An image-based film mark detection method for a dicing machine

Abstract

The present invention discloses an image-based film mark detection method for a dicing machine, comprising the following steps: S1, setting the maximum width and contrast of the knife mark to be extracted; S2, after the machine cutting action is completed, the lens moves to the knife entry and exit to collect images; S3, using a Gaussian filter to obtain the first-order partial derivative and the second-order partial derivative of the image; S4, calculating the sub-pixel coordinates, response and direction of each pixel point according to the Hessian matrix; S5, performing hysteresis threshold and link operations on the above results to form a number of contour lines; S6, performing linear regression on the contour lines, and then screening out the curves that meet the knife marks, and calculating their widths. The present invention adopts the above-mentioned film mark detection method to distinguish the presence or absence of knife marks, identify the width of knife marks, and finally obtain the depth information of the blade cutting into the base film, and adjust the cutting depth of the next knife, so as to achieve accurate and manual control of the cutting depth.

Description

Technical Field

[0001] The present invention relates to the technical field of dicing machines, and in particular to an image-based film mark detection method for dicing machines. Background Art

[0002] When cutting LED chips or wafer products, the dicing machine usually has high requirements for the cutting depth. Generally, the blade will just cut into the bottom film, or cut into the bottom film by 5-10 microns. This can just cut through the product, effectively control the back collapse, and maintain the excellent characteristics of the product. However, the horizontal accuracy of the dicing machine's working disk is difficult to meet the above standards, generally more than 3 microns. The larger the size of the working disk, the greater the error, and the more difficult it is to ensure the accuracy. This is reflected in the use of the machine. When cutting the product, the knife marks formed by the blade at the bottom film where the knife enters and exits are not consistent. There may be no knife marks on the left and there are on the right. The knife marks on the left may cut deeper than those on the right. In this case, it is generally adjusted based on human experience, which not only increases the probability of error, but also affects the efficiency of the machine and the product yield. Summary of the invention

[0003] The present invention proposes a film mark detection method to solve the problem that the cutting depth must be controlled by manual experience. To achieve the above purpose, the present invention provides the following technical solutions:

[0004] A film mark detection method based on an image for a dicing machine comprises the following steps:

[0005] S1. Set the maximum width and contrast of the knife marks to be extracted;

[0006] S2. After the machine cutting action is completed, the camera moves to the entry and exit positions to collect images;

[0007] S3, using Gaussian filter to obtain the first-order partial derivative and second-order partial derivative of the image;

[0008] S4, calculating the sub-pixel coordinates, response and direction of each pixel point according to the Hessian matrix;

[0009] S5, performing hysteresis threshold and linking operations on the above results, thereby forming a number of contour lines;

[0010] S6. Perform linear regression on the contour line, then filter out the curve that meets the knife mark and calculate its width.

[0011] Preferably, step S1 also includes calculating the standard deviation of the Gaussian filter based on the maximum width and contrast of the knife mark, and setting a weak threshold and a strong threshold for linking pixel points, wherein the weak threshold and the strong threshold are both constants.

[0012] Preferably, step S4 is specifically:

[0013] S41, constructing a Hessian matrix for each pixel, calculating the eigenvalue and eigenvector of the Hessian matrix, and saving the eigenvalue with the largest absolute value and its corresponding eigenvector;

[0014] Hesse matrix

[0015] Among them, r xx 、r xy 、r yy To obtain the second-order partial derivative of the image using a Gaussian filter;

[0016] S42. Calculate the sub-pixel coordinates of the current pixel based on the eigenvalue and eigenvector obtained in step S41. Pixels with eigenvalues ​​greater than the strong threshold are marked as 2, and pixels between the strong threshold and the weak threshold are marked as 1. Save the eigenvalues ​​corresponding to the pixels and their corresponding eigenvectors.

[0017] Preferably, step S5 is specifically:

[0018] Sort the pixels marked as 2 in descending order according to their eigenvalues, link the pixels with values ​​greater than the weak threshold starting from the points with large eigenvalues ​​along the direction of the eigenvector, and finally form multiple contour lines, which are saved.

[0019] Preferably, step S6 specifically comprises:

[0020] S61, screening out contour lines that are larger than half of the image width, fitting straight lines to these contour lines using the least squares method, and retaining contour lines whose angles between the directions of the fitting straight lines and the horizontal direction are within ±2°;

[0021] S62, calculating the mean absolute error between the contour line obtained in step S61 and the fitting straight line, and selecting the contour line whose mean absolute error is less than 5;

[0022] S63, for the contour line obtained in step S62, search for the point with the largest absolute value of gradient at each point along two directions of the feature vector, so as to obtain the width of the left and right sides of the current point;

[0023] S64, averaging the widths of each point obtained in step S63 as the final output width.

[0024] The present invention adopts the above-mentioned structure of the image-based film mark detection method. By equipping it with a high-magnification microscope, it collects the cut mark images of the entry and exit positions of the knife in real time, distinguishes the presence or absence of the knife mark, identifies the width of the knife mark, and finally obtains the depth information of the blade cutting into the base film, and adjusts the cutting depth of the next knife, thereby achieving accurate and manual control of the cutting depth. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is the film trace image to be identified;

[0026] Figure 2 is the eigenvalue image obtained by the Hessian matrix;

[0027] Figure 3 It is the contour map after the link operation;

[0028] Figure 4 is the center line and width of the extracted knife mark;

[0029] Figure 5 To determine the width schematic diagram by searching for the maximum gradient in two directions along the normal direction of the contour point;

[0030] Figure 6 is another film trace image to be identified;

[0031] Figure 7 Schematic diagram of detected contour lines and their widths;

[0032] Figure 8 Determine the width diagram for searching the edge;

[0033] Fig. 9 Schematic diagram for calculating cutting depth from output width for conical inserts. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0035] The detection system of the scheme of the present invention is built on the motion and image recognition system of the dicing machine. The motion system of the dicing machine is mainly composed of X, Y, Z, and T axes, wherein the T axis represents the rotation axis with a workbench on it, the T axis is fixed on the X axis, the X axis can move left and right, the microscope is fixed on the Y axis, the Y axis can move forward and backward, and the Z axis is fixed on the Y axis and can move up and down. The detection system cooperates with the software algorithm to realize the discrimination of the presence or absence of film marks and the corresponding width calculation. The algorithm first calculates the first-order partial derivative and the second-order partial derivative of the image, and then calculates the precise position, eigenvalue, and direction of each point according to the Hessian matrix, and finally extracts the center and width of the knife mark according to the above information. It mainly detects the range of 0-20 microns that the dicing machine blade cuts into the film, and there is a clear conversion relationship between width and depth.

[0036] The specific implementation steps of the plan are as follows:

[0037] S1. Set the maximum width and contrast of the knife marks to be extracted, and calculate the standard deviation of the Gaussian filter based on this, and set the weak threshold and strong threshold used to link the pixels. The weak threshold and strong threshold are both constants. The closer to the pixel, the larger the gray value, and the farther away from the pixel, the smaller the gray value. The strong and weak thresholds represent the distance relationship between the pixel points.

[0038] S2. After the machine cutting action is completed, the camera moves to the point where the cutting knife just entered and exited to collect images;

[0039] S3, use the Gaussian filter edge operator to obtain the image r x ,r y ,r xx ,r xy ,r yy (first-order and second-order partial derivatives); Gaussian filter relies on Gaussian function to complete the operation, and its corresponding formula is as follows:

[0040] One-dimensional Gaussian function and its first-order and second-order partial derivatives

[0041]

[0042]

[0043]

[0044] Two-dimensional Gaussian function and its first and second order partial derivatives

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051] Gaussian filtering is the process of weighted averaging the entire image. The value of each pixel is obtained by weighted averaging itself and other pixel values ​​in the neighborhood (Gaussian window). The specific operation of Gaussian filtering is: use a template (or convolution, mask) to scan each pixel in the image, and replace the value of the central pixel of the template with the weighted average gray value of the pixels in the neighborhood determined by the template.

[0052] S4. Construct the Hessian matrix for each pixel Find the eigenvalues ​​and eigenvectors of the matrix, and save the eigenvalue with the largest absolute value and its corresponding eigenvector; the second-order Hessian matrix can only get two eigenvalues, so only the eigenvalue with the largest absolute value needs to be saved;

[0053] S5. Construct a linear equation At+b=0 based on the eigenvalue and eigenvector of step S4. If the equation t has a solution and t∈[-0.5,0.5], calculate the sub-pixel coordinates of the current pixel. If the eigenvalue is greater than the strong threshold, it is marked as 2, and if it is greater than the weak threshold, it is marked as 1. Save the eigenvalue and its corresponding eigenvector.

[0054]

[0055]

[0056]

[0057]

[0058] (p x , p y )=(tn x tn y )

[0059]

[0060] The above formulas are: ① Hessian matrix of function f, ② second-order directional derivative of function f, ③ second-order Taylor expansion of function z, ④ first-order derivative of function z, ⑤ sub-pixel position where function z takes extreme value, ⑥ analytical expression of t. These functions mainly obtain coordinates with sub-pixel accuracy by seeking extreme value.

[0061] S6, sort the pixel points marked as 2 in step S5 in descending order according to the eigenvalue, link the pixel points greater than the weak threshold from the point with the largest eigenvalue along the direction of the eigenvector (bidirectional), and finally form multiple contour lines, and save these contour lines; the pixel points of the strong threshold must be linked, and the weak threshold pixel points are used to determine whether the pixel points are adjacent to the strong threshold pixel points, and if so, they are also linked;

[0062] Note: To observe the outline of a scratch, you need to start from the first or last point of the outline and observe the direction in both horizontal and vertical directions. Therefore, the feature vector generated by the pixel points on the scratch also exists in two directions.

[0063] S7, filter out contour lines that are larger than half of the image width, fit straight lines to these contour lines using the least squares method, and retain contour lines whose angles between the fitted straight line and the horizontal direction are within ±2°;

[0064] S8, calculating the mean absolute error between the contour line and the fitted straight line in step S7, and selecting the contour line with an error less than 5;

[0065] S9, for the contour line obtained in step S8, search for the point with the largest absolute value of gradient (within half of the maximum width) at each point along the two directions of the feature vector, so as to obtain the width of the left and right sides of the current point;

[0066] S10, averaging the widths of each point obtained in step S9 as the final output width.

[0067] The tip of the dicing machine blade is usually V-shaped or conical, so each blade has a corresponding cutting formula. Fig. 9 As shown in the figure, the conical blade can further calculate the cutting depth OM by measuring the cutting width PP' (output width). The next time you make a cutting compensation, you only need to adjust the cutting depth according to the compensated cutting width.

[0068] The above are specific embodiments of the present invention, but the protection scope of the present invention should not be limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention, so the protection scope of the present invention should be based on the protection scope defined in the claims.

Claims

1. A film mark detection method based on an image for a dicing machine, characterized in that: The steps include: S1. Set the maximum width and contrast of the knife marks to be extracted; S2. After the machine cutting action is completed, the camera moves to the entry and exit positions to collect images; S3, using Gaussian filter to obtain the first-order partial derivative and second-order partial derivative of the image; S4, calculating the sub-pixel coordinates, response and direction of each pixel point according to the Hessian matrix; S5. Construct a linear equation At+b=0 according to the eigenvalue and eigenvector of step S4. If the equation t has a solution and t∈[-0.5,0.5], calculate the sub-pixel coordinates of the current pixel. If the eigenvalue is greater than the strong threshold, mark it as 2, and if it is greater than the weak threshold, mark it as 1. Save the eigenvalue and its corresponding eigenvector. Perform hysteresis threshold and link operations on the above results to form several contour lines. S6, performing linear regression on the contour line, then screening out the curve that meets the knife mark, and calculating its width; Step S6 is specifically as follows: S61, screening out contour lines that are larger than half of the image width, fitting straight lines to these contour lines using the least squares method, and retaining contour lines whose angles between the directions of the fitting straight lines and the horizontal direction are within ±2°; S62, calculating the average absolute error between the contour line obtained in step S61 and the fitting straight line, and selecting the contour line with an error less than 5; S63, for the contour line obtained in step S62, search for the point with the largest absolute value of gradient at each point along two directions of the feature vector, so as to obtain the width of the left and right sides of the current point; S64, averaging the widths of each point obtained in step S63 as the final output width.

2. The film mark detection method according to claim 1, characterized in that: Step S1 also includes calculating the standard deviation of the Gaussian filter according to the maximum width and contrast of the knife mark, and setting a weak threshold and a strong threshold for linking pixel points, wherein the weak threshold and the strong threshold are both constants.

3. The film mark detection method according to claim 2, characterized in that: Step S4 is specifically as follows: S41, constructing a Hessian matrix for each pixel, calculating the eigenvalue and eigenvector of the Hessian matrix, and saving the eigenvalue with the largest absolute value and its corresponding eigenvector; Hesse matrix Among them, r xx 、r xy 、r yy To obtain the second-order partial derivative of the image using a Gaussian filter; S42. Calculate the sub-pixel coordinates of the current pixel based on the eigenvalue and eigenvector obtained in step S41. Pixels whose eigenvalues ​​are greater than the strong threshold are marked as 2, and pixels between the strong threshold and the weak threshold are marked as 1. Save the eigenvalues ​​corresponding to the pixels and their corresponding eigenvectors.

4. The film mark detection method according to claim 3, characterized in that: Step S5 is specifically as follows: Sort the pixels marked as 2 in descending order according to their eigenvalues, link the pixels with values ​​greater than the weak threshold starting from the points with large eigenvalues ​​along the direction of the eigenvector, and finally form multiple contour lines, which are saved.

Images