3D glass online monitoring platform based on machine vision

Abstract

The present invention discloses a 3D glass online monitoring platform based on machine vision, which belongs to the technical field of glass production detection platform, and includes an image acquisition module, a motion control unit and an auxiliary unit, wherein the image acquisition module is connected to a data analysis module, the data analysis module is connected to a defect classification module, and the image acquisition module includes a 3D camera and an image acquisition card for collecting images captured by the 3D camera. The present invention can further optimize the judgment of scratches, stains, bubbles and pits that originally rely on two-dimensional data by combining the depth information obtained by the 3D camera to further optimize the judgment of scratches, stains, bubbles and pits, realize three-dimensional recognition and classification of defects on glass, improve the efficiency of calculation and recognition, and solve the problem that the detection of machine vision in current glass production relies on two-dimensional information judgment and the lack of high accuracy.

Description

Technical Field

[0001] The present invention relates to an online monitoring platform, in particular to a 3D glass online monitoring platform based on machine vision, and belongs to the technical field of glass production detection platforms. Background Art

[0002] In the process of glass production, it is necessary to detect defects such as defects, stones and rope marks in the glass. Today's machine vision can also detect various defects and color characteristics on the glass surface. Machine vision in the glass production process can also detect coated thin glass, ceramic glass and TFT-LCD glass. Many scholars are also studying the optimization and improvement of machine vision glass detection algorithms to improve the detection effect and improve the accuracy and efficiency of detection. Machine vision has been valued and promoted by manufacturers in glass inspection due to its unique detection advantages. However, there are also many technical difficulties in machine vision glass detection, such as how to improve the effect and clarity of defect collection, how to improve the measurement accuracy and efficiency of machine vision to achieve real-time proportional detection, and how to use the characteristics of defects to learn and accurately classify defects. The steps for machine vision to solve glass detection are generally to build a machine vision subject, obtain a good visual image, and then use the collected image for segmentation, feature extraction, regional morphology and edge extraction, as well as defect classification and judgment. However, the detection means of machine vision in current glass production can only rely on two-dimensional data such as defect area, perimeter, circularity, aspect ratio, elongation, Euler number, inclination and filling degree to classify defects on the glass surface into scratches, stains, bubbles and pits. The calculation amount is large and the process is complicated. The calculation parameters are large, the output results are many, and the multi-process calculation and result output affect the efficiency of judgment. In addition, the accuracy of two-dimensional information judgment is not high. With the development of machine vision, 3D cameras can be used to scan and calculate the three-dimensional contours in the image, and the information of the three-dimensional contours in the image can be calculated and output. Therefore, the present invention proposes a 3D glass online monitoring platform based on machine vision, which can further optimize the judgment of scratches, stains, bubbles and pits that originally relied on two-dimensional data combined with the depth information obtained by the 3D camera, realize the three-dimensional recognition and classification of defects on the glass, and optimize the calculation process at the same time, improve the efficiency of calculation and recognition, and improve the accuracy of defect recognition and splitting. Summary of the invention

[0003] The main purpose of the present invention is to solve the problem that the detection means of machine vision in current glass production can only rely on two-dimensional data such as defect area, perimeter, circularity, aspect ratio, elongation, Euler number, inclination and filling degree to divide the defects on the glass surface into scratches, stains, bubbles and pits, which has a large amount of calculation and a complex process, large calculation parameters, many output results, multi-process calculation and result output affecting the judgment efficiency, and the two-dimensional information judgment accuracy is not high. A 3D glass online monitoring platform based on machine vision is provided.

[0004] The purpose of the present invention can be achieved by adopting the following technical solutions:

[0005] A 3D glass online monitoring platform based on machine vision includes an image acquisition module, a motion control unit and an auxiliary unit, wherein the image acquisition module is connected to a data analysis module, wherein the data analysis module is connected to a defect classification module, wherein the image acquisition module includes a 3D camera and an image acquisition card for acquiring images captured by the 3D camera, wherein the data analysis module includes a filter block, wherein the filter block is connected to a 3D information calculation block, wherein the 3D information analysis block includes a defect volume calculation block, a surrounding surface calculation block, an in-body tilt depth calculation block, a stereo extension calculation block and a stereo inclination angle calculation block, wherein the functions of each module are as follows:

[0006] The filter block uses the filling degree and depth index to filter out tiny interference points and surface stains respectively, indicating that defects can be wiped away;

[0007] The defect volume calculation block uses the product of the unit pixel volume in the defect and the number of pixel blocks to represent the defect volume, and uses the ratio of the defect volume to the volume of the entire glass as the filter bubble ratio to filter out tiny bubbles. The analysis formula of the defect volume calculation block is:

[0008] (1) Defect volume formula:

[0009]

[0010] Where: V qx is recorded as the defect volume, i, j and z are recorded as the horizontal and vertical segment numbers of the glass surface defect area and the vertical depth segment number of the glass, n, m and l are recorded as the horizontal and vertical segment numbers of the glass surface defect area and the vertical depth segment number of the glass, respectively, and V0 is recorded as the volume of the unit volume pixel;

[0011] (2) Follicle ratio formula:

[0012]

[0013] Where: α is the filter bubble ratio, L is the length of the glass in the area to be tested, D is the width of the glass in the area to be tested, and H is the thickness of the glass in the area to be tested;

[0014] The enclosure area calculation block uses the product of the maximum perimeter and the maximum depth on the horizontal section of the defect contour as the enclosure area, and uses the size of the enclosure area to reflect the area of ​​the largest circumscribed special-shaped cylinder of the defect. As a three-dimensional basic attribute of the defect, the formula of the enclosure area is:

[0015] S yx =C max *H max ;

[0016] Where: S yx Recorded as the enclosing area, C max H is the maximum circumference of the defect contour on the horizontal section. max Recorded as the maximum depth of the defect profile;

[0017] The internal inclination depth is obtained by multiplying the ratio of the major axis and the minor axis of the circumscribed minimum ellipse having the same second-order center distance as the maximum cross-section on the horizontal section of the defect by the average depth of the defect. The obtained value is recorded as the internal inclination depth, which is used to judge the flatness of the defect. The evaluation formula of the internal inclination depth is:

[0018]

[0019] Where: I bp Recorded as the internal tilt degree, L cz and L dz are respectively recorded as the major and minor axis lengths of the circumscribed minimum ellipse with the same second-order center distance as the defect, Recorded as the average depth of the defect;

[0020] The three-dimensional extension calculation block uses the ratio of the volume of the smallest cuboid circumscribing the defect to the defect volume as the three-dimensional extension, which is used to determine the three-dimensional extension degree of the defect. The evaluation formula of the three-dimensional extension is:

[0021]

[0022] Where: I ys Recorded as the three-dimensional extension, V wj,min It is recorded as the volume of the smallest circumscribed cuboid of the defect;

[0023] The three-dimensional inclination angle calculation block uses the angle between the major axis of the circumscribed minimum ellipse with the same second-order center distance as the maximum cross section on the horizontal section of the defect and the horizontal direction, multiplied by the arc tangent value corresponding to the ratio of the maximum depth to the length of the major semi-axis. The obtained value is recorded as the three-dimensional inclination angle, which is used to represent the three-dimensional direction of the defect. The evaluation formula of the three-dimensional inclination angle is:

[0024]

[0025] Where: θ lt is denoted as the solid inclination angle, It is recorded as the inverse tangent value corresponding to the ratio of the maximum depth to the length of the major semi-axis.

[0026] As a further solution of the present invention, the motion control unit includes a lower computer, a PLC motion controller, a photoelectric encoder and an actuator, the actuator includes a marking machine and a sorting machine, and the auxiliary unit includes a dustproof light shield, an alarm and a server.

[0027] As a further scheme of the present invention, the defect classification module classifies the defects filtered out by the filter block as filtered defects, records the defects with a follicle ratio greater than the standard microbubble follicle ratio and less than or equal to the standard bubble follicle ratio as bubbles, records the defects with an enclosing area greater than the standard enclosing area, a follicle ratio greater than the standard bubble follicle ratio and an in-body inclination depth as large-area flat concave defects or large-area flat internal hollow defects, and records the defects with an enclosing area less than or equal to the standard enclosing area, an in-body inclination depth less than or equal to the standard in-body inclination depth, a three-dimensional extension greater than the standard three-dimensional extension and a three-dimensional inclination angle greater than the standard three-dimensional inclination angle as scratches, and the standard microbubble follicle ratio, standard bubble follicle ratio, standard enclosing area, standard in-body inclination depth and standard three-dimensional inclination angle are obtained by analyzing the historical data of glass inspection.

[0028] As a further solution of the present invention, the defect classification module is provided with a glass quality score, and the glass quality score is negatively correlated with the defect volume, the surrounding area, the internal tilt depth, the three-dimensional extension and the three-dimensional tilt angle. The evaluation formula of the glass quality score is:

[0029] P zl =λ1*V qx +λ2*S yx +λ3*I bp +λ4*I ys +λ5*θ lt +ξ;

[0030] Where: P zl is recorded as the glass quality score, and λ1, λ2, λ3, λ4 and λ5 are recorded as V qx , S yx ,I bp ,I ys and θ lt The linear transformation coefficient of ξ is recorded as the transformation constant additional value of the glass quality score. λ1, λ2, λ3, λ4, λ5 and ξ are all obtained by analyzing historical data using numerical analysis software.

[0031] Enter the glass quality score as an independent variable into the function Where f is the function expression, x is the independent variable of function f, and using f(P zl ) value is compared with 0.5 to realize the classification of glass quality score. When 0≤f(P zl )<0.5, the glass quality is divided into one category, when 0.5≤f(P zl )≤1, the glass quality is divided into two categories.

[0032] Beneficial technical effects of the present invention: According to the machine vision-based 3D glass online monitoring platform of the present invention, through the setting of the 3D camera, the 3D camera can be used in combination with the deep learning technology to identify the three-dimensional information of defects on the glass, and the volume, surrounding area, internal tilt depth, three-dimensional extension and three-dimensional inclination angle of the defects can be calculated in combination with the three-dimensional information of the defects, which can help the defect analysis module to establish a basis for data analysis, and provide a basis for identification for three-dimensional classification of defects in combination with three-dimensional data information, so as to facilitate the identification of three-dimensional features of defects, and can further optimize the judgment of scratches, stains, bubbles and pitting that originally relied on two-dimensional data by combining the depth information obtained by the 3D camera, so as to realize three-dimensional recognition and classification of defects on the glass, and at the same time optimize the calculation process, improve the efficiency of calculation and recognition, and improve the accuracy of defect recognition and splitting. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The figure is a structural block diagram of a 3D glass online monitoring platform based on machine vision according to the present invention. DETAILED DESCRIPTION

[0034] In order to make the technical solution of the present invention more clear and specific to those skilled in the art, the present invention is further described in detail below in conjunction with embodiments and drawings, but the implementation manner of the present invention is not limited thereto.

[0035] like Figure 1 As shown, the 3D glass online monitoring platform based on machine vision provided in this embodiment includes an image acquisition module, a motion control unit and an auxiliary unit, the image acquisition module is connected to a data analysis module, the data analysis module is connected to a defect classification module, the image acquisition module includes a 3D camera and an image acquisition card for acquiring images taken by the 3D camera, the data analysis module includes a filter block, the filter block is connected to a 3D information calculation block, the 3D information analysis block includes a defect volume calculation block, a surrounding surface calculation block, an in-body tilt depth calculation block, a three-dimensional extension calculation block and a three-dimensional inclination angle calculation block, wherein the functions of each module are as follows:

[0036] The filter block uses the filling degree and depth index to filter out tiny interference points and surface stains respectively, indicating that defects can be wiped away;

[0037] The defect volume calculation block uses the product of the unit pixel volume in the defect and the number of pixel blocks to represent the defect volume, and uses the ratio of the defect volume to the volume of the entire glass as the filter bubble ratio to filter out tiny bubbles. The analysis formula of the defect volume calculation block is:

[0038] (1) Defect volume formula:

[0039]

[0040] Where: V qx is recorded as the defect volume, i, j and z are recorded as the horizontal and vertical segment numbers of the glass surface defect area and the vertical depth segment number of the glass, n, m and l are recorded as the horizontal and vertical segment numbers of the glass surface defect area and the vertical depth segment number of the glass, respectively, and V0 is recorded as the volume of the unit volume pixel;

[0041] By calculating the defect volume, the defects can be filtered for the first time in combination with the defect volume, and very small bubbles can be filtered out, laying a data foundation for the bubble ratio formula.

[0042] (2) Follicle ratio formula:

[0043]

[0044] Where: α is the filter bubble ratio, L is the length of the glass in the area to be tested, D is the width of the glass in the area to be tested, and H is the thickness of the glass in the area to be tested;

[0045] By setting the bubble ratio formula, the relative sizes of the volume of the glass and the volume of the bubbles can be combined to filter out small bubbles that have little impact on the overall performance of the glass, thus preventing glasses with small bubbles from being mistakenly filtered out as defective glass.

[0046] The enclosure area calculation block uses the product of the maximum perimeter and the maximum depth on the horizontal section of the defect contour as the enclosure area, and uses the size of the enclosure area to reflect the area of ​​the largest circumscribed special-shaped cylinder of the defect. As a three-dimensional basic attribute of the defect, the formula of the enclosure area is:

[0047] S yx =C max *H max ;

[0048] Where: S yx Recorded as the enclosing area, C max H is the maximum circumference of the defect contour on the horizontal section. max Recorded as the maximum depth of the defect profile;

[0049] By setting the enclosing area formula, the defects of the glass can be displayed in combination with the maximum depth of the defect and the maximum circumference of the horizontal section of the defect outline, showing a basic three-dimensional attribute of the glass defects, which serves as a basis for judgment and lays a data foundation for subsequent classification and identification.

[0050] The internal inclination depth is obtained by multiplying the ratio of the major axis and the minor axis of the circumscribed minimum ellipse having the same second-order center distance as the maximum cross-section on the horizontal section of the defect by the average depth of the defect. The obtained value is recorded as the internal inclination depth, which is used to judge the flatness of the defect. The evaluation formula of the internal inclination depth is:

[0051]

[0052] Where: I bp Recorded as the internal tilt degree, L cz and L dz are respectively recorded as the major and minor axis lengths of the circumscribed minimum ellipse with the same second-order center distance as the defect, Recorded as the average depth of the defect;

[0053] By setting the in-body tilt depth, the three-dimensional flatness of the defect can be comprehensively displayed by combining the average depth within the defect contour with the ratio of the major axis to the minor axis of the circumscribed minimum ellipse with the same second-order center distance in the largest cross-section on the horizontal section of the defect. A larger value can show that the defect is relatively flat, which may be a crack or a narrow and flat scratch.

[0054] The three-dimensional extension calculation block uses the ratio of the volume of the smallest cuboid circumscribing the defect to the defect volume as the three-dimensional extension, which is used to determine the three-dimensional extension degree of the defect. The evaluation formula of the three-dimensional extension is:

[0055]

[0056] Where: I ys Recorded as the three-dimensional extension, V wj,min It is recorded as the volume of the smallest circumscribed cuboid of the defect;

[0057] The calculation of the three-dimensional elongation can determine the extension degree of the defect in the depth direction, indicate the impact range of the defect, and indirectly reflect the impact of the defect on the glass quality.

[0058] The three-dimensional inclination angle calculation block uses the angle between the major axis of the circumscribed minimum ellipse with the same second-order center distance as the maximum cross section on the horizontal section of the defect and the horizontal direction, multiplied by the arc tangent value corresponding to the ratio of the maximum depth to the length of the major semi-axis. The obtained value is recorded as the three-dimensional inclination angle, which is used to represent the three-dimensional direction of the defect. The evaluation formula of the three-dimensional inclination angle is:

[0059]

[0060] Where: θ lt is denoted as the solid inclination angle, It is recorded as the inverse tangent value corresponding to the ratio of the maximum depth to the length of the major semi-axis.

[0061] By setting the three-dimensional inclination angle, it is easy to show the plane spreading degree and depth expansion degree of the defect, so as to show the three-dimensional expansion degree of the defect.

[0062] The machine vision-based 3D glass online monitoring platform proposed in the present invention can use the 3D camera in combination with deep learning technology to identify the three-dimensional information of defects on the glass through the setting of the 3D camera, and calculate the volume, surrounding area, internal tilt depth, three-dimensional extension and three-dimensional inclination angle of the defects in combination with the three-dimensional information of the defects, which can help the defect analysis module to establish a basis for data analysis, and provide a basis for identification for three-dimensional classification of defects in combination with three-dimensional data information, so as to facilitate the identification of three-dimensional features of defects, and can further optimize the judgment of scratches, stains, bubbles and pits that originally relied on two-dimensional data in combination with the depth information obtained by the 3D camera, so as to realize three-dimensional recognition and classification of defects on the glass, and at the same time optimize the calculation process, improve the efficiency of calculation and recognition, and improve the accuracy of defect recognition and splitting.

[0063] The motion control unit includes a lower computer, a PLC motion controller, a photoelectric encoder and an actuator, the actuator includes a marking machine and a sorting machine, and the auxiliary unit includes a dustproof light shield, an alarm and a server.

[0064] The setting of the control unit can help the detection platform to realize the automation of equipment detection and operation, improve the efficiency and automation of detection, and the setting of the marking machine and sorting machine can facilitate the sorting and classification of glass, and label it for easy identification. The setting of the dustproof hood can prevent the influence of external excess light sources on the image captured by the 3D camera, and can also prevent dust from falling on the glass and increasing the calculation burden of recognition. The setting of the alarm can sound an alarm when glass with obvious defects and poor quality is identified, and the defective glass can be processed. The setting of the server can enable the service information of multiple devices to be shared, transmitted and calculated.

[0065] The defect classification module classifies defects filtered out by the filter block as filtered defects, records defects with a follicle ratio greater than a standard microbubble-follicle ratio and less than or equal to a standard bubble-follicle ratio as bubbles, records defects with an enclosing area greater than a standard enclosing area, a follicle ratio greater than a standard bubble-follicle ratio and an in-body inclination depth as large-area flat sunken defects or large-area flat internal hollow defects, and records defects with an enclosing area less than or equal to a standard enclosing area, an in-body inclination depth less than or equal to a standard in-body inclination depth, a three-dimensional extension greater than a standard three-dimensional extension, and a three-dimensional inclination angle greater than a standard three-dimensional inclination angle as scratches, and the standard microbubble-follicle ratio, the standard bubble-follicle ratio, the standard enclosing area, the standard in-body inclination depth and the standard three-dimensional inclination angle are obtained by analyzing historical glass inspection data.

[0066] The defect classification module sets a glass quality score, which is negatively correlated with the defect volume, surrounding area, internal tilt depth, three-dimensional extension, and three-dimensional tilt angle. The evaluation formula for the glass quality score is:

[0067] P zl =λ1*V qx +λ2*S yx +λ3*I bp +λ4*I ys +λ5*θ lt +ξ;

[0068] Where: P zl is recorded as the glass quality score, and λ1, λ2, λ3, λ4 and λ5 are recorded as V qx , S yx ,I bp ,I ys and θ lt The linear transformation coefficient of ξ is recorded as the transformation constant additional value of the glass quality score. λ1, λ2, λ3, λ4, λ5 and ξ are all obtained by analyzing historical data using numerical analysis software.

[0069] Enter the glass quality score as an independent variable into the function Where f is the function expression, x is the independent variable of function f, and using f(P zl ) value is compared with 0.5 to realize the classification of glass quality score. When 0≤f(P zl )<0.5, the glass quality is divided into one category, when 0.5≤f(P zl )≤1, the glass quality is divided into two categories.

[0070] By classifying the glass quality scores, it is possible to provide reference indicators for glass quality assessment according to the classification of the glass quality scores, facilitate optimization of the glass defect detection algorithm, and realize the classification of glass quality.

[0071] In summary, in this embodiment, according to the 3D glass online monitoring platform based on machine vision of this embodiment, by calculating the defect volume, the defect can be first filtered in combination with the defect volume, and very small bubbles can be filtered out, laying a data foundation for the bubble ratio formula. By setting the bubble ratio formula, the small bubbles that have little impact on the overall performance of the glass can be filtered out in combination with the relative size between the volume of the glass and the volume of the bubble, so as to avoid the glass cup with small bubbles being mistakenly considered as defective glass and filtered out. By setting the surrounding area formula, the defects of the glass can be displayed in combination with the maximum depth of the defect and the maximum perimeter on the horizontal section of the defect contour, showing a basic three-dimensional attribute of the glass defect, which can be used as a basis for judgment to lay a data foundation for subsequent classification and identification. By setting the in-body tilt depth, the three-dimensional flatness of the defect can be comprehensively displayed in combination with the average depth within the defect contour and the ratio of the major axis and the minor axis of the circumscribed minimum ellipse with the same second-order center distance on the maximum cross-section of the defect horizontal section. A larger value can show that the defect is relatively flat, which may be a crack or a narrow and flat scratch. The calculation of the three-dimensional extension can determine the extension degree of the defect in the depth direction, indicating the scope of the defect, and indirectly reflecting the degree of influence of the defect on the glass quality. The setting of the three-dimensional inclination angle can conveniently show the plane spreading degree and the depth expansion degree of the defect, so as to show the three-dimensional expansion degree of the defect. The setting of the control unit can help the detection platform to realize the automatic processing of equipment detection and operation, improve the efficiency and automation of detection, and the setting of the marking machine and the sorting machine can facilitate the sorting and classification of glass, and label it for easy identification. The setting of the dustproof hood can prevent the influence of external redundant light sources on the image captured by the 3D camera, and can also prevent dust from falling on the glass, increasing the calculation burden of recognition. The setting of the alarm can sound an alarm when the glass with obvious defects and poor glass quality is identified, and the defective glass can be processed. The setting of the server can enable the service information of multiple devices to be shared, transmitted and calculated. By classifying the glass quality score, it can provide reference indicators for the quality evaluation of glass according to the classification of the glass quality score, which is convenient for optimizing the algorithm of glass defect detection and realizing the classification of glass quality.

[0072] The above description is only a further embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and concepts of the present invention within the scope disclosed by the present invention, which belong to the protection scope of the present invention.

Claims

1. A 3D glass online monitoring platform based on machine vision, characterized in that: It includes an image acquisition module, a motion control unit and an auxiliary unit. The image acquisition module is connected to a data analysis module, which is connected to a defect classification module. The image acquisition module includes a 3D camera and an image acquisition card for acquiring images taken by the 3D camera. The data analysis module includes a filter block, which is connected to a 3D information calculation block. The 3D information analysis block includes a defect volume calculation block, an enclosing area calculation block, an in-vivo tilt depth calculation block, a stereo extension calculation block and a stereo inclination angle calculation block. The functions of each module are as follows: The filter block uses the filling degree and depth index to filter out tiny interference points and surface stains respectively, indicating that defects can be wiped away; The defect volume calculation block uses the product of the unit pixel volume in the defect and the number of pixel blocks to represent the defect volume, and uses the ratio of the defect volume to the volume of the entire glass as the filter bubble ratio to filter out tiny bubbles. The analysis formula of the defect volume calculation block is: (1) Defect volume formula: Where: V qx is recorded as the defect volume, i, j and z are recorded as the horizontal and vertical segment numbers of the glass surface defect area and the vertical depth segment number of the glass, n, m and l are recorded as the horizontal and vertical segment numbers of the glass surface defect area and the vertical depth segment number of the glass, respectively, and V0 is recorded as the volume of the unit volume pixel; (2) Follicle ratio formula: Where: α is the filter bubble ratio, L is the length of the glass in the area to be tested, D is the width of the glass in the area to be tested, and H is the thickness of the glass in the area to be tested; The enclosure area calculation block uses the product of the maximum perimeter and the maximum depth on the horizontal section of the defect contour as the enclosure area, and uses the size of the enclosure area to reflect the area of ​​the largest circumscribed special-shaped cylinder of the defect. As a three-dimensional basic attribute of the defect, the formula of the enclosure area is: S yx =C max *H max ; Where: S yx Recorded as the enclosing area, C max H is the maximum circumference of the defect contour on the horizontal section. max Recorded as the maximum depth of the defect profile; The internal tilt depth is obtained by multiplying the ratio of the major axis and the minor axis of the circumscribed minimum ellipse having the same second-order center distance as the maximum cross-section on the horizontal section of the defect by the average depth of the defect. The obtained value is recorded as the internal tilt depth, which is used to judge the flatness of the defect. The evaluation formula of the internal tilt depth is: Where: I bp Recorded as the internal tilt degree, L cz and L dz are respectively recorded as the major and minor axis lengths of the circumscribed minimum ellipse with the same second-order center distance as the defect, Recorded as the average depth of defects; The three-dimensional extension calculation block uses the ratio of the volume of the smallest cuboid circumscribing the defect to the defect volume as the three-dimensional extension, which is used to determine the three-dimensional extension degree of the defect. The evaluation formula of the three-dimensional extension is: Where: I ys Recorded as the three-dimensional extension, V wj,min It is recorded as the volume of the smallest circumscribed cuboid of the defect; The three-dimensional inclination angle calculation block uses the angle between the major axis of the circumscribed minimum ellipse with the same second-order center distance as the maximum cross section on the horizontal section of the defect and the horizontal direction, multiplied by the arc tangent value corresponding to the ratio of the maximum depth to the length of the major semi-axis. The obtained value is recorded as the three-dimensional inclination angle, which is used to represent the three-dimensional direction of the defect. The evaluation formula of the three-dimensional inclination angle is: Where: θ lt is denoted as the solid inclination angle, It is recorded as the inverse tangent value corresponding to the ratio of the maximum depth to the length of the major semi-axis.

2. The 3D glass online monitoring platform based on machine vision according to claim 1, characterized in that: The motion control unit includes a lower computer, a PLC motion controller, a photoelectric encoder and an actuator, the actuator includes a marking machine and a sorting machine, and the auxiliary unit includes a dustproof light shield, an alarm and a server.

3. The 3D glass online monitoring platform based on machine vision according to claim 1, characterized in that: The defect classification module classifies defects filtered out by the filter block as filtered defects, records defects with a follicle ratio greater than a standard microbubble-follicle ratio and less than or equal to a standard bubble-follicle ratio as bubbles, records defects with an enclosing area greater than a standard enclosing area, a follicle ratio greater than a standard bubble-follicle ratio and an in-body inclination depth as large-area flat sunken defects or large-area flat internal hollow defects, and records defects with an enclosing area less than or equal to a standard enclosing area, an in-body inclination depth less than or equal to a standard in-body inclination depth, a three-dimensional extension greater than a standard three-dimensional extension, and a three-dimensional inclination angle greater than a standard three-dimensional inclination angle as scratches, and the standard microbubble-follicle ratio, the standard bubble-follicle ratio, the standard enclosing area, the standard in-body inclination depth and the standard three-dimensional inclination angle are obtained by analyzing historical glass inspection data.

4. The 3D glass online monitoring platform based on machine vision according to claim 1, characterized in that: The defect classification module sets a glass quality score, which is negatively correlated with the defect volume, surrounding area, internal tilt depth, three-dimensional extension, and three-dimensional tilt angle. The evaluation formula for the glass quality score is: P zl =λ1*V qx +λ2*S yx +λ3*I bp +λ4*I ys +λ5*θ lt +ξ; Where: P zl is recorded as the glass quality score, and λ1, λ2, λ3, λ4 and λ5 are recorded as V qx , S yx ,I bp ,I ys and θ lt The linear transformation coefficient, ξ is recorded as the additional value of the transformation constant of the glass quality score, λ1, λ2, λ3, λ4, λ5 and ξ are obtained by analyzing historical data using numerical analysis software; Enter the glass quality score as an independent variable into the function Where f is the function expression, x is the independent variable of function f, and using f(P zl ) value is compared with 0.5 to realize the classification of glass quality score. When 0≤f(P zl )<0.5, the glass quality is divided into one category, when 0.5≤f(P zl )≤1, the glass quality is divided into two categories.

Images