Data management method and system for realizing bonded silver wire based on convolutional neural model

Abstract

The invention relates to the technical field of bonding silver wires, and discloses a data management method and system for realizing bonding silver wires based on a convolutional neural model, and the method comprises the steps: respectively carrying out the quality sublimation processing of silver wire imaging data and bonding process data, obtaining target imaging data and target process data, carrying out the sorting processing of the target imaging data, and obtaining a sorting result; a microscopic topological image and a macroscopic posture image are obtained; carrying out feature mining on the microscopic topological imaging and the macroscopic posture imaging to obtain microscopic topological features and macroscopic contour features, and carrying out feature coupling processing on the microscopic topological features and the macroscopic contour features to obtain high-order coupling features; calculating an index confidence gradient corresponding to the process index, and analyzing a silver wire quality effect corresponding to the bonded silver wire; and analyzing the abnormal key elements corresponding to the bonding silver wire, and executing abnormal processing on the bonding silver wire to obtain a processing result. The main purpose of the invention is to solve the problem of low data management efficiency of bonding silver wires.

Description

Technical Field

[0001] The invention relates to a data management method and system for silver bonding wires based on a convolutional neural model, and belongs to the technical field of silver bonding wires. Background Art

[0002] Silver bonding wire is a key material widely used in the field of electronic packaging. With the rapid development of the modern electronic manufacturing industry, the functions of electronic products are becoming increasingly complex and the performance requirements are constantly increasing. The quality and production efficiency of silver bonding wire are crucial to ensure the reliability and performance of electronic products. Therefore, efficient and accurate data management of the production process of silver bonding wire has become an urgent problem to be solved.

[0003] Most of the existing data management methods for silver bonding wires are based on traditional data analysis and processing methods. First, technicians will collect various data in the production process, including imaging data of silver bonding wires (such as microstructure images and macroscopic appearance images obtained through a microscope) and bonding process data (such as temperature, pressure, time and other parameters). Then, simple statistical analysis methods are used to organize and preliminarily analyze these data. According to experience and preset rules, the data are screened and classified to extract some key data features that are considered to have an impact on quality. Based on these analysis results, corresponding programs are written or control strategies are formulated to guide the production process and quality control. However, this method mainly relies on manual experience and simple processing of limited data features. It is impossible to effectively mine and utilize a large amount of nonlinear and high-dimensional data in the complex production process of silver bonding wires. Moreover, with slight changes in the production environment and process, these data management methods based on fixed rules are difficult to adapt in time, resulting in low accuracy in evaluating the quality of silver bonding wires and optimizing the production process. It cannot meet the requirements of modern electronic manufacturing for high-quality and high-stability production of silver bonding wires. Therefore, a method that can improve the data management efficiency of silver bonding wires is needed. Summary of the invention

[0004] The present invention provides a method and system for realizing data management of silver bonding wires based on a convolutional neural model, the main purpose of which is to solve the problem of low efficiency of data management of silver bonding wires.

[0005] To achieve the above object, the present invention provides a method for data management of silver bonding wire based on a convolutional neural model, comprising: Acquire silver wire imaging data and bonding process data corresponding to the bonding silver wire, perform mass sublimation processing on the silver wire imaging data and the bonding process data respectively to obtain target imaging data and target process data, and perform sorting processing on the target imaging data to obtain microscopic topological imaging and macroscopic body imaging; The microscopic topological imaging, the macroscopic body imaging and the target process data are respectively input as input data into a pre-trained convolutional neural model, and the microscopic topological imaging and the macroscopic body imaging are respectively subjected to feature mining by using the deformable convolutional layer in the convolutional neural model to obtain microscopic topological features and macroscopic contour features, and the microscopic topological features and the macroscopic contour features are subjected to feature coupling processing by using the bilinear fusion layer in the convolutional neural model to obtain high-order coupling features; Querying the process index corresponding to the bonding silver wire, calculating the index confidence gradient corresponding to the process index according to the target process data, and analyzing the silver wire quality effect corresponding to the bonding silver wire by combining the index confidence gradient and the high-order coupling feature; When the quality of the silver wire does not meet the preset quality, based on the indicator confidence gradient and the high-order coupling characteristics, the abnormal key factors corresponding to the bonding silver wire are analyzed, and based on the abnormal key factors, abnormal processing of the bonding silver wire is performed to obtain a processing result.

[0006] Optionally, the performing quality sublimation processing on the silver wire imaging data and the bonding process data respectively to obtain target imaging data and target process data comprises: Performing data cleaning processing on the bonding process data to obtain cleaning process data; Performing visual parameter optimization on the silver wire imaging data to obtain optimized imaging data; Performing data repair processing on the cleaning process data to obtain repair process data; Performing image gain processing on the optimized imaging data to obtain gain imaging data; The repair process data and the gain imaging data are respectively subjected to data screening processing to obtain target imaging data and target process data.

[0007] Optionally, performing image gain processing on the optimized imaging data to obtain gain imaging data includes: Performing filtering processing on the optimized imaging data to obtain filtered imaging data; Identifying an imaging contour corresponding to the filtered imaging data, and analyzing contour integrity corresponding to the imaging contour; According to the contour integrity, performing contour fitting processing on the filtered imaging data to obtain fitting imaging data; The fitting imaging data is sharpened and enhanced to obtain gain imaging data.

[0008] Optionally, the sorting and processing of the target imaging data to obtain microscopic topological imaging and macroscopic body imaging includes: Identify the number of image pixels corresponding to each image in the target imaging data, and set a pixel matrix corresponding to each image in the target imaging data based on the number of image pixels; Performing grayscale processing on each image in the target imaging data to obtain a target grayscale image; Based on the pixel matrix, calculating the horizontal grayscale variation rate and the vertical grayscale variation rate corresponding to the target grayscale image; Combining the horizontal grayscale variation rate and the vertical grayscale variation rate, the grayscale gradient amplitude and gradient direction angle corresponding to the target grayscale image can be calculated by the following formula: ; Among them, A represents the grayscale gradient amplitude corresponding to the target grayscale image, Represents the gradient direction angle corresponding to the target grayscale image, represents the horizontal grayscale variation rate, represents the vertical grayscale change rate, represents the inverse tangent function; Based on the grayscale gradient amplitude and the gradient direction angle, the target imaging data is sorted and processed to obtain microscopic topological imaging and macroscopic body imaging.

[0009] Optionally, the method of using the deformable convolution layer in the convolutional neural model to perform feature mining on the microscopic topological imaging and the macroscopic body imaging to obtain microscopic topological features and macroscopic contour features includes: Identifying the topological texture in the microscopic topological imaging, and calculating the texture complexity corresponding to the topological texture; Detecting a body boundary in the macroscopic body imaging, and calculating a boundary curvature rate corresponding to the body boundary; In combination with the texture complexity and the boundary curvature, setting a convolution scale parameter of the deformable convolution layer; According to the convolution scale parameter, the deformable convolution layer is used to perform feature mining on the microscopic topological imaging and the macroscopic body imaging respectively to obtain microscopic topological features and macroscopic contour features.

[0010] Optionally, calculating the texture complexity corresponding to the topological texture includes: Identify the texture grayscale value and grayscale level corresponding to the topological texture, and construct a texture grayscale co-occurrence matrix corresponding to the topological texture according to the texture grayscale value; Based on the texture grayscale co-occurrence matrix, counting the matrix occurrence frequency corresponding to the texture grayscale value; Combining the grayscale level and the matrix occurrence frequency, the texture complexity corresponding to the topological texture is calculated by the following formula: ; Among them, D represents the texture complexity corresponding to the topological texture, The coordinates are The matrix occurrence frequency corresponding to the texture grayscale value, i and j represent the coordinates corresponding to the texture grayscale value, and E represents the grayscale level.

[0011] Optionally, calculating the boundary curvature rate corresponding to the posture boundary includes: Detecting boundary pixel points in the body boundary and determining pixel coordinate values ​​corresponding to the boundary pixel points; According to the pixel coordinate value, the local curvature corresponding to the boundary pixel point can be calculated by the following formula: ; in, Represents the local curvature rate corresponding to the boundary pixel point, and Indicates the pixel coordinate value corresponding to the dth pixel point in the boundary pixel point, and Indicates the pixel coordinate value corresponding to the d-1th pixel point in the boundary pixel point. and Indicates the pixel coordinate value corresponding to the d+1th pixel point in the boundary pixel point; The boundary curvature rate corresponding to the body boundary is calculated according to the local curvature rate.

[0012] Optionally, the microscopic topological features and the macroscopic contour features are subjected to feature coupling processing using a bilinear fusion layer in the convolutional neural model to obtain high-order coupling features, including: Using the pooling function in the bilinear fusion layer, the microscopic topological features and the macroscopic contour features are respectively subjected to feature dimensionality reduction processing to obtain reduced-dimensionality topological features and reduced-dimensionality contour features; Using the vector conversion layer in the bilinear fusion layer to respectively vectorize the reduced dimension topological feature and the reduced dimension contour feature to obtain a reduced dimension topological feature vector and a reduced dimension contour feature vector; Calculating the bilinear outer product between the reduced dimension topological feature vector and the reduced dimension profile feature vector using the bilinear function in the bilinear fusion layer; According to the bilinear outer product, construct a bilinear matrix corresponding to the reduced dimension topological feature vector and the reduced dimension profile feature vector; According to the bilinear matrix, a high-order coupling feature vector corresponding to the reduced-dimensional topological feature vector and the reduced-dimensional contour feature vector is determined, and the high-order coupling feature vector is characterized to obtain a high-order coupling feature.

[0013] Optionally, calculating the indicator confidence gradient corresponding to the process indicator according to the target process data includes: Extracting the index process data corresponding to the process index from the target process data; Quantifying the index process data to obtain an index process quantization value, and calculating a weight coefficient corresponding to the index process quantization value; Combining the weight coefficient and the process quantization value of the indicator, the indicator confidence gradient corresponding to the process indicator is calculated by the following formula: ; Among them, G represents the confidence gradient of the indicator corresponding to the process indicator, Indicates the weight coefficient corresponding to the e-th quantized value in the indicator process quantized value, represents the e-th indicator process quantization value, e represents the serial number corresponding to the indicator process quantization value, and n represents the number of indicator process quantization values. represents the reference quantization value, represents the attenuation factor, Indicates the time point corresponding to the nth quantized value in the process quantized value of the indicator. Indicates the time point corresponding to the e-th quantized value in the indicator process quantized value.

[0014] A data management system for bonding silver wire is implemented based on a convolutional neural model, the system comprising: A data processing module, used for acquiring silver wire imaging data and bonding process data corresponding to the bonding silver wire, performing mass sublimation processing on the silver wire imaging data and the bonding process data respectively to obtain target imaging data and target process data, and performing sorting processing on the target imaging data to obtain microscopic topological imaging and macroscopic body imaging; A feature coupling module, used to input the microscopic topological imaging, the macroscopic body imaging and the target process data as input data into a pre-trained convolutional neural model, respectively, and use the deformable convolution layer in the convolutional neural model to perform feature mining on the microscopic topological imaging and the macroscopic body imaging to obtain microscopic topological features and macroscopic contour features, and use the bilinear fusion layer in the convolutional neural model to perform feature coupling processing on the microscopic topological features and the macroscopic contour features to obtain high-order coupling features; A silver wire quality and efficiency analysis module, used to query the process index corresponding to the bonding silver wire, calculate the index confidence gradient corresponding to the process index according to the target process data, and analyze the silver wire quality and efficiency corresponding to the bonding silver wire by combining the index confidence gradient and the high-order coupling feature; The exception processing module is used to analyze the abnormal key factors corresponding to the bonding silver wire based on the indicator confidence gradient and the high-order coupling characteristics when the quality of the silver wire does not meet the preset quality, and based on the abnormal key factors, perform abnormal processing on the bonding silver wire to obtain a processing result.

[0015] Compared with the problems described in the background technology, the present invention can remove noise interference and erroneous data by performing quality sublimation processing on the silver wire imaging data and the bonding process data respectively, so as to improve the data quality corresponding to the silver wire imaging data and the bonding process data, and lay a foundation for improving the accuracy of subsequent analysis of the quality of the bonding silver wire. The present invention can adaptively capture the complex shape changes and detail features of different scales in the imaging by utilizing the deformable convolution layer in the convolutional neural model to perform feature mining on the microscopic topological imaging and the macroscopic body imaging, and effectively improve the characterization capability of microscopic and macroscopic features, so as to provide a basis for subsequent more accurate analysis and application. By providing more characteristic information, the present invention calculates the indicator confidence gradient corresponding to the process indicator according to the target process data, and the reliability of the process indicator can be understood through the indicator confidence gradient, thereby improving the accuracy of the subsequent analysis of the silver wire quality and efficiency corresponding to the bonding silver wire. It should be understood that when the silver wire quality and efficiency do not meet the preset quality and efficiency, it means that the quality of the bonding silver wire does not meet the standard. The present invention analyzes the abnormal key elements corresponding to the bonding silver wire based on the indicator confidence gradient and the high-order coupling characteristics, and then obtains the abnormal influence cause of the bonding silver wire, thereby providing a basis for the subsequent abnormal handling of the bonding silver wire. Therefore, the present invention proposes a data management method and system for bonding silver wire based on a convolutional neural model to improve the data management efficiency of bonding silver wire. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of a method for implementing data management of silver bonding wires based on a convolutional neural model provided by an embodiment of the present invention; Figure 2 A functional module diagram of a data management system for bonding silver wires based on a convolutional neural model provided in one embodiment of the present invention.

[0017] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0018] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0019] The embodiment of the present application provides a data management method for bonding silver wires based on a convolutional neural model. The execution subject of the data management method for bonding silver wires based on a convolutional neural model includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the data management method for bonding silver wires based on a convolutional neural model can be executed by software or hardware installed on a terminal device or a server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0020] Embodiment 1: Reference Figure 1 FIG. 1 is a flow chart of a method for implementing data management of silver bonding wires based on a convolutional neural model according to an embodiment of the present invention. In this embodiment, the method for implementing data management of silver bonding wires based on a convolutional neural model includes: S1. Acquire silver wire imaging data and bonding process data corresponding to the bonding silver wire, perform mass sublimation processing on the silver wire imaging data and the bonding process data respectively to obtain target imaging data and target process data, perform sorting processing on the target imaging data to obtain microscopic topological imaging and macroscopic body imaging.

[0021] The present invention can remove noise interference and erroneous data by performing quality sublimation processing on the silver wire imaging data and the bonding process data respectively, so as to improve the data quality corresponding to the silver wire imaging data and the bonding process data, and lay a foundation for improving the accuracy of subsequent analysis of the quality of the bonding silver wire. It should be explained that the bonding silver wire is a material widely used in the field of electronic packaging, and the silver wire imaging data is a product visualization image data set corresponding to the bonding silver wire and a data set collected during the bonding process. The target imaging data and the target process data are respectively data obtained after the data quality corresponding to the silver wire imaging data and the bonding process data is improved.

[0022] In detail, the silver wire imaging data and the bonding process data are subjected to quality sublimation processing respectively to obtain target imaging data and target process data, including: Performing data cleaning processing on the bonding process data to obtain cleaning process data; Performing visual parameter optimization on the silver wire imaging data to obtain optimized imaging data; Performing data repair processing on the cleaning process data to obtain repair process data; Performing image gain processing on the optimized imaging data to obtain gain imaging data; The repair process data and the gain imaging data are respectively subjected to data screening processing to obtain target imaging data and target process data.

[0023] It should be explained that the cleaning process data is the data obtained after removing the data generated by the noise interference in the bonding process data, the optimized imaging data is the image data obtained after adjusting the image brightness, contrast and clarity in the silver wire imaging data, the repair process data is the data obtained after repairing the erroneous data in the cleaning process data, and the gain imaging data is the image data obtained after various enhancement processing of the optimized imaging data.

[0024] Furthermore, the data cleaning processing of the bonding process data can be achieved through the box plot method; the visual parameter optimization of the silver wire imaging data can be achieved through the histogram equalization method; the data repair processing of the cleaning process data can be achieved through the mean interpolation method; the data screening processing of the repair process data and the gain imaging data can be achieved through the normal distribution method, and the data can be screened according to the mean and standard deviation corresponding to the data.

[0025] Further, as an optional embodiment of the present invention, performing image gain processing on the optimized imaging data to obtain gain imaging data includes: Performing filtering processing on the optimized imaging data to obtain filtered imaging data; Identifying an imaging contour corresponding to the filtered imaging data, and analyzing contour integrity corresponding to the imaging contour; According to the contour integrity, performing contour fitting processing on the filtered imaging data to obtain fitting imaging data; The fitting imaging data is sharpened and enhanced to obtain gain imaging data.

[0026] It should be explained that the filtered imaging data is the data obtained after suppressing the noise interference in the optimized imaging data, the imaging contour is the image shape boundary in the filtered imaging data, the contour integrity is a description of the degree of completeness corresponding to the imaging contour, and the fitted imaging data is the data obtained after fitting and optimizing the incomplete contour in the filtered imaging data.

[0027] Furthermore, the filtering processing of the optimized imaging data can be achieved through a low-pass filter; the imaging contour corresponding to the filtered imaging data can be identified through an edge detection algorithm, such as a Sobel operator; the contour integrity corresponding to the imaging contour can be analyzed through the connectivity of the contour, such as starting from a pixel point on the contour and traversing all pixel points according to the connectivity rule. If it is found that some pixel points that should originally belong to the contour cannot be reached during the traversal process, then it can be judged that the contour is not connected and the contour integrity is incomplete; based on the contour integrity, the filtered imaging data can be contour fitted by a curve fitting method to obtain fitted imaging data; the sharpening enhancement of the fitted imaging data can be achieved by an unsharp mask method.

[0028] The present invention can classify the target imaging data into microscopic topological imaging and macroscopic body imaging by sorting and processing the target imaging data, thereby effectively improving the accuracy and comprehensiveness of subsequent quality assessment of bonding silver wires. It should be explained that the microscopic topological imaging and the macroscopic body imaging are the microscopic structure detail images and large-scale structure images in the target imaging data.

[0029] In detail, the target imaging data is sorted and processed to obtain microscopic topological imaging and macroscopic body imaging, including: Identify the number of image pixels corresponding to each image in the target imaging data, and set a pixel matrix corresponding to each image in the target imaging data based on the number of image pixels; Performing grayscale processing on each image in the target imaging data to obtain a target grayscale image; Based on the pixel matrix, calculating the horizontal grayscale variation rate and the vertical grayscale variation rate corresponding to the target grayscale image; Combining the horizontal grayscale variation rate and the vertical grayscale variation rate, the grayscale gradient amplitude and gradient direction angle corresponding to the target grayscale image can be calculated by the following formula: ; Among them, A represents the grayscale gradient amplitude corresponding to the target grayscale image, Represents the gradient direction angle corresponding to the target grayscale image, represents the horizontal grayscale variation rate, represents the vertical grayscale change rate, represents the inverse tangent function; Based on the grayscale gradient amplitude and the gradient direction angle, the target imaging data is sorted and processed to obtain microscopic topological imaging and macroscopic body imaging.

[0030] It should be explained that the image pixel number is the total number of pixels corresponding to each image in the target imaging data, the pixel matrix is ​​a two-dimensional array structure for storing pixel information of each image in the target imaging data, the target grayscale image is an image in which each image in the target imaging data is represented by only a single color, the horizontal grayscale variation rate and the vertical grayscale variation rate respectively represent the severity of the grayscale changes in the horizontal and vertical directions corresponding to the target grayscale image, the grayscale gradient amplitude represents the comprehensive intensity of the grayscale changes corresponding to the target grayscale image, and the gradient direction angle represents the main direction of the grayscale changes corresponding to the target grayscale image.

[0031] Furthermore, the identification of the number of image pixels corresponding to each image in the target imaging data can be achieved by a pixel identification tool, and the pixel identification tool is compiled by a scripting language, such as a JS scripting language; based on the number of image pixels, the pixel matrix corresponding to each image in the target imaging data can be set by a direct allocation method, such as a 3×3 matrix; the grayscale processing of each image in the target imaging data can be achieved by an average value method, and the average value of the pixel RGB channel value is calculated as the grayscale value; based on the pixel matrix, the horizontal grayscale variation rate and the vertical grayscale variation rate corresponding to the target grayscale image can be calculated by a custom difference method; based on the grayscale gradient amplitude and the Gradient direction angle, sorting and processing the target imaging data to obtain microscopic topological imaging and macroscopic body imaging. If the grayscale gradient amplitude of a pixel point is higher than the set threshold and its gradient direction angle is relatively messy, then the pixel point is classified as microscopic topological imaging. For example, in the image, those pixels showing the grain boundary or tiny texture of the bonding silver wire will be sorted into the microscopic topological imaging. If the grayscale gradient amplitude of a pixel point is lower than the set threshold and its gradient direction angle is relatively consistent, then the pixel point is classified as macroscopic body imaging. For example, in the image, those pixels showing the overall shape and general outline of the bonding silver wire will be sorted into the macroscopic body imaging.

[0032] S2. Input the microscopic topological imaging, the macroscopic body imaging and the target process data as input data into a pre-trained convolutional neural model respectively, use the deformable convolution layer in the convolutional neural model to perform feature mining on the microscopic topological imaging and the macroscopic body imaging respectively to obtain microscopic topological features and macroscopic contour features, use the bilinear fusion layer in the convolutional neural model to perform feature coupling processing on the microscopic topological features and the macroscopic contour features to obtain high-order coupling features.

[0033] The present invention utilizes the deformable convolution layer in the convolutional neural model to perform feature mining on the microscopic topological imaging and the macroscopic body imaging respectively, so as to adaptively capture the complex shape changes and detail features of different scales in the imaging, effectively improve the characterization capability of microscopic and macroscopic features, and provide more feature information for subsequent more accurate analysis and application. It should be explained that the convolutional neural model is a tool for deep analysis and feature extraction of data, the deformable convolution layer is an improved convolution layer for extracting features in the convolutional neural model, and the microscopic topological features and the macroscopic contour features are respectively the microscopic structure characterization and the overall shape characterization of the object corresponding to the microscopic topological imaging and the macroscopic body imaging.

[0034] In detail, the deformable convolution layer in the convolutional neural model is used to perform feature mining on the microscopic topological imaging and the macroscopic body imaging to obtain microscopic topological features and macroscopic contour features, including: Identifying the topological texture in the microscopic topological imaging, and calculating the texture complexity corresponding to the topological texture; Detecting a body boundary in the macroscopic body imaging, and calculating a boundary curvature rate corresponding to the body boundary; In combination with the texture complexity and the boundary curvature, setting a convolution scale parameter of the deformable convolution layer; According to the convolution scale parameter, the deformable convolution layer is used to perform feature mining on the microscopic topological imaging and the macroscopic body imaging respectively to obtain microscopic topological features and macroscopic contour features.

[0035] It should be explained that the topological texture is a specific texture structure in microscopic topological imaging, the texture complexity indicates the complexity of the topological texture, the body boundary is the external morphological boundary of the object in macroscopic body imaging, the boundary curvature indicates the degree of bending change corresponding to the body boundary, and the convolution scale parameter is the scale control factor when the deformable convolution layer performs feature extraction, including the sampling step and sampling range.

[0036] Furthermore, the topological texture in the microscopic topological imaging can be identified by a texture filter; the body boundary in the macroscopic body imaging can be detected by an edge detection operator; the convolution scale of the deformed convolution layer is set in combination with the texture complexity and the boundary curvature. If the texture complexity is high and the boundary curvature is large, it indicates that the microscopic topological texture is rich and the boundary is tortuous and changeable. At this time, the sampling step size of the deformed convolution layer can be set to a smaller value, such as the original step size is 4 pixels, which is adjusted to 2 pixels, and the sampling range is appropriately expanded, such as expanding outward by 20% on the basis of the original range, so as to more accurately capture the complex details and boundary features in the microscopic topological imaging; according to the convolution scale parameter, the convolution kernels of different sizes in the deformed convolution layer are used to perform feature mining on the microscopic topological imaging and the macroscopic body imaging respectively to obtain microscopic topological features and macroscopic contour features.

[0037] Further, as an optional embodiment of the present invention, the calculating the texture complexity corresponding to the topological texture includes: Identify the texture grayscale value and grayscale level corresponding to the topological texture, and construct a texture grayscale co-occurrence matrix corresponding to the topological texture according to the texture grayscale value; Based on the texture grayscale co-occurrence matrix, counting the matrix occurrence frequency corresponding to the texture grayscale value; Combining the grayscale level and the matrix occurrence frequency, the texture complexity corresponding to the topological texture is calculated by the following formula: ; Among them, D represents the texture complexity corresponding to the topological texture, The coordinates are The matrix occurrence frequency corresponding to the texture grayscale value, i and j represent the coordinates corresponding to the texture grayscale value, and E represents the grayscale level.

[0038] It should be explained that the texture grayscale value is the grayscale intensity corresponding to the topological texture, the grayscale level is the grayscale grade corresponding to the topological texture, usually 256, and the texture grayscale co-occurrence matrix is ​​a statistical matrix of the grayscale distribution relationship corresponding to the topological texture. Furthermore, the texture grayscale value and grayscale level corresponding to the topological texture can be identified by an identification function, and the identification function is compiled by a programming language; the texture grayscale co-occurrence matrix corresponding to the topological texture can be constructed by a matrix function, such as a zero matrix function.

[0039] Further, as an optional embodiment of the present invention, the calculating the boundary curvature rate corresponding to the body posture boundary includes: Detecting boundary pixel points in the body boundary and determining pixel coordinate values ​​corresponding to the boundary pixel points; According to the pixel coordinate value, the local curvature corresponding to the boundary pixel point can be calculated by the following formula: ; in, Represents the local curvature rate corresponding to the boundary pixel point, and Indicates the pixel coordinate value corresponding to the dth pixel point in the boundary pixel point, and Indicates the pixel coordinate value corresponding to the d-1th pixel point in the boundary pixel point. and Indicates the pixel coordinate value corresponding to the d+1th pixel point in the boundary pixel point; The boundary curvature rate corresponding to the body boundary is calculated according to the local curvature rate.

[0040] It should be explained that the boundary pixel points are component pixel points in the body boundary, and the local curvature is a measure of the curvature change corresponding to the boundary pixel points. Furthermore, the detection of the boundary pixel points in the body boundary can be achieved by the Canny edge detection operator; the pixel coordinate value corresponding to the boundary pixel point is the position corresponding to the pixel point; the difference in the local curvature between adjacent points in the boundary pixel points is calculated, and the difference is summed, and the result of the sum is divided by the length corresponding to the body boundary to obtain the boundary curvature corresponding to the body boundary.

[0041] The present invention utilizes the bilinear fusion layer in the convolutional neural model to perform feature coupling processing on the micro topological features and the macro contour features, which can effectively fuse feature information of different scales, thereby improving the performance in tasks related to the comprehensive analysis of micro and macro features. It should be explained that the bilinear fusion layer is a neural network layer based on bilinear operations, which is used to realize the fusion processing of two variables, and the high-order coupling features are high-level features obtained after the micro topological features and the macro contour features are fused.

[0042] In detail, the bilinear fusion layer in the convolutional neural model is used to perform feature coupling processing on the microscopic topological features and the macroscopic contour features to obtain high-order coupling features, including: Using the pooling function in the bilinear fusion layer, the microscopic topological features and the macroscopic contour features are respectively subjected to feature dimensionality reduction processing to obtain reduced-dimensionality topological features and reduced-dimensionality contour features; Using the vector conversion layer in the bilinear fusion layer to respectively vectorize the reduced dimension topological feature and the reduced dimension contour feature to obtain a reduced dimension topological feature vector and a reduced dimension contour feature vector; Calculating the bilinear outer product between the reduced dimension topological feature vector and the reduced dimension profile feature vector using the bilinear function in the bilinear fusion layer; According to the bilinear outer product, construct a bilinear matrix corresponding to the reduced dimension topological feature vector and the reduced dimension profile feature vector; According to the bilinear matrix, a high-order coupling feature vector corresponding to the reduced-dimensional topological feature vector and the reduced-dimensional contour feature vector is determined, and the high-order coupling feature vector is characterized to obtain a high-order coupling feature.

[0043] It should be explained that the pooling function is a function used to perform feature dimensionality reduction processing on the micro topological features and the macro contour features, which can effectively reduce the number of features and retain the most significant features, such as the maximum pooling function. The reduced dimensionality topological features and the reduced dimensionality contour features are representative features of the micro topological features and the macro contour features, respectively. The vector conversion layer is a neural layer used to convert the reduced dimensionality topological features and the reduced dimensionality contour features into corresponding expression vectors, which is composed of linear transformation functions; the bilinear function is a function used to calculate the bilinear outer product between the reduced dimensionality topological feature vector and the reduced dimensionality contour feature vector, such as the outer product bilinear function (TensorProduct - like Bilinear Function), the bilinear outer product represents the second-order mutual information between the reduced dimensionality topological feature vector and the reduced dimensionality contour feature vector, the bilinear matrix is ​​a square matrix composed of the bilinear outer products corresponding to the reduced dimensionality topological feature vector and the reduced dimensionality contour feature vector, and further, the high-order coupled feature vectors corresponding to the reduced dimensionality topological feature vector and the reduced dimensionality contour feature vector can be determined by performing matrix decomposition on the bilinear matrix, such as the singular value decomposition method; the characterization processing of the high-order coupled feature vector can be achieved by an autoencoder.

[0044] S3. Query the process index corresponding to the bonding silver wire, calculate the index confidence gradient corresponding to the process index according to the target process data, and analyze the silver wire quality and efficiency corresponding to the bonding silver wire by combining the index confidence gradient and the high-order coupling characteristics.

[0045] The present invention calculates the indicator confidence gradient corresponding to the process indicator based on the target process data, and the indicator confidence gradient can be used to understand the reliability of the process indicator, thereby improving the accuracy of subsequent analysis of the quality and efficiency of the silver wire corresponding to the bonding silver wire. It should be explained that the process indicator is the processing indicator corresponding to the bonding silver wire, and the indicator confidence gradient represents the reliability of the process indicator.

[0046] In detail, the calculating the indicator confidence gradient corresponding to the process indicator according to the target process data includes: Extracting the index process data corresponding to the process index from the target process data; Quantifying the index process data to obtain an index process quantization value, and calculating a weight coefficient corresponding to the index process quantization value; Combining the weight coefficient and the process quantization value of the indicator, the indicator confidence gradient corresponding to the process indicator is calculated by the following formula: ; Among them, G represents the confidence gradient of the indicator corresponding to the process indicator, Indicates the weight coefficient corresponding to the e-th quantized value in the indicator process quantized value, represents the e-th indicator process quantization value, e represents the serial number corresponding to the indicator process quantization value, and n represents the number of indicator process quantization values. represents the reference quantization value, represents the attenuation factor, Indicates the time point corresponding to the nth quantized value in the process quantized value of the indicator. Indicates the time point corresponding to the e-th quantized value in the indicator process quantized value.

[0047] It should be explained that the indicator process data is the data corresponding to the process indicator extracted from the target process data, the indicator process quantization value is the expression value corresponding to the indicator process data, the weight coefficient represents the importance corresponding to the indicator process quantization value, and the attenuation factor reflects the attenuation degree of the impact of the data point deviation on the cumulative deviation value over time.

[0048] Furthermore, the quantization processing of the index process data can be achieved through a normalization method; the calculation of the weight coefficient corresponding to the index process quantization value can be achieved through an entropy weight method; and the attenuation factor can be calculated by inputting the corresponding historical process index data into an exponential attenuation model.

[0049] The present invention analyzes the silver wire quality efficiency corresponding to the bonding silver wire by combining the indicator confidence gradient and the high-order coupling characteristics, and can obtain the comprehensive performance of the bonding silver wire in terms of quality and usage efficiency, thereby laying a foundation for the subsequent analysis of abnormal key factors. It should be explained that the silver wire quality efficiency is the sum of the quality and usage efficiency corresponding to the bonding silver wire. Further, the silver wire quality efficiency corresponding to the bonding silver wire is analyzed in combination with the indicator confidence gradient and the high-order coupling characteristics. For example, the indicator confidence gradient shows that the diameter indicator of the bonding silver wire has large fluctuations and instability during the processing (the gradient changes frequently and the amplitude is large). At the same time, the high-order coupling characteristics show that the synergistic relationship between the microstructural characteristics of the silver wire and the macroscopic physical properties deviates from the ideal range, resulting in low silver wire quality efficiency, which is manifested as poor contact and increased resistance in the bonding of electronic devices, affecting the signal transmission quality and the overall performance of the product.

[0050] S4. When the quality of the silver wire does not meet the preset quality, based on the indicator confidence gradient and the high-order coupling characteristics, analyze the abnormal key factors corresponding to the bonding silver wire, and based on the abnormal key factors, perform abnormal processing on the bonding silver wire to obtain a processing result.

[0051] It should be understood that when the quality of the silver wire does not meet the preset quality, it means that the quality of the bonding silver wire does not meet the standard. The present invention analyzes the abnormal key factors corresponding to the bonding silver wire based on the indicator confidence gradient and the high-order coupling characteristics, and then obtains the abnormal influence reasons of the bonding silver wire, thereby providing a basis for the subsequent abnormal processing of the bonding silver wire. It should be explained that the preset quality is the indicator standard reference corresponding to the silver wire quality, and the abnormal key factor is the performance deviation key factor corresponding to the bonding silver wire, such as the diameter accuracy factor. Further, based on the indicator confidence gradient and the high-order coupling characteristics, the abnormal key factors corresponding to the bonding silver wire are analyzed. For example, if the indicator confidence gradient shows that the resistance index fluctuation of the bonding silver wire is abnormal and the high-order coupling characteristics indicate that the microscopic crystal structure of the silver wire is poorly correlated with the processing temperature, then the unstable processing temperature is the abnormal key factor, which affects the integrity of the crystal structure and thus causes resistance fluctuations; query the abnormal processing method corresponding to the abnormal key factor, and set the corresponding abnormal processing parameters in combination with the actual abnormal scenario, so as to perform abnormal processing of the bonding silver wire and obtain the processing result.

[0052] Compared with the problems described in the background technology, the present invention can remove noise interference and erroneous data by performing quality sublimation processing on the silver wire imaging data and the bonding process data respectively, so as to improve the data quality corresponding to the silver wire imaging data and the bonding process data, and lay a foundation for improving the accuracy of subsequent analysis of the quality of the bonding silver wire. The present invention can adaptively capture the complex shape changes and detail features of different scales in the imaging by utilizing the deformable convolution layer in the convolutional neural model to perform feature mining on the microscopic topological imaging and the macroscopic body imaging, and effectively improve the characterization capability of microscopic and macroscopic features, so as to provide a basis for subsequent more accurate analysis and The application provides more characteristic information. The present invention calculates the indicator confidence gradient corresponding to the process indicator according to the target process data. The indicator confidence gradient can be used to understand the reliability of the process indicator, thereby improving the accuracy of the subsequent analysis of the silver wire quality and efficiency corresponding to the bonding silver wire. It should be understood that when the silver wire quality and efficiency do not meet the preset quality and efficiency, it means that the quality of the bonding silver wire does not meet the standard. The present invention analyzes the abnormal key elements corresponding to the bonding silver wire based on the indicator confidence gradient and the high-order coupling characteristics, and then obtains the abnormal influence cause of the bonding silver wire, thereby providing a basis for the subsequent abnormal handling of the bonding silver wire. Therefore, the present invention proposes a data management method for bonding silver wire based on a convolutional neural model to improve the data management efficiency of bonding silver wire.

[0053] Embodiment 2: like Figure 2 As shown, it is a functional module diagram of a data management system for bonding silver wires based on a convolutional neural model provided by one embodiment of the present invention.

[0054] The data management system 100 for bonding silver wires based on a convolutional neural model described in the present invention can be installed in an electronic device. According to the functions implemented, the data management system 100 for bonding silver wires based on a convolutional neural model can include a data processing module 101, a feature coupling module 102, a silver wire quality and efficiency analysis module 103, and an exception handling module 104. The module described in the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0055] In this embodiment, the functions of each module / unit are as follows: The data processing module 101 is used to obtain silver wire imaging data and bonding process data corresponding to the bonding silver wire, perform mass sublimation processing on the silver wire imaging data and the bonding process data respectively to obtain target imaging data and target process data, and perform sorting processing on the target imaging data to obtain microscopic topological imaging and macroscopic body imaging; The feature coupling module 102 is used to input the microscopic topological imaging, the macroscopic body imaging and the target process data as input data into a pre-trained convolutional neural model, respectively, and use the deformable convolution layer in the convolutional neural model to perform feature mining on the microscopic topological imaging and the macroscopic body imaging to obtain microscopic topological features and macroscopic contour features, and use the bilinear fusion layer in the convolutional neural model to perform feature coupling processing on the microscopic topological features and the macroscopic contour features to obtain high-order coupling features; The silver wire quality and efficiency analysis module 103 is used to query the process index corresponding to the bonding silver wire, calculate the index confidence gradient corresponding to the process index according to the target process data, and analyze the silver wire quality and efficiency corresponding to the bonding silver wire by combining the index confidence gradient and the high-order coupling feature; The exception processing module 104 is used to analyze the abnormal key factors corresponding to the bonding silver wire based on the indicator confidence gradient and the high-order coupling characteristics when the quality of the silver wire does not meet the preset quality, and perform abnormal processing on the bonding silver wire based on the abnormal key factors to obtain a processing result.

[0056] In detail, each module described in the data management system 100 for implementing silver wire bonding based on a convolutional neural model in the embodiment of the present application is used in the same manner as described above. Figure 1 The technical means described in the text is the same as the data management method for bonding silver wire based on a convolutional neural model, and can produce the same technical effects, so I will not go into details here.

[0057] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A data management method for bonding silver wire based on a convolutional neural model, characterized in that: The method comprises: Acquire silver wire imaging data and bonding process data corresponding to the bonding silver wire, perform mass sublimation processing on the silver wire imaging data and the bonding process data respectively to obtain target imaging data and target process data, and perform sorting processing on the target imaging data to obtain microscopic topological imaging and macroscopic body imaging; The microscopic topological imaging, the macroscopic body imaging and the target process data are respectively input as input data into a pre-trained convolutional neural model, and the microscopic topological imaging and the macroscopic body imaging are respectively subjected to feature mining by using the deformable convolutional layer in the convolutional neural model to obtain microscopic topological features and macroscopic contour features, and the microscopic topological features and the macroscopic contour features are subjected to feature coupling processing by using the bilinear fusion layer in the convolutional neural model to obtain high-order coupling features; Querying the process index corresponding to the bonding silver wire, calculating the index confidence gradient corresponding to the process index according to the target process data, and analyzing the silver wire quality effect corresponding to the bonding silver wire by combining the index confidence gradient and the high-order coupling feature; When the quality of the silver wire does not meet the preset quality, based on the indicator confidence gradient and the high-order coupling characteristics, the abnormal key factors corresponding to the bonding silver wire are analyzed, and based on the abnormal key factors, abnormal processing of the bonding silver wire is performed to obtain a processing result.

2. The method for data management of silver bonding wire based on convolutional neural model according to claim 1, characterized in that: The step of performing quality sublimation processing on the silver wire imaging data and the bonding process data respectively to obtain target imaging data and target process data comprises: Performing data cleaning processing on the bonding process data to obtain cleaning process data; Performing visual parameter optimization on the silver wire imaging data to obtain optimized imaging data; Performing data repair processing on the cleaning process data to obtain repair process data; Performing image gain processing on the optimized imaging data to obtain gain imaging data; The repair process data and the gain imaging data are respectively subjected to data screening processing to obtain target imaging data and target process data.

3. The method for data management of silver bonding wire based on convolutional neural model as claimed in claim 2, characterized in that: The performing image gain processing on the optimized imaging data to obtain gain imaging data comprises: Performing filtering processing on the optimized imaging data to obtain filtered imaging data; Identifying an imaging contour corresponding to the filtered imaging data, and analyzing contour integrity corresponding to the imaging contour; According to the contour integrity, performing contour fitting processing on the filtered imaging data to obtain fitting imaging data; The fitting imaging data is sharpened and enhanced to obtain gain imaging data.

4. The method for data management of silver bonding wire based on convolutional neural model according to claim 1, characterized in that: The step of sorting and processing the target imaging data to obtain microscopic topological imaging and macroscopic body imaging includes: Identify the number of image pixels corresponding to each image in the target imaging data, and set a pixel matrix corresponding to each image in the target imaging data based on the number of image pixels; Performing grayscale processing on each image in the target imaging data to obtain a target grayscale image; Based on the pixel matrix, calculating the horizontal grayscale variation rate and the vertical grayscale variation rate corresponding to the target grayscale image; Combining the horizontal grayscale variation rate and the vertical grayscale variation rate, the grayscale gradient amplitude and gradient direction angle corresponding to the target grayscale image can be calculated by the following formula: ； Among them, A represents the grayscale gradient amplitude corresponding to the target grayscale image, Represents the gradient direction angle corresponding to the target grayscale image, represents the horizontal grayscale variation rate, represents the vertical grayscale change rate, represents the inverse tangent function; Based on the grayscale gradient amplitude and the gradient direction angle, the target imaging data is sorted and processed to obtain microscopic topological imaging and macroscopic body imaging.

5. The method for data management of silver bonding wire based on convolutional neural model according to claim 1, characterized in that: The method of using the deformable convolution layer in the convolutional neural model to perform feature mining on the microscopic topological imaging and the macroscopic body imaging to obtain microscopic topological features and macroscopic contour features includes: Identifying the topological texture in the microscopic topological imaging, and calculating the texture complexity corresponding to the topological texture; Detecting a body boundary in the macroscopic body imaging, and calculating a boundary curvature rate corresponding to the body boundary; In combination with the texture complexity and the boundary curvature, setting a convolution scale parameter of the deformable convolution layer; According to the convolution scale parameter, the deformable convolution layer is used to perform feature mining on the microscopic topological imaging and the macroscopic body imaging respectively to obtain microscopic topological features and macroscopic contour features.

6. The method for data management of silver bonding wire based on convolutional neural model as claimed in claim 5, characterized in that: The calculating the texture complexity corresponding to the topological texture includes: Identify the texture grayscale value and grayscale level corresponding to the topological texture, and construct a texture grayscale co-occurrence matrix corresponding to the topological texture according to the texture grayscale value; Based on the texture grayscale co-occurrence matrix, counting the matrix occurrence frequency corresponding to the texture grayscale value; Combining the grayscale level and the matrix occurrence frequency, the texture complexity corresponding to the topological texture is calculated by the following formula: ； Among them, D represents the texture complexity corresponding to the topological texture, The coordinates are The matrix occurrence frequency corresponding to the texture grayscale value, i and j represent the coordinates corresponding to the texture grayscale value, and E represents the grayscale level.

7. The method for data management of silver bonding wire based on convolutional neural model as claimed in claim 5, characterized in that: The calculating the boundary curvature rate corresponding to the body posture boundary includes: Detecting boundary pixel points in the body boundary and determining pixel coordinate values ​​corresponding to the boundary pixel points; According to the pixel coordinate value, the local curvature corresponding to the boundary pixel point can be calculated by the following formula: ； in, Represents the local curvature rate corresponding to the boundary pixel point, and Indicates the pixel coordinate value corresponding to the dth pixel point in the boundary pixel point, and Indicates the pixel coordinate value corresponding to the d-1th pixel point in the boundary pixel point. and Indicates the pixel coordinate value corresponding to the d+1th pixel point in the boundary pixel point; The boundary curvature rate corresponding to the body boundary is calculated according to the local curvature rate.

8. The method for data management of silver bonding wire based on convolutional neural model according to claim 1, characterized in that: The microscopic topological features and the macroscopic contour features are subjected to feature coupling processing by using the bilinear fusion layer in the convolutional neural model to obtain high-order coupling features, including: Using the pooling function in the bilinear fusion layer, the microscopic topological features and the macroscopic contour features are respectively subjected to feature dimensionality reduction processing to obtain reduced-dimensionality topological features and reduced-dimensionality contour features; Using the vector conversion layer in the bilinear fusion layer to respectively vectorize the reduced dimension topological feature and the reduced dimension contour feature to obtain a reduced dimension topological feature vector and a reduced dimension contour feature vector; Calculating the bilinear outer product between the reduced dimension topological feature vector and the reduced dimension profile feature vector using the bilinear function in the bilinear fusion layer; According to the bilinear outer product, construct a bilinear matrix corresponding to the reduced dimension topological feature vector and the reduced dimension profile feature vector; According to the bilinear matrix, a high-order coupling feature vector corresponding to the reduced-dimensional topological feature vector and the reduced-dimensional contour feature vector is determined, and the high-order coupling feature vector is characterized to obtain a high-order coupling feature.

9. The method for data management of silver bonding wire based on convolutional neural model according to claim 1, characterized in that: The step of calculating the indicator confidence gradient corresponding to the process indicator according to the target process data includes: Extracting the index process data corresponding to the process index from the target process data; Quantifying the index process data to obtain an index process quantization value, and calculating a weight coefficient corresponding to the index process quantization value; Combining the weight coefficient and the process quantization value of the indicator, the indicator confidence gradient corresponding to the process indicator is calculated by the following formula: ； Among them, G represents the confidence gradient of the indicator corresponding to the process indicator, Indicates the weight coefficient corresponding to the e-th quantized value in the indicator process quantized value, represents the e-th indicator process quantization value, e represents the serial number corresponding to the indicator process quantization value, and n represents the number of indicator process quantization values. represents the reference quantization value, represents the attenuation factor, Indicates the time point corresponding to the nth quantized value in the process quantized value of the indicator. Indicates the time point corresponding to the e-th quantized value in the indicator process quantized value.

10. A data management system for bonding silver wire based on a convolutional neural model, characterized in that: The system comprises: A data processing module, used for acquiring silver wire imaging data and bonding process data corresponding to the bonding silver wire, performing mass sublimation processing on the silver wire imaging data and the bonding process data respectively to obtain target imaging data and target process data, and performing sorting processing on the target imaging data to obtain microscopic topological imaging and macroscopic body imaging; A feature coupling module, used to input the microscopic topological imaging, the macroscopic body imaging and the target process data as input data into a pre-trained convolutional neural model, respectively, and perform feature mining on the microscopic topological imaging and the macroscopic body imaging using the deformable convolution layer in the convolutional neural model to obtain microscopic topological features and macroscopic contour features, and perform feature coupling processing on the microscopic topological features and the macroscopic contour features using the bilinear fusion layer in the convolutional neural model to obtain high-order coupling features; A silver wire quality and efficiency analysis module, used to query the process index corresponding to the bonding silver wire, calculate the index confidence gradient corresponding to the process index according to the target process data, and analyze the silver wire quality and efficiency corresponding to the bonding silver wire by combining the index confidence gradient and the high-order coupling feature; The exception processing module is used to analyze the abnormal key factors corresponding to the bonding silver wire based on the indicator confidence gradient and the high-order coupling characteristics when the quality of the silver wire does not meet the preset quality, and based on the abnormal key factors, perform abnormal processing on the bonding silver wire to obtain a processing result.

Images