Processing Method of Circuit Board for Automotive Millimeter-Wave Radar

Abstract

This application relates to the field of circuit board processing, and specifically relates to a processing method for a circuit board used in automotive millimeter-wave radar. During the quality inspection process of the holes in the circuit board for automotive millimeter-wave radar after drilling, by collecting the hole detection images of the circuit board for automotive millimeter-wave radar after drilling, and introducing image processing and analysis algorithms based on vision and deep learning at the backend to analyze the hole detection images, so as to identify the implicit features such as the roughness of the hole wall, burrs at the hole opening and burrs in the images of the circuit board, thereby judging whether there are quality defect problems in the processed holes of the circuit board. In this way, the processing quality inspection of the circuit board for automotive millimeter-wave radar can be carried out in an automated manner to ensure that each hole meets the product standards and improve the intelligent level of the processing of the circuit board for automotive millimeter-wave radar.

Description

Technical Field

[0001] This application relates to the field of circuit board processing, and specifically relates to a processing method for a circuit board used in automotive millimeter-wave radar. Background Art

[0002] With the development of autonomous driving technology, millimeter-wave radar, as one of the key sensors, its performance directly affects the reliability and safety of the autonomous driving system. The circuit board of millimeter-wave radar is one of the core components of millimeter-wave radar, and its processing quality directly determines the performance of the radar system. Drilling is a crucial step in circuit board processing, and the drilling quality is directly related to the reliability and stability of the circuit board. The existing circuit board required for 77GHz millimeter-wave radar uses polytetrafluoroethylene fiberglass copper clad laminate, which has a high coefficient of thermal expansion and poor heat dissipation performance. At the same time, during the processing of the existing millimeter-wave radar circuit board, when drilling the circuit board, the roughness of the hole wall is large, and obvious burrs at the hole opening will be formed, failing to meet the product quality requirements.

[0003] In view of the above technical problems, Chinese Patent CN113015340B discloses a drilling method for a circuit board of automotive millimeter-wave radar and a circuit board of automotive millimeter-wave radar. By adding a drilling backing plate, phenolic board and phenolic board, the friction during drilling is reduced, thereby reducing the roughness of the hole wall. In addition, during the drilling process of the circuit board of automotive millimeter-wave radar, the drill bit first drills into a high heat dissipation lubricating aluminum sheet to reduce the direct impact on the circuit board during drilling, thereby reducing the burrs and flash at the hole opening.

[0004] In the above drilling method for the circuit board of automotive millimeter-wave radar, although the drilling quality of the circuit board of automotive millimeter-wave radar can be significantly improved through specific drilling methods and material selection, there may still be some potential defect problems. For example, although high heat dissipation lubricating aluminum sheet and phenolic board are used, if the drilling process control is not strict, the roughness of the hole wall may still be large. And the bonding between different materials (such as ceramic powder filled polytetrafluoroethylene fiberglass board and modified epoxy resin fiberglass high-speed board) may be uneven or not firm, which will also cause the product quality to fail to meet the requirements. Therefore, although the existing patent considers reducing burrs and flash in the design, these problems may still occur in actual operation.

[0005] Therefore, an optimized processing scheme for a circuit board used in automotive millimeter-wave radar is expected, which can perform automatic quality inspection on the circuit board used in automotive millimeter-wave radar produced by processing. Summary of the Invention

[0006] This application is made in consideration of the above problems. An object of this application is to provide a processing method for a circuit board used in automotive millimeter-wave radar.

[0007] An embodiment of the present application provides a processing method for a circuit board for an automotive millimeter-wave radar, including: placing the circuit board for the automotive millimeter-wave radar to be drilled on a drilling backing plate, and placing a lubricating aluminum sheet above the circuit board for the automotive millimeter-wave radar to be drilled to obtain a composite circuit board containing the lubricating aluminum sheet, wherein the lubricating aluminum sheet and the circuit board for the automotive millimeter-wave radar to be drilled have the same size, and the positions of the lubricating aluminum sheet and the circuit board for the automotive millimeter-wave radar to be drilled overlap; clamping both sides of the composite circuit board containing the lubricating aluminum sheet with a phenolic board, and fixing the phenolic board through a fixture; drilling the composite circuit board containing the lubricating aluminum sheet with a drill bit to obtain a circuit board for an automotive millimeter-wave radar after drilling; it further includes: performing quality inspection on the holes of the circuit board for an automotive millimeter-wave radar after drilling;

[0008] Wherein, performing quality inspection on the holes of the circuit board for an automotive millimeter-wave radar after drilling includes:

[0009] Obtaining a hole detection image of the circuit board for an automotive millimeter-wave radar after drilling;

[0010] Performing contrast normalization processing on the hole detection image to obtain an enhanced hole detection image;

[0011] Passing the enhanced hole detection image through a hole state semantic information transfer and aggregation module based on a cross-level pyramid structure to obtain a hole multi-scale feature information aggregation feature map;

[0012] Performing hole state semantic feature selection and enhancement processing based on a compression-suppression structure on the hole multi-scale feature information aggregation feature map to obtain a hole state semantic enhanced expression feature;

[0013] Based on the hole state semantic enhanced expression feature, determining whether the holes of the circuit board for an automotive millimeter-wave radar after drilling are qualified.

[0014] For example, in the processing method for a circuit board for an automotive millimeter-wave radar according to an embodiment of the present application, wherein passing the enhanced hole detection image through a hole state semantic information transfer and aggregation module based on a cross-level pyramid structure to obtain a hole multi-scale feature information aggregation feature map includes:

[0015] Passing the enhanced hole detection image through a convolutional layer with a convolution kernel of 3×3 to obtain a shallow feature map of the enhanced hole detection image;

[0016] Passing the shallow feature map of the enhanced hole detection image through a hole detection image shallow residual information fusion and enhancement module to obtain a shallow residual fusion feature map of the enhanced hole detection image;

[0017] The enhanced hole detection image shallow residual fusion feature map is passed through a convolutional layer with a convolution kernel of 3×3 to obtain an enhanced hole detection image deep feature map;

[0018] The enhanced hole detection image deep feature map is passed through a hole detection image deep residual information fusion and enhancement module to obtain an enhanced hole detection image deep residual fusion feature map;

[0019] The enhanced hole detection image deep residual fusion feature map is input into a scale modulation module based on the SPPF layer to obtain an enhanced hole detection image deep modulation feature map;

[0020] The enhanced hole detection image deep modulation feature map is upsampled to obtain an upsampled enhanced hole detection image deep modulation feature map;

[0021] The upsampled enhanced hole detection image deep modulation feature map and the enhanced hole detection image shallow residual fusion feature map are fused to obtain an enhanced hole detection image multi-scale feature map;

[0022] The enhanced hole detection image multi-scale feature map is passed through a hole detection image multi-scale residual information fusion and enhancement module to obtain the hole multi-scale feature information aggregation feature map.

[0023] For example, in the processing method of a circuit board for an automotive millimeter-wave radar according to an embodiment of the present application, wherein the enhanced hole detection image shallow feature map is passed through a hole detection image shallow residual information fusion and enhancement module to obtain an enhanced hole detection image shallow residual fusion feature map, including:

[0024] The enhanced hole detection image shallow feature map is subjected to channel dimensionality reduction modulation through a point convolutional layer to obtain a channel dimensionality reduction modulation enhanced hole detection image shallow feature map;

[0025] The channel dimensionality reduction modulation enhanced hole detection image shallow feature map is subjected to feature extraction through a convolutional layer with a convolution kernel of 3×3 to obtain an enhanced hole detection image shallow implicit information feature map;

[0026] The enhanced hole detection image shallow implicit information feature map is subjected to channel dimensionality increase modulation through a point convolutional layer to obtain a channel modulation enhanced hole detection image shallow feature map;

[0027] The channel modulation enhanced hole detection image shallow feature map and the channel dimensionality reduction modulation enhanced hole detection image shallow feature map are fused to obtain the enhanced hole detection image shallow residual fusion feature map.

[0028] For example, in the processing method of a circuit board for an automotive millimeter-wave radar according to an embodiment of the present application, wherein, inputting the enhanced hole detection image deep residual fusion feature map into a scale modulation module based on the SPPF layer to obtain an enhanced hole detection image deep modulation feature map includes:

[0029] Passing the enhanced hole detection image deep residual fusion feature map through a scale modulation module based on a point convolution layer to obtain a scale-adjusted enhanced hole detection image deep residual fusion feature map;

[0030] Performing max pooling processing on each feature matrix along the channel dimension in the scale-adjusted enhanced hole detection image deep residual fusion feature map to obtain an enhanced hole detection image deep residual fusion feature vector;

[0031] Based on the enhanced hole detection image deep residual fusion feature vector, performing channel weighting and enhancement on the scale-adjusted enhanced hole detection image deep residual fusion feature map to obtain the enhanced hole detection image deep modulation feature map.

[0032] For example, in the processing method of a circuit board for an automotive millimeter-wave radar according to an embodiment of the present application, wherein, performing a hole state semantic feature selection and enhancement process based on a compression-suppression structure on the hole multi-scale feature information aggregation feature map to obtain a hole state semantic enhanced expression feature includes:

[0033] Passing the hole multi-scale feature information aggregation feature map through a hole state semantic feature selection and enhancement module based on a compression-suppression structure to obtain a hole state semantic feature enhanced expression feature map as the hole state semantic enhanced expression feature.

[0034] For example, in the processing method of a circuit board for an automotive millimeter-wave radar according to an embodiment of the present application, wherein, passing the hole multi-scale feature information aggregation feature map through a hole state semantic feature selection and enhancement module based on a compression-suppression structure to obtain a hole state semantic feature enhanced expression feature map as the hole state semantic enhanced expression feature includes:

[0035] Calculating the global mean of each feature matrix along the channel dimension of the hole multi-scale feature information aggregation feature map to perform channel compression on the hole multi-scale feature information aggregation feature map to obtain a hole multi-scale aggregation feature compressed information representation vector;

[0036] Performing one-dimensional convolutional encoding on the hole multi-scale aggregation feature compressed information representation vector to obtain a hole multi-scale aggregation feature compressed information inter-correlation representation feature vector;

[0037] Concatenate the hole multi-scale aggregated feature compressed information representation vector and the correlation representation feature vector between the hole multi-scale aggregated feature compressed information to obtain a hole multi-scale aggregated feature compressed information multi-modal representation vector;

[0038] Input the hole multi-scale aggregated feature compressed information multi-modal representation vector into a compressed information feature extraction module to obtain a hole multi-scale aggregated feature compressed information multi-modal correlation feature vector, where the compressed information feature extraction module is a multi-layer perceptron including two fully connected layers and a SiLU activation function;

[0039] Use the Sigmoid function to perform a normalization operation on the hole multi-scale aggregated feature compressed information multi-modal correlation feature vector to obtain a hole multi-scale feature information aggregation weight feature vector;

[0040] Based on the hole multi-scale feature information aggregation weight feature vector, perform feature amplification and suppression operations on the hole multi-scale feature information aggregation feature map to obtain the hole state semantic feature enhanced expression feature map.

[0041] For example, in the processing method of a circuit board for an automotive millimeter-wave radar according to an embodiment of the present application, where inputting the hole multi-scale aggregated feature compressed information multi-modal representation vector into a compressed information feature extraction module to obtain a hole multi-scale aggregated feature compressed information multi-modal correlation feature vector includes:

[0042] Use the negative of the feature value at each position in the hole multi-scale aggregated feature compressed information multi-modal correlation feature vector as the exponent of the natural constant to calculate the exponential function value with the natural constant as the base for each position to obtain a hole multi-scale aggregated feature compressed information multi-modal correlation class support feature vector;

[0043] Calculate the reciprocal of the sum of the feature value at each position in the hole multi-scale aggregated feature compressed information multi-modal correlation class support feature vector and the constant one to obtain the hole multi-scale feature information aggregation weight feature vector.

[0044] For example, in the processing method of a circuit board for an automotive millimeter-wave radar according to an embodiment of the present application, where based on the hole multi-scale feature information aggregation weight feature vector, performing feature amplification and suppression operations on the hole multi-scale feature information aggregation feature map to obtain the hole state semantic feature enhanced expression feature map includes:

[0045] Multiply the feature value at each position in the hole multi-scale feature information aggregation weight feature vector by each feature matrix along the channel dimension of the hole multi-scale feature information aggregation feature map for each position to obtain the hole state semantic feature enhanced expression feature map.

[0046] For example, in the processing method of a circuit board for an automotive millimeter-wave radar according to an embodiment of the present application, based on the semantic enhanced expression features of the hole state, determining whether the hole quality of the circuit board for the automotive millimeter-wave radar after drilling is qualified includes:

[0047] Pass the semantic feature enhanced expression feature map of the hole state through a hole quality detector based on a classifier to obtain a detection result, and the detection result is used to indicate whether the hole quality of the circuit board for the automotive millimeter-wave radar after drilling is qualified.

[0048] For example, in the processing method of a circuit board for an automotive millimeter-wave radar according to an embodiment of the present application, it further includes a training step: for training the hole state semantic information transfer and aggregation module based on the cross-level pyramid structure, the hole state semantic feature selection and enhancement module based on the compression-suppression structure, and the hole quality detector based on the classifier;

[0049] Among them, the training step includes:

[0050] Obtain training data, where the training data includes training hole detection images and real detection results;

[0051] Perform contrast normalization processing on the training hole detection images to obtain training enhanced hole detection images;

[0052] Pass the training enhanced hole detection images through the hole state semantic information transfer and aggregation module based on the cross-level pyramid structure to obtain a training hole multi-scale feature information aggregation feature map;

[0053] Pass the training hole multi-scale feature information aggregation feature map through the hole state semantic feature selection and enhancement module based on the compression-suppression structure to obtain a training hole state semantic feature enhanced expression feature map;

[0054] Pass the training hole state semantic feature enhanced expression feature map through the hole quality detector based on the classifier to obtain a training detection result;

[0055] Calculate the cross-entropy loss function value between the training detection result and the real detection result to obtain a classification loss function value;

[0056] Calculate the hole state semantic feature enhanced expression loss function value of the training hole state semantic feature enhanced expression feature map;

[0057] Using the weighted sum of the classification loss function value and the enhanced expression loss function value of the hole state semantic features as the loss function value, train the hole state semantic information transmission and aggregation module based on the cross-level pyramid structure, the hole state semantic feature selection and enhancement module based on the compression-suppression structure, and the hole quality detector based on the classifier.

[0058] In the method for processing a circuit board for an automotive millimeter-wave radar according to an embodiment of the present application, during the quality inspection of the holes in the circuit board for an automotive millimeter-wave radar after drilling, by collecting the hole detection images of the circuit board for an automotive millimeter-wave radar after drilling, and introducing an image processing and analysis algorithm based on vision and deep learning at the backend to analyze the hole detection images, so as to identify the implicit features such as the roughness of the hole wall, the burrs at the hole opening, and the burrs in the image of the circuit board, thereby determining whether there are quality defect problems in the processed holes of the circuit board. In this way, the processing quality inspection of the circuit board for an automotive millimeter-wave radar can be carried out in an automated manner to ensure that each hole meets the product standards and improve the intelligent level of the processing of the circuit board for an automotive millimeter-wave radar. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] To more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application will be briefly introduced below. Obviously, the drawings in the following description only relate to some embodiments of the present application and do not limit the present application.

[0060] Figure 1 Shows the application architecture schematic diagram of the method for processing a circuit board for an automotive millimeter-wave radar in an embodiment of the present application;

[0061] Figure 2 Shows the flowchart of the quality inspection step in the method for processing a circuit board for an automotive millimeter-wave radar in an embodiment of the present application;

[0062] Figure 3 Shows the flowchart of the sub-step S530 in the method for processing a circuit board for an automotive millimeter-wave radar in an embodiment of the present application;

[0063] Figure 4 Shows the flowchart of the sub-step S540 in the method for processing a circuit board for an automotive millimeter-wave radar in an embodiment of the present application;

[0064] Figure 5 Shows the structural schematic diagram of the processing system of the circuit board for an automotive millimeter-wave radar in an embodiment of the present application; and

[0065] Figure 6 Shows the application scenario diagram of the method for processing a circuit board for an automotive millimeter-wave radar in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only partial embodiments of the present application, rather than all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts also belong to the scope of protection of the present application.

[0067] The terms used in this specification are those general terms that are currently widely used in the art in consideration of the functions of the present application. However, these terms may change according to the intentions of those of ordinary skill in the art, precedents, or new technologies in the art. In addition, specific terms may be selected, and in such cases, their detailed meanings will be described in the detailed description of the present application. Therefore, the terms used in the specification should not be understood as simple names, but based on the meanings of the terms and the overall description of the present application.

[0068] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or the server. The modules are merely illustrative, and different aspects of the system and method may use different modules.

[0069] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations before or below do not necessarily need to be performed precisely in sequence. Instead, various steps may be processed in reverse order or simultaneously as needed. At the same time, other operations may also be added to these processes, or one or several operations may be removed from these processes.

[0070] Figure 1 The application architecture diagram of the processing method of the circuit board for automotive millimeter-wave radar in the embodiments of the present application is shown, including a server 100 and a terminal device 200.

[0071] The terminal device 200 and the server 100 can be connected through the Internet to achieve communication with each other. Optionally, the above Internet uses standard communication technologies and / or protocols. The Internet is usually the Internet, but can also be any network, including but not limited to any combination of a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a private network or a virtual private network. In some embodiments, technologies and / or formats including HyperText Markup Language (HTML), Extensible Markup Language (XML), etc. are used to represent the data exchanged through the network. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec), etc. can be used to encrypt all or some of the links. In other embodiments, custom and / or proprietary data communication technologies can also be used to replace or supplement the above data communication technologies.

[0072] The server 100 can provide various network services for the terminal device 200. Among them, the server 100 can be a single server, a server cluster composed of several servers, or a cloud computing center. Specifically, the server 100 can include a processor 110 (Center Processing Unit, CPU), a memory 120, an input device 130, an output device 140, etc. The input device 130 can include a keyboard, a mouse, a touch screen, etc. The output device 140 can include a display device, such as a liquid crystal display (LCD), a cathode ray tube (CRT), etc.

[0073] The memory 120 can include a read-only memory (ROM) and a random access memory (RAM), and provide the program instructions and data stored in the memory 120 to the processor 110. In the embodiments of the present application, the memory 120 can be used to store the program of the processing method of the circuit board for automotive millimeter-wave radar in the embodiments of the present application.

[0074] The processor 110 is configured to execute the steps of the processing method of any circuit board for automotive millimeter-wave radar in the embodiments of the present application according to the obtained program instructions by calling the program instructions stored in the memory 120.

[0075] In addition, the application architecture diagram in the embodiments of the present application is for more clearly illustrating the technical solutions in the embodiments of the present application, and does not constitute a limitation on the technical solutions provided in the embodiments of the present application. Of course, for other application architectures and business applications, the technical solutions provided in the embodiments of the present application are equally applicable to similar problems.

[0076] The following provides a non-limiting description of the processing method of the circuit board for automotive millimeter-wave radar according to at least one embodiment of the present application through several examples or embodiments. As described below, different features in these specific examples or embodiments can be combined with each other without conflict, so as to obtain new examples or embodiments, and these new examples or embodiments also fall within the scope of protection of the present application.

[0077] In view of the above technical problems, in the technical solution of the present application, a processing method of a circuit board for automotive millimeter-wave radar is proposed, which can add a quality inspection link in the last step of the production process of the circuit board for automotive millimeter-wave radar, that is, perform quality inspection on the holes of the circuit board for automotive millimeter-wave radar after drilling. Specifically, the technical concept of the present application is to collect the hole detection image of the circuit board for automotive millimeter-wave radar after drilling during the quality inspection process of the holes of the circuit board for automotive millimeter-wave radar after drilling, and introduce an image processing and analysis algorithm based on vision and deep learning at the backend to analyze the hole detection image, so as to identify the implicit features such as the roughness of the hole wall, burrs at the hole opening and burrs in the image, thereby judging whether there are quality defect problems in the processed holes of the circuit board. In this way, the processing quality inspection of the circuit board for automotive millimeter-wave radar can be carried out in an automated manner to ensure that each hole meets the product standards and improve the processing intelligence level of the circuit board for automotive millimeter-wave radar.

[0078] Correspondingly, the processing method of the circuit board for automotive millimeter-wave radar includes: placing the circuit board for automotive millimeter-wave radar to be drilled on a drilling backing plate, and placing a lubricating aluminum sheet above the circuit board for automotive millimeter-wave radar to be drilled to obtain a composite circuit board containing the lubricating aluminum sheet, wherein the size of the lubricating aluminum sheet is the same as that of the circuit board for automotive millimeter-wave radar to be drilled, and the positions of the lubricating aluminum sheet and the circuit board for automotive millimeter-wave radar to be drilled overlap; clamping both sides of the composite circuit board containing the lubricating aluminum sheet with a phenolic board, and fixing the phenolic board through a fixture; using a drill bit to drill the composite circuit board containing the lubricating aluminum sheet to obtain a circuit board for automotive millimeter-wave radar after drilling.

[0079] Further, the processing method of the circuit board for automotive millimeter-wave radar further includes: performing quality inspection on the holes of the circuit board for automotive millimeter-wave radar after drilling.

[0080] Specifically, as Figure 2 shown, performing quality inspection on the holes of the circuit board for automotive millimeter-wave radar after drilling includes: S510, obtaining a hole detection image of the circuit board for automotive millimeter-wave radar after drilling; S520, performing contrast normalization processing on the hole detection image to obtain an enhanced hole detection image; S530, passing the enhanced hole detection image through a hole state semantic information transfer and aggregation module based on a cross-level pyramid structure to obtain a hole multi-scale feature information aggregation feature map; S540, performing hole state semantic feature selection and enhancement processing based on a compression-suppression structure on the hole multi-scale feature information aggregation feature map to obtain a hole state semantic enhanced expression feature; S550, based on the hole state semantic enhanced expression feature, determining whether the holes of the circuit board for automotive millimeter-wave radar after drilling are qualified.

[0081] Specifically, in the technical solution of the present application, during the quality inspection of the holes of the circuit board for automotive millimeter-wave radar after drilling, first, a hole detection image of the circuit board for automotive millimeter-wave radar after drilling is obtained. It should be understood that during the acquisition process of the hole detection image, there may be problems such as insufficient local contrast in the original image. At the same time, images taken under different conditions may have problems such as uneven illumination or contrast differences, resulting in unclear details such as hole edges, making it difficult to perform subsequent feature recognition and extraction. Based on this, in the technical solution of the present application, it is necessary to perform contrast normalization processing on the hole detection image to obtain an enhanced hole detection image. By performing contrast normalization processing on the hole detection image, local contrast features of the image can be extracted to reduce the influence of global contrast changes in the image, making the features of the holes more prominent. Specifically. The steps of contrast normalization processing are to subtract the local mean of each pixel value of the hole detection image and divide it by the local standard deviation to achieve image contrast normalization.

[0082] Then, since the hole features and semantic information of the circuit board need to be presented at different scales and levels of the image. For example, shallow features can capture local color, edges, and textures of the circuit board holes in the image, while deep features can capture global and semantic information about the circuit board holes in the image. Therefore, it is necessary to capture the hole multi-scale fusion features of the circuit board in the enhanced hole detection image through a pyramid network.

[0083] In the pyramid network, the underlying network is mainly responsible for extracting the low-level features and spatial detail information of the image, while the high-level network is responsible for extracting more abstract high-level features. However, in the traditional network structure, the location information extracted by the underlying network is difficult to directly affect the high-level network, resulting in the distortion of the location information in the extraction of different-level features of the enhanced hole detection image. In addition, in the pyramid network, the transmission of features usually requires a large amount of computational effort. As the number of network layers increases, the computational effort grows exponentially, which may lead to the training and inference processes becoming very time-consuming. Based on this, in order to address the problem that the shallow location information in the pyramid network is difficult to affect the deep features and causes a large amount of computational effort required for transmitting feature information, in the technical solution of this application, the enhanced hole detection image is further passed through a hole state semantic information transmission and aggregation module based on a cross-level pyramid structure to obtain a hole multi-scale feature information aggregation feature map. In particular, in the technical solution of this application, the hole state semantic information transmission and aggregation module based on the cross-level pyramid structure includes a feature extraction network and a feature fusion network. In the feature fusion network, the information transmission of different-level features of the hole image is achieved by means of residual fusion and skip connection, so as to effectively fuse the shallow features and deep features in the enhanced hole detection image. This can effectively transmit the hole location information and shallow feature information extracted at the bottom layer to the high layer, enabling the deep features to better fuse the shallow features of the bottom layer, thereby improving the expression ability and accuracy of the hole features. In addition, by using the pyramid structure to perform residual feature extraction and information transmission on the hole features at different levels, the computational effort in the feature transmission process is reduced. By extracting features and aggregating information at different scales, the efficient transmission and utilization of feature information are achieved, while unnecessary computational overhead is reduced, and the computational efficiency and speed of the network are improved.

[0084] Accordingly, as Figure 3As shown, in step S530, the enhanced hole detection image is passed through a hole state semantic information transfer and aggregation module based on a cross-level pyramid structure to obtain a hole multi-scale feature information aggregation feature map, including: S531, passing the enhanced hole detection image through a convolutional layer with a convolution kernel of 3×3 to obtain a shallow feature map of the enhanced hole detection image; S532, passing the shallow feature map of the enhanced hole detection image through a shallow residual information fusion and enhancement module for hole detection images to obtain a shallow residual fusion feature map of the enhanced hole detection image; S533, passing the shallow residual fusion feature map of the enhanced hole detection image through a convolutional layer with a convolution kernel of 3×3 to obtain a deep feature map of the enhanced hole detection image; S534, passing the deep feature map of the enhanced hole detection image through a deep residual information fusion and enhancement module for hole detection images to obtain a deep residual fusion feature map of the enhanced hole detection image; S535, inputting the deep residual fusion feature map of the enhanced hole detection image into a scale modulation module based on the SPPF layer to obtain a deep modulation feature map of the enhanced hole detection image; S536, performing upsampling on the deep modulation feature map of the enhanced hole detection image to obtain an upsampled deep modulation feature map of the enhanced hole detection image; S537, fusing the upsampled deep modulation feature map of the enhanced hole detection image with the shallow residual fusion feature map of the enhanced hole detection image to obtain a multi-scale feature map of the enhanced hole detection image; S538, passing the multi-scale feature map of the enhanced hole detection image through a multi-scale residual information fusion and enhancement module for hole detection images to obtain the hole multi-scale feature information aggregation feature map.

[0085] Among them, in step S532, passing the shallow feature map of the enhanced hole detection image through a shallow residual information fusion and enhancement module for hole detection images to obtain a shallow residual fusion feature map of the enhanced hole detection image includes: passing the shallow feature map of the enhanced hole detection image through a point convolutional layer for channel dimension reduction modulation to obtain a channel dimension reduction modulated shallow feature map of the enhanced hole detection image; passing the channel dimension reduction modulated shallow feature map of the enhanced hole detection image through a convolutional layer with a convolution kernel of 3×3 for feature extraction to obtain a shallow implicit information feature map of the enhanced hole detection image; passing the shallow implicit information feature map of the enhanced hole detection image through a point convolutional layer for channel dimension increase modulation to obtain a shallow feature map of the enhanced hole detection image after channel modulation; fusing the shallow feature map of the enhanced hole detection image after channel modulation with the channel dimension reduction modulated shallow feature map of the enhanced hole detection image to obtain the shallow residual fusion feature map of the enhanced hole detection image.

[0086] Among them, in step S535, inputting the deep residual fusion feature map of the enhanced hole detection image into a scale modulation module based on the SPPF layer to obtain a deep modulation feature map of the enhanced hole detection image includes: passing the deep residual fusion feature map of the enhanced hole detection image through a scale modulation module based on a point convolution layer to obtain a scale-adjusted deep residual fusion feature map of the enhanced hole detection image; performing maximum pooling processing on each feature matrix along the channel dimension in the scale-adjusted deep residual fusion feature map of the enhanced hole detection image to obtain a deep residual fusion feature vector of the enhanced hole detection image; and performing channel-weighted enhancement on the scale-adjusted deep residual fusion feature map of the enhanced hole detection image based on the deep residual fusion feature vector of the enhanced hole detection image to obtain the deep modulation feature map of the enhanced hole detection image.

[0087] Furthermore, considering that during the process of hole quality detection of the circuit board, the tiny defect features in the hole characterization information, such as the degree of hole wall roughness, the appearance of burrs and flash at the hole opening, etc., are crucial. Therefore, it is necessary to assign higher feature weights to these feature information related to tiny defects to ensure that they are given sufficient attention during the network processing. Based on this, in the technical solution of this application, the aggregated feature map of the hole multi-scale feature information is further passed through a hole state semantic feature selection and enhancement module based on a compression-suppression structure to obtain a feature map with enhanced expression of the hole state semantic features. Specifically, through the processing of the hole state semantic feature selection and enhancement module based on the compression-suppression structure, higher feature weights can be assigned to the tiny defect features, so that these features can be utilized multiple times at different levels of the network, thereby strengthening the learning and expression of these features. Moreover, it suppresses the flow of background feature information irrelevant to defect recognition, reduces the problem of loss of effective information of tiny defects in the circuit board holes as the network deepens, and the influence of the low contrast difference between tiny defects and the background, and improves the learning and expression ability of the deep network for tiny defect features.

[0088] Specifically, the hole state semantic feature selection and enhancement module based on the compression-suppression structure can, based on the idea of the compression and suppression structure, use global average pooling to compress the information of the hole multi-scale feature information aggregation feature map, capture the correlation information between the hole multi-scale compressed information through one-dimensional convolutional encoding, and then extract the hole implicit features through a multi-layer perceptron composed of two fully connected layers and the SiLU activation function. Then, the Sigmoid function is used for normalization to obtain the weight values of different channels in the hole multi-scale feature information aggregation feature map. Finally, the feature amplification and suppression operations are performed on the hole multi-scale feature information aggregation feature map in combination with the weight values to enhance the network's feature selection and recognition capabilities for subtle defects and quality problems in the holes. Specifically, compared with the traditional ReLU activation function, the SiLU activation function has the characteristics of having no upper bound but having a lower bound, being smooth, and non-monotonic, which can improve the training effect and performance of deep networks.

[0089] Correspondingly, in step S540, the hole state semantic feature selection and enhancement processing based on the compression-suppression structure is performed on the hole multi-scale feature information aggregation feature map to obtain the hole state semantic enhanced expression feature, including: passing the hole multi-scale feature information aggregation feature map through the hole state semantic feature selection and enhancement module based on the compression-suppression structure to obtain the hole state semantic feature enhanced expression feature map as the hole state semantic enhanced expression feature.

[0090] Specifically, as Figure 4As shown, the multi-scale feature information aggregation feature map of the holes is passed through a hole state semantic feature selection and enhancement module based on a compression-suppression structure to obtain a hole state semantic feature enhanced expression feature map as the enhanced expression of the hole state semantics, including: S541, calculating the global mean of each feature matrix of the multi-scale feature information aggregation feature map of the holes along the channel dimension to compress the multi-scale feature information aggregation feature map of the holes along the channel dimension to obtain a multi-scale aggregation feature compression information representation vector of the holes; S542, performing one-dimensional convolutional encoding on the multi-scale aggregation feature compression information representation vector of the holes to obtain an associated representation feature vector between the multi-scale aggregation feature compression information of the holes; S543, concatenating the multi-scale aggregation feature compression information representation vector of the holes and the associated representation feature vector between the multi-scale aggregation feature compression information of the holes to obtain a multi-modal representation vector of the multi-scale aggregation feature compression information of the holes; S544, inputting the multi-modal representation vector of the multi-scale aggregation feature compression information of the holes into a compressed information feature extraction module to obtain a multi-modal associated feature vector of the multi-scale aggregation feature compression information of the holes, where the compressed information feature extraction module is a multi-layer perceptron including two fully connected layers and a SiLU activation function; S545, using the Sigmoid function to perform a normalization operation on the multi-modal associated feature vector of the multi-scale aggregation feature compression information of the holes to obtain a multi-scale feature information aggregation weight feature vector; S546, based on the multi-scale feature information aggregation weight feature vector, performing a feature amplification and suppression operation on the multi-scale feature information aggregation feature map of the holes to obtain the hole state semantic feature enhanced expression feature map.

[0091] Wherein, in step S544, inputting the multi-modal representation vector of the multi-scale aggregation feature compression information of the holes into a compressed information feature extraction module to obtain a multi-modal associated feature vector of the multi-scale aggregation feature compression information of the holes includes: taking the negative of the feature value at each position in the multi-modal associated feature vector of the multi-scale aggregation feature compression information of the holes as the exponent of the natural constant to calculate the exponential function value with the natural constant as the base at each position to obtain a multi-modal associated class support feature vector of the multi-scale aggregation feature compression information of the holes; calculating the reciprocal of the sum of the feature value at each position in the multi-modal associated class support feature vector of the multi-scale aggregation feature compression information of the holes and the constant one to obtain the multi-scale feature information aggregation weight feature vector.

[0092] Among them, in step S546, based on the hole multi-scale feature information aggregation weight feature vector, performing feature amplification and suppression operations on the hole multi-scale feature information aggregation feature map to obtain the hole state semantic feature enhanced expression feature map includes: multiplying the feature values at each position in the hole multi-scale feature information aggregation weight feature vector by each feature matrix along the channel dimension of the hole multi-scale feature information aggregation feature map to obtain the hole state semantic feature enhanced expression feature map.

[0093] In a specific example, passing the hole multi-scale feature information aggregation feature map through a hole state semantic feature selection and enhancement module based on a compression-suppression structure to obtain the hole state semantic feature enhanced expression feature map as the hole state semantic enhanced expression feature includes: passing the hole multi-scale feature information aggregation feature map through the hole state semantic feature selection and enhancement module based on the compression-suppression structure and processing it according to the following hole state semantic feature selection and enhancement formula to obtain the hole state semantic feature enhanced expression feature map; where the hole state semantic feature selection and enhancement formula is: , , Among them, is the hole multi-scale feature information aggregation feature map, represents the hole multi-scale feature information aggregation feature map of the channel, and the feature value at the point with coordinates is and are the height and width of the hole multi-scale feature information aggregation feature map respectively, is the hole multi-scale aggregation feature compression information representation vector, is a one-dimensional convolutional encoding, is a concatenation processing of vectors, is the hole multi-scale aggregation feature compression information multi-modal representation vector, is a multi-layer perceptron, represents multiplication by position, is the hole state semantic feature enhanced expression feature map.

[0094] Furthermore, the enhanced expression feature map of the hole state semantic features is then passed through a hole quality detector based on a classifier to obtain a detection result, which is used to indicate whether the hole quality of the circuit board for automotive millimeter-wave radar after drilling is qualified. That is to say, the enhanced expression feature map of the hole state semantic features is used to detect the hole quality of the circuit board, so as to identify implicit features such as the roughness of the hole wall, burrs at the hole opening, and burrs in the image, thereby determining whether there are quality defects in the processed holes of the circuit board. In this way, the processing quality of the circuit board for automotive millimeter-wave radar can be detected in an automated manner to ensure that each hole meets the product standards and improve the intelligent level of the processing of the circuit board for automotive millimeter-wave radar.

[0095] Correspondingly, in step S550, based on the enhanced expression features of the hole state semantics, determining whether the hole quality of the circuit board for automotive millimeter-wave radar after drilling is qualified includes: passing the enhanced expression feature map of the hole state semantic features through a hole quality detector based on a classifier to obtain a detection result, which is used to indicate whether the hole quality of the circuit board for automotive millimeter-wave radar after drilling is qualified.

[0096] It should be understood that the role of the classifier is to use the given categories and known training data to learn classification rules and classifiers, and then classify (or predict) unknown data. Logistic regression, SVM, etc. are commonly used to solve binary classification problems. For multi-class classification problems, logistic regression or SVM can also be used, but it requires multiple binary classifications to form a multi-class classification, which is prone to errors and has low efficiency. Commonly used multi-class classification methods include the Softmax classification function.

[0097] Furthermore, in the technical solution of the present application, the processing method of the circuit board for automotive millimeter-wave radar further includes a training step: for training the hole state semantic information transfer and aggregation module based on the cross-level pyramid structure, the hole state semantic feature selection and enhancement module based on the compression-suppression structure, and the hole quality detector based on the classifier.

[0098] Among them, the training steps include: obtaining training data, where the training data includes training hole detection images and real detection results; performing contrast normalization on the training hole detection images to obtain training enhanced hole detection images; passing the training enhanced hole detection images through the hole state semantic information transmission and aggregation module based on the cross-level pyramid structure to obtain a training hole multi-scale feature information aggregation feature map; passing the training hole multi-scale feature information aggregation feature map through the hole state semantic feature selection and enhancement module based on the compression-suppression structure to obtain a training hole state semantic feature enhanced expression feature map; passing the training hole state semantic feature enhanced expression feature map through the hole quality detector based on the classifier to obtain a training detection result; calculating the cross-entropy loss function value between the training detection result and the real detection result to obtain a classification loss function value; calculating the hole state semantic feature enhanced expression loss function value of the training hole state semantic feature enhanced expression feature map; and training the hole state semantic information transmission and aggregation module based on the cross-level pyramid structure, the hole state semantic feature selection and enhancement module based on the compression-suppression structure, and the hole quality detector based on the classifier with the weighted sum of the classification loss function value and the hole state semantic feature enhanced expression loss function value as the loss function value.

[0099] Preferably, a new loss function value is further introduced in addition to the classification loss function value, where the new loss function value is called the hole state semantic feature enhancement expression loss function value. Specifically, calculating the hole state semantic feature enhancement expression loss function value of the training hole state semantic feature enhancement expression feature map includes the following steps: expanding the training hole state semantic feature enhancement expression feature map into a training hole state semantic feature enhancement expression feature vector; calculating a training hole state semantic feature enhancement expression sum matrix and a training hole state semantic feature enhancement expression difference matrix based on the training hole state semantic feature enhancement expression feature vector, where the feature values at the position of the training hole state semantic feature enhancement expression sum matrix and the training hole state semantic feature enhancement expression difference matrix are respectively the mean value and half of the absolute value of the difference between the feature value and the feature value of the training hole state semantic feature enhancement expression feature vector; multiplying the training hole state semantic feature enhancement expression feature vector as a query vector with the training hole state semantic feature enhancement expression sum matrix and the training hole state semantic feature enhancement expression difference matrix respectively to obtain a training hole state semantic feature enhancement expression query sum vector and a training hole state semantic feature enhancement expression query difference vector; calculating the vector inner product of the training hole state semantic feature enhancement expression feature vector and the training hole state semantic feature enhancement expression query difference vector to obtain a first hole state semantic feature enhancement expression feature loss term; multiplying the training hole state semantic feature enhancement expression sum matrix and the training hole state semantic feature enhancement expression difference matrix, and calculating the norm of the product matrix to obtain a second hole state semantic feature enhancement expression feature loss term; subtracting the product of a predetermined weight hyperparameter and the second hole state semantic feature enhancement expression feature loss term from the first hole state semantic feature enhancement expression feature loss term to obtain the hole state semantic feature enhancement expression loss function value.

[0100] Among them, the calculation process of the new loss function value, such as the hole state semantic feature enhancement expression loss function value, is specifically represented by the following loss calculation formula: , where ,

[0101] Among them, is the training hole state semantic feature enhancement expression feature vector, and are respectively the training hole state semantic feature enhancement expression sum matrix and the training hole state semantic feature enhancement expression difference matrix, and are respectively the of the training hole state semantic feature enhancement expression sum matrix and the training hole state semantic feature enhancement expression difference matrix The eigenvalue of the position, and are respectively the eigenvalue and the eigenvalue of the feature vector for the enhanced expression of the semantic features of the training hole state, is matrix multiplication, is to calculate the norm of the matrix, is a predetermined weight hyperparameter, is the value of the loss function for the enhanced expression of the semantic features of the hole state.

[0102] Here, considering that the multi-scale feature information aggregation feature map of the training holes expresses the training enhanced hole detection image based on the multi-scale and multi-depth image semantic features, when the multi-scale feature information aggregation feature map of the training holes passes through the hole state semantic feature selection and enhancement module based on the compression-suppression structure, the enhanced expression feature map of the training hole state semantic features will also have classification and regression recognition difficulties attributed to the complex image semantic representation due to the feature selective enhancement differences of the image semantic features at different depths and different scales based on the compression-suppression structure, thus affecting the accuracy of the classification result.

[0103] Therefore, the present application performs the query-based composition of the inner product space of the details of the enhanced expression feature vector of the training hole state semantic features through the structured feature representation of the short-range-long-range cross-scale detail link of the enhanced expression feature vector of the training hole state semantic features after unfolding the enhanced expression feature map of the training hole state semantic features, so as to approximate the low-rank independent observable composition formed by the link details provided by the structured detail interaction of the enhanced expression feature vector of the training hole state semantic features. In this way, by training with the loss function of the enhanced expression of the hole state semantic features, the distributed detail group of the enhanced expression feature map of the training hole state can be decomposed based on the detail complexity to promote the classification and regression decomposition recognition of the enhanced expression feature map of the training hole state semantic features based on the complex feature representation, and improve the accuracy of the detection result obtained by the hole quality detector based on the classifier. In this way, the quality defect problems of the holes in the circuit board processing can be more accurately identified and detected to ensure that each hole meets the product standard and improve the intelligent level of the circuit board processing for automotive millimeter-wave radars.

[0104] Based on the above embodiments, refer to Figure 5As shown in the figure, it is a schematic structural diagram of a processing system 800 for a circuit board used in an automotive millimeter-wave radar according to an embodiment of the present application. The processing system 800 for the circuit board used in the automotive millimeter-wave radar includes: a composite circuit board manufacturing module 810, configured to place the circuit board of the automotive millimeter-wave radar to be drilled on a drilling backing plate, and place a lubricating aluminum sheet above the circuit board of the automotive millimeter-wave radar to be drilled to obtain a composite circuit board containing the lubricating aluminum sheet, wherein the size between the lubricating aluminum sheet and the circuit board of the automotive millimeter-wave radar to be drilled is the same, and the positions between the lubricating aluminum sheet and the circuit board of the automotive millimeter-wave radar to be drilled overlap; a fixing module 820, configured to clamp both sides of the composite circuit board containing the lubricating aluminum sheet using a phenolic board, and fix the phenolic board through a fixture; a drilling module 830, configured to drill the composite circuit board containing the lubricating aluminum sheet using a drill bit to obtain a circuit board for an automotive millimeter-wave radar after drilling; a quality inspection module 840, configured to perform quality inspection on the holes of the circuit board for an automotive millimeter-wave radar after drilling.

[0105] In one example, in the above-mentioned processing system 800 for the circuit board used in the automotive millimeter-wave radar, the quality inspection module 840 is configured to: obtain a hole detection image of the circuit board for the automotive millimeter-wave radar after drilling; perform contrast normalization processing on the hole detection image to obtain an enhanced hole detection image; pass the enhanced hole detection image through a hole state semantic information transfer aggregation module based on a cross-level pyramid structure to obtain a hole multi-scale feature information aggregation feature map; perform hole state semantic feature selection and enhancement processing based on a compression-suppression structure on the hole multi-scale feature information aggregation feature map to obtain a hole state semantic enhanced expression feature; based on the hole state semantic enhanced expression feature, determine whether the holes of the circuit board for the automotive millimeter-wave radar after drilling are qualified.

[0106] Here, those skilled in the art can understand that the specific functions and operations of each module in the above-mentioned processing system 800 for the circuit board used in the automotive millimeter-wave radar have been described in detail in the description of the processing method for the circuit board used in the automotive millimeter-wave radar above, and therefore, the repeated description thereof will be omitted. Figures 2 to 4 For the application scenario diagram of the processing method for the circuit board used in the automotive millimeter-wave radar according to an embodiment of the present application. As

[0107] Figure 6 shown, in this application scenario, first, obtain a hole detection image of the circuit board for the automotive millimeter-wave radar after drilling (for example, Figure 6 the D shown in the figure), and then, input the hole detection image into a server deployed with a processing algorithm for the circuit board used in the automotive millimeter-wave radar (for example, Figure 6 the one shown in the figure), and then, input the hole detection image into a server deployed with a processing algorithm for the circuit board used in the automotive millimeter-wave radar (for example, Figure 6In the S) shown, the server can process the hole detection image using the processing algorithm of the circuit board for automotive millimeter-wave radar to obtain a detection result indicating whether the hole quality of the circuit board for automotive millimeter-wave radar after drilling is qualified.

[0108] Based on the above embodiments, an electronic device of another exemplary embodiment is further provided in the embodiments of the present application. In some possible embodiments, the electronic device in the embodiments of the present application may include a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the processing method of the circuit board for automotive millimeter-wave radar in the above embodiments can be implemented.

[0109] For example, taking the server 100 in the present application Figure 1 as an example for illustration, the processor in this electronic device is the processor 110 in the server 100, and the memory in this electronic device is the memory 120 in the server 100.

[0110] The embodiments of the present application also provide a computer-readable storage medium, on which computer-executable instructions are stored. When the computer-executable instructions are run by a processor, the processing method of the circuit board for automotive millimeter-wave radar according to the embodiments of the present application described with reference to the above drawings can be executed. The computer-readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0111] The embodiments of the present application also provide a computer program product or a computer program, which includes computer-executable instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer-executable instructions from the computer-readable storage medium, and the processor executes the computer-executable instructions, so that the computer device executes the processing method of the circuit board for automotive millimeter-wave radar according to the embodiments of the present application.

[0112] Those skilled in the art can understand that the content disclosed in the present application can have various variations and improvements. For example, the various devices or components described above can be implemented by hardware, software, firmware, or some or all of the combinations of the three.

[0113] In addition, although the present application makes various references to certain units in the system according to embodiments of the present application, any number of different units may be used and run on the client and / or server. The units are merely illustrative, and different aspects of the system and method may use different units.

[0114] Those of ordinary skill in the art will understand that all or part of the steps in the above methods can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a magnetic disk, or an optical disc, etc. Optionally, all or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiments can be implemented in the form of hardware or in the form of a software functional module. The present application is not limited to any specific form of combination of hardware and software.

[0115] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with their meaning in the context of the relevant art, and should not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0116] The above is an illustration of the present application and should not be considered a limitation thereof. Although several exemplary embodiments of the present application have been described, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the present application.

Claims

1. A method for processing a circuit board for an automotive millimeter wave radar, comprising: The automotive millimeter-wave radar circuit board to be drilled is placed on a drilling pad, and a lubricating aluminum sheet is placed above the automotive millimeter-wave radar circuit board to be drilled to obtain a composite circuit board containing the lubricating aluminum sheet, wherein the lubricating aluminum sheet is consistent in size with the automotive millimeter-wave radar circuit board to be drilled, and the positions of the lubricating aluminum sheet and the automotive millimeter-wave radar circuit board to be drilled overlap; a phenolic plate is used to clamp both sides of the composite circuit board containing the lubricating aluminum sheet, and the phenolic plate is fixed by a clamp; a drill is used to drill the composite circuit board containing the lubricating aluminum sheet to obtain a circuit board for automotive millimeter-wave radar after drilling; it is characterized in that it also includes: quality inspection of the holes of the circuit board for automotive millimeter-wave radar after drilling; Wherein, the quality inspection of the holes of the circuit board for the automotive millimeter wave radar after drilling includes: Acquire a hole detection image of the circuit board for the automotive millimeter-wave radar after the drilling; Performing contrast normalization processing on the hole detection image to obtain an enhanced hole detection image; The enhanced hole detection image is passed through a hole state semantic information transmission aggregation module based on a cross-level pyramid structure to obtain a hole multi-scale feature information aggregation feature map; Performing hole state semantic feature selection and enhancement processing based on a compression-suppression structure on the hole multi-scale feature information aggregation feature map to obtain a hole state semantic enhancement expression feature; Based on the semantic enhancement expression feature of the hole state, determining whether the hole quality of the circuit board for the automotive millimeter wave radar after drilling is qualified; The enhanced hole detection image is passed through a hole state semantic information transmission aggregation module based on a cross-level pyramid structure to obtain a hole multi-scale feature information aggregation feature map, including: Passing the enhanced hole detection image through a convolution layer with a convolution kernel of 3×3 to obtain a shallow feature map of the enhanced hole detection image; The enhanced hole detection image shallow feature map is passed through a hole detection image shallow residual information fusion enhancement module to obtain an enhanced hole detection image shallow residual fusion feature map; The shallow residual fusion feature map of the enhanced hole detection image is passed through a convolution layer with a convolution kernel of 3×3 to obtain a deep feature map of the enhanced hole detection image; The enhanced hole detection image deep feature map is passed through a hole detection image deep residual information fusion enhancement module to obtain an enhanced hole detection image deep residual fusion feature map; Inputting the enhanced hole detection image deep layer residual fusion feature map into the scale modulation module based on the SPPF layer to obtain the enhanced hole detection image deep layer modulation feature map; Upsampling the enhanced hole detection image deep modulation feature map to obtain an upsampled enhanced hole detection image deep modulation feature map; The upsampled enhanced hole detection image deep modulation feature map is fused with the enhanced hole detection image shallow residual fusion feature map to obtain an enhanced hole detection image multi-scale feature map; The enhanced hole detection image multi-scale feature map is passed through a hole detection image multi-scale residual information fusion enhancement module to obtain the hole multi-scale feature information aggregation feature map; Among them, the hole state semantic feature selection and enhancement processing based on the compression-suppression structure is performed on the hole multi-scale feature information aggregation feature map to obtain the hole state semantic enhancement expression feature, including: Calculating the global mean of each feature matrix of the hole multi-scale feature information aggregation feature map along the channel dimension to perform channel compression on the hole multi-scale feature information aggregation feature map to obtain a hole multi-scale aggregation feature compression information representation vector; Performing one-dimensional convolution coding on the hole multi-scale aggregate feature compression information representation vector to obtain a feature vector representing the correlation between the hole multi-scale aggregate feature compression information; Cascading the hole multi-scale aggregation feature compression information representation vector and the hole multi-scale aggregation feature compression information association representation feature vector to obtain the hole multi-scale aggregation feature compression information multi-modal representation vector; Inputting the hole multi-scale aggregation feature compression information multimodal representation vector into a compression information feature extraction module to obtain a hole multi-scale aggregation feature compression information multimodal association feature vector, wherein the compression information feature extraction module is a multi-layer perceptron including two fully connected layers and a SiLU activation function; Using a Sigmoid function, a normalization operation is performed on the multi-modal correlation feature vector of the hole multi-scale aggregation feature compression information to obtain a hole multi-scale feature information aggregation weight feature vector; Based on the hole multi-scale feature information aggregation weight feature vector, feature amplification and suppression operations are performed on the hole multi-scale feature information aggregation feature map to obtain a hole state semantic feature enhancement expression feature map as the hole state semantic enhancement expression feature.

2. The method for processing a circuit board for an automotive millimeter wave radar according to claim 1, characterized in that: The enhanced hole detection image shallow feature map is passed through a hole detection image shallow residual information fusion enhancement module to obtain an enhanced hole detection image shallow residual fusion feature map, including: The shallow feature map of the enhanced hole detection image is subjected to channel dimensionality reduction modulation through a point convolution layer to obtain a shallow feature map of the enhanced hole detection image; The shallow feature map of the channel dimension reduction modulation enhanced hole detection image is subjected to feature extraction through a convolution layer with a convolution kernel of 3×3 to obtain a shallow implicit information feature map of the enhanced hole detection image; The shallow layer hidden information feature map of the enhanced hole detection image is subjected to channel dimensionality up-modulation through a point convolution layer to obtain a shallow layer feature map of the enhanced hole detection image after channel modulation; The shallow feature map of the channel-modulated enhanced hole detection image and the shallow feature map of the channel-reduced dimensionally modulated enhanced hole detection image are fused to obtain the shallow residual fusion feature map of the enhanced hole detection image.

3. The method for processing a circuit board for an automotive millimeter wave radar according to claim 2, characterized in that: Inputting the enhanced hole detection image deep layer residual fusion feature map into the scale modulation module based on the SPPF layer to obtain the enhanced hole detection image deep layer modulation feature map, including: The enhanced hole detection image deep residual fusion feature map is passed through a scale modulation module based on a point convolution layer to obtain a scale-adjusted enhanced hole detection image deep residual fusion feature map; Performing maximum pooling processing on each feature matrix along the channel dimension in the deep residual fusion feature map of the scale-adjusted enhanced hole detection image to obtain a deep residual fusion feature vector of the enhanced hole detection image; Based on the enhanced hole detection image deep residual fusion feature vector, the scale-adjusted enhanced hole detection image deep residual fusion feature map is channel-weighted enhanced to obtain the enhanced hole detection image deep modulation feature map.

4. The method for processing a circuit board for an automotive millimeter wave radar according to claim 3, characterized in that: Inputting the hole multi-scale aggregation feature compression information multi-modal representation vector into the compression information feature extraction module to obtain the hole multi-scale aggregation feature compression information multi-modal correlation feature vector, including: Using the negative number of the eigenvalue of each position in the multi-modal association eigenvector of the hole multi-scale aggregation feature compression information as the exponent of the natural constant to calculate the exponential function value based on the natural constant according to the position to obtain the multi-modal association class support eigenvector of the hole multi-scale aggregation feature compression information; The inverse of the sum of the eigenvalues ​​of each position in the multi-modal association class support feature vector of the hole multi-scale aggregation feature compression information and the constant one is calculated to obtain the hole multi-scale feature information aggregation weight feature vector.

5. The method for processing a circuit board for an automotive millimeter wave radar according to claim 4, characterized in that: Based on the hole multi-scale feature information aggregation weight feature vector, feature amplification and suppression operations are performed on the hole multi-scale feature information aggregation feature map to obtain the hole state semantic feature enhanced expression feature map, including: The eigenvalues ​​of each position in the hole multi-scale feature information aggregation weight feature vector are multiplied by position with each feature matrix along the channel dimension of the hole multi-scale feature information aggregation feature map to obtain the hole state semantic feature enhanced expression feature map.

6. The method for processing a circuit board for an automotive millimeter wave radar according to claim 5, characterized in that: Based on the semantic enhancement expression feature of the hole state, determining whether the hole quality of the circuit board for the automotive millimeter wave radar after drilling is qualified includes: The hole state semantic feature enhanced expression feature map is passed through a classifier-based hole quality detector to obtain a detection result, and the detection result is used to indicate whether the hole quality of the circuit board for automotive millimeter wave radar after drilling is qualified.

7. The method for processing a circuit board for an automotive millimeter wave radar according to claim 6, characterized in that: It also includes a training step: for training the hole state semantic information transfer aggregation module based on the cross-level pyramid structure, the hole state semantic feature selection and enhancement module based on the compression-suppression structure, and the hole quality detector based on the classifier; Wherein, the training step includes: Acquire training data, wherein the training data includes training hole detection images and actual detection results; Performing contrast normalization processing on the training hole detection image to obtain a training enhanced hole detection image; The training enhanced hole detection image is passed through the hole state semantic information transmission aggregation module based on the cross-level pyramid structure to obtain a training hole multi-scale feature information aggregation feature map; The training hole multi-scale feature information aggregation feature map is passed through the hole state semantic feature selection and enhancement module based on the compression-suppression structure to obtain a training hole state semantic feature enhanced expression feature map; Passing the training hole state semantic feature enhanced expression feature map through the classifier-based hole quality detector to obtain a training detection result; Calculating a cross entropy loss function value between the training detection result and the true detection result to obtain a classification loss function value; Calculating the hole state semantic feature enhanced expression loss function value of the training hole state semantic feature enhanced expression feature graph; Based on the weighted sum of the classification loss function value and the hole state semantic feature enhancement expression loss function value as the loss function value, the hole state semantic information transfer aggregation module based on the cross-level pyramid structure, the hole state semantic feature selection and enhancement module based on the compression-suppression structure, and the classifier-based hole quality detector are trained.

Images