A resin plug hole feature extraction method and resin plug hole quality evaluation method based on GPU image processing

Abstract

The present application belongs to the field of optical measurement, and particularly relates to a resin plug hole feature extraction method and a resin plug hole quality evaluation method based on GPU image processing. First, a 2D image of a resin plug hole plate is obtained through an optical imaging system, and image preprocessing and sub-pixel level edge positioning are performed on a GPU parallel computing platform. Then, polynomial interpolation fitting technology is used to achieve sub-pixel accurate positioning of the hole boundary, and the hole area is segmented based on the principle of maximum inter-class variance. In combination with a least square circle fitting algorithm, automatic identification and classification of defects such as hole bubbles, hole cavities and hole depressions are achieved. Finally, a resin plug hole quality evaluation system is established by statistically analyzing the defect distribution proportion and area consistency. The present application takes advantage of GPU parallel computing, solves the problems of low efficiency and poor accuracy of traditional manual detection, and realizes micron-level precision and fully automated resin plug hole quality detection, thereby significantly improving the detection efficiency and reliability.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of optical measurement, and particularly relates to a resin plug hole feature extraction method based on GPU image processing and a resin plug hole quality evaluation method. BACKGROUND

[0002] The resin plug hole technology is widely used in high-end electronic manufacturing fields such as high-density interconnection (HDI) boards, thick copper boards and BGA packaging, and the quality thereof directly affects the reliability and performance of products. The traditional detection method mainly relies on artificial visual inspection or a 2D optical detection system based on CPU. The artificial detection observes the plug hole filling state through a microscope or a magnifying glass to identify defects such as bubbles and missed plugs; and the automatic detection system usually roughly analyzes the plug hole contour by using a conventional image processing algorithm (such as threshold segmentation and edge detection). In the prior art, some schemes attempt to combine machine learning for defect classification, but are limited by the calculation efficiency and are difficult to realize real-time and high-precision full-surface detection. In addition, the traditional method is based on pixel-level edge positioning and cannot meet the process requirements of sub-micron precision, especially in the identification of complex defects (such as micron-level bubbles and non-uniform depressions).

[0003] The traditional detection method has significant limitations: first, the artificial detection is low in efficiency and strong in subjectivity, the result is affected by the experience of the operator, and the repeatability is poor, which is difficult to meet the demand of mass production; second, the image processing algorithm based on CPU is slow and cannot process high-resolution images in real time, resulting in insufficient detection coverage; third, the existing technology is insufficient in the identification precision of sub-pixel level defects (such as fuzzy hole boundary and micron-level cavities), and lacks a quantitative evaluation system, which is difficult to distinguish defects with similar appearances (such as hole depression and hole cavity). In addition, the traditional method does not fully utilize the advantages of GPU parallel computing and cannot realize high-speed and high-precision full-automatic detection, which restricts the quality control level in the high-end electronic manufacturing field. SUMMARY

[0004] The purpose of the present application is to provide a resin plug hole feature extraction method and a resin plug hole quality evaluation method based on GPU image processing, so as to solve the technical problems that the traditional resin plug hole detection method relies on artificial visual inspection or CPU image processing, is low in efficiency, strong in subjectivity, insufficient in precision and cannot realize sub-pixel level defect identification.

[0005] The present application achieves the above-mentioned purpose by the following technical solutions:

[0006] In a first aspect, the present application provides a resin plug hole feature extraction method based on GPU image processing, which comprises the following steps:

[0007] S1, build the running environment of GPU and initialize, get the GPU computing platform, use the GPU computing platform to receive the full-width 2D image data of the to-be-tested board and the corresponding gray scale diagram;

[0008] S2, after denoising and edge enhancement on the gray scale diagram, extract the pixel-level edge points;

[0009] S3, based on the contour extraction algorithm, analyze the pixel-level edge points to obtain the integer pixel contour set of each hole;

[0010] S4, for each edge point in the integer pixel contour set, use a quadratic polynomial to fit the gray value in a field window based on a set size, obtain the extreme point and determine the sub-pixel level hole boundary position to form a sub-pixel contour set of the resin hole.

[0011] Further, in step S3, the circular contour analysis is specifically:

[0012] Using the scale invariant l of a circle 2 :4πS=1to judge the pixel-level edge points, wherein l is the circumference of the contour, S is the area of the contour, to obtain the integer pixel contour set of each resin hole.

[0013] Further, in step S4, for each edge point in the integer pixel contour set, using a quadratic polynomial to fit the gray value in a field window based on a set size, obtaining the extreme point, including:

[0014] Obtain the edge point (x0, y0), select a neighborhood window of a set size pixel, use a quadratic polynomial to fit the gray value, the expression is:

[0015] I(x,y)=a(x-x0) 2 +b(y-y0) 2 +c(x-x0)(y-y0)+d(x-x0)+e(y-y0)

[0016] Wherein, I(x, y) is the gray value of the coordinates (x, y) in the image, a and b are the second-order change coefficients of the gray value in x and y directions respectively, c is the cross change coefficient, d and e are the first-order change coefficients in x and y directions respectively;

[0017] According to the partial derivative equation set of the polynomial, the extreme point is found, the expression is:

[0018]

[0019] Wherein, is the partial derivative of x, is the partial derivative of y;

[0020] The above formula is converted into a matrix formula to solve Δx and Δy, and the expression is:

[0021]

[0022] x subpix = x0+ Δx

[0023] y subpix = y0+ Δy

[0024] where Δx is a solution of the partial derivative of x equal to 0, Δy is a solution of the partial derivative of y equal to 0, x subpix is a sub-pixel X coordinate, and y subpix is a sub-pixel Y coordinate.

[0025] Further, before step S1, the method further comprises: acquiring full-width 2D image data of the to-be-tested board by an imaging system comprising a line-scan camera, an optical lens, and a light source box.

[0026] In a second aspect, the present application provides a resin hole quality evaluation method, which comprises the following steps:

[0027] S10, obtaining a sub-pixel contour set of each resin hole of the to-be-tested board by using the resin hole feature extraction method as described above;

[0028] S11, determining a segmentation threshold of the sub-pixel contour area range of each hole based on the principle of maximum inter-class variance;

[0029] S12, acquiring a binary image in the sub-pixel boundary area, and recording a segmentation area greater than the segmentation threshold as a bright area feature, and recording a segmentation area less than the segmentation threshold as a dark area feature;

[0030] S13, extracting an integral pixel contour of the bright area and the dark area, performing the step S4 to obtain a sub-pixel contour of the bright area and the dark area, and counting a gray value in the sub-pixel contour area of the bright area and the dark area;

[0031] S14, performing least square fitting on the sub-pixel contours of the bright area and the dark area to obtain a fitting circle equation;

[0032] S15, calculating a residual sum of squares as a fitting goodness of the circle according to the fitting circle equation and actual sub-pixel contour boundary information of the bright area and the dark area;

[0033] S16, determining a hole bubble defect of the to-be-tested board according to the gray value and the fitting goodness of the bright area, and determining a hole cavity defect and / or a hole recess defect of the to-be-tested board according to the gray value and the fitting goodness of the dark area;

[0034] S17, calculate the defect number of the to-be-tested board, obtain the global defect proportion of the to-be-tested board, and divide the to-be-tested board into regions to obtain the distribution information of different defects of resin vias in each region and the regional defect proportion distribution information;

[0035] S18, according to whether the distribution information of different defects of the to-be-tested board and the regional defect proportion distribution information satisfy a consistent distribution rule, evaluate the process capability of the resin via plate making process of the to-be-tested board.

[0036] Further, in step S16, the hole bubble defect of the to-be-tested board is determined according to the gray value of the bright region and the goodness of fit, comprising:

[0037] The gray value of the bright region and the goodness of fit are judged, and the gray value greater than a set value and the goodness of fit less than a set value are recorded as a hole bubble defect.

[0038] Further, in step S16, the hole cavity defect and / or hole recess defect of the to-be-tested board is determined according to the gray value of the dark region and the goodness of fit, comprising:

[0039] The gray value of the dark region and the goodness of fit are judged, and the gray value less than a corresponding set value and the goodness of fit less than a corresponding set value are recorded as a hole cavity defect, and the gray value less than a corresponding set value and the goodness of fit greater than a corresponding set value are recorded as a hole recess defect.

[0040] Further, steps S10-S18 are executed through a GPU computing platform, specifically comprising:

[0041] Load the image data to the GPU global memory of the GPU computing platform;

[0042] Call the GPU kernel function of the GPU computing platform to process the image segmentation, contour fitting and defect analysis in parallel;

[0043] Return the analysis result to the host memory to generate a report.

[0044] Further, the GPU computing platform comprises one or more GPU processors, and the memory in communication connection with the GPU processor, the memory is used to store the program instructions for executing the resin via quality evaluation method of any one of the above, and the GPU processor is used to execute the program instructions to realize the resin via quality evaluation method.

[0045] The beneficial effects of the present application are:

[0046] 1.The present application realizes efficient extraction and accurate analysis of resin plug hole features through GPU parallel computing platform, significantly improving detection efficiency and accuracy. First, sub-pixel edge positioning technology is used to achieve micron-level accurate positioning of the hole boundary through polynomial interpolation fitting, solving the problem of insufficient accuracy of traditional pixel-level detection. Second, the image segmentation method based on the maximum inter-class variance principle combined with the least squares circle fitting algorithm can accurately identify and distinguish complex defects such as hole bubbles, hole cavities and hole depressions, greatly improving the accuracy of defect identification.

[0047] 2.The intelligent quality evaluation system established in the present application realizes quantitative evaluation and process feedback of resin plug hole quality. By statistically analyzing the global distribution proportion and regional consistency characteristics of defects, not only can the quality grade of single board be determined in real time, but also data support can be provided for process optimization. Compared with traditional manual detection, this method eliminates the influence of subjective factors, and the detection results have high repeatability. This method can be widely used in quality control of high-end electronic products such as HDI board and BGA packaging, effectively preventing problems such as board explosion and false welding caused by resin plug hole defects, significantly improving the reliability and yield of products, and has important industrial application value. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 A flowchart of a resin plug hole feature extraction method based on GPU image processing proposed for Embodiment 1 of the present application;

[0049] Figure 2 A flowchart of a resin plug hole quality evaluation method proposed for Embodiment 2 of the present application;

[0050] Figure 3 Another flowchart of a resin plug hole quality evaluation method proposed for Embodiment 2 of the present application. DETAILED DESCRIPTION

[0051] It is necessary to point out that the following detailed description is only used to further illustrate the present application, and cannot be understood as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application according to the above application content.

[0052] Embodiment 1

[0053] In combination Figure 1 , the present embodiment proposes a resin plug hole feature extraction method based on GPU image processing. This method is used in a 2D optical inspection system, which uses a line scan camera, an optical lens and an imaging system of a light source box to realize automatic scanning and detection movement of the resin plug hole board to be tested through a high-precision mechanical and electrical control system, and to obtain full-width 2D image data of the board to be tested (resin plug hole board). The method includes the following steps:

[0054] S1, build the running environment of GPU and initialize to obtain a GPU computing platform, use the GPU computing platform to receive full-surface 2D image data of a to-be-tested board and a corresponding grayscale image.

[0055] S2, reduce noise by Gaussian blur on the grayscale image, enhance edge features by using an edge enhancement filter, and finally obtain pixel-level edge points by a Canny edge detector.

[0056] S3, perform circle contour analysis on the pixel-level edge points based on a contour extraction algorithm to obtain a whole-pixel contour set of each resin plug hole.

[0057] Further preferably, in step S3, the circle contour analysis is specifically:

[0058] adopting a circle scale invariant l 2 :4πS=1to judge the pixel-level edge points, wherein l is the circumference of the contour, and S is the area of the contour, to obtain a whole-pixel contour set of each resin plug hole.

[0059] S4, for each edge point (x0, y0) in the whole-pixel contour set, use a quadratic polynomial to fit the grayscale value in a field window based on a set size, dynamically adjust the window size for different apertures (5x5 pixels for a standard hole and 3x3 pixels for a micro hole), obtain an extreme value point and determine the sub-pixel level hole boundary position to form a sub-pixel contour set of the resin plug hole.

[0060] The hole boundary sub-pixel positioning method described above is used in this embodiment to analyze the whole-pixel boundary of the plug hole, perform polynomial interpolation near the hole boundary, and use the extreme value point of the interpolation function to determine the sub-pixel level hole boundary position.

[0061] Further preferably, in step S4, for each edge point in the whole-pixel contour set, a quadratic polynomial is used to fit the grayscale value in a field window based on a set size to obtain an extreme value point, which includes:

[0062] An edge point (x0, y0) is obtained, a neighborhood window of a set size of pixels is selected, a quadratic polynomial is used to fit the grayscale value, and the expression is:

[0063] I(x,y)=a(x-x0) 2 +b(y-y0) 2 +c(x-x0)(y-y0)+d(x-x0)+e(y-y0)

[0064] Wherein, I(x, y) is the gray value at coordinate (x, y) in the image, a and b are the second order change coefficients of the gray value in x and y directions respectively, c is the cross change coefficient, d and e are the first order change coefficients in x and y directions respectively;

[0065] The extreme point is found according to the partial derivative equation of the polynomial, and the expression is:

[0066]

[0067] Wherein, The partial derivative of x is, The partial derivative of y is;

[0068] The above formula is converted into a matrix formula to solve Δx and Δy, and the expression is:

[0069]

[0070] x subpix =x0+Δx

[0071] y subpix =y0+Δy

[0072] Wherein, Δx is the solution of the partial derivative of x equaling to 0, Δy is the solution of the partial derivative of y equaling to 0, x subpix is the sub-pixel X coordinate, and y subpix is the sub-pixel Y coordinate.

[0073] According to the above embodiment 1, the above extraction method first acquires the full-width 2D image of the to-be-measured plate by using a line scanning camera, performs image preprocessing through a GPU platform, and extracts pixel-level edge points by using an edge detection algorithm; then the hole boundary is accurately positioned at a sub-pixel level by using a quadratic polynomial fitting technology, and effective contours are screened in combination with scale invariant analysis of a circle to form a set of hole contour profiles with sub-pixel accuracy. Compared with the traditional manual detection and CPU image processing method, the application utilizes the parallel computing advantage of the GPU to significantly improve the detection efficiency and positioning accuracy, effectively solves the problems of low efficiency, strong subjectivity and difficulty in realizing sub-pixel level detection in the prior art, and provides a reliable data basis for subsequent defect identification and quality evaluation.

[0074] Embodiment 2

[0075] In combination Figure 2 and Figure 3 , on the basis of the above embodiment 1, the present embodiment proposes a resin hole quality evaluation method, which is a hole bubble defect judgment, hole depression and hole cavity judgment and hole quality evaluation after the above hole boundary sub-pixel positioning, and the method comprises the following steps:

[0076] S10, using the resin plug hole feature extraction method as in embodiment 1 to obtain the sub-pixel contour set of each resin plug hole of the to-be-tested board.

[0077] S11, in the sub-pixel contour area of each plug hole, according to the principle of maximum inter-class variance, the segmentation threshold of the hole region is calculated, and the principle is: the cumulative mean M of the gray level K and the global mean MG of the image are calculated: Solve when The value of K is the segmentation threshold to be solved.

[0078] S12, obtain the binary image in the sub-pixel boundary area, and record the segmentation area greater than the segmentation threshold as a bright area feature, and record the segmentation area less than the segmentation threshold as a dark area feature.

[0079] S13, extract the integral pixel contour of the bright area and the dark area, execute step S4 to obtain the sub-pixel contour of the bright area and the dark area, and count the gray value in the sub-pixel contour area of the bright area and the dark area.

[0080] S14, the least square fitting is performed on the sub-pixel contour of the bright area and the dark area to obtain a fitting circle equation, and more specifically, the least square fitting is performed on the sub-pixel contour of the bright area to obtain the center and the radius of the fitting circle, and the least square fitting circle curve R 2 =(x-A) 2 +(y-B) 2 is expanded to obtain R 2 =x 2 +y 2 -2Ax-2By+A 2 +B 2 , only need to solve V=∑(x 2 +y 2 -2Ax-2By+A 2 +B 2 -R 2 ) 2 , the corresponding values of the coefficients A, B and R are obtained when V is the smallest, and the coefficient result matrix is easily obtained by taking the partial derivative of the above formula. The coefficient matrix equation is solved, and the coefficients A, B, C and D are obtained, which correspond to the fitting circle equation.

[0081] S15, according to the fitted circle equation and the actual bright area and dark area sub-pixel contour boundary information, the residual sum of squares is calculated as the goodness of fit of the circle.

[0082] S16, determining the hole bubble defect of the to-be-tested board according to the gray value and the goodness of fit of the bright area, and determining the hole cavity defect and / or hole recess defect of the to-be-tested board according to the gray value and the goodness of fit of the dark area, more specifically: judging the gray value and the goodness of fit of the bright area, and recording the hole bubble defect when the gray value is greater than a set value and the goodness of fit is less than a set value. Judging the gray value and the goodness of fit of the dark area, and recording the hole cavity defect when the gray value is less than the corresponding set value and the goodness of fit is less than the corresponding set value, and recording the hole recess defect when the gray value is less than the corresponding set value and the goodness of fit is greater than the corresponding set value.

[0083] S17, calculating the number of defects of the to-be-tested board, obtaining the global defect proportion of the to-be-tested board, and dividing the to-be-tested board into regions to obtain the distribution information of different defects of the resin plug hole in each region and the regional defect proportion distribution information.

[0084] S18, evaluating the process capability of the resin plug hole plate making process of the to-be-tested board according to whether the distribution information of different defects of the to-be-tested board and the regional defect proportion distribution information satisfy the consistent distribution rule.

[0085] In specific implementation, first, the global defect rate is calculated based on the sub-pixel contour data, which reflects the overall plate quality level; second, the defect distribution heat map is generated by spatial grid division, which intuitively displays the defect aggregation area and is used to identify local process abnormalities; and finally, the process capability index is output by combining the statistical process control method, realizing the quantitative evaluation of process stability.

[0086] In step S18, the global percentage data of each kind of defect number divided by the total number of resin plug holes * 100% is calculated, indicating the overall defect proportion of the current plug hole plate. The to-be-tested board is divided into regions, and different defect data are one-to-one corresponding according to the divided regions, so that the distribution of different defects of the resin plug hole in each region can be obtained, and the number of each kind of defect in the region is divided by the total number of resin plug holes in the region * 100%. The regional defect proportion distribution is obtained. According to the defect proportion distribution of different kinds of defects in each region, it is judged whether the consistent distribution rule is satisfied, which can feedback the process capability of the current resin plug hole plate making process.

[0087] In specific implementation, steps S10-S18 are executed by a GPU computing platform, specifically including: loading image data to the GPU global memory of the GPU computing platform; calling the GPU kernel function of the GPU computing platform to process image segmentation, contour fitting and defect analysis in parallel; returning the analysis result to the host memory to generate a report. The GPU computing platform includes one or more GPU processors and a memory in communication connection with the GPU processors, the memory being used to store program instructions for executing the resin plug hole quality evaluation method as described above, and the GPU processor being used to execute the program instructions to implement the resin plug hole quality evaluation method.

[0088] According to the above embodiment 2, the full flow acceleration of the resin plug hole detection in the above evaluation method is realized through the GPU computing platform, which embodies the complete technical implementability. The GPU execution process of steps S10-S18 in embodiment 2 corresponds to the GPU parallel processing model: first, the collected 2D image data is loaded into the GPU global memory, the image segmentation, contour fitting and defect analysis are performed in parallel through the writing of specific kernel functions (Kernel), and finally the results are returned to the host memory to generate a report. The technical points are reflected in three aspects: 1) task allocation mechanism based on CUDA architecture, parallel processing of image data is realized through thread blocks / grid blocks (such as each thread processing a 128x128 pixel sub-image); 2) special kernel function design, covering the full algorithm chain from preprocessing to defect analysis; 3) CPU-GPU collaborative architecture, in which the GPU processor is responsible for computationally intensive tasks, and the memory stores program instructions to form a complete hardware solution. This design reduces the time consumption of sub-pixel feature extraction and quality evaluation by more than 90% compared with the traditional CPU scheme, verifying the industrial applicability of the technical scheme.

[0089] It can be understood that the accurate extraction of plug hole geometric features is realized through embodiment 1, forming a sub-pixel contour set containing parameters such as center coordinates and radius size; these high-precision feature data are directly used as input sources for quality evaluation in embodiment 2. In embodiment 2, based on the feature set, the in-hole region segmentation and defect classification are further implemented: through the features of the light / dark regions divided by the maximum inter-class variance method, combined with the least squares circle fitting algorithm, the intelligent discrimination of defects such as bubbles, cavities and depressions is realized. The two embodiments form a progressive technical chain - embodiment 1 solves the problem of "how to accurately measure", and embodiment 2 solves the problem of "how to scientifically evaluate", which together build a complete solution from feature extraction to quality evaluation, significantly improving the automation degree and result reliability of resin plug hole detection.

[0090] In combination with Figure 3 , the quality evaluation process realizes efficient and automatic detection based on the GPU parallel computing architecture, and the core technical process can be divided into four stages: first, the geometric parameters (center coordinates, radius, etc.) of the plug hole are extracted through the sub-pixel contour set, and the size out-of-tolerance (hole large / small) is automatically determined in combination with the preset process standard; second, the in-hole region is dynamically segmented based on the maximum inter-class variance principle, the light / dark region features are extracted, the least squares method is used to fit the actual contour, and the residual sum of squares is calculated as the shape evaluation index by comparing the fitted circle equation; the third stage realizes intelligent classification of defects by combining the gray value and the shape index: for example, if the light area gray value is greater than 200 and the residual (residual sum of squares) is less than 0.05 mm 2 , it is determined as a bubble defect, and if the dark area gray value is less than 100 and the residual is less than 0.08 mm 2Empty, residual > 0.15 mm 2 Recessed; finally, spatial statistics are performed to calculate the global defect rate (total defect number / total hole number x 100%) and generate a 5x5 grid defect distribution heat map. For example, when the defect rate of any grid exceeds 8% or the overall defect rate > 5%, a process alarm is triggered. The entire process is accelerated by a CUDA kernel function, and the defect classification algorithm is optimized by parallel reduction, which controls the single board evaluation time within 30 seconds, 15 times more efficient than the traditional method.

[0091] In the above embodiments, all or part of them can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of them can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated.

[0092] The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be magnetic media (for example, floppy disk, hard disk, magnetic tape), optical media (for example, DVD), or semiconductor media (for example, solid state disk (SSD)) and the like.

[0093] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art will understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent replacements for part of the technical features; and these modifications or replacements do not drive the essence of the corresponding technical solutions out of the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for evaluating a resin plug hole quality, characterized by, The method comprises the following steps: S10, obtaining a sub-pixel contour set of each resin plug hole of the to-be-tested board by using a resin plug hole feature extraction method; The resin plug hole feature extraction method comprises the following steps: S1, building a GPU running environment and initializing to obtain a GPU computing platform, using the GPU computing platform to receive full-surface 2D image data of the to-be-tested board and a corresponding grayscale image; S2, extracting pixel-level edge points after denoising and edge enhancement of the grayscale image; S3, performing circular contour analysis on the pixel-level edge points based on a contour extraction algorithm to obtain an integer pixel contour set of each plug hole; S4, fitting the gray value in a neighborhood window based on a set size using a quadratic polynomial for each edge point in the integer pixel contour set to obtain extreme points and determine the sub-pixel level hole boundary position to form a sub-pixel contour set of the resin plug hole; S11, determining a segmentation threshold of the sub-pixel contour area range of each plug hole based on the maximum inter-class variance principle; S12, obtaining a binary image in the sub-pixel boundary area, and recording a segmentation area greater than the segmentation threshold as a bright area feature and a segmentation area less than the segmentation threshold as a dark area feature; S13, extracting the integer pixel contour of the bright area and the dark area, performing step S4 to obtain the sub-pixel contour of the bright area and the dark area, and counting the gray value in the sub-pixel contour area of the bright area and the dark area; S14, performing least squares fitting on the sub-pixel contour of the bright area and the dark area to obtain a fitting circle equation; S15, calculating the residual sum of squares as the fitting goodness of the circle according to the fitted circle equation and the actual sub-pixel contour boundary information of the bright area and the dark area; S16, determining the hole bubble defect of the to-be-tested board according to the gray value and the fitting goodness of the bright area, and determining the hole cavity defect and / or hole recess defect of the to-be-tested board according to the gray value and the fitting goodness of the dark area; S17, calculating the number of defects of the to-be-tested board to obtain the global defect proportion of the to-be-tested board, and dividing the to-be-tested board into regions to obtain the distribution information of different defects of the resin plug holes in each region and the regional defect proportion distribution information; S18, evaluating the resin plug hole plate manufacturing process capability of the to-be-tested board according to whether the distribution information of different defects of the to-be-tested board and the regional defect proportion distribution information satisfy the consistent distribution rule.

2. The resin plug quality evaluation method according to claim 1, characterized by, In step S16, the determination of the hole bubble defect of the to-be-tested board according to the gray value and the fitting goodness of the bright area comprises: judging the gray value and the fitting goodness of the bright area, and recording the gray value greater than a set value and the fitting goodness less than a set value as the hole bubble defect.

3. The resin plug quality evaluation method according to claim 1, characterized by, In step S16, the determination of the hole cavity defect and / or hole recess defect of the to-be-tested board according to the gray value and the fitting goodness of the dark area comprises: judging the gray value and the fitting goodness of the dark area, and recording the gray value less than a corresponding set value and the fitting goodness less than a corresponding set value as the hole cavity defect, and recording the gray value less than a corresponding set value and the fitting goodness greater than a corresponding set value as the hole recess defect.

4. The resin plug quality evaluation method according to claim 1, characterized by, Steps S10-S18 are executed by the GPU computing platform, and specifically comprise: loading image data to a GPU global memory of a GPU computing platform; calling a GPU kernel function of the GPU computing platform to process image segmentation, contour fitting and defect analysis in parallel; returning analysis results to a host memory to generate a report.

5. The resin plug quality evaluation method according to claim 1, characterized by, In step S3, the circle contour analysis specifically includes: Using circular scale invariants judging the pixel-level edge points, wherein is a perimeter of the contour, is an area of the contour, obtaining a set of integer pixel contours for each resin plug hole.

6. The resin plug quality evaluation method according to claim 1, characterized by, In step S4, for each edge point in the set of integer-pixel contours, a quadratic polynomial is used to fit the gray value in a neighborhood window based on a set size, and an extreme point is obtained, including: Acquiring edge points A neighborhood window of a certain size in pixels is chosen and a quadratic polynomial is fitted to the gray values, expressed as ； wherein is the gray value at coordinates in the image, and are second order variation coefficients of the gray value in and directions, is the cross variation coefficient, and are first order variation coefficients in and directions; The extreme point is found according to a partial derivative equation set of the polynomial, and the expression is: ； wherein is the derivative of with respect to is the derivative of with respect to The above equation is converted to a matrix equation to solve for and The expression is: ； wherein is the solution to the partial derivative equals 0, is the solution to the partial derivative equals 0, is the subpixel coordinate, is the subpixel coordinate.

7. The resin plug quality evaluation method according to claim 1, characterized by, Before step S1, the method further includes: acquiring full-width 2D image data of the to-be-tested board through an imaging system including a line-scan camera, an optical lens and a light source box.

8. The resin plug quality evaluation method according to claim 4, characterized by, The GPU computing platform includes one or more GPU processors, and the memory connected in communication with the GPU processors, the memory being used to store program instructions for executing the resin plug hole quality evaluation method in any one of claims 1 to 3, and the GPU processors being used to execute the program instructions to implement the resin plug hole quality evaluation method.

Images