Fabrication system and method of ultra-thick 5g antenna module

Abstract

The application relates to the field of intelligent processing, and particularly discloses a manufacturing system and method for a super-thick 5G antenna module, which uses artificial intelligence technology based on a deep neural network model to perform feature coding and extraction on a detection positioning image and a reference positioning image of a super-thick structure, so as to obtain a classification label for indicating whether the positioning of the super-thick structure is accurate, thereby constructing a drilling processing scheme for a Teflon circuit board to analyze the detection image and the reference image of the super-thick structure, improving the positioning accuracy of the super-thick structure, and reducing labor costs.

Description

Technical Field

[0001] This application relates to the field of intelligent manufacturing, and more specifically to a system and method for manufacturing an ultra-thick 5G antenna module. Background Technology

[0002] As the fifth generation of mobile communication technology, 5G networks, due to their high speed, low latency, and low power consumption, will penetrate into the Internet of Things and various industries in the future, deeply integrating with industrial facilities, medical instruments, and transportation vehicles to effectively meet the diverse business needs of vertical industries such as industry, healthcare, and transportation. Simultaneously, with the rapid arrival of the 5G network era, the demand for its core component, the 5G antenna module, is also increasing. To meet its special performance requirements, some antenna modules have been designed with a thickness exceeding 11.5mm. For such ultra-thick plates, significant technical bottlenecks are faced in processes such as lamination, drilling, electroplating, circuitry, and CNC machining. Ultra-thick plate lamination technology is a method for fabricating ultra-thick structures, used to manufacture ultra-thick 5G antenna modules. Its specific steps include: material preparation, design and fabrication of conductor layers, preparation of ultra-thick structural layers, lamination, cutting and processing, connection and packaging, testing, and debugging. During cutting and processing, the material to be cut needs to be positioned to ensure the cutting effect. However, conventional positioning methods usually require operators with certain skills and experience to ensure accurate positioning and fixation of the ultra-thick structure. Operators unfamiliar with positioning methods may lead to inaccurate or unstable positioning.

[0003] Therefore, we look forward to an optimized manufacturing solution for ultra-thick 5G antenna modules. Summary of the Invention

[0004] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a system and method for fabricating an ultra-thick 5G antenna module. This system utilizes artificial intelligence technology based on a deep neural network model to encode and extract features from the detection and positioning images and reference positioning images of the ultra-thick structure, obtaining classification labels indicating the accuracy of the ultra-thick structure's positioning. By constructing a drilling and processing scheme for Teflon circuit boards to analyze the detection and reference images of the ultra-thick structure, the accuracy of ultra-thick structure positioning is improved, and labor costs are reduced.

[0005] According to one aspect of this application, a system for fabricating an ultra-thick 5G antenna module is provided, comprising:

[0006] The image acquisition module is used to acquire detection and positioning images and reference positioning images of the ultra-thick structure;

[0007] The preprocessing module is used to preprocess the detected positioning image to obtain a processed positioning image;

[0008] An image enhancement module is used to pass the processed positioning image through a high-definition image generator based on a generative adversarial network to obtain an enhanced positioning image;

[0009] A dual-stream detection module is used to pass the enhanced positioning image and the reference positioning image through a dual-stream detection model containing a first image encoder and a second image encoder to obtain a detection feature map and a reference feature map;

[0010] The difference module is used to calculate the difference between the detected feature map and the reference feature map to obtain a difference feature map;

[0011] An optimization module is used to optimize the difference feature map to obtain an optimized difference feature map;

[0012] The classification result generation module is used to pass the optimized differential feature map through a classifier to obtain a classification result, which is used to indicate whether the positioning of the ultra-thick structure is accurate.

[0013] In the aforementioned fabrication system for ultra-thick 5G antenna modules, the adversarial generative network includes a generator and a discriminator.

[0014] The image enhancement module is further configured to input the processed localization image into the high-definition image generator based on the generative adversarial network so that the generator of the high-definition image generator outputs the enhanced localization image through deconvolution encoding.

[0015] In the fabrication system of the ultra-thick 5G antenna module described above, the first image encoder and the second image encoder have the same mesh structure, and the first image encoder and the second image encoder are convolutional neural network models containing deep and shallow feature encoding modules.

[0016] In the aforementioned fabrication system for the ultra-thick 5G antenna module, the dual-stream detection module includes:

[0017] Detection of image coding units, including:

[0018] The first shallow feature extraction subunit is used to extract the detection shallow feature map from the Mth layer of the first convolutional neural network model, where M is greater than or equal to 1 and less than or equal to 6.

[0019] The first deep feature extraction subunit is used to extract detection deep feature maps from the Nth layer of the first convolutional neural network model, where N / M is greater than or equal to 5 and less than or equal to 10.

[0020] The first fusion subunit is used to fuse the detection shallow feature map and the detection deep feature map using the deep and shallow feature fusion module of the first convolutional neural network model to obtain a detection feature map.

[0021] In the aforementioned manufacturing system for the ultra-thick 5G antenna module, the dual-stream detection module further includes:

[0022] Reference image coding unit, including:

[0023] The second shallow feature extraction subunit is used to extract a reference shallow feature map from the Mth layer of the second convolutional neural network model, where M is greater than or equal to 1 and less than or equal to 6.

[0024] The second deep feature extraction subunit is used to extract a reference deep feature map from the Nth layer of the second convolutional neural network model, where N / M is greater than or equal to 5 and less than or equal to 10.

[0025] The second fusion subunit is used to fuse the reference shallow feature map and the reference deep feature map using the deep and shallow feature fusion module of the first convolutional neural network model to obtain a reference feature map.

[0026] In the aforementioned manufacturing system for an ultra-thick 5G antenna module, the differential module includes:

[0027] The difference feature map between the detected feature map and the reference feature map is calculated using the following difference formula;

[0028] The difference formula is as follows:

[0029]

[0030] Wherein, F1 represents the detected feature map, F2 represents the reference feature map, and F represents the differential feature map. This indicates differentiation based on position.

[0031] In the aforementioned manufacturing system for ultra-thick 5G antenna modules, the optimization module includes:

[0032] A spatial probabilization unit is used to fuse the detected feature map and the reference feature map based on affine subspace probabilization to obtain a fused feature map.

[0033] An optimization unit is used to fuse the fused feature map and the differential feature map to obtain an optimized differential feature map.

[0034] In the aforementioned fabrication system for ultra-thick 5G antenna modules, the spatial probabilistic unit includes:

[0035] The feature map unpacking subunit is used to unpack the detection feature map and the reference feature map to obtain the detection feature vector and the reference feature vector;

[0036] The standardization subunit is used to standardize the detection feature vector and the reference feature vector to obtain a standardized detection feature vector and a standardized reference feature vector.

[0037] The covariance subunit is used to calculate the covariance matrix between the standardized detection feature vector and the standardized reference feature vector.

[0038] A decomposition subunit is used to perform eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues ​​and multiple eigenvalue vectors corresponding to the multiple eigenvalues;

[0039] The basis extraction subunit is used to extract the feature value vectors corresponding to the two largest feature values ​​from the plurality of feature vectors as the first dimension feature vector and the second dimension feature vector, and the first dimension feature vector and the second dimension feature vector serve as the basis of the affine subspace;

[0040] A two-dimensional arrangement subunit is used to arrange the first-dimensional feature vector and the second-dimensional feature vector according to column vectors to obtain an affine subspace matrix;

[0041] The matrix multiplication subunit is used to perform matrix multiplication of the detected feature vector and the reference feature vector with the affine subspace matrix respectively to map the detected feature vector and the reference feature vector to the affine subspace matrix to obtain the affine transformation-like detected feature vector and the affine transformation-like reference feature vector.

[0042] The positional summation subunit is used to calculate the positional summation of the affine transformation detection feature vector and the affine transformation reference feature vector to obtain the fused feature vector;

[0043] The dimension reconstruction subunit is used to perform dimension reconstruction on the fused feature vector to obtain a fused feature map.

[0044] In the aforementioned manufacturing system for ultra-thick 5G antenna modules, the classification result generation module includes:

[0045] The expansion unit is used to expand the optimized difference feature map into a classification feature vector according to the row vector or column vector;

[0046] A fully connected encoding unit is used to perform fully connected encoding on the classification feature vector using the fully connected layer of the classifier to obtain an encoded classification feature vector;

[0047] The classification result generation unit is used to input the encoded classification feature vector into the Softmax classification function of the classifier to obtain the classification result.

[0048] According to another aspect of this application, a method for manufacturing an ultra-thick 5G antenna module is also provided, comprising:

[0049] Acquire detection and localization images and reference localization images of ultra-thick structures;

[0050] The detected positioning image is preprocessed to obtain the processed positioning image;

[0051] The processed localization image is then passed through a high-definition image generator based on a generative adversarial network to obtain an enhanced localization image;

[0052] The enhanced localization image and the reference localization image are processed by a two-stream detection model that includes a first image encoder and a second image encoder to obtain a detection feature map and a reference feature map.

[0053] The difference between the detected feature map and the reference feature map is calculated to obtain a difference feature map;

[0054] The differential feature map is optimized to obtain an optimized differential feature map;

[0055] The optimized differential feature map is passed through a classifier to obtain a classification result, which is used to indicate whether the localization of the ultra-thick structure is accurate.

[0056] Compared with existing technologies, the fabrication system and method for ultra-thick 5G antenna modules provided in this application use artificial intelligence technology based on deep neural network models to encode and extract features from the detection and positioning images and reference positioning images of ultra-thick structures to obtain classification labels indicating whether the positioning of ultra-thick structures is accurate. In this way, by constructing a drilling and processing scheme for Teflon circuit boards to analyze the detection and reference images of ultra-thick structures, the accuracy of ultra-thick structure positioning is improved and labor costs are reduced. Attached Figure Description

[0057] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0058] Figure 1 The figure shows a block diagram of a system for fabricating an ultra-thick 5G antenna module according to an embodiment of this application.

[0059] Figure 2 The figure illustrates a system architecture diagram of a fabrication system for an ultra-thick 5G antenna module according to an embodiment of this application.

[0060] Figure 3 The figure shows a block diagram of the classification result generation module in the fabrication system of an ultra-thick 5G antenna module according to an embodiment of this application.

[0061] Figure 4 The illustration shows a flowchart of a method for manufacturing an ultra-thick 5G antenna module according to an embodiment of this application.

[0062] Figure 5 A block diagram of an electronic device according to an embodiment of this application is illustrated. Detailed Implementation

[0063] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0064] Application Overview

[0065] As mentioned in the background section, positioning of the material to be cut is necessary during cutting and processing to ensure the cutting effect. However, conventional positioning methods usually require operators with certain skills and experience to ensure accurate positioning and fixation of ultra-thick structures. Operators unfamiliar with positioning methods may cause inaccurate or unstable positioning. Therefore, an optimized manufacturing solution for ultra-thick 5G antenna modules is needed.

[0066] To address the aforementioned technical issues, a fabrication scheme for an ultra-thick 5G antenna module is proposed. This scheme utilizes artificial intelligence technology based on a deep neural network model to encode and extract features from the detection and positioning images and reference positioning images of the ultra-thick structure. This yields classification labels indicating the accuracy of the ultra-thick structure's positioning. Furthermore, by constructing a Teflon circuit board drilling process, the detection and reference images of the ultra-thick structure are analyzed, thereby improving the accuracy of ultra-thick structure positioning and reducing labor costs.

[0067] Currently, deep learning and neural networks have been widely applied in fields such as computer vision, natural language processing, and speech signal processing. Furthermore, deep learning and neural networks have demonstrated near-human or even superior performance in areas such as image classification, object detection, semantic segmentation, and text translation.

[0068] In recent years, the development of deep learning and neural networks has provided new ideas and solutions for the fabrication of ultra-thick 5G antenna modules.

[0069] Specifically, firstly, detection and reference positioning images of the ultra-thick structure are acquired. The reference positioning image serves as a standard in the fabrication system. It provides an ideal positioning result for comparison and contrast with the detection image. By comparing it with the reference positioning image, it can be determined whether the positioning of the ultra-thick structure meets the design requirements. The detection and reference positioning images provide a benchmark for evaluating the positioning accuracy of the ultra-thick structure. By comparing the detection and reference images, positioning errors and offsets can be determined.

[0070] Then, the detected localization image is preprocessed to obtain a processed localization image. Preprocessing removes noise, blur, and other unwanted interference from the image, thereby improving image clarity and quality. This helps reduce the impact of noise in the image on localization accuracy and makes the localized target more explicit and identifiable. By adjusting the brightness and contrast of the image, preprocessing enhances the difference between the localized target and the background, making it more prominent. This helps improve the accuracy of the localization algorithm, enabling it to better distinguish the target from the background and accurately locate the target's position. Preprocessing can standardize the image, giving it consistent features and attributes. This helps eliminate differences between different images, making them comparable. Through standardization, images acquired under different acquisition devices and conditions can have consistent features, thereby improving the consistency and reliability of localization.

[0071] Next, the processed localization image is passed through a high-resolution image generator based on a generative adversarial network (GAN) to obtain an enhanced localization image. The GAN-based high-resolution image generator can learn the features and structure of large amounts of image data to generate higher quality and clearer images. By inputting the processed localization image into the generator, an enhanced localization image with higher resolution and clearer details can be obtained. This helps improve the accuracy of the localization algorithm, making the target location more precise and reliable. During the processing of the localization image, some details and information may be lost. By using the high-resolution image generator, this lost information can be compensated for as much as possible, and the integrity of the image can be restored. This helps improve the quality and usability of the localization image, making it more suitable for subsequent analysis and processing. The high quality and clarity of the enhanced localization image contribute to improving the accuracy of the localization algorithm. A clearer image can provide more details and features, enabling the localization algorithm to more accurately identify and locate the target. This is crucial for the precise localization of ultra-thick structures, reducing localization errors and improving fabrication quality.

[0072] Specifically, the adversarial generative network includes a generator and a discriminator; the processed localization image is passed through a high-definition image generator based on the adversarial generative network to obtain an enhanced localization image, which is further used to input the processed localization image into the high-definition image generator based on the adversarial generative network so that the generator of the high-definition image generator outputs the enhanced localization image through deconvolution coding.

[0073] Then, the enhanced localization image and the reference localization image are passed through a two-stream detection model containing a first image encoder and a second image encoder to obtain detection feature maps and reference feature maps. The first and second image encoders learn the features and structure of the image data, converting the input image into a feature representation. These encoders can extract key features from the image, such as edges, textures, and shapes, thereby capturing important information about the target. By using a two-stream detection model, features of both the enhanced and reference localization images can be extracted simultaneously for subsequent comparison and analysis. Using a first and second image encoder with the same grid structure ensures that the two images have a consistent structure in their feature representations. This helps align the features of the enhanced and reference localization images, allowing them to be compared and matched in the same feature space. Through feature alignment, the differences and accuracy of localization can be evaluated more accurately.

[0074] Next, the difference between the detected feature map and the reference feature map is calculated to obtain a differential feature map. The differential feature map can be used to evaluate the accuracy and precision of the positioning. By calculating the difference between the detected feature map and the reference feature map, a differential feature map reflecting the target position offset and positioning error can be obtained. This differential feature map can be used to quantify the positioning precision and assist in judging the accuracy of the positioning.

[0075] Furthermore, the differential feature map is processed by a classifier to obtain a classification result, which is used to indicate whether the positioning of the ultra-thick structure is accurate. In this way, by constructing a drilling and processing scheme for Teflon circuit boards, and analyzing the detection and reference images of the ultra-thick structure, the accuracy of ultra-thick structure positioning is improved, and labor costs are reduced.

[0076] Specifically, in the task of locating and detecting ultra-thick structures, the purpose of calculating the differential feature map is to capture the differences between the detection image and the reference image, thereby extracting surface details and features. However, the differential feature map usually contains some noise and irrelevant information, which may negatively affect the final location and detection results of ultra-thick structures. To reduce noise and irrelevant information in the differential feature map and enhance the relevant features of the surface, it is necessary to fuse the detection feature map and the reference feature map. By fusing the two feature maps, their beneficial information can complement each other, enhancing the visibility and recognition ability of the surface. Then, by fusing the fused feature map and the differential feature map, the different information of the fused feature map and the differential feature map can be comprehensively utilized to provide a more comprehensive and richer feature representation. This, in turn, improves the robustness and accuracy of the location and detection task.

[0077] Specifically, fusing the detected feature map and the reference feature map based on affine subspace probabilistics to obtain a fused feature map includes: expanding the detected feature map and the reference feature map to obtain a detected feature vector and a reference feature vector; standardizing the detected feature vector and the reference feature vector to obtain a standardized detected feature vector and a standardized reference feature vector; calculating the covariance matrix between the standardized detected feature vector and the standardized reference feature vector; performing eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues ​​and multiple eigenvalue vectors corresponding to the multiple eigenvalues; and extracting the eigenvalue vectors corresponding to the two largest eigenvalues ​​from the multiple feature vectors as the first dimension feature. The first and second dimension feature vectors are used as the basis of the affine subspace. The first and second dimension feature vectors are arranged as column vectors to obtain the affine subspace matrix. The detected feature vector and the reference feature vector are multiplied by the affine subspace matrix to map them onto the affine subspace matrix to obtain a quasi-affine transformation detected feature vector and a quasi-affine transformation reference feature vector. The positional sum of the quasi-affine transformation detected feature vector and the quasi-affine transformation reference feature vector is calculated to obtain a fused feature vector. The fused feature vector is then dimensionally reconstructed to obtain a fused feature map.

[0078] In the technical solution of this application, after expanding the detection feature map and the reference feature map to obtain the detection feature vector and the reference feature vector, the detection feature vector and the reference feature vector are standardized to map them into a probability space. Then, based on the principal component analysis and eigenvalue decomposition, the basis vectors of the affine subspace are obtained. After obtaining the basis vectors of the affine subspace, the detection feature vector and the reference feature vector are mapped to a common affine subspace. Then, the position-wise correlation response is used to densely connect the detection feature vector and the reference feature vector after the affine transformation to obtain the correlation feature matrix. In this way, the main information and structure of the data in the detection feature vector and the reference feature vector can be effectively extracted, reducing the complexity and redundancy of the data. At the same time, the data dimension and transformation method can be flexibly adjusted to adapt to different data types and scenarios.

[0079] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0080] Exemplary System

[0081] Figure 1 The diagram illustrates a block diagram of a system for fabricating an ultra-thick 5G antenna module according to an embodiment of this application. Figure 1 As shown, the system 100 for manufacturing an ultra-thick 5G antenna module according to an embodiment of this application includes: an image acquisition module 110 for acquiring a detection positioning image and a reference positioning image of the ultra-thick structure; a preprocessing module 120 for preprocessing the detection positioning image to obtain a processed positioning image; an image enhancement module 130 for passing the processed positioning image through a high-definition image generator based on a generative adversarial network to obtain an enhanced positioning image; a dual-stream detection module 140 for passing the enhanced positioning image and the reference positioning image through a dual-stream detection model including a first image encoder and a second image encoder to obtain a detection feature map and a reference feature map; a differential module 150 for calculating the difference between the detection feature map and the reference feature map to obtain a differential feature map; an optimization module 160 for optimizing the differential feature map to obtain an optimized differential feature map; and a classification result generation module 170 for passing the optimized differential feature map through a classifier to obtain a classification result, wherein the classification result is used to indicate whether the positioning of the ultra-thick structure is accurate.

[0082] Figure 2 The illustration shows a system architecture diagram of a fabrication system for an ultra-thick 5G antenna module according to an embodiment of this application. Figure 2As shown, in this system architecture, firstly, a detection localization image and a reference localization image of the ultra-thick structure are acquired. Then, the detection localization image is preprocessed to obtain a processed localization image. Next, the processed localization image is passed through a high-resolution image generator based on a generative adversarial network to obtain an enhanced localization image. Then, the enhanced localization image and the reference localization image are passed through a two-stream detection model containing a first image encoder and a second image encoder to obtain a detection feature map and a reference feature map. Next, the difference between the detection feature map and the reference feature map is calculated to obtain a differential feature map. Then, the differential feature map is optimized to obtain an optimized differential feature map. Finally, the optimized differential feature map is passed through a classifier to obtain a classification result, which indicates whether the localization of the ultra-thick structure is accurate.

[0083] In the aforementioned ultra-thick 5G antenna module fabrication system 100, the image acquisition module 110 is used to acquire detection and positioning images and reference positioning images of the ultra-thick structure. As mentioned in the background section, during cutting and processing, the material to be cut needs to be positioned to ensure the cutting effect. However, conventional positioning methods usually require operators to have certain skills and experience to ensure accurate positioning and fixation of the ultra-thick structure. Operators unfamiliar with positioning methods may cause inaccurate or unstable positioning. Therefore, an optimized fabrication scheme for ultra-thick 5G antenna modules is desired.

[0084] To address the aforementioned technical issues, a fabrication scheme for an ultra-thick 5G antenna module is proposed. This scheme utilizes artificial intelligence technology based on a deep neural network model to encode and extract features from the detection and positioning images and reference positioning images of the ultra-thick structure. This yields classification labels indicating the accuracy of the ultra-thick structure's positioning. Furthermore, by constructing a Teflon circuit board drilling process, the detection and reference images of the ultra-thick structure are analyzed, thereby improving the accuracy of ultra-thick structure positioning and reducing labor costs.

[0085] Currently, deep learning and neural networks have been widely applied in fields such as computer vision, natural language processing, and speech signal processing. Furthermore, deep learning and neural networks have demonstrated near-human or even superior performance in areas such as image classification, object detection, semantic segmentation, and text translation.

[0086] In recent years, the development of deep learning and neural networks has provided new ideas and solutions for the fabrication of ultra-thick 5G antenna modules.

[0087] Specifically, firstly, detection and reference positioning images of the ultra-thick structure are acquired. The reference positioning image serves as a standard in the fabrication system. It provides an ideal positioning result for comparison and contrast with the detection image. By comparing it with the reference positioning image, it can be determined whether the positioning of the ultra-thick structure meets the design requirements. The detection and reference positioning images provide a benchmark for evaluating the positioning accuracy of the ultra-thick structure. By comparing the detection and reference images, positioning errors and offsets can be determined.

[0088] In the aforementioned ultra-thick 5G antenna module fabrication system 100, the preprocessing module 120 is used to preprocess the detected positioning image to obtain a processed positioning image. Preprocessing can remove noise, blur, and other unnecessary interference from the image, thereby improving image clarity and quality. This helps reduce the impact of noise in the image on positioning accuracy and makes the positioning target more explicit and identifiable. By adjusting the brightness and contrast of the image, preprocessing can enhance the difference between the positioning target and the background, making it more prominent. This helps improve the accuracy of the positioning algorithm, enabling it to better distinguish the target from the background and accurately locate the target's position. Preprocessing can standardize the image, giving it consistent features and attributes. This helps eliminate differences between different images, making them comparable. Through standardization, images acquired under different acquisition devices and conditions can have consistent characteristics, thereby improving the consistency and reliability of positioning.

[0089] In the aforementioned ultra-thick 5G antenna module fabrication system 100, the image enhancement module 130 is used to process the processed positioning image through a high-definition image generator based on a generative adversarial network (GAN) to obtain an enhanced positioning image. The GAN-based high-definition image generator can generate higher quality and clearer images by learning the features and structure of a large amount of image data. By inputting the processed positioning image into the generator, an enhanced positioning image with higher resolution and clearer details can be obtained. This helps improve the accuracy of the positioning algorithm, making the target location more explicit and reliable. During the processing of the positioning image, some details and information may be lost. By using the high-definition image generator, this lost information can be compensated for as much as possible, and the integrity of the image can be restored. This helps improve the quality and usability of the positioning image, making it more suitable for subsequent analysis and processing. The high quality and clarity of the enhanced positioning image help improve the accuracy of the positioning algorithm. A clearer image can provide more details and features, enabling the positioning algorithm to more accurately identify and locate the target. This is crucial for the precise positioning of ultra-thick structures, reducing positioning errors and improving fabrication quality.

[0090] Specifically, in this embodiment, the Generative Adversarial Network (GAN) includes a generator and a discriminator; the image enhancement module is further configured to input the processed localization image into the high-definition image generator based on the GAN so that the generator of the high-definition image generator outputs the enhanced localization image through deconvolution coding.

[0091] In the aforementioned ultra-thick 5G antenna module fabrication system 100, the dual-stream detection module 140 is used to process the enhanced positioning image and the reference positioning image through a dual-stream detection model containing a first image encoder and a second image encoder to obtain detection feature maps and reference feature maps. The first and second image encoders learn the features and structure of the image data, converting the input image into a feature representation. These encoders can extract key features from the image, such as edges, textures, and shapes, thereby capturing important information about the target. By using the dual-stream detection model, features of both the enhanced positioning image and the reference positioning image can be extracted simultaneously for subsequent comparison and analysis. By using the first and second image encoders with the same grid structure, it can be ensured that the two images have a consistent structure in their feature representations. This helps to align the features of the enhanced positioning image and the reference positioning image, allowing them to be compared and matched in the same feature space. Through feature alignment, the differences and accuracy of positioning can be evaluated more accurately.

[0092] Specifically, in the embodiments of this application, the first image encoder and the second image encoder have the same mesh structure, and the first image encoder and the second image encoder are convolutional neural network models containing deep and shallow feature encoding modules.

[0093] Specifically, in this embodiment, the dual-stream detection module includes: a detection image encoding unit, comprising: a first shallow feature extraction subunit, configured to extract a shallow detection feature map from the Mth layer of the first convolutional neural network model, wherein M is greater than or equal to 1 and less than or equal to 6; a first deep feature extraction subunit, configured to extract a deep detection feature map from the Nth layer of the first convolutional neural network model, wherein N / M is greater than or equal to 5 and less than or equal to 10; and a first fusion subunit, configured to fuse the shallow detection feature map and the deep detection feature map using a deep-shallow feature fusion module of the first convolutional neural network model to obtain a detection feature map.

[0094] Specifically, in this embodiment, the dual-stream detection module further includes: a reference image encoding unit, comprising: a second shallow feature extraction subunit, configured to extract a reference shallow feature map from the Mth layer of the second convolutional neural network model, wherein M is greater than or equal to 1 and less than or equal to 6; a second deep feature extraction subunit, configured to extract a reference deep feature map from the Nth layer of the second convolutional neural network model, wherein N / M is greater than or equal to 5 and less than or equal to 10; and a second fusion subunit, configured to fuse the reference shallow feature map and the reference deep feature map using the deep and shallow feature fusion module of the first convolutional neural network model to obtain a reference feature map.

[0095] In the aforementioned ultra-thick 5G antenna module fabrication system 100, the differential module 150 is used to calculate the difference between the detection feature map and the reference feature map to obtain a differential feature map. The differential feature map can be used to evaluate the accuracy and precision of positioning. By calculating the difference between the detection feature map and the reference feature map, a differential feature map reflecting the target position offset and positioning error can be obtained. This differential feature map can be used to quantify the positioning precision and assist in judging the accuracy of positioning.

[0096] Specifically, in this embodiment, the difference module 150 includes: calculating a difference feature map between the detected feature map and the reference feature map using the following difference formula; wherein, the difference formula is:

[0097]

[0098] Wherein, F1 represents the detected feature map, F2 represents the reference feature map, and F represents the differential feature map. This indicates differentiation based on position.

[0099] In the aforementioned ultra-thick 5G antenna module fabrication system 100, the optimization module 160 is used to optimize the differential feature map to obtain an optimized differential feature map. Furthermore, considering that in the positioning and detection task of ultra-thick structures, the purpose of calculating the differential feature map is to capture the difference between the detection image and the reference image, thereby extracting surface details and features. However, differential feature maps usually contain some noise and irrelevant information, which may negatively affect the final positioning and detection results of ultra-thick structures. To reduce noise and irrelevant information in the differential feature map and enhance the relevant features of the surface, it is necessary to fuse the detection feature map and the reference feature map. By fusing the two feature maps, their beneficial information can complement each other, enhancing the visibility and recognition ability of the surface. Then, by fusing the fused feature map and the differential feature map, the different information from the fused feature map and the differential feature map is comprehensively utilized to provide a more comprehensive and richer feature representation. This, in turn, improves the robustness and accuracy of the positioning and detection task.

[0100] Specifically, fusing the detected feature map and the reference feature map based on affine subspace probabilistics to obtain a fused feature map includes: expanding the detected feature map and the reference feature map to obtain a detected feature vector and a reference feature vector; standardizing the detected feature vector and the reference feature vector to obtain a standardized detected feature vector and a standardized reference feature vector; calculating the covariance matrix between the standardized detected feature vector and the standardized reference feature vector; performing eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues ​​and multiple eigenvalue vectors corresponding to the multiple eigenvalues; and extracting the eigenvalue vectors corresponding to the two largest eigenvalues ​​from the multiple feature vectors as the first dimension feature. The first and second dimension feature vectors are used as the basis of the affine subspace. The first and second dimension feature vectors are arranged as column vectors to obtain the affine subspace matrix. The detected feature vector and the reference feature vector are multiplied by the affine subspace matrix to map them onto the affine subspace matrix to obtain a quasi-affine transformation detected feature vector and a quasi-affine transformation reference feature vector. The positional sum of the quasi-affine transformation detected feature vector and the quasi-affine transformation reference feature vector is calculated to obtain a fused feature vector. The fused feature vector is then dimensionally reconstructed to obtain a fused feature map.

[0101] In the technical solution of this application, after expanding the detection feature map and the reference feature map to obtain the detection feature vector and the reference feature vector, the detection feature vector and the reference feature vector are standardized to map them into a probability space. Then, based on the principal component analysis and eigenvalue decomposition, the basis vectors of the affine subspace are obtained. After obtaining the basis vectors of the affine subspace, the detection feature vector and the reference feature vector are mapped to a common affine subspace. Then, the position-wise correlation response is used to densely connect the detection feature vector and the reference feature vector after the affine transformation to obtain the correlation feature matrix. In this way, the main information and structure of the data in the detection feature vector and the reference feature vector can be effectively extracted, reducing the complexity and redundancy of the data. At the same time, the data dimension and transformation method can be flexibly adjusted to adapt to different data types and scenarios.

[0102] Specifically, in this embodiment of the application, the optimization module 160 includes: a spatial probabilization unit, used to fuse the detection feature map and the reference feature map based on affine subspace probabilization to obtain a fused feature map; and an optimization unit, used to fuse the fused feature map and the differential feature map to obtain an optimized differential feature map.

[0103] Specifically, in this embodiment, the spatial probabilistic unit includes: a feature map unrolling subunit, used to unroll the detection feature map and the reference feature map to obtain a detection feature vector and a reference feature vector; a standardization subunit, used to standardize the detection feature vector and the reference feature vector to obtain a standardized detection feature vector and a standardized reference feature vector; a covariance subunit, used to calculate the covariance matrix between the standardized detection feature vector and the standardized reference feature vector; a decomposition subunit, used to perform eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues ​​and multiple eigenvalue vectors corresponding to the multiple eigenvalues; and a basis extraction subunit, used to extract the eigenvalue vectors corresponding to the two largest eigenvalues ​​from the multiple feature vectors as the first dimension feature vector and the second dimension feature vector. The system comprises: a vector, wherein the first-dimensional feature vector and the second-dimensional feature vector serve as the basis of the affine subspace; a two-dimensional arrangement subunit, used to arrange the first-dimensional feature vector and the second-dimensional feature vector as column vectors to obtain an affine subspace matrix; a matrix multiplication subunit, used to perform matrix multiplication of the detected feature vector and the reference feature vector with the affine subspace matrix respectively to map the detected feature vector and the reference feature vector to the affine subspace matrix to obtain a quasi-affine transformation detected feature vector and a quasi-affine transformation reference feature vector; a positional addition subunit, used to calculate the positional sum of the quasi-affine transformation detected feature vector and the quasi-affine transformation reference feature vector to obtain a fused feature vector; and a dimension reconstruction subunit, used to perform dimension reconstruction on the fused feature vector to obtain a fused feature map.

[0104] In the aforementioned ultra-thick 5G antenna module fabrication system 100, the classification result generation module 170 is used to pass the optimized differential feature map through a classifier to obtain a classification result. The classification result is used to indicate whether the positioning of the ultra-thick structure is accurate. Thus, by constructing a Teflon circuit board drilling processing scheme to analyze the detection image and reference image of the ultra-thick structure, the accuracy of ultra-thick structure positioning is improved, and labor costs are reduced.

[0105] Figure 3 The diagram illustrates a block diagram of a classification result generation module in a fabrication system for an ultra-thick 5G antenna module according to an embodiment of this application. For example... Figure 3As shown, the classification result generation module 170 includes: an expansion unit 171, used to expand the optimized difference feature map into a classification feature vector according to row vectors or column vectors; a fully connected encoding unit 172, used to perform fully connected encoding on the classification feature vector using the fully connected layer of the classifier to obtain an encoded classification feature vector; and a classification result generation unit 173, used to input the encoded classification feature vector into the Softmax classification function of the classifier to obtain the classification result.

[0106] In summary, the fabrication system for the ultra-thick 5G antenna module according to the embodiments of this application has been explained. It uses artificial intelligence technology based on a deep neural network model to encode and extract features from the detection and positioning images and reference positioning images of the ultra-thick structure to obtain classification labels indicating whether the positioning of the ultra-thick structure is accurate. In this way, by constructing a drilling and processing scheme for Teflon circuit boards to analyze the detection and reference images of the ultra-thick structure, the accuracy of the ultra-thick structure positioning is improved and the labor cost is reduced.

[0107] Exemplary methods

[0108] Figure 4 The illustration shows a flowchart of a system method for fabricating an ultra-thick 5G antenna module according to an embodiment of this application. Figure 4 As shown, the fabrication system method for an ultra-thick 5G antenna module according to an embodiment of this application includes the following steps: S110, acquiring a detection positioning image and a reference positioning image of the ultra-thick structure; S120, preprocessing the detection positioning image to obtain a processed positioning image; S130, passing the processed positioning image through a high-definition image generator based on a generative adversarial network to obtain an enhanced positioning image; S140, passing the enhanced positioning image and the reference positioning image through a dual-stream detection model including a first image encoder and a second image encoder to obtain a detection feature map and a reference feature map; S150, calculating the difference between the detection feature map and the reference feature map to obtain a differential feature map; S160, optimizing the differential feature map to obtain an optimized differential feature map; and S170, passing the optimized differential feature map through a classifier to obtain a classification result, wherein the classification result is used to indicate whether the positioning of the ultra-thick structure is accurate.

[0109] Here, those skilled in the art will understand that the specific operations of each step in the above-described method for manufacturing ultra-thick 5G antenna modules have been referenced above. Figures 1 to 3 The manufacturing system for the ultra-thick 5G antenna module is described in detail in the system description, and therefore, its repeated description will be omitted.

[0110] As described above, the ultra-thick 5G antenna module manufacturing system 100 according to the embodiments of this application can be implemented in various terminal devices, such as ultra-thick 5G antenna module manufacturing system servers. In one example, the ultra-thick 5G antenna module manufacturing system 100 according to the embodiments of this application can be integrated into the terminal device as a software module and / or hardware module. For example, the ultra-thick 5G antenna module manufacturing system 100 can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the ultra-thick 5G antenna module manufacturing system 100 can also be one of many hardware modules of the terminal device.

[0111] Alternatively, in another example, the manufacturing system 100 for the ultra-thick 5G antenna module and the terminal device can also be separate devices, and the manufacturing system 100 for the ultra-thick 5G antenna module can be connected to the terminal device via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0112] In summary, the fabrication system method for an ultra-thick 5G antenna module according to the embodiments of this application has been clarified. It uses artificial intelligence technology based on a deep neural network model to encode and extract features from the detection and positioning images and reference positioning images of the ultra-thick structure to obtain classification labels indicating whether the positioning of the ultra-thick structure is accurate. In this way, by constructing a drilling processing scheme for Teflon circuit boards to analyze the detection and reference images of the ultra-thick structure, the accuracy of ultra-thick structure positioning is improved and labor costs are reduced.

[0113] Exemplary electronic devices

[0114] Below, for reference Figure 5 This describes an electronic device according to embodiments of the present application. Figure 5 This is a block diagram of an electronic device according to an embodiment of this application. Figure 5 As shown, the electronic device 10 includes one or more processors 11 and memory 12.

[0115] The processor 11 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0116] The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may execute the program instructions to implement the functions and / or other desired functions in the fabrication system method of the ultra-thick 5G antenna module of the various embodiments of this application described above. Various contents, such as detection and positioning images of the ultra-thick structure and reference positioning images of the ultra-thick structure, may also be stored in the computer-readable storage medium.

[0117] In one example, the electronic device 10 may also include an input device 13 and an output device 14, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0118] The input device 13 may include, for example, a keyboard, a mouse, etc.

[0119] The output device 14 can output various information to the outside, including whether the positioning of the ultra-thick structure is accurate, etc. The output device 14 may include, for example, a display, speaker, printer, and communication network and its connected remote output devices, etc.

[0120] Of course, for the sake of simplicity, Figure 5 Only some of the components of the electronic device 10 relevant to this application are shown in this illustration; components such as buses, input / output interfaces, etc., are omitted. In addition, the electronic device 10 may include any other suitable components depending on the specific application.

[0121] Exemplary computer program products and computer-readable storage media

[0122] In addition to the methods and devices described above, embodiments of this application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps of the functionalities in the manufacturing system method for an ultra-thick 5G antenna module according to various embodiments of this application as described in the "Exemplary Methods" section of this specification.

[0123] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0124] Furthermore, embodiments of this application may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps of the functionalities in the fabrication system method for an ultra-thick 5G antenna module according to various embodiments of this application as described in the "Exemplary Methods" section above.

[0125] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0126] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0127] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0128] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0129] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0130] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A manufacturing system for an ultra-thick 5G antenna module, characterized in that, include: The image acquisition module is used to acquire detection and positioning images and reference positioning images of the ultra-thick structure; The preprocessing module is used to preprocess the detected positioning image to obtain a processed positioning image; An image enhancement module is used to pass the processed positioning image through a high-definition image generator based on a generative adversarial network to obtain an enhanced positioning image; A dual-stream detection module is used to pass the enhanced positioning image and the reference positioning image through a dual-stream detection model containing a first image encoder and a second image encoder to obtain a detection feature map and a reference feature map; The difference module is used to calculate the difference between the detected feature map and the reference feature map to obtain a difference feature map; An optimization module is used to optimize the difference feature map to obtain an optimized difference feature map; The classification result generation module is used to pass the optimized differential feature map through a classifier to obtain a classification result, which is used to indicate whether the positioning of the ultra-thick structure is accurate. The optimization module includes: A spatial probabilization unit is used to fuse the detected feature map and the reference feature map based on affine subspace probabilization to obtain a fused feature map. An optimization unit is used to fuse the fused feature map and the differential feature map to obtain an optimized differential feature map; The spatial probabilistic unit includes: The feature map unpacking subunit is used to unpack the detection feature map and the reference feature map to obtain the detection feature vector and the reference feature vector; The standardization subunit is used to standardize the detection feature vector and the reference feature vector to obtain a standardized detection feature vector and a standardized reference feature vector. The covariance subunit is used to calculate the covariance matrix between the standardized detection feature vector and the standardized reference feature vector. A decomposition subunit is used to perform eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues ​​and multiple eigenvalue vectors corresponding to the multiple eigenvalues; The basis extraction subunit is used to extract the feature vectors corresponding to the two largest feature values ​​from the plurality of feature vectors as the first dimension feature vector and the second dimension feature vector, and the first dimension feature vector and the second dimension feature vector serve as the basis of the affine subspace; A two-dimensional arrangement subunit is used to arrange the first-dimensional feature vector and the second-dimensional feature vector according to column vectors to obtain an affine subspace matrix; The matrix multiplication subunit is used to perform matrix multiplication of the detected feature vector and the reference feature vector with the affine subspace matrix respectively to map the detected feature vector and the reference feature vector to the affine subspace matrix to obtain the affine transformation-like detected feature vector and the affine transformation-like reference feature vector. The positional summation subunit is used to calculate the positional summation of the affine transformation detection feature vector and the affine transformation reference feature vector to obtain the fused feature vector; The dimension reconstruction subunit is used to perform dimension reconstruction on the fused feature vector to obtain a fused feature map.

2. The fabrication system for an ultra-thick 5G antenna module according to claim 1, characterized in that, The adversarial generative network includes a generator and a discriminator; The image enhancement module is further configured to input the processed localization image into the high-definition image generator based on the generative adversarial network so that the generator of the high-definition image generator outputs the enhanced localization image through deconvolution encoding.

3. The fabrication system for an ultra-thick 5G antenna module according to claim 2, characterized in that, The first image encoder and the second image encoder have the same mesh structure, and the first image encoder and the second image encoder are convolutional neural network models that include deep and shallow feature encoding modules.

4. The fabrication system for an ultra-thick 5G antenna module according to claim 3, characterized in that, The dual-stream detection module includes: Detection of image coding units, including: The first shallow feature extraction subunit is used to extract the detection shallow feature map from the Mth layer of the first convolutional neural network model, where M is greater than or equal to 1 and less than or equal to 6. The first deep feature extraction subunit is used to extract detection deep feature maps from the Nth layer of the first convolutional neural network model, where N / M is greater than or equal to 5 and less than or equal to 10. The first fusion subunit is used to fuse the detection shallow feature map and the detection deep feature map using the deep and shallow feature fusion module of the first convolutional neural network model to obtain a detection feature map.

5. The fabrication system for an ultra-thick 5G antenna module according to claim 4, characterized in that, The dual-stream detection module further includes: Reference image coding unit, including: The second shallow feature extraction subunit is used to extract a reference shallow feature map from the Mth layer of the second convolutional neural network model, where M is greater than or equal to 1 and less than or equal to 6. The second deep feature extraction subunit is used to extract a reference deep feature map from the Nth layer of the second convolutional neural network model, where N / M is greater than or equal to 5 and less than or equal to 10. The second fusion subunit is used to fuse the reference shallow feature map and the reference deep feature map using the deep and shallow feature fusion module of the first convolutional neural network model to obtain a reference feature map.

6. The fabrication system for an ultra-thick 5G antenna module according to claim 5, characterized in that, The differential module includes: The difference feature map between the detected feature map and the reference feature map is calculated using the following difference formula; The difference formula is as follows: in, This represents the detection feature map. This refers to the reference feature map. This represents the difference feature map. This indicates differentiation based on position.

7. The fabrication system for an ultra-thick 5G antenna module according to claim 6, characterized in that, The classification result generation module includes: The expansion unit is used to expand the optimized difference feature map into a classification feature vector according to the row vector or column vector; A fully connected encoding unit is used to perform fully connected encoding on the classification feature vector using the fully connected layer of the classifier to obtain an encoded classification feature vector; The classification result generation unit is used to input the encoded classification feature vector into the Softmax classification function of the classifier to obtain the classification result.

Images