PCB manufacturing quality analysis method, device and equipment

Abstract

The application provides a PCB manufacturing quality analysis method, device and equipment, and relates to the technical field of PCB manufacturing quality analysis. In the application, first, a first video frame and a second video frame are determined; second, manufacturing quality semantic representations of the first video frame and the second video frame at multiple semantic depths are mined; then, for each semantic depth, a quality semantic correlation vector at the semantic depth is analyzed based on the manufacturing quality semantic representations of the first video frame and the second video frame at the semantic depth; finally, a target quality semantic correlation vector is determined according to the multiple quality semantic correlation vectors analyzed at the multiple semantic depths, and manufacturing quality analysis data is determined based on the target quality semantic correlation vector. Based on the above, the problem of relatively low reliability of PCB manufacturing quality analysis in the prior art can be improved.

Description

Technical Field

[0001] This application relates to the field of PCB manufacturing quality analysis technology, and more specifically, to a PCB manufacturing quality analysis method, apparatus, and equipment. Background Technology

[0002] With the continuous advancement of technology, electronic devices are demanding increasingly higher performance and quality, especially in the manufacturing process of printed circuit boards (PCBs), which are core components of electronic products. As a fundamental component in electronic products, the quality of PCBs directly affects the overall performance and stability of the equipment. Therefore, how to effectively analyze PCB manufacturing quality and improve the accuracy and reliability of the analysis has become an important research topic in the industry.

[0003] Currently, traditional PCB manufacturing quality analysis mainly relies on manual inspection and traditional image processing techniques. Manual inspection methods are not only time-consuming but also susceptible to subjective factors and fatigue, resulting in relatively low accuracy and reliability of the results. While traditional image processing techniques can handle a certain degree of image feature recognition, their adaptability to complex scenarios is limited, failing to fully extract multi-layered and multi-dimensional quality information from PCB images, and often struggling to effectively identify subtle quality issues. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a PCB manufacturing quality analysis method, apparatus and equipment to improve the problem of relatively low reliability of PCB manufacturing quality analysis in the prior art.

[0005] To achieve the above objectives, this application adopts the following technical solution: A PCB manufacturing quality analysis method, comprising: A first video frame and a second video frame are determined, wherein the first video frame is formed by performing an image acquisition operation on a target PCB board, and the second video frame is formed by performing an image acquisition operation on a reference PCB board, wherein the manufacturing quality of the reference PCB board meets the application requirements. The production quality semantic representations of the first video frame and the second video frame at multiple semantic depths are extracted. For each of the plurality of semantic depths, a quality semantic representation is created based on the first video frame and the second video frame respectively corresponding to the semantic depth, and the quality semantic association vector under the semantic depth is analyzed. Based on the multiple quality semantic association vectors determined and analyzed at the multiple semantic depths, a target quality semantic association vector is determined, and based on the target quality semantic association vector, the manufacturing quality analysis data of the target PCB board is determined.

[0006] In a preferred embodiment of this application, the step of mining the manufacturing quality semantic representations of the first video frame and the second video frame at multiple semantic depths in the above-mentioned PCB manufacturing quality analysis method includes: The semantic mining unit in the target production quality analysis model includes multiple semantic mining sub-units, which mine production quality semantic representations at different semantic depths, thereby outputting the production quality semantic representations of the first video frame and the second video frame at multiple semantic depths respectively. Wherein, the size of the production quality semantic representation of the first video frame or the second video frame at the target semantic depth has a target ratio with the size of the corresponding first video frame or the second video frame, and the target ratio is less than 1.

[0007] In a preferred embodiment of this application, in the aforementioned PCB manufacturing quality analysis method, the step of mining manufacturing quality semantic representations at different semantic depths through multiple semantic mining sub-units included in the semantic mining unit of the target manufacturing quality analysis model, and thereby outputting the manufacturing quality semantic representations of the first video frame and the second video frame at multiple semantic depths, includes: For the first semantic mining subunit included in the semantic mining unit of the target production quality analysis model, the first video frame and the second video frame are convolved by the convolutional network layer in the semantic mining subunit to form the corresponding first convolutional semantic representation and second convolutional semantic representation. The first convolutional semantic representation and the second convolutional semantic representation are then self-attention processed by the self-attention network layer in the semantic mining subunit to form the production quality semantic representations of the first video frame and the second video frame at the first semantic depth. For each semantic mining subunit from the second to the last in the semantic mining unit, the production quality semantic representations at the previous semantic depth are convolved by the convolutional network layer in the semantic mining subunit to form the corresponding first convolutional semantic representation and second convolutional semantic representation. The first convolutional semantic representation and the second convolutional semantic representation are then self-attentioned by the self-attention network layer in the semantic mining subunit to form the production quality semantic representations of the first video frame and the second video frame at the current semantic depth.

[0008] In a preferred embodiment of this application, in the above-described PCB manufacturing quality analysis method, the step of analyzing the quality semantic association vector at each of the plurality of semantic depths, based on the manufacturing quality semantic representations corresponding to the first video frame and the second video frame at the respective semantic depths, includes: A correlation analysis is performed on the production quality semantic representation of the first video frame and the production quality semantic representation of the second video frame at the semantic depth to obtain the correlation parameters of the first video frame and the second video frame at the semantic depth. Based on the association parameters, a weighted summation calculation is performed on the production quality semantic representation corresponding to the first video frame at the semantic depth, and a quality semantic association vector at the semantic depth is output, wherein the association parameters are a parameter distribution matrix.

[0009] In a preferred embodiment of this application, in the above-described PCB manufacturing quality analysis method, the step of performing a correlation analysis on the manufacturing quality semantic representations of the first video frame and the second video frame at the semantic depth to obtain the correlation parameters of the first video frame and the second video frame at the semantic depth includes: The first mapping matrix carried by the association mining unit in the target production quality analysis model is used to map the production quality semantic representation of the second video frame at the semantic depth to form the corresponding first mapping semantic representation. The association mining unit carries multiple second mapping matrices to map the production quality semantic representation of the first video frame at the semantic depth, forming multiple corresponding second mapping semantic representations. The association mining unit also carries a third mapping matrix, which is used to map the production quality semantic representation of the first video frame at the semantic depth to form a corresponding third mapping semantic representation. The third mapping semantic representation is used to replace the production quality semantic representation of the first video frame at the semantic depth and is weighted and summed with the association parameters to obtain the corresponding quality semantic association vector. The transpose of each of the first mapping semantic representations and the transpose of each of the plurality of second mapping semantic representations is multiplied by a dot product to obtain a plurality of local association parameters; The mean of the multiple local correlation parameters is calculated to obtain the correlation parameters of the first video frame and the second video frame at the semantic depth.

[0010] In a preferred embodiment of this application, the steps of determining a target quality semantic association vector based on the multiple quality semantic association vectors analyzed at the multiple semantic depths, and determining the manufacturing quality analysis data of the target PCB board based on the target quality semantic association vector, include: Based on the size relationship between the corresponding semantic depths, the multiple quality semantic association vectors are connected to form a target quality semantic association vector with the size of the second video frame. The target quality semantic association vector is processed by a fully connected layer to form a corresponding fully connected quality semantic vector. Based on the fully connected quality semantic vector, the manufacturing quality analysis data of the target PCB board is determined.

[0011] In a preferred embodiment of this application, in the above-described PCB manufacturing quality analysis method, the step of connecting the multiple quality semantic association vectors based on the size relationship between corresponding semantic depths, so that the multiple quality semantic association vectors at the multiple semantic depths are connected to form a target quality semantic association vector with the size of the second video frame, includes: In each stage of the connection processing, the production quality semantic representation with the target size formed by the connection processing of the previous stage is semantically expanded to form a first production quality semantic representation. In the first stage of the connection processing, after the quality semantic association vector corresponding to the current stage is semantically expanded, it is merged with the production quality semantic representations of the first video frame and the second video frame at the semantic depth corresponding to the first stage to form a corresponding merged production quality semantic representation. The merged production quality semantic representation is then semantically compressed to obtain a production quality semantic representation with the target size, which is determined as the production quality semantic representation with the target size formed by the connection processing of the first stage, and the semantic depth corresponding to the first stage belongs to the last semantic depth. The quality semantic association vector corresponding to the current stage is semantically expanded so that the quality semantic association vector is adjusted to a second production quality semantic representation with the target size; The first and second manufacturing quality semantic representations are summed to obtain the third manufacturing quality semantic representation. The third production quality semantic representation is merged with the production quality semantic representations of the first video frame and the second video frame at the semantic depth corresponding to the current stage to form the merged production quality semantic representation corresponding to the current stage. The semantic compression process is performed on the merged production quality semantic representation corresponding to the current stage to obtain a production quality semantic representation with a target size, and this is determined as the production quality semantic representation with a target size formed by the connection process of the current stage. The production quality semantic representation with the target size formed by the connection processing in the last stage is determined as the corresponding target quality semantic association vector.

[0012] In a preferred embodiment of this application, in the above-described PCB manufacturing quality analysis method, the step of merging the third manufacturing quality semantic representation with the manufacturing quality semantic representations of the first video frame and the second video frame at the corresponding semantic depth in the current stage to form a merged manufacturing quality semantic representation for the current stage includes: For each level of semantic compression processing in the multi-level semantic compression processing, the output semantic representation of the previous level of semantic compression processing and the production quality semantic representation of the second video frame at the corresponding semantic depth in the current stage are concatenated, and the concatenated semantic representation is downsampled, and the downsampled semantic representation is subjected to self-attention processing to obtain the output semantic representation of the current level of semantic compression processing. The output semantic representation of the previous level of semantic compression processing of the first level is the third production quality semantic representation. For each level of semantic expansion processing in the multi-level semantic expansion processing, the output semantic representation of the previous level of semantic expansion processing and the production quality semantic representation of the first video frame at the corresponding semantic depth at the current stage are concatenated, and the concatenated semantic representation is upsampled. Then, the upsampled semantic representation is subjected to self-attention processing to obtain the output semantic representation of the current level of semantic expansion processing. The output semantic representation of the previous level of semantic expansion processing of the first level is the output semantic representation of the last level of semantic compression processing. The output semantic representation of the semantic expansion process at the last level is determined as the quality semantic representation of the merging process corresponding to the current stage.

[0013] This application also provides a PCB manufacturing quality analysis device, including: The video frame determination module is used to determine a first video frame and a second video frame, wherein the first video frame is formed by performing an image acquisition operation on a target PCB board, and the second video frame is formed by performing an image acquisition operation on a reference PCB board, wherein the manufacturing quality of the reference PCB board meets the application requirements. The semantic mining module is used to mine the production quality semantic representations of the first video frame and the second video frame at multiple semantic depths. The semantic association module is used to create a quality semantic representation based on the first video frame and the second video frame respectively corresponding to the semantic depth for each of the plurality of semantic depths, and analyze the quality semantic association vector under the semantic depth. The quality analysis module is used to determine a target quality semantic association vector based on multiple quality semantic association vectors analyzed at the multiple semantic depths, and to determine the manufacturing quality analysis data of the target PCB board based on the target quality semantic association vector.

[0014] Based on the above, this application also provides an electronic device, including: Memory, used to store computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the PCB manufacturing quality analysis method described above.

[0015] The PCB manufacturing quality analysis method, apparatus, and equipment provided in this application first determine a first video frame and a second video frame; second, they mine the manufacturing quality semantic representations of the first and second video frames at multiple semantic depths; then, for each semantic depth, based on the manufacturing quality semantic representations of the first and second video frames at each semantic depth, they analyze the quality semantic association vector at that semantic depth; finally, based on the multiple quality semantic association vectors determined and analyzed at multiple semantic depths, they determine a target quality semantic association vector, and based on the target quality semantic association vector, they determine the manufacturing quality analysis data. Based on the above, because it mines manufacturing quality semantic representations at multiple semantic depths, the mined semantic information can cover everything from shallow to deep levels and from concrete to abstract, achieving a full representation of global semantic information, thus improving the reliability of quality analysis. In addition, since the semantic information of the reference PCB board is also combined for association mining, different quality semantic association vectors can be mined for target PCB boards with different qualities. This enables high-precision representation of semantic information of different qualities, which can further improve the reliability of quality analysis and improve the problem of relatively low reliability of PCB manufacturing quality analysis in the existing technology. Attached Figure Description

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings.

[0017] Figure 1 A structural block diagram of an electronic device provided in an embodiment of this application.

[0018] Figure 2This is a flowchart illustrating the PCB manufacturing quality analysis method provided in an embodiment of this application.

[0019] Figure 3 This is a schematic diagram illustrating the connection processing of quality semantic association vectors provided in an embodiment of this application.

[0020] Figure 4 This is a schematic diagram illustrating the merging process of the semantic representation of production quality provided in an embodiment of this application.

[0021] Figure 5 This is a block diagram of the PCB manufacturing quality analysis device provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0024] like Figure 1 As shown in the illustration, this application provides an electronic device. The electronic device may include a memory, a processor, and a PCB manufacturing quality analysis device.

[0025] Specifically, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, the memory and the processor can be electrically connected via one or more communication buses or signal lines. The PCB manufacturing quality analysis device includes at least one software functional module stored in the memory in the form of software or firmware. The processor is used to execute executable computer programs stored in the memory, such as the software functional modules and computer programs included in the PCB manufacturing quality analysis device, to implement the PCB manufacturing quality analysis method provided in this application embodiment.

[0026] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0027] Furthermore, the processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0028] Understandable. Figure 1 The structure shown is for illustrative purposes only; the electronic device may also include components that are more advanced than those shown. Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices.

[0029] Combination Figure 2 This application also provides a PCB manufacturing quality analysis method applicable to the aforementioned electronic device. The method steps defined in the relevant process of the PCB manufacturing quality analysis method can be implemented by the electronic device. The following will describe... Figure 2 The specific process shown will be explained in detail.

[0030] Step S110: Determine the first video frame and the second video frame.

[0031] In this embodiment, the electronic device can determine a first video frame and a second video frame. The first video frame is formed by performing an image acquisition operation on a target PCB board (e.g., by performing an image acquisition operation on the manufactured target PCB board using an image acquisition device), and the second video frame is formed by performing an image acquisition operation on a reference PCB board (e.g., by performing an image acquisition operation on the manufactured reference PCB board using an image acquisition device). The manufacturing quality of the reference PCB board meets application requirements (e.g., a PCB board with good manufacturing quality is determined).

[0032] Step S120: Mine out the production quality semantic representations of the first video frame and the second video frame at multiple semantic depths.

[0033] In this embodiment, after determining the first video frame and the second video frame, the electronic device can mine the production quality semantic representations of the first and second video frames at multiple semantic depths. That is, on the one hand, the production quality semantic representations of the first video frame at multiple semantic depths can be mined, such as the production quality semantic representations at the first semantic depth, the second semantic depth, and the third semantic depth. On the other hand, the production quality semantic representations of the second video frame at multiple semantic depths can be mined, such as the production quality semantic representations at the first semantic depth, the second semantic depth, and the third semantic depth. Furthermore, the semantic representations can be in vector form; that is, the semantic features mined from the first and second video frames are represented in vector form.

[0034] Step S130: For each of the plurality of semantic depths, based on the first video frame and the second video frame respectively corresponding to the semantic depth, create a quality semantic representation and analyze the quality semantic association vector under the semantic depth.

[0035] In this embodiment, after obtaining the production quality semantic representations at multiple semantic depths, the electronic device can analyze the quality semantic association vector at each semantic depth based on the production quality semantic representations of the first and second video frames at that semantic depth. For example, the production quality semantic representations of the first and second video frames at the first semantic depth can be correlated to obtain a quality semantic association vector at the first semantic depth. Similarly, the production quality semantic representations of the first and second video frames at the second semantic depth can be correlated to obtain a quality semantic association vector at the second semantic depth. Likewise, the production quality semantic representations of the first and second video frames at the third semantic depth can be correlated to obtain a quality semantic association vector at the third semantic depth. Therefore, by performing association mining at different semantic depths, the accuracy of association mining can be improved.

[0036] Step S140: Based on the multiple quality semantic association vectors determined and analyzed under the multiple semantic depths, a target quality semantic association vector is determined, and based on the target quality semantic association vector, the manufacturing quality analysis data of the target PCB board is determined.

[0037] In this embodiment, after obtaining multiple quality semantic association vectors at multiple semantic depths, the electronic device can determine a target quality semantic association vector based on the analyzed quality semantic association vectors at the multiple semantic depths, and determine the manufacturing quality analysis data of the target PCB board based on the target quality semantic association vector. That is, multiple quality semantic association vectors at different semantic depths can be fused first to obtain a target quality semantic association vector that can fully represent global semantic information. Then, quality analysis is performed based on the target quality semantic association vector to obtain manufacturing quality analysis data.

[0038] Based on the above, by mining manufacturing quality semantic representations at multiple semantic depths, the mined semantic information can cover everything from shallow to deep levels and from concrete to abstract, achieving a full representation of global semantic information. Therefore, the reliability of quality analysis can be improved. Furthermore, by combining semantic information from a reference PCB board for correlation mining, different quality semantic correlation vectors can be mined for target PCB boards with different qualities. This enables high-precision representation of semantic information for different qualities, further improving the reliability of quality analysis and addressing the relatively low reliability of PCB manufacturing quality analysis in existing technologies.

[0039] It should be noted that for step S120, the specific method of mining the production quality semantic representation of the first video frame and the second video frame at multiple semantic depths is not limited and can be selected according to actual needs.

[0040] For example, in an alternative implementation, in order to improve the accuracy of semantic information mining at multiple semantic depths, so that the resulting production-quality semantic representations at multiple semantic depths can capture more semantic information, the above step S120 may further include step S121, as follows.

[0041] Step S121: The semantic mining unit in the target production quality analysis model is used to mine production quality semantic representations at different semantic depths through multiple semantic mining sub-units, thereby outputting the production quality semantic representations of the first video frame and the second video frame at multiple semantic depths.

[0042] In other words, the semantic mining unit within the target production quality analysis model (a neural network model with a training process that, during training, can be trained based on sample video frames, reference video frames, and the quality labels corresponding to the sample video frames) can mine production quality semantic representations at different semantic depths through multiple semantic mining sub-units. This outputs the production quality semantic representations of the first and second video frames at various semantic depths. For example, one semantic mining sub-unit can correspond to a production quality semantic representation at one semantic depth. Furthermore, the size of the production quality semantic representation of the first or second video frame at the target semantic depth has a target ratio to the size of the corresponding first or second video frame, where the target ratio is less than 1. That is, during semantic mining, as the semantic depth increases, the size of the mined production quality semantic representation can decrease; for example, the target ratio can be 1 / 2, 1 / 4, 1 / 8, 1 / 16, etc.

[0043] It is understood that in step S121 above, the specific method of outputting the production quality semantic representations of the first video frame and the second video frame at multiple semantic depths is not limited. For example, in an alternative implementation, in order to fully extract the effective semantic information in the video frame and make the semantic representation of the obtained production quality semantic representation more accurate, step S121 above may further include the following: First, for the first semantic mining subunit included in the semantic mining unit of the target production quality analysis model, the first video frame and the second video frame are convolved by the convolutional network layer in the semantic mining subunit to form the corresponding first convolutional semantic representation and second convolutional semantic representation (it should be noted that in some embodiments, pooling and activation processing can also be performed after convolution). Then, the first convolutional semantic representation and the second convolutional semantic representation are self-attention processed by the self-attention network layer in the semantic mining subunit (refer to relevant prior art) to form the production quality semantic representations of the first video frame and the second video frame at the first semantic depth, that is, the production quality semantic representations of the first video frame and the second video frame at the first semantic depth are obtained, and the two production quality semantic representations have the same size. Secondly, for each semantic mining subunit from the second to the last in the semantic mining unit, the production quality semantic representations at the previous semantic depth are convolved in the convolutional network of the semantic mining subunit to form corresponding first and second convolutional semantic representations (it should be noted that in some embodiments, pooling and activation processing can also be performed after convolution). Then, the first and second convolutional semantic representations are self-attention processed by the self-attention network layer of the semantic mining subunit to form the production quality semantic representations of the first and second video frames at the current semantic depth. For example, the production quality semantic representations of the first and second video frames at the second semantic depth are obtained, and the two production quality semantic representations have the same size; the production quality semantic representations of the first and second video frames at the third semantic depth are obtained, and the two production quality semantic representations have the same size.

[0044] In other words, in the steps described above, cascaded semantic mining subunits are used to mine two video frames separately (as the depth of convolutional processing increases, more high-level and abstract features can be captured), thereby obtaining production-quality semantic representations with different semantic depths. Furthermore, by performing self-attention processing, important internal features can be highlighted.

[0045] It should be noted that for step S130, the specific method for analyzing the quality semantic association vector at the semantic depth is not limited and can be selected according to actual needs.

[0046] For example, in an alternative implementation, in order to achieve reliable association mining so that the obtained quality semantic association vector can effectively represent the association semantic information between two quality semantic representations, the above step S130 may further include the following steps S131 and S132, the specific contents of each step are as follows.

[0047] Step S131: Perform correlation analysis on the production quality semantic representation of the first video frame and the production quality semantic representation of the second video frame at the semantic depth to obtain the correlation parameters of the first video frame and the second video frame at the semantic depth.

[0048] In this embodiment, a correlation analysis can be performed on the production quality semantic representations of the first video frame and the second video frame at the semantic depth to obtain correlation parameters between them at the semantic depth. These correlation parameters are parameter distribution matrices. For example, the two production quality semantic representations can be dot-producted to obtain a corresponding dot-product matrix. This dot-product matrix can then be normalized to obtain a parameter distribution matrix, which can be used to characterize the correlation between the two production quality semantic representations.

[0049] Step S132: Based on the association parameters, perform a weighted summation calculation on the production quality semantic representation corresponding to the first video frame at the semantic depth, and output the quality semantic association vector at the semantic depth.

[0050] In this embodiment, after obtaining the association parameters, a weighted summation calculation can be performed on the production quality semantic representation corresponding to the first video frame at the semantic depth based on the association parameters, outputting a quality semantic association vector at the semantic depth. That is, since the association parameters can characterize the association relationship between two production quality semantic representations, weighted summation calculation based on the association relationship allows for the focus and characterization of semantic information in the production quality semantic representation that is related to another production quality semantic representation, thereby obtaining a quality semantic association vector that can reliably characterize the association information.

[0051] It is understood that the specific method of performing correlation analysis in step S131 above is not limited. For example, in an alternative implementation, in order to capture richer information and more complex relationships during the correlation analysis process and to make the correlation analysis more reliable, step S131 above may further include the following: First, the production quality semantic representation corresponding to the second video frame at the semantic depth can be mapped using the first mapping matrix carried by the association mining unit in the target production quality analysis model to form the corresponding first mapping semantic representation. For example, the first mapping matrix and the production quality semantic representation can be multiplied to achieve the corresponding linear transformation, thereby obtaining the corresponding first mapping semantic representation. Secondly, the production quality semantic representation corresponding to the first video frame at the semantic depth can be mapped using multiple second mapping matrices carried by the association mining unit to form multiple corresponding second mapping semantic representations (that is, the first second mapping semantic representation can be multiplied with the production quality semantic representation to achieve a first linear transformation, thereby obtaining a first second mapping semantic representation; the second second mapping semantic representation can be multiplied with the production quality semantic representation to achieve a second linear transformation, thereby obtaining a second second mapping semantic representation; thus, multiple linear transformations of production quality semantic information can be achieved, capturing more semantic information and relationships). The association mining unit also carries a third mapping matrix, which is used to map (e.g., multiply) the production quality semantic representation corresponding to the first video frame at the semantic depth to form a corresponding third mapping semantic representation. The third mapping semantic representation is used to replace the production quality semantic representation corresponding to the first video frame at the semantic depth and is weighted and summed with the association parameters to obtain the corresponding quality semantic association vector. Then, the transpose of each of the first mapping semantic representation and the transpose of the plurality of second mapping semantic representations can be multiplied by a dot product to obtain a plurality of local association parameters. For example, the transpose of the first mapping semantic representation and the transpose of the first second mapping semantic representation can be multiplied by a dot product to obtain the first local association parameter, and the transpose of the first mapping semantic representation and the transpose of the second second mapping semantic representation can be multiplied by a dot product to obtain the second local association parameter. In this way, various association analyses focusing on different semantic information can be realized. Finally, the average of the multiple local correlation parameters can be calculated to obtain the correlation parameters of the first video frame and the second video frame at the semantic depth. In other words, the results of multiple correlation analyses that focus on different semantic information can be fused to obtain correlation parameters that can characterize the correlation analysis results from a global and holistic perspective.

[0052] It should be noted that for step S140, the specific method for determining the manufacturing quality analysis data of the target PCB board is not limited and can be selected according to actual needs.

[0053] For example, in an alternative implementation, in order to obtain a reliable target quality semantic association vector and thus ensure the reliability of the quality analysis, the above step S140 may further include steps S141 and S142, the specific contents of each step are as follows.

[0054] Step S141: Based on the size relationship between the corresponding semantic depths, the multiple quality semantic association vectors are connected to form a target quality semantic association vector with the size of the second video frame.

[0055] In this embodiment, the multiple quality semantic association vectors can be concatenated based on the size relationship between their corresponding semantic depths, so that the multiple quality semantic association vectors at the multiple semantic depths are connected to form a target quality semantic association vector with the size of the second video frame. Thus, concatenating based on the corresponding semantic depths improves the reliability of the concatenation process. For example, if quality semantic association vectors with large differences in semantic depth are concatenated during arbitrary concatenation, it may lead to poor semantic matching.

[0056] Step S142: Perform full-connection processing on the target quality semantic association vector to form a corresponding full-connection quality semantic vector, and determine the manufacturing quality analysis data of the target PCB board based on the full-connection quality semantic vector.

[0057] In this embodiment, after obtaining the target quality semantic association vector, a fully connected processing can be performed on the target quality semantic association vector to form a corresponding fully connected quality semantic vector. Based on the fully connected quality semantic vector, the manufacturing quality analysis data of the target PCB board can be determined. For example, in one implementation, the fully connected quality semantic vector can be processed based on a linear regression function to obtain a value in a continuous interval, such as a value between 0 and 1, where a higher value indicates higher quality. In another implementation, the fully connected quality semantic vector can be processed based on a classification function (such as softmax) to obtain a corresponding probability distribution, such as the probability of high quality, the probability of average quality, and the probability of low quality. When processing based on a linear regression function, the fully connected quality semantic vector can be a 1*1 vector. When processing based on a classification function, the fully connected quality semantic vector can be a 1*n vector, where the value of n is equal to the predetermined number of quality categories.

[0058] It is understood that the specific method of connecting the multiple quality semantic association vectors in step S141 above is not limited. For example, in an alternative implementation, in order to reliably connect quality semantic association vectors of different semantic depths to achieve full fusion of semantic information, step S141 above may further include the following steps S141a, S141b, S141c, S141d, S141e, and S141f. The specific contents of each step are as follows (which can be combined with...). Figure 3 (The content shown).

[0059] Step S141a: In the connection processing of each stage, the production quality semantic representation with target size formed by the connection processing of the previous stage is semantically expanded to form the first production quality semantic representation.

[0060] In this embodiment, in each stage of the connection processing, the production quality semantic representation with a target size formed by the connection processing of the previous stage can be semantically expanded to form a first production quality semantic representation. Specifically, in the first stage of the connection processing, after semantically expanding the quality semantic association vector corresponding to the current stage, it is merged with the production quality semantic representations of the first video frame and the second video frame at the semantic depth corresponding to the first stage, forming a merged production quality semantic representation. This merged production quality semantic representation is then semantically compressed to obtain a production quality semantic representation with the target size, which is determined as the production quality semantic representation with the target size formed by the first stage of the connection processing. The semantic depth corresponding to the first stage belongs to the last semantic depth (i.e., connection processing is performed sequentially from the last semantic depth to the first semantic depth to ensure that the span of the semantic depth in the connection processing is not too large). For example, semantic expansion processing can refer to upsampling to increase the vector size, and semantic compression processing can refer to downsampling to decrease the vector size.

[0061] Step S141b involves semantic expansion of the quality semantic association vector corresponding to the current stage, adjusting the quality semantic association vector to a second production quality semantic representation with the target size.

[0062] In this embodiment, the quality semantic association vector corresponding to the current stage can also be semantically expanded (e.g., upsampled) to adjust the quality semantic association vector to a second quality semantic representation with a target size. For example, if the current stage is the second stage, the semantic depth of the corresponding quality semantic association vector is the penultimate semantic depth; if the current stage is the last stage, the semantic depth of the corresponding quality semantic association vector is the first semantic depth.

[0063] Step S141c: Summing the first manufacturing quality semantic representation and the second manufacturing quality semantic representation to obtain the third manufacturing quality semantic representation.

[0064] In this embodiment, after obtaining the first and second production quality semantic representations, the first and second production quality semantic representations can be summed to obtain a third production quality semantic representation, thus achieving the fusion of the two semantic information types. Alternatively, in other embodiments, after summing, normalization can be performed to obtain the third production quality semantic representation.

[0065] Step S141d: The third production quality semantic representation is merged with the production quality semantic representations of the first video frame and the second video frame at the corresponding semantic depth in the current stage to form the merged production quality semantic representation corresponding to the current stage.

[0066] In this embodiment, after obtaining the third production quality semantic representation, the third production quality semantic representation can be merged with the production quality semantic representations of the first video frame and the second video frame at the semantic depth corresponding to the current stage, respectively, to form the merged production quality semantic representation corresponding to the current stage. In other words, through the above steps, the results of the previous stage can be fused with the results of the association mining corresponding to the current stage, and then the original mining results can be further fused to achieve a large-scale fusion of global semantic information, thereby ensuring the semantic richness of the formed merged production quality semantic representation.

[0067] Step S141e: Perform semantic compression processing on the merged production quality semantic representation corresponding to the current stage to obtain a production quality semantic representation with a target size, and determine it as the production quality semantic representation with a target size formed by the connection processing of the current stage.

[0068] In this embodiment of the application, after forming the merged production quality semantic representation, the merged production quality semantic representation corresponding to the current stage can be semantically compressed (as shown in the sampling below) to obtain a production quality semantic representation with a target size, and determined as the production quality semantic representation with a target size formed by the connection processing of the current stage.

[0069] Step S141f: The production quality semantic representation with target size formed by the connection processing in the last stage is determined as the corresponding target quality semantic association vector.

[0070] In this embodiment of the application, after processing each stage based on the aforementioned steps, if the production quality semantic representation with a target size is obtained from the last stage of connection processing, the production quality semantic representation with a target size formed by the last stage of connection processing can be determined as the corresponding target quality semantic association vector. In this way, the gradual fusion of semantic information at various semantic depths can be achieved.

[0071] It is understood that the specific method of merging in step S141d above is not limited. For example, in an alternative implementation, in order to fully integrate the three production quality semantic representations, step S141d may further include the following (which can be combined with...). Figure 4 (Content shown) First, for each level of semantic compression processing in the multi-level semantic compression process, the output semantic representation of the previous level of semantic compression processing and the production quality semantic representation of the second video frame at the corresponding semantic depth in the current stage are concatenated. The concatenated semantic representation is then downsampled, and the downsampled semantic representation is subjected to self-attention processing to obtain the output semantic representation of the current level of semantic compression processing. The output semantic representation of the previous level of semantic compression processing in the first level is the third production quality semantic representation. In other words, in the process of multi-level semantic compression processing, the production quality semantic representation of the second video frame at the corresponding semantic depth in the current stage is gradually fused. Secondly, for each level of semantic expansion processing in the multi-level semantic expansion process, the output semantic representation of the previous level of semantic expansion processing and the production quality semantic representation of the first video frame at the corresponding semantic depth in the current stage are concatenated. The concatenated semantic representation is then upsampled, and the upsampled semantic representation is subjected to self-attention processing to obtain the output semantic representation of the current level of semantic expansion processing. The output semantic representation of the previous level of semantic expansion processing in the first level is the output semantic representation of the last level of semantic compression processing. In other words, in the process of multi-level semantic expansion processing, the production quality semantic representation of the first video frame at the corresponding semantic depth in the current stage is gradually fused. In this way, by realizing the gradual fusion of the semantic information of the first video frame and the second video frame in different semantic processing, the fusion accuracy can be improved. Then, the output semantic representation of the last level of semantic expansion processing is determined as the merging quality semantic representation corresponding to the current stage.

[0072] Combination Figure 5 This application also provides a PCB manufacturing quality analysis device applicable to the aforementioned electronic devices. The PCB manufacturing quality analysis device may include a video frame determination module, a semantic mining module, a semantic association module, and a quality analysis module.

[0073] The video frame determination module is used to determine a first video frame and a second video frame. The first video frame is formed by acquiring an image of a target PCB board, and the second video frame is formed by acquiring an image of a reference PCB board. The manufacturing quality of the reference PCB board meets the application requirements. In this embodiment, the video frame determination module can be used to perform... Figure 2 The relevant content regarding the video frame determination module in step S110 shown can be found in the previous description of step S110.

[0074] The semantic mining module is used to mine the production quality semantic representations of the first video frame and the second video frame at multiple semantic depths. In this embodiment, the semantic mining module can be used to perform... Figure 2 The relevant content regarding the semantic mining module in step S120 shown can be found in the previous description of step S120.

[0075] The semantic association module is used to, for each of the plurality of semantic depths, create quality semantic representations based on the first video frame and the second video frame respectively corresponding to the semantic depth at that semantic depth, and analyze the quality semantic association vector at that semantic depth. In this embodiment of the application, the semantic association module can be used to perform... Figure 2 The relevant content regarding the semantic association module in step S130 shown can be found in the previous description of step S130.

[0076] The quality analysis module is used to determine a target quality semantic association vector based on multiple quality semantic association vectors analyzed at the multiple semantic depths, and to determine the manufacturing quality analysis data of the target PCB board based on the target quality semantic association vector. In this embodiment, the quality analysis module can be used to perform... Figure 2 The relevant content regarding the quality analysis module in step S140 shown can be found in the preceding description of step S140.

[0077] In this embodiment of the application, corresponding to the above-described PCB manufacturing quality analysis method applied to the electronic device, a computer-readable storage medium is also provided, which stores a computer program that executes the various steps of the PCB manufacturing quality analysis method when the computer program is run.

[0078] The steps executed by the aforementioned computer program during runtime will not be described in detail here; please refer to the explanation of the PCB manufacturing quality analysis method above.

[0079] In summary, the PCB manufacturing quality analysis method, apparatus, and equipment provided in this application first determine a first video frame and a second video frame; second, they mine the manufacturing quality semantic representations of the first and second video frames at multiple semantic depths; then, for each semantic depth, based on the manufacturing quality semantic representations of the first and second video frames at each semantic depth, they analyze the quality semantic association vector at that semantic depth; finally, based on the multiple quality semantic association vectors determined and analyzed at multiple semantic depths, they determine a target quality semantic association vector, and based on the target quality semantic association vector, they determine the manufacturing quality analysis data. Based on the above, because it mines manufacturing quality semantic representations at multiple semantic depths, the mined semantic information can cover everything from shallow to deep levels and from concrete to abstract, achieving a full representation of global semantic information. Therefore, it can improve the reliability of quality analysis. In addition, since the semantic information of the reference PCB board is also combined for association mining, different quality semantic association vectors can be mined for target PCB boards with different qualities. This enables high-precision representation of semantic information of different qualities, which can further improve the reliability of quality analysis and improve the problem of relatively low reliability of PCB manufacturing quality analysis in the existing technology.

[0080] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0081] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0082] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0083] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for analyzing PCB manufacturing quality, characterized in that, include: A first video frame and a second video frame are determined, wherein the first video frame is formed by performing an image acquisition operation on a target PCB board, and the second video frame is formed by performing an image acquisition operation on a reference PCB board, wherein the manufacturing quality of the reference PCB board meets the application requirements. The production quality semantic representations of the first video frame and the second video frame at multiple semantic depths are extracted. For each of the plurality of semantic depths, a quality semantic representation is created based on the first video frame and the second video frame respectively corresponding to the semantic depth, and the quality semantic association vector under the semantic depth is analyzed. In each stage of the connection processing, the production quality semantic representation with the target size formed by the connection processing of the previous stage is semantically expanded to form a first production quality semantic representation. In the first stage of the connection processing, after the quality semantic association vector corresponding to the current stage is semantically expanded, it is merged with the production quality semantic representations of the first video frame and the second video frame at the semantic depth corresponding to the first stage to form a corresponding merged production quality semantic representation. The merged production quality semantic representation is then semantically compressed to obtain a production quality semantic representation with the target size, which is determined as the production quality semantic representation with the target size formed by the connection processing of the first stage, and the semantic depth corresponding to the first stage belongs to the last semantic depth. The quality semantic association vector corresponding to the current stage is semantically expanded so that the quality semantic association vector is adjusted to a second production quality semantic representation with the target size; The first and second manufacturing quality semantic representations are summed to obtain the third manufacturing quality semantic representation. The third production quality semantic representation is merged with the production quality semantic representations of the first video frame and the second video frame at the semantic depth corresponding to the current stage to form the merged production quality semantic representation corresponding to the current stage. The semantic compression process is performed on the merged production quality semantic representation corresponding to the current stage to obtain a production quality semantic representation with a target size, and this is determined as the production quality semantic representation with a target size formed by the connection process of the current stage. The production quality semantic representation with the target size formed by the connection processing in the last stage is determined as the corresponding target quality semantic association vector; The target quality semantic association vector is processed by a fully connected layer to form a corresponding fully connected quality semantic vector. Based on the fully connected quality semantic vector, the manufacturing quality analysis data of the target PCB board is determined.

2. The PCB manufacturing quality analysis method according to claim 1, characterized in that, The step of mining the production quality semantic representations of the first video frame and the second video frame at multiple semantic depths includes: The semantic mining unit in the target production quality analysis model includes multiple semantic mining sub-units, which mine production quality semantic representations at different semantic depths, thereby outputting the production quality semantic representations of the first video frame and the second video frame at multiple semantic depths respectively. Wherein, the size of the production quality semantic representation of the first video frame or the second video frame at the target semantic depth has a target ratio with the size of the corresponding first video frame or the second video frame, and the target ratio is less than 1.

3. The PCB manufacturing quality analysis method according to claim 2, characterized in that, The step of mining production quality semantic representations at different semantic depths through multiple semantic mining sub-units included in the semantic mining unit of the target production quality analysis model, and outputting the production quality semantic representations of the first video frame and the second video frame at multiple semantic depths respectively, includes: For the first semantic mining subunit included in the semantic mining unit of the target production quality analysis model, the first video frame and the second video frame are convolved by the convolutional network layer in the semantic mining subunit to form the corresponding first convolutional semantic representation and second convolutional semantic representation. The first convolutional semantic representation and the second convolutional semantic representation are then self-attention processed by the self-attention network layer in the semantic mining subunit to form the production quality semantic representations of the first video frame and the second video frame at the first semantic depth. For each semantic mining subunit from the second to the last in the semantic mining unit, the production quality semantic representations at the previous semantic depth are convolved by the convolutional network layer in the semantic mining subunit to form the corresponding first convolutional semantic representation and second convolutional semantic representation. The first convolutional semantic representation and the second convolutional semantic representation are then self-attentioned by the self-attention network layer in the semantic mining subunit to form the production quality semantic representations of the first video frame and the second video frame at the current semantic depth.

4. The PCB manufacturing quality analysis method according to claim 1, characterized in that, The step of creating quality semantic representations based on the first video frame and the second video frame respectively at each of the plurality of semantic depths, and analyzing the quality semantic association vector at the semantic depth, includes: A correlation analysis is performed on the production quality semantic representation of the first video frame and the production quality semantic representation of the second video frame at the semantic depth to obtain the correlation parameters of the first video frame and the second video frame at the semantic depth. Based on the association parameters, a weighted summation calculation is performed on the production quality semantic representation corresponding to the first video frame at the semantic depth, and a quality semantic association vector at the semantic depth is output, wherein the association parameters are a parameter distribution matrix.

5. The PCB manufacturing quality analysis method according to claim 4, characterized in that, The step of performing correlation analysis on the production quality semantic representations of the first video frame and the second video frame at the semantic depth to obtain the correlation parameters of the first video frame and the second video frame at the semantic depth includes: The first mapping matrix carried by the association mining unit in the target production quality analysis model is used to map the production quality semantic representation of the second video frame at the semantic depth to form the corresponding first mapping semantic representation. The association mining unit carries multiple second mapping matrices to map the production quality semantic representation of the first video frame at the semantic depth, forming multiple corresponding second mapping semantic representations. The association mining unit also carries a third mapping matrix, which is used to map the production quality semantic representation of the first video frame at the semantic depth to form a corresponding third mapping semantic representation. The third mapping semantic representation is used to replace the production quality semantic representation of the first video frame at the semantic depth and is weighted and summed with the association parameters to obtain the corresponding quality semantic association vector. The transpose of each of the first mapping semantic representations and the transpose of each of the plurality of second mapping semantic representations is multiplied by a dot product to obtain a plurality of local association parameters; The mean of the multiple local correlation parameters is calculated to obtain the correlation parameters of the first video frame and the second video frame at the semantic depth.

6. The PCB manufacturing quality analysis method according to claim 1, characterized in that, The step of merging the third production quality semantic representation with the production quality semantic representations of the first video frame and the second video frame at the semantic depth corresponding to the current stage to form the merged production quality semantic representation corresponding to the current stage includes: For each level of semantic compression processing in the multi-level semantic compression processing, the output semantic representation of the previous level of semantic compression processing and the production quality semantic representation of the second video frame at the corresponding semantic depth in the current stage are concatenated, and the concatenated semantic representation is downsampled, and the downsampled semantic representation is subjected to self-attention processing to obtain the output semantic representation of the current level of semantic compression processing. The output semantic representation of the previous level of semantic compression processing of the first level is the third production quality semantic representation. For each level of semantic expansion processing in the multi-level semantic expansion processing, the output semantic representation of the previous level of semantic expansion processing and the production quality semantic representation of the first video frame at the corresponding semantic depth at the current stage are concatenated, and the concatenated semantic representation is upsampled. Then, the upsampled semantic representation is subjected to self-attention processing to obtain the output semantic representation of the current level of semantic expansion processing. The output semantic representation of the previous level of semantic expansion processing of the first level is the output semantic representation of the last level of semantic compression processing. The output semantic representation of the semantic expansion process at the last level is determined as the quality semantic representation of the merging process corresponding to the current stage.

7. A PCB manufacturing quality analysis device, characterized in that, include: The video frame determination module is used to determine a first video frame and a second video frame, wherein the first video frame is formed by performing an image acquisition operation on a target PCB board, and the second video frame is formed by performing an image acquisition operation on a reference PCB board, wherein the manufacturing quality of the reference PCB board meets the application requirements. The semantic mining module is used to mine the production quality semantic representations of the first video frame and the second video frame at multiple semantic depths. The semantic association module is used to create a quality semantic representation based on the first video frame and the second video frame respectively corresponding to the semantic depth for each of the plurality of semantic depths, and analyze the quality semantic association vector under the semantic depth. The quality analysis module is used to perform semantic expansion processing on the production quality semantic representation with a target size formed by the connection processing of the previous stage in each stage of connection processing, forming a first production quality semantic representation. Specifically, in the first stage of connection processing, after semantic expansion processing of the quality semantic association vector corresponding to the current stage, it is merged with the production quality semantic representations of the first video frame and the second video frame at the semantic depth corresponding to the first stage, forming a merged production quality semantic representation. This merged production quality semantic representation is then semantically compressed to obtain a production quality semantic representation with a target size, which is determined to be the production quality semantic representation with a target size formed by the connection processing of the first stage, and the semantic depth corresponding to the first stage belongs to the last semantic depth. Semantic expansion processing is then performed on the quality semantic association vector corresponding to the current stage, adjusting the quality semantic association vector to a second production quality semantic representation with a target size. The process involves: summing the first and second production quality semantic representations to obtain a third production quality semantic representation; merging the third production quality semantic representation with the production quality semantic representations of the first and second video frames at the current stage's corresponding semantic depth to form a merged production quality semantic representation for the current stage; performing semantic compression on the merged production quality semantic representation for the current stage to obtain a production quality semantic representation of the target size, and determining it as the production quality semantic representation of the target size formed by the connection processing of the current stage; determining the production quality semantic representation of the target size formed by the connection processing of the last stage as the corresponding target quality semantic association vector; performing full connection processing on the target quality semantic association vector to form a corresponding fully connected quality semantic vector; and determining the production quality analysis data of the target PCB board based on the fully connected quality semantic vector.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the PCB manufacturing quality analysis method according to any one of claims 1-6.

Images