of a specific embodiment of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the specific order or sequential order shown in the process depicted in the drawings is not necessarily required to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0062] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for constructing a scene-to-image cross-domain mapping sample set driven by attribute features, characterized in that: include: The basic sample set is augmented by using the image feature transformation method to obtain a material library for sample synthesis; The basic sample set includes simulation target materials and real-shot environment materials; Perform spatial position registration on the actual target and the actual environment in the actual application scenario to obtain the position information of the actual target in the environmental coordinate system; Select corresponding simulation target materials and real-life environment materials from the material library based on actual application scenarios; Based on the position information of the actual target in the environmental coordinate system, the selected simulated target material is threshold segmented and fused with the real-shot environmental material to obtain a high-confidence synthetic sample.

2. The method according to claim 1, characterized in that Before the basic sample set is augmented by the image feature transformation method to obtain a material library for sample synthesis, the method further includes: A simulation modeling method is adopted to establish a target model and generate a number of simulation target materials; the simulation target materials include a target type, a target scale and a target direction; A real-shot collection method is adopted to obtain real-shot environment materials; the real-shot environment materials include scene information, time of day information, lighting information and weather information.

3. The method according to claim 1, characterized in that The method of using the image feature transformation method to augment the basic sample set to obtain a material library for sample synthesis includes: Analyzing the mapping relationship between scene conditions and image features on the basic sample set to obtain augmented image features; the augmented image features are image features that are strongly coupled with the intelligent detection and recognition algorithm; Based on the augmented image features, image preprocessing and image feature transformation are performed on the basic sample set to obtain a material library for sample synthesis.

4. The method according to claim 1, characterized in that: The spatial position registration of the actual target and the actual environment in the actual application scenario to obtain the position information of the actual target in the environment coordinate system includes: Determine the actual goals and actual environment based on the actual application scenarios; Extract features from actual targets and actual environments to obtain key features; Based on the key features, the actual target and the actual environment are registered with feature points to obtain the transformation matrix from the target coordinate system to the environment coordinate system; Based on the transformation matrix, the coordinates of the actual target are converted to the environmental coordinate system to obtain the position information of the actual target in the environmental coordinate system after spatial registration.

5. The method according to claim 4, characterized in that The key features include the rotation angle, translation vector and scaling ratio of the actual target relative to the actual environment.

6. The method according to claim 5, characterized in that The step of performing feature point registration on the actual target and the actual environment based on the key features to obtain a transformation matrix from the target coordinate system to the environment coordinate system includes: Based on the position information of the actual target in the target coordinate system and the rotation angle of the actual target relative to the actual environment, a rotation matrix for converting from the target coordinate system to the environment coordinate system is obtained; Based on the translation vector, scaling factor, and rotation matrix, obtain the transformation matrix from the target coordinate system to the environment coordinate system.

7. The method according to claim 1, characterized in that The step of performing threshold segmentation on the selected simulated target material based on the position information of the actual target in the environmental coordinate system and fusing it with the real-shot environmental material to obtain a high-confidence synthetic sample includes: Based on the position information of the actual target in the environment coordinate system, the gradient of the local area in the image of the simulated target material is calculated and the segmentation threshold is determined; Segmenting the target part and the background part of the image of the simulated target material based on the segmentation threshold; Determine the edge of the segmented simulation target material based on the Sobel detection operator and perform edge smoothing optimization to obtain the smoothed simulation target material; The smoothed simulation target material is synthesized with the real-shot environment material to generate a simulation-real-shot synthesis sample; The contrast of the simulated real-shot synthetic samples is adjusted to obtain high-confidence synthetic samples.

8. The method according to claim 7, characterized in that The step of determining the edge of the segmented simulation target material based on the Sobel detection operator and performing edge smoothing optimization to obtain the smoothed simulation target material comprises: The Sobel detection operator is used to convolve the simulation target material along the X-axis and the Y-axis respectively to determine the image edge of the simulation target material; The edge of the image of the simulation target material is smoothed and optimized through Gaussian kernel filtering to obtain the smoothed simulation target material.

9. The method according to claim 8, characterized in that The step of adjusting the contrast of the simulated real shot synthetic sample to obtain a high confidence synthetic sample comprises: Perform grayscale histogram statistics on the simulated real-shot synthetic samples; Based on the grayscale histogram, the contrast of the simulated real-shot synthetic samples is adjusted using the histogram equalization method to obtain high-confidence synthetic samples.

10. A device for constructing a scene-to-image cross-domain mapping sample set driven by attribute features, characterized in that: include: A sample augmentation module, used to augment the basic sample set by using an image feature transformation method to obtain a material library for sample synthesis; the basic sample set includes simulated target materials and real-shot environment materials; The spatial registration module is used to perform spatial registration of the actual target and the actual environment in the actual application scenario to obtain the position information of the actual target in the environmental coordinate system; The actual material acquisition module is used to select the corresponding simulation target material and real-shot environment material from the material library based on the actual application scenario; The material fusion module is used to perform threshold segmentation on the selected simulated target material based on the position information of the actual target in the environmental coordinate system and fuse it with the real-shot environmental material to obtain a high-confidence synthetic sample.

Images