Tunnel face structural plane segmentation method and system
By using a multi-dimensional reshaping and fusion segmentation model, the problems of low efficiency of traditional methods and high computational load of deep learning models in tunnel construction are solved, achieving accurate and efficient segmentation of the tunnel face structure and improving recognition accuracy and real-time performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA CONSTR RAILWAY INVESTMENT & CONSTR GRP CO LTD
- Filing Date
- 2026-01-14
- Publication Date
- 2026-05-01
AI Technical Summary
In tunnel construction, existing technologies rely on traditional methods that are inefficient due to reliance on manual observation and are easily affected by subjective factors. Image recognition technology has limited accuracy under complex conditions, and deep learning models have a large computational load that makes it difficult to meet real-time requirements and achieve accurate and efficient segmentation of the tunnel face structure.
A multi-dimensional reshaping and fusion segmentation model is adopted, which combines a two-stage specific feature extraction module and a full-grained feature mapping and decoding module with specific feature recognition and mapping decoding processing to achieve accurate segmentation of the face of the tunnel.
It improves the ability to extract structural surface features under complex conditions, realizes accurate and efficient segmentation of the tunnel face structure, and enhances the accuracy and real-time performance of recognition.
Smart Images
Figure CN121962906A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field of face segmentation, and particularly relates to a method and system for segmenting the structural surface of a face. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] During tunnel construction, the identification of the structural surfaces at the tunnel face is crucial for construction safety and quality control. Traditional methods mainly rely on manual observation and experience-based judgment, which is inefficient and susceptible to subjective factors, leading to inaccurate identification results. With the development of computer vision and artificial intelligence technologies, image-based intelligent recognition methods have gradually become a research hotspot for structural surface identification.
[0004] Existing image recognition technologies face several challenges when applied to tunnel face structure identification. Firstly, traditional image processing algorithms struggle to effectively extract features from complex tunnel face images, resulting in limited recognition accuracy. Secondly, while deep learning methods improve accuracy to some extent, their high model complexity and computational demands make them unsuitable for real-time requirements. Furthermore, the specific scenario of tunnel face structure segmentation presents challenges such as data noise, varying lighting conditions, and the complexity and diversity of the structures.
[0005] In summary, improving the feature extraction capability of tunnel face structures under complex conditions, and further achieving accurate and efficient identification of tunnel face structures, is a problem that needs to be solved. Summary of the Invention
[0006] To overcome the shortcomings of the prior art, the present invention provides a method and system for segmenting the structural surface of the tunnel face, thereby achieving accurate and efficient segmentation of the structural surface of the tunnel face.
[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for segmenting the structural surface of a working face, comprising: The image of the working face is acquired and then input into the trained segmentation model to obtain the segmentation result of the working face structure. The segmentation model includes multiple two-stage specific feature extraction modules; the processing procedure of the two-stage specific feature extraction modules for the input face image is as follows: The face image is reshaped in different dimensions to obtain face images of different dimensions after reshaping. The face images of different dimensions after reshaping are then fused to obtain fused features. The fused features are subjected to specific feature recognition to obtain specific feature recognition results; The specific feature recognition results output by each of the two-stage specific feature extraction modules are subjected to feature mapping and decoding processing to obtain output features. Pixel point recognition is performed based on the output features to obtain the face structure segmentation result.
[0008] In a second aspect, the present invention provides a face segmentation system for a working face structure, comprising: The acquisition module is configured to acquire images of the working face. Segmentation module: It is configured to input the acquired face image into the trained segmentation model to obtain the face structure segmentation result; The segmentation model includes multiple two-stage specific feature extraction modules; the processing procedure of the two-stage specific feature extraction modules for the input face image is as follows: The face image is reshaped in different dimensions to obtain face images of different dimensions after reshaping. The face images of different dimensions after reshaping are then fused to obtain fused features. The fused features are subjected to specific feature recognition to obtain specific feature recognition results; The specific feature recognition results output by each of the two-stage specific feature extraction modules are subjected to feature mapping and decoding processing to obtain output features. Pixel point recognition is performed based on the output features to obtain the face structure segmentation result.
[0009] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.
[0010] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.
[0011] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.
[0012] The above one or more technical solutions have the following beneficial effects: In this invention, the face image is reshaped in different dimensions to obtain face images of different dimensions after reshaping. The face images of different dimensions after reshaping are then fused. By reshaping the image dimensions, image features can be extracted from different scales, enhancing the recognition of face details. The specific feature recognition results output by each two-stage specific feature extraction module are respectively subjected to feature mapping decoding to obtain output features. Pixel recognition is performed based on the output features, which can achieve effective segmentation of structural surfaces in the face image.
[0013] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0014] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0015] Figure 1 This is a schematic diagram of the network structure of the two-stage specific feature extraction module in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the network structure of the two-stage specific feature extraction module in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the network structure of the full-granularity feature mapping decoding module in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the specific feature fusion attention network structure in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the final features output by the segmentation model in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the network structure of feature extraction module one and feature extraction module three in Embodiment 1 of the present invention; Figure 7 This is a schematic diagram of the network structure of feature extraction module 2 and feature extraction module 4 in Embodiment 1 of the present invention. Detailed Implementation
[0016] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0017] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.
[0018] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.
[0019] Example 1 This embodiment discloses a method for segmenting the structural surface of a tunnel face, including: The image of the working face is acquired and then input into the trained segmentation model to obtain the segmentation result of the working face structure. The segmentation model includes multiple two-stage specific feature extraction modules connected in sequence; the processing procedure of the two-stage specific feature extraction modules for the input face image is as follows: The face image is reshaped in different dimensions to obtain face images of different dimensions after reshaping. The face images of different dimensions after reshaping are then fused to obtain fused features. Specific feature identification is performed on the fused features to obtain the specific feature identification results; The specific feature recognition results output by each of the two-stage specific feature extraction modules are subjected to feature mapping and decoding processing to obtain output features. Pixel point recognition is performed based on the output features to obtain the face structure segmentation results.
[0020] The segmentation model proposed in this embodiment will be described in detail below: like Figure 4 As shown, the segmentation model includes four two-stage specific feature extraction modules and one full-granularity feature mapping decoding module. First, it receives the face image as input and converts it into vector data that the segmentation model can process, denoted as... , The dimension is .Will The input is processed by the two-stage specific feature extraction module 1, and the output is the feature. ,feature The dimension is .Will The input is processed by the two-stage specific feature extraction module 2, and the output is the feature. ,feature The dimension is .Will The input is processed by the two-stage specific feature extraction module three, and the output is the feature. ,feature The dimension is .Will The input is processed by the two-stage specific feature extraction module 4, and the output is the feature. ,feature The dimension is Features ,feature ,feature ,feature Input the full-granularity feature mapping decoding module and output the feature. ,feature The dimension is .feature This is the final output feature.
[0021] Accurate segmentation of structural surfaces in facet images: using the final features output by the segmentation model. As a basis, The dimension is The image size is also , Each data point in the image represents a single pixel. Each data point has only two choices: 0 or 1. 1 represents the foreground (structural plane), and 0 represents the background (other parts). If the data is 0, the RGB value of the corresponding image pixel becomes (0, 0, 0); if the data is 1, the RGB value remains unchanged. The final output is the modified image, achieving segmentation. This utilizes features. Multiplying the image by the original image preserves the original values for regions corresponding to 1 and sets the values for regions corresponding to 0 to 0. This method can effectively segment structural surface regions in an image.
[0022] For example, the output 0 is Figure 5 In the middle, the purple part is 1, and the white part is 1. In the end, the purple part pixels will be changed to RGB values (0, 0, 0), while the white part will retain its original RGB values without change.
[0023] Two-stage specific feature extraction module, such as Figure 2 As shown, the two-stage specific feature extraction module includes a progressive fusion attention module and a specific feature recognition module. The two-stage specific feature extraction module receives image input and converts it into a processable feature vector. This vector is then input to the progressive fusion attention module to obtain its feature output. This feature output is then input to the specific feature recognition module to obtain its feature output.
[0024] In the progressive fusion attention module, feature input is received. The feature input dimension is Input features Input feature extraction module one, output the first feature First feature The dimension is .
[0025] Input features Input feature extraction module two, output the second feature Second feature The dimension is .
[0026] Input features Input dimension reshaping module one, output the third feature Third feature The dimension is The first feature Second feature Multiplying them together, the output yields the fourth feature. Fourth feature The dimension is The third feature and the fourth feature Multiplying them together, the output yields the fifth feature. Fifth feature The dimension is The fifth feature Input dimension reshaping module two, output the sixth feature Sixth feature The dimension is .
[0027] Input features Input feature extraction module two, output the seventh feature The seventh feature The dimension is .
[0028] Input features Input to feature extraction module four, output the eighth feature. The eighth feature The dimension is .
[0029] Input features Input dimension reshaping module three, output the ninth feature. Ninth feature The dimension is The seventh feature and the eighth feature Multiplying them together, the output yields the tenth feature. The tenth feature The dimension is The ninth feature and the tenth feature Multiplying them together, the output yields the eleventh feature. Eleventh feature The dimension is .
[0030] The eleventh feature Input the fourth dimension reshaping module, and output the twelfth feature. The twelfth feature The dimension is .
[0031] The sixth feature and input features Adding them together yields the thirteenth characteristic. Thirteenth feature The dimension is The twelfth feature and input features Adding them together yields the fourteenth feature. , The dimension is The thirteenth feature and Adding them together yields the fifteenth characteristic of the output. The fifteenth feature The dimension is .
[0032] Among them, feature extraction module one and feature extraction module three are multi-convolutional branch feature extraction networks, such as Figure 6 As shown. The input is denoted as... It is divided into three branches. In the first branch, the ordinary convolution module one contains an ordinary convolution module and a pooling module, which takes the input... The input is from a regular convolutional module, and the output is... The second ordinary convolutional module contains one ordinary convolutional module and one pooling module, which converts features... Input into ordinary convolution module 2, output is The third ordinary convolutional module contains one ordinary convolutional module and one pooling module, which converts features... Inputting into a regular convolutional module 3, the output is... In the second branch, the input will be... The input is a dilated convolutional module, and the output is the feature. In the third branch, the input will be... The input is processed by the dilated convolution module, and the output is the feature. Features ,feature and characteristics The input concatenation module outputs features. The fourth ordinary convolutional module contains one ordinary convolutional module and one pooling module, which converts features... The input is taken into a regular convolutional module 4, and the output is the feature. Features In the input dimension reshaping module, the output is the feature. This is the final output of this module.
[0033] Among them, feature extraction module two and feature extraction module four are skip pyramid feature extraction networks, such as Figure 7 As shown, the input is denoted as The first ordinary convolution module contains one ordinary convolution module and one pooling module, which will process the input... The input is taken into a regular convolutional module 1, and the output is the feature. Features The input is the dilated convolution module 1, and the output is the feature. Features The input is processed by a high-voidity dilated convolution module, and the output is the feature. Features The input to the upsampling module 1 is used to obtain the features. Features and characteristics Add them together and output the features. Features The input is processed by the second dilated convolution module, and the output is the feature. Features The input to the second upsampling module is used to obtain the features. The second ordinary convolutional module contains one ordinary convolutional module and one pooling module, which converts features... The input is processed by a regular convolutional module 2, and the output is the feature. Input With features Adding together yields the features Features Input the dimension reshaping module to obtain features. This is the final output of this module.
[0034] This embodiment utilizes multi-dimensional fusion to enable the model to comprehensively learn data features from multiple perspectives. Coarse-grained features are suitable for segmentation region localization, while fine-grained features are suitable for accurate pixel classification. Multi-dimensional fusion can improve the model's final segmentation performance.
[0035] In the specific feature extraction module, the fifteenth feature receives the feature output from the progressive fusion attention module. The fifteenth feature The input is the first fully connected layer, and the output is the sixteenth feature. The sixteenth feature The dimension is The sixteenth feature The input is fed into the convolutional layer, and the output yields the seventeenth feature. The seventeenth feature The dimension is The seventeenth feature The input activation function outputs the eighteenth feature. The eighteenth feature The dimension is The eighteenth feature The input is the second fully connected layer, and the output is the nineteenth feature. The nineteenth feature The dimension is .
[0036] In tunnel scenes, structural surfaces in images typically lack a specific pattern, requiring identification based on their unique features. This module, by fusing features from different scales into features of the same scale—essentially a feature fusion process—improves the sensitivity for identifying structural surfaces in images.
[0037] Build a full-granularity feature mapping decoding module. The full-granularity feature mapping decoding module is as follows: Figure 3 As shown. The full-granularity feature map decoding module receives four inputs, namely... , , , . The dimension is , The dimension is , The dimension is , The dimension is .Will The input is a fully connected layer 1, and the output is the feature. ,feature The dimension is .Will The input is processed through a fully connected layer 2, and the output is the feature. ,feature The dimension is .Will The input is obtained from the third fully connected layer, and the output is the feature. ,feature The dimension is .Will The input is a fully connected layer 4, and the output is the feature. ,feature The dimension is Features The input is the first upsampling layer, and the output is the feature. ,feature The dimension is Features The input is obtained from the second upsampling layer, and the output is the feature. ,feature The dimension is Features The input is obtained from the third upsampling layer, and the output is the feature. ,feature The dimension is Features The input is from the fourth upsampling layer, and the output is the feature. ,feature The dimension is Features ,feature ,feature ,feature The input feature concatenation module outputs the features. ,feature The dimension is Features The input is processed by a multilayer perceptron module, and the output is the feature. ,feature The dimension is .
[0038] In terms of spatial consistency, large-scale features, by fusing small-scale details, can make the recognition results approximate the actual location of the structural surfaces in the original image as closely as possible. Typically, the color gamut of images within a tunnel varies considerably, with some images appearing grayish and others yellowish, which is related to the color of the supplementary lighting. Furthermore, the color gamut varies significantly across different regions within a single image. Therefore, in terms of generalization, enabling the model to learn features at different scales can improve its generalization ability across tunnel face images with different color gamuts.
[0039] To train the segmentation model, a dataset was constructed by collecting tunnel face data and structural surface segmentation data, including: S11: Set up two 200W LED projector stands about 5m in front of the working face, one on each side, to evenly illuminate the working face during shooting and reduce the impact of shadows and uneven lighting on image quality.
[0040] S12: Mount the camera on a tripod, assuming the tripod is 6 meters directly in front of the tunnel face. Adjust the camera's aperture, shutter speed, and ISO according to the lighting conditions inside the tunnel to ensure proper exposure and image sharpness. Save the images obtained from the phone to a storage device.
[0041] S13: Select images from the captured images, choosing complete images of the tunnel face. Professional geologists then use Labelme annotation software to annotate the structural surfaces. Store the original images and annotation labels on a storage device.
[0042] S14: Data augmentation: First, geometric transformations are employed, including mirror flipping (horizontal and vertical flipping), random rotation within the range of -45° to 45°, and random cropping. The number of images subjected to geometric transformations accounts for 40% of the total data volume after enhancement.
[0043] Second, color transformation was performed by adjusting parameters such as brightness, contrast, saturation, and hue of the images. The number of images with color transformation accounted for 30% of the total data volume after enhancement.
[0044] Third, Gaussian noise and salt-and-pepper noise were added to the images, with the number of images with added noise accounting for 20% of the total data volume after enhancement.
[0045] Fourth, light processing is performed, including operations such as mean white balance, grayscale assumption, adaptive correction, mixed color cast correction, and histogram equalization. The number of images processed by light accounts for 10% of the total data volume after enhancement.
[0046] S15: After data collection, data representation, and data augmentation, a total of 3,000 images and 3,000 labels were obtained. These were divided into a test set and a training set at a ratio of 80% and 20%, respectively. The training set included 2,400 images and 2,400 JSON label data, while the test set included 600 images and 600 JSON label data.
[0047] Example 2 The purpose of this embodiment is to provide a face segmentation system for the working face, including: The acquisition module is configured to acquire images of the working face. Segmentation module: It is configured to input the acquired face image into the trained segmentation model to obtain the face structure segmentation result; The segmentation model includes multiple two-stage specific feature extraction modules; the processing procedure of the two-stage specific feature extraction modules for the input face image is as follows: The face image is reshaped in different dimensions to obtain face images of different dimensions after reshaping. The face images of different dimensions after reshaping are then fused to obtain fused features. The fused features are subjected to specific feature recognition to obtain specific feature recognition results; The specific feature recognition results output by each of the two-stage specific feature extraction modules are subjected to feature mapping and decoding processing to obtain output features. Pixel point recognition is performed based on the output features to obtain the face structure segmentation result.
[0048] In further embodiments, the following is also provided: An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When executed by the processor, the computer instructions perform the method described in Embodiment 1. For brevity, further details are omitted here.
[0049] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.
[0050] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.
[0051] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.
[0052] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.
[0053] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.
[0054] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.
[0055] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.
[0056] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.
[0057] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0058] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.
Claims
1. A method for segmenting the structural surface of a tunnel face, characterized in that, include: The image of the working face is acquired and then input into the trained segmentation model to obtain the segmentation result of the working face structure. The segmentation model includes multiple two-stage specific feature extraction modules; the processing procedure of the two-stage specific feature extraction modules for the input face image is as follows: The face image is reshaped in different dimensions to obtain face images of different dimensions after reshaping. The face images of different dimensions after reshaping are then fused to obtain fused features. The fused features are subjected to specific feature recognition to obtain specific feature recognition results; The specific feature recognition results output by each of the two-stage specific feature extraction modules are subjected to feature mapping and decoding processing to obtain output features. Pixel point recognition is performed based on the output features to obtain the face structure segmentation result.
2. The method for segmenting the structural surface of a tunnel face as described in claim 1, characterized in that, The tunnel face image is reshaped in different dimensions to obtain reshaped tunnel face images of different dimensions. These reshaped tunnel face images of different dimensions are then fused to obtain fused features, specifically: Feature extraction and dimension reshaping were performed on the face images respectively; The extracted features and the reshaped features are multiplied together, and the dimensions of the multiplication result are reshaped again. The features after dimensional reshaping are fused with the face image to obtain fused features.
3. The method for segmenting the structural surface of a tunnel face as described in claim 1, characterized in that, The fused features are subjected to specific feature recognition to obtain specific feature recognition results, specifically as follows: The fused features are processed sequentially through a fully connected layer 1, a convolutional layer, an activation function, and a fully connected layer 2 to obtain the specific feature recognition result.
4. The method for segmenting the structural surface of a tunnel face as described in claim 1, characterized in that, The specific feature recognition results output by each of the two-stage specific feature extraction modules are subjected to feature mapping and decoding processing to obtain the output features, specifically: The specific feature recognition results output by each two-stage specific feature extraction module are processed through a fully connected layer and an upsampling layer, respectively. The features processed by the upsampling layer are concatenated, and the concatenated features are input into a multilayer perceptron to obtain the output features.
5. The method for segmenting the structural surface of a tunnel face as described in claim 1, characterized in that, The training of the segmentation model specifically includes: Images of the tunnel face were collected and structural surfaces were labeled to create a dataset. Data augmentation was performed on the dataset through geometric transformation, color transformation, noise addition, and lighting processing. The segmentation model is trained using the data augmented dataset to obtain a trained segmentation model.
6. The method for segmenting the structural surface of a tunnel face as described in claim 2, characterized in that, Feature extraction is performed on the input face image using a multi-convolutional branch feature extraction network and a jump pyramid feature extraction network, respectively.
7. A face segmentation system for a tunnel face, characterized in that, include: The acquisition module is configured to acquire images of the working face. Segmentation module: It is configured to input the acquired face image into the trained segmentation model to obtain the face structure segmentation result; The segmentation model includes multiple two-stage specific feature extraction modules; the processing procedure of the two-stage specific feature extraction modules for the input face image is as follows: The face image is reshaped in different dimensions to obtain face images of different dimensions after reshaping. The face images of different dimensions after reshaping are then fused to obtain fused features. The fused features are subjected to specific feature recognition to obtain specific feature recognition results; The specific feature recognition results output by each of the two-stage specific feature extraction modules are subjected to feature mapping and decoding processing to obtain output features. Pixel point recognition is performed based on the output features to obtain the face structure segmentation result.
8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-6.
9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-6.
10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-6.