Wafer fiducial point positioning method, inspection apparatus, and computer program product

Abstract

The present application relates to the technical field of image processing, and particularly relates to a wafer reference point positioning method, a detection device and a computer program product. The present application obtains a reference point region image including a reference point, obtains a pixel point set conforming to a preset gray scale ratio rule from the reference point region image, confirms a vertical candidate point set and a horizontal candidate point set according to the pixel point set, respectively fits a vertical line and a horizontal line based on the vertical candidate point set and the horizontal candidate point set, and finally calculates an intersection point as a reference point position, thereby solving the problem of inaccurate positioning of the reference point caused by image blurring and distortion due to position differences of multiple imaging units, and significantly improving the accuracy and reliability of the positioning of the reference point. Point fitting does not depend on the overall shape of the auxiliary figure corresponding to the reference point and the continuity of the line segment in the auxiliary figure, but is based on the vertical line candidate point set and the horizontal line candidate point set for point fitting, thereby ensuring that the positioning result of the reference point is accurate and reliable.

Description

Technical Field

[0001] This invention relates to the field of image inspection technology, and in particular to a wafer reference point positioning method, inspection equipment, and computer program product. Background Technology

[0002] In the semiconductor industry, where precision requirements are extremely high, a mark point is a reference marker with a specific shape (such as a cross, circle, rectangle, or rhombus) that is pre-fabricated on the wafer surface or at a specific location. Its core function is to provide a precise spatial coordinate reference for wafer positioning, calibration, inspection, and process steps. Therefore, the accuracy of the mark point is a core prerequisite for high-precision manufacturing, inspection, positioning, and system calibration.

[0003] In existing detection equipment, TDI (Time Delay Integration) imaging units are used to convert light signals into image pixel intensity. Multiple TDI imaging units cause imaging inconsistencies due to differences in spatial position, resulting in huge differences in contrast, brightness, and sharpness of the images. This causes the line segments of the reference points in the image to appear blurred or distorted, which negatively affects the accuracy and reliability of the reference point acquisition. Summary of the Invention To address the aforementioned problems, this invention provides a wafer reference point positioning method, a testing device, and a computer program product.

[0004] The present invention provides a method for locating wafer reference points, comprising the following steps: acquiring a wafer image; generating a reference point region image based on the wafer image according to a preset method, the reference point region image including the reference point; acquiring a set of pixels in the reference point region image that conform to a preset grayscale comparison rule; confirming a set of vertical candidate points and a set of horizontal candidate points based on the set of pixels; fitting the vertical candidate point set and the horizontal candidate point set respectively to obtain a vertical line and a horizontal line; calculating the intersection of the vertical line and the horizontal line as the final position of the reference point; wherein, confirming the set of vertical candidate points and the set of horizontal candidate points based on the set of pixels includes the following steps: establishing a reference coordinate system based on the reference point region image, the reference coordinate system being one of a pixel coordinate system, an image coordinate system, a camera coordinate system, and a world coordinate system; Based on the reference coordinate system, the midpoints of pairs of points with the same vertical coordinate in the pixel set are calculated to obtain the vertical candidate point set; based on the reference coordinate system, the midpoints of pairs of points with the same horizontal coordinate in the pixel set are calculated to obtain the horizontal candidate point set.

[0005] Preferably, the step of confirming the vertical candidate point set and the horizontal candidate point set based on the pixel set includes the following steps: establishing a reference coordinate system based on the reference point region image, wherein the reference coordinate system is one of the pixel coordinate system, image coordinate system, camera coordinate system, and world coordinate system; calculating the midpoint of pairs of points with the same vertical coordinate in the pixel set based on the reference coordinate system to obtain the vertical candidate point set; and calculating the midpoint of pairs of points with the same horizontal coordinate in the pixel set based on the reference coordinate system to obtain the horizontal candidate point set.

[0006] Preferably, obtaining the set of pixels in the reference point region image that conforms to a preset grayscale comparison rule includes: traversing each pixel in the reference point region image row by row or column by column; during the traversal of each pixel, for the currently traversed pixel, identifying pixels in its preset neighborhood that conform to the preset grayscale comparison rule as target points; constructing a set of pixels that conform to the preset grayscale comparison rule based on each target point; or, based on the auxiliary graphic information corresponding to the reference point, determining the geometric direction corresponding to the line segment of the auxiliary graphic in the reference point region image; traversing each pixel in its preset neighborhood along the geometric direction reference line; during the traversal of each pixel, for the currently traversed pixel, identifying pixels in its preset neighborhood that conform to the preset grayscale comparison rule as target points; constructing a set of pixels that conform to the preset grayscale comparison rule based on each target point.

[0007] Preferably, the preset grayscale comparison rule includes: obtaining the grayscale values ​​of each pixel within a preset neighborhood of the pixel; comparing the magnitudes of the grayscale values ​​of each pixel and selecting the point with the maximum grayscale value as the target point; or, obtaining the grayscale values ​​of each pixel within a preset neighborhood of the pixel; calculating the gradient magnitude of each pixel and selecting the point with the maximum gradient magnitude as the target point.

[0008] Preferably, the step of fitting the vertical line and the horizontal line based on the vertical candidate point set and the horizontal candidate point set respectively includes the following steps: removing outliers from the vertical candidate point set and the horizontal candidate point set to obtain a vertical line noise reduction point set and a horizontal line noise reduction point set; fitting the vertical line noise reduction point set using the least squares method, RANSAC algorithm, or Hough transform to obtain the vertical line; and fitting the horizontal line noise reduction point set to obtain the horizontal line.

[0009] Preferably, a first direction line is fitted based on the set of vertical candidate points, and it is determined whether the distance from each point in the set of vertical candidate points to the first direction line exceeds a threshold. If so, the point is removed, and the remaining points are used as the set of vertical line noise reduction points. A second direction line is fitted based on the set of horizontal candidate points, and it is determined whether the distance from each point in the set of horizontal candidate points to the second direction line exceeds a threshold. If so, the point is removed, and the remaining points are used as the set of horizontal line noise reduction points.

[0010] Preferably, the step of acquiring the wafer image and generating a reference point region image based on the wafer image according to a preset method, wherein the reference point region image includes the reference point, includes the following steps: acquiring the wafer image, inputting the wafer image into a pre-trained large model to obtain coarse localization information; acquiring the region of interest of the reference point based on the coarse localization information, and cropping the region of interest to obtain the reference point region image.

[0011] Preferably, the training of the large model includes the following steps: acquiring wafer reference point images through multiple sets of imaging units at different locations; performing enhancement processing on the wafer reference point images, labeling the reference points, and generating a training dataset, wherein the enhancement processing includes at least one of image translation, rotation, brightness transformation, and image distortion; and inputting the training dataset into the initial large model for training to obtain the large model.

[0012] To solve the above-mentioned technical problems, the present invention provides another technical solution as follows: a detection device for performing the wafer reference point positioning method described above, comprising a coarse positioning module for acquiring a wafer image and generating a reference point region image according to a preset method based on the wafer image, the reference point region image including the reference point; a pixel set extraction module for acquiring a set of pixels in the reference point region image that conforms to a preset grayscale comparison rule; and a candidate point set generation module for confirming a vertical candidate point set and a horizontal candidate point set based on the pixel set; wherein, the confirmation of the vertical candidate point set and the horizontal candidate point set based on the pixel set includes... The method includes: establishing a reference coordinate system based on the reference point region image, wherein the reference coordinate system is one of pixel coordinate system, image coordinate system, camera coordinate system, and world coordinate system; calculating the midpoint of pairs of points with the same ordinate in the pixel set based on the reference coordinate system to obtain the vertical candidate point set; calculating the midpoint of pairs of points with the same abscissa in the pixel set based on the reference coordinate system to obtain the horizontal candidate point set; a fitting module for fitting based on the vertical candidate point set and the horizontal candidate point set respectively to obtain a vertical line and a horizontal line; and a positioning module for calculating the intersection of the vertical line and the horizontal line as the final position of the reference point.

[0013] To solve the above-mentioned technical problems, the present invention provides another technical solution as follows: a computer program product, the computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the wafer reference point positioning method described above.

[0014] Compared with the prior art, the wafer reference point positioning method, detection equipment, and computer program product provided by the present invention have the following beneficial effects: 1. This invention provides a wafer reference point positioning method. The method involves acquiring a reference point region image including the reference point, obtaining a set of pixels conforming to a preset grayscale comparison rule from the reference point region image, identifying a vertical candidate point set and a horizontal candidate point set based on this pixel set, fitting vertical and horizontal lines based on the vertical and horizontal candidate point sets respectively, and finally calculating the intersection of the vertical and horizontal lines as the reference point position. This solves the problem of inaccurate reference point positioning caused by image blurring and distortion due to positional differences among multiple imaging units, significantly improving the accuracy and reliability of reference point positioning. Generating a reference point region image containing the reference point based on the wafer image allows for obtaining the approximate position of the reference point. The reference point falling within the reference point region image provides a focused area for subsequent more precise positioning, improving processing efficiency. By acquiring a set of pixels conforming to a preset grayscale comparison rule in the reference point region image, key pixels of the auxiliary graphic line segments corresponding to the reference point can be captured in blurred images, unaffected by line segment distortion or blurring caused by inconsistencies in TDI imaging. By identifying vertical and horizontal candidate point sets based on pixel sets and fitting straight lines to each set, the influence of distortion in the auxiliary graphic line segments corresponding to the reference point is offset. Using the intersection of the vertical and horizontal lines as the reference point position, the final precise coordinates of the reference point are obtained through geometric calculation. Since point fitting does not depend on the overall shape of the auxiliary graphic corresponding to the reference point or the continuity of line segments within the auxiliary graphic, but rather on the precise analysis of the pixel set to determine the reference point's position, even if the auxiliary graphic line segments corresponding to the reference point are distorted or blurred, point fitting based on the obtained vertical and horizontal candidate point sets can still yield accurate and reliable positioning results, unaffected by the distortion or blurring of the auxiliary graphic line segments corresponding to the reference point, effectively improving the accuracy and reliability of reference point positioning.

[0015] 2. The method for determining the vertical and horizontal candidate point sets based on a set of pixels provided in this embodiment includes the following steps: establishing a reference coordinate system based on a reference point region image, wherein the reference coordinate system is one of a pixel coordinate system, an image coordinate system, a camera coordinate system, or a world coordinate system; calculating the midpoint of paired points with the same vertical coordinate in the pixel set based on the reference coordinate system to obtain the vertical candidate point set; and calculating the midpoint of paired points with the same horizontal coordinate in the pixel set based on the reference coordinate system to obtain the horizontal candidate point set. By establishing a reference coordinate system based on a reference point region image, each pixel in the reference point region image has a corresponding coordinate value. This can be achieved regardless of whether a pixel coordinate system, or an image coordinate system, camera coordinate system, or world coordinate system obtained by pixel coordinate system transformation, is used. Calculating the midpoint of paired points with the same vertical coordinate and paired points with the same horizontal coordinate respectively utilizes the geometric symmetry of midpoint calculation to reduce the positional deviation of individual pixels caused by image blurring or noise, making the vertical and horizontal candidate point sets closer to the true distribution of the auxiliary graphic line segments corresponding to the reference point, further improving the fitting accuracy of vertical and horizontal lines.

[0016] 3. The method of obtaining a set of pixels conforming to a preset grayscale comparison rule in a reference point region image provided in this embodiment of the invention includes: traversing each pixel in the reference point region image row by row or column by column, and for each pixel, identifying pixels within its preset neighborhood that conform to the preset grayscale comparison rule as target points and constructing a pixel set. The row-by-row traversal method is highly versatile and applicable to reference points of different auxiliary graphics, comprehensively covering the reference point region image and avoiding omission of potentially conforming pixels. Based on the auxiliary graphic information corresponding to the reference point, the geometric direction corresponding to the line segment of the auxiliary graphic in the reference point region image is determined; along the geometric direction, each pixel within the preset neighborhood of the reference line corresponding to the geometric direction is traversed, and for each pixel, identifying pixels within its preset neighborhood that conform to the preset grayscale comparison rule as target points and constructing a pixel set. The method of traversing along the geometric direction corresponding to the line segment of the auxiliary graphic reduces interference from irrelevant pixels, improves screening efficiency, and enhances the targeting of the pixel set. Both of the above methods can improve the accuracy of the pixel set conforming to the preset grayscale comparison rule.

[0017] 4. The preset grayscale comparison rules provided in this embodiment of the invention include: obtaining the grayscale values ​​of each pixel within a preset neighborhood of the pixel; comparing the magnitude of the grayscale values ​​of each pixel and selecting the grayscale maximum point as the target point; or, obtaining the grayscale values ​​of each pixel within a preset neighborhood of the pixel; calculating the gradient magnitude of each pixel and selecting the point with the largest gradient magnitude as the target point. The grayscale maximum point is the pixel with the highest grayscale value within the preset neighborhood, which can highlight the grayscale difference between the auxiliary graphic line segment corresponding to the reference point and the background. When the grayscale maximum point is used as the target point, the core pixel corresponding to the auxiliary graphic line segment can be accurately located. The gradient magnitude reflects the degree of drastic change in pixel grayscale. The maximum gradient magnitude point usually corresponds to the edge of the line segment. When the maximum gradient magnitude point is used as the target point, even when the auxiliary graphic line segment corresponding to the reference point is blurred or distorted, its edge features can still be effectively captured, overcoming the problem of unclear line segment boundaries caused by the difference in image clarity in the reference point area. Both grayscale comparison rules can accurately extract key pixels of the auxiliary graphic line segments corresponding to the reference point from the reference point region image, ensuring that the obtained pixel set can truly reflect the distribution of the auxiliary graphic line segments corresponding to the reference point, and providing a reliable data foundation for subsequent steps.

[0018] 5. The method provided in this embodiment of the invention, which involves fitting vertical and horizontal candidate point sets to obtain vertical and horizontal lines respectively, includes the following steps: removing outliers from the vertical and horizontal candidate point sets to obtain a set of denoised vertical and horizontal lines; fitting the set of denoised vertical lines using the least squares method, RANSAC algorithm, or Hough transform to obtain a vertical line; and fitting the set of denoised horizontal lines to obtain a horizontal line. By first removing outliers and then using algorithms such as least squares, RANSAC, and Hough transform to fit line segments, the problem of candidate point position deviation caused by noise and pixel fluctuations in the image is specifically addressed. The point fitting algorithm, combined with the previous outlier denoising processing, can further improve the accuracy of the horizontal and vertical lines generated by point fitting, avoid interference from outliers on the fitting results, and ensure accurate positioning of the reference point's coordinates using the vertical and horizontal lines.

[0019] 6. In this embodiment of the invention, a first direction line is fitted based on a set of vertical candidate points. Points in the set of vertical candidate points whose distance to the first direction line exceeds a threshold are removed, and the remaining points are used as the set of denoised points for the vertical line. Similarly, a second direction line is fitted based on a set of horizontal candidate points. Points in the set of horizontal candidate points whose distance to the second direction line exceeds a threshold are removed, and the remaining points are used as the set of denoised points for the horizontal line. Since the first direction line is fitted based on the set of vertical candidate points, theoretically, the points in the set of vertical candidate points will not deviate too far from the first direction line. Similarly, since the second direction line is fitted based on the set of horizontal candidate points, theoretically, the points in the set of horizontal candidate points will not deviate too far from the second direction line. If the distance from a point in the set of vertical candidate points to the first direction line exceeds a threshold, or the distance from a point in the set of horizontal candidate points to the second direction line exceeds a threshold, it means that the point is an anomaly. This allows for the rapid screening of anomalies, purification of the point set, reduction of computational complexity, and improvement of the accuracy of the reference point positioning. Compared to complex denoising algorithms, this method is simpler, improving processing efficiency while ensuring denoising effect, and quickly obtaining reliable sets of denoised points for both vertical and horizontal lines as the basic data for fitting.

[0020] 7. The wafer image acquisition method provided in this embodiment of the invention generates a reference point region image based on the wafer image according to a preset method. The reference point region image includes reference points and includes the following steps: acquiring the wafer image; inputting the wafer image into a pre-trained large model to obtain coarse localization information; obtaining the region of interest (ROI) of the reference points based on the coarse localization information; and cropping the ROI to obtain the reference point region image. The reference point region image obtained by the above operation focuses on the local area where the reference points are located, eliminating irrelevant information interference from other non-reference point areas in the wafer image, reducing the complexity and computational load of subsequent processing, and improving the localization speed and accuracy.

[0021] 8. The training of the large model provided in this embodiment of the invention includes the following steps: acquiring wafer reference point images through multiple sets of imaging units at different locations; enhancing the wafer reference point images, labeling the reference points, and generating a training dataset, wherein the enhancement processing includes at least one of image translation, rotation, brightness transformation, and image distortion; inputting the training dataset into the initial large model for training to obtain the large model. Acquiring wafer reference point images through multiple sets of imaging units at different locations and performing enhancement processing generates a training set that enables the large model to quickly learn the reference point features under different imaging angles and image states based on fewer samples, enhancing the adaptability of the large model. The training dataset covers various image variation scenarios, giving the large model stronger generalization ability. Even when there are fluctuations or differences in the image, it can still stably output accurate coarse localization results, improving the reliability of coarse localization and providing a reliable foundation for subsequent accurate localization of reference points.

[0022] 9. The detection device and computer program product provided in the embodiments of the present invention have the same beneficial effects as the wafer reference point positioning method described above, and will not be repeated here. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of the wafer reference point positioning method provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a set of pixels conforming to a preset grayscale comparison rule in the wafer reference point positioning method provided in the first embodiment of the present invention; Figure 3 This is a schematic diagram of the candidate point set of the wafer reference point positioning method provided in the first embodiment of the present invention; Figure 4 This is a flowchart of step S3 of the wafer reference point positioning method provided in the first embodiment of the present invention; Figure 5 This is a flowchart of step S2 of the wafer reference point positioning method provided in the first embodiment of the present invention; Figure 6 This is a flowchart of step S4 of the wafer reference point positioning method provided in the first embodiment of the present invention; Figure 7 This is a flowchart of step S41 of the wafer reference point positioning method provided in the first embodiment of the present invention; Figure 8 This is a flowchart of step S1 of the wafer reference point positioning method provided in the first embodiment of the present invention; Figure 9 This is a flowchart of the large model training steps of the wafer reference point positioning method provided in the first embodiment of the present invention; Figure 10 This is a flowchart of step S2 of the wafer reference point positioning method provided in the second embodiment of the present invention; Figure 11 This is a structural block diagram of the detection device provided in the third embodiment of the present invention.

[0025] Explanation of reference numerals in the attached diagram: 1. Coarse positioning module; 2. Pixel point set extraction module; 3. Candidate point set generation module; 4. Fitting module; 5. Positioning module; 100. Detection equipment. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0027] The terms “vertical,” “horizontal,” “left,” “right,” “up,” “down,” “upper left,” “upper right,” “lower left,” “lower right,” “lower left,” “lower right,” and similar expressions used in this article are for illustrative purposes only.

[0028] In the semiconductor field, a mark point is a reference marker (such as a cross, circle, rectangle, or rhombus) with a specific shape, pre-fabricated on the wafer surface or at a specific location. Its core function is to provide a precise spatial coordinate reference for wafer positioning, calibration, inspection, and process steps. The coordinate position of a mark point can be determined by the intersection of the auxiliary graphic line segments corresponding to the mark point, or by feature points of the auxiliary graphic such as the center point. Traditional mark point positioning methods rely on the continuity of the auxiliary graphic line segments corresponding to the mark point. However, this is easily distorted by process differences, imaging noise, and other factors, causing the auxiliary graphic line segments corresponding to the mark point in the image to become twisted, blurred, or broken. This disrupts the overall continuity of the auxiliary graphic line segments corresponding to the mark point, thus affecting the accuracy of the mark point positioning.

[0029] In the field of image detection, the pixel coordinate system refers to a direct coordinate system established with the top-left corner of the image as the origin and using pixels as the unit. - The x-coordinate of a pixel and the vertical axis These are the column number and row number in the image array, respectively. The number of columns and rows represents the number of pixels.

[0030] An image coordinate system is a transitional coordinate system that transforms pixel coordinates into physical dimensions. It is used to eliminate pixel discreteness and describe the physical location of image pixels. The image coordinate system uses the center point of the image plane as its origin. shaft and The axes are parallel to the two perpendicular sides of the image plane, respectively. Represent its coordinate values, define : Physical size of 1 pixel in the direction (unit: mm / pixel); : The physical size of 1 pixel in the direction (unit: mm / pixel); The coordinates of the center point of the image plane in the pixel coordinate system. The conversion formula is: .

[0031] The camera coordinate system uses the optical center of the camera as its origin. shaft and The axes are parallel to the image coordinate system. shaft and The optical axis of the camera is Shaft, using This indicates its coordinate value.

[0032] The world coordinate system is the absolute coordinate system of the objective three-dimensional world, also known as the objective coordinate system. (Using ( () represents its coordinate values. The camera coordinate system can be transformed into the world coordinate system through rotation and translation.

[0033] A wafer coordinate system is a coordinate system established for precisely positioning any point on the wafer surface. Its origin is usually located at the geometric center of the wafer, and it has a perpendicular axis. shaft and Axis. Optionally, the origin of the wafer coordinate system is located at the center of the wafer. The axis is parallel to the horizontal direction of the wafer. Axis perpendicular to The axes form a right-handed coordinate system, with the wafer notch or flat edge as the 0° reference, used to identify wafer orientation and rotation alignment.

[0034] Please combine Figure 1 and Figure 2 The first embodiment of the present invention provides a wafer reference point positioning method, comprising the following steps: Step S1: Obtain a wafer image. Based on the wafer image, generate a reference point region image according to a preset method. The reference point region image includes reference points. Step S2: Obtain the set of pixels in the reference point region image that conform to the preset grayscale comparison rules; Step S3: Determine the vertical candidate point set and the horizontal candidate point set based on the pixel point set; Step S4: Fit the vertical candidate point set and the horizontal candidate point set respectively to obtain the vertical line and the horizontal line; Step S5: Calculate the intersection of the vertical and horizontal lines as the final position of the reference point.

[0035] Understandably, the above steps solve the problem of inaccurate reference point positioning caused by image blurring and distortion due to positional differences among multiple imaging units, significantly improving the accuracy and reliability of reference point positioning. Generating a reference point region image containing the reference point based on the wafer image allows for the approximate location of the reference point. The reference point falling within the reference point region image provides a focused area for subsequent, more precise positioning, improving processing efficiency. By acquiring a set of pixels in the reference point region image that conforms to preset grayscale comparison rules, key pixels of the auxiliary graphic line segments corresponding to the reference point can be captured in blurred images, unaffected by line segment distortion or blurring caused by inconsistencies in TDI imaging. By confirming vertical and horizontal candidate point sets based on the pixel set and fitting straight lines respectively, the influence of morphological distortion of the auxiliary graphic line segments corresponding to the reference point is offset. Using the intersection of the vertical and horizontal lines as the reference point position, the final precise coordinates of the reference point are obtained through geometric calculation. Since point fitting does not depend on the overall shape of the auxiliary graphic corresponding to the reference point or the continuity of line segments within the auxiliary graphic, but rather on the precise analysis of the pixel set to determine the reference point's location. Even if the auxiliary graphic line segments corresponding to the reference point are distorted or blurred, point fitting based on the obtained vertical and horizontal candidate point sets can still yield accurate and reliable positioning results, unaffected by the distortion or blurring of the auxiliary graphic line segments corresponding to the reference point, effectively improving the accuracy and reliability of reference point positioning.

[0036] Optionally, the auxiliary graphic corresponding to the reference point can be a cross line, a ring, a frame, a rhombus, a circle, a rectangle, etc. When the auxiliary graphic corresponding to the reference point is a cross line, the cross line can include two line segments, one of which forms a 45° angle with the horizontal axis of the wafer coordinate system, and the other line segment forms a 135° angle with the horizontal axis of the wafer coordinate system. The coordinate position of the reference point is defined by the intersection of the line segments forming 45° and 135° angles with the horizontal axis of the wafer coordinate system. Due to the diagonal structure of 45° and 135°, a strict symmetry about the vertical and horizontal axes is naturally formed, which facilitates the midpoint calculation operation for points with the same horizontal or vertical coordinates. In scenarios where the image contrast, brightness, or sharpness fluctuates due to the angle difference between different TDI imaging units, the cross line segment feature based on a fixed angle can still be stably identified, enhancing the adaptability of the wafer reference point positioning method to imaging differences.

[0037] Optionally, a reference point region image can be generated based on the wafer image using a preset method. This reference point region image can be obtained using methods such as template matching, large model recognition and positioning, or Hough transform. Other methods are not limited here.

[0038] Please combine Figures 2 to 4 Furthermore, step S3 includes the following steps: Step S31: Establish a reference coordinate system based on the reference point area image. The reference coordinate system can be one of the pixel coordinate system, image coordinate system, camera coordinate system, or world coordinate system. Step S32: Based on the reference coordinate system, calculate the midpoint of pairs of points with the same ordinate in the pixel set to obtain a set of vertical candidate points; Step S33: Based on the reference coordinate system, calculate the midpoint of pairs of points with the same horizontal coordinate in the pixel point set to obtain a set of horizontal candidate points.

[0039] Understandably, establishing a reference coordinate system based on the reference point region image aims to ensure that each pixel in the reference point region image has a corresponding coordinate value. This can be achieved regardless of whether a pixel coordinate system, or an image coordinate system, camera coordinate system, or world coordinate system transformed from pixel coordinates, is used. Midpoints are calculated for pairs of points with the same ordinate and pairs of points with the same abscissa. The geometric symmetry of midpoint calculation reduces the impact of positional deviations caused by image blur or noise on individual pixels, making the vertical and horizontal candidate point sets closer to the true distribution of the auxiliary graphic line segments corresponding to the reference points, further improving the fitting accuracy of vertical and horizontal lines.

[0040] It should be noted that the order of steps S32 and S33 can be interchanged.

[0041] For ease of understanding, we will use the auxiliary graphics corresponding to the reference points as 45° and 135° line segments, and employ the pixel coordinate system as the reference coordinate system for illustration. Please refer to... Figure 2 and Figure 3 In the reference point region image, the horizontal axis of the pixel coordinate system is u, and the vertical axis is v. Figure 2 In this context, each pixel in the pixel set is represented by a hollow circle. For example, as shown... Figure 2 As shown, one point is represented by a hollow circle R1. Figure 3 In the diagram, each point in the vertical candidate point set and the horizontal candidate point set is represented by a solid circle. For example, as shown... Figure 3 As shown, one point is represented by a solid circle C1. Step S3 first establishes a pixel coordinate system based on the reference point region image. In the pixel coordinate system, each pixel has a definite coordinate value for subsequent calculation. Based on the pixel coordinate system, the midpoints of pairs of points with the same vertical coordinate and the same horizontal coordinate are calculated respectively. The midpoints of the pairs of points with the same vertical coordinate are taken as vertical candidate points, and the midpoints of the pairs of points with the same horizontal coordinate are taken as horizontal candidate points, finally obtaining the vertical candidate point set and the horizontal candidate point set.

[0042] Please see Figure 5 Furthermore, step S2 includes: Step S21: Traverse each pixel in the reference point region image row by row or column by column. During the traversal of each pixel, for the currently traversed pixel, identify the pixel in its preset neighborhood that conforms to the preset grayscale comparison rule as the target point. Step S22: Based on each target point, construct a set of pixels that conforms to the preset grayscale comparison rules.

[0043] Understandably, the line-by-line traversal method is highly versatile and applicable to reference points of different auxiliary graphics. It can fully cover the reference point area image and avoid missing potentially matching pixels that conform to the preset grayscale comparison rules.

[0044] Optionally, the reference point region image can be divided into different regions, and traversal can be performed separately in each different region to improve traversal speed. For example, based on the horizontal and vertical axes of the reference coordinate system, the reference point region image can be divided into four regions, which correspond to the first quadrant, second quadrant, third quadrant, and fourth quadrant of the reference coordinate system, respectively.

[0045] Furthermore, the preset grayscale comparison rules include: obtaining the grayscale values ​​of each pixel within a preset neighborhood of the pixel; comparing the magnitude of the grayscale values ​​of each pixel and selecting the point with the maximum grayscale value as the target point; or, obtaining the grayscale values ​​of each pixel within a preset neighborhood of the pixel; calculating the gradient magnitude of each pixel and selecting the point with the maximum gradient magnitude as the target point.

[0046] Understandably, a grayscale maxima is the pixel with the highest grayscale value within a preset neighborhood. This highlights the grayscale difference between the auxiliary graphic line segment corresponding to the reference point and the background. When using a grayscale maxima as the target point, it can accurately locate the core pixel corresponding to the auxiliary graphic line segment. Gradient magnitude reflects the drastic change in pixel grayscale. The point with the maximum gradient magnitude usually corresponds to the edge of the line segment. When using the point with the maximum gradient magnitude as the target point, it can still effectively capture the edge features even when the auxiliary graphic line segment corresponding to the reference point is blurred or distorted, overcoming the problem of unclear line segment boundaries caused by differences in image clarity in the reference point area. Both grayscale comparison rules can accurately extract the key pixels of the auxiliary graphic line segment corresponding to the reference point from the reference point area image, ensuring that the obtained pixel set can truly reflect the distribution of the auxiliary graphic line segment corresponding to the reference point, providing a reliable data foundation for subsequent steps.

[0047] Optionally, the preset neighborhood can be in the form of 3×3 (3 rows and 3 columns, a total of 9 pixels) or 5×5 (5 rows and 5 columns, a total of 25 pixels), and other specifications are not limited.

[0048] Optionally, the gradient magnitude is calculated based on the partial derivatives of the pixel's grayscale values ​​along the horizontal and vertical axes of the reference coordinate system, as shown in the formula: .in, It is the rate of change of grayscale along the horizontal axis. It is the rate of change of grayscale along the vertical axis. It is the gradient magnitude of that pixel.

[0049] Please see Figure 6 Furthermore, step S4 includes the following steps: Step S41: Remove outliers from the vertical candidate point set and the horizontal candidate point set to obtain the vertical line noise reduction point set and the horizontal line noise reduction point set. Step S42: Using the least squares method, RANSAC algorithm, or Hough transform, fit the set of noise reduction points for the vertical line to obtain the vertical line; fit the set of noise reduction points for the horizontal line to obtain the horizontal line.

[0050] Understandably, by first removing outliers and then using algorithms such as least squares, RANSAC, and Hough transform to fit line segments, the problem of candidate point position deviation caused by noise and pixel fluctuations in the image can be specifically addressed. The point fitting algorithm, combined with the noise reduction processing of outliers in the early stage, can further improve the accuracy of the horizontal and vertical lines generated by the point fitting, avoid interference from outliers on the fitting results, and ensure that the position coordinates of the reference point are accurately located using the vertical and horizontal lines.

[0051] Please see Figure 7 Furthermore, step S41 includes the following steps: Step S411: Fit a first direction line based on the set of vertical candidate points, and determine whether the distance from each point in the set of vertical candidate points to the first direction line exceeds the threshold. If so, remove the point and use the remaining points as the set of vertical line noise reduction points. Step S412: Fit a second direction line based on the set of horizontal candidate points, and determine whether the distance from each point in the set of horizontal candidate points to the second direction line exceeds the threshold. If so, remove the point and use the remaining points as the set of noise reduction points for the horizontal line.

[0052] It should be noted that the order of steps S411 and S412 can be interchanged.

[0053] Understandably, since the first direction line is fitted based on the set of vertical candidate points, the points in the vertical candidate point set should theoretically not deviate too far from the first direction line. Similarly, since the second direction line is fitted based on the set of horizontal candidate points, the points in the horizontal candidate point set should theoretically not deviate too far from the second direction line. If the distance from a point in the vertical candidate point set to the first direction line exceeds a threshold, or the distance from a point in the horizontal candidate point set to the second direction line exceeds a threshold, then that point is considered an outlier. This allows for the rapid filtering of outliers, purifying the point set, reducing computational complexity, and improving the accuracy of benchmark point positioning. Compared to complex denoising algorithms, this method is simpler, improving processing efficiency while ensuring denoising effectiveness, and quickly obtaining reliable sets of denoised vertical and horizontal points as the basis for fitting the vertical and horizontal lines.

[0054] Please see Figure 8 Furthermore, step S1 includes the following steps: Step S11: Obtain the wafer image and input the wafer image into the pre-trained large model to obtain coarse localization information; Step S12: Obtain the region of interest of the reference point based on the coarse positioning information, and crop the region of interest to obtain the reference point region image.

[0055] Understandably, the reference point region image obtained by the above operation focuses on the local area where the reference point is located, eliminating irrelevant information interference from other non-reference point regions in the wafer image, reducing the complexity and computational load of subsequent processing, and improving positioning speed and accuracy.

[0056] It's important to note that the Region of Interest (ROI) refers to the area in an image that needs to be processed, delineated using shapes such as rectangles, circles, ellipses, or irregular polygons, in machine vision and image processing. Various operators and functions are commonly used in machine vision software such as Halcon, OpenCV, and Matlab to determine the ROI and proceed with further image processing.

[0057] Please see Figure 9 Furthermore, the training of large models includes the following steps: Step S01: Acquire wafer reference point images using multiple imaging units at different locations; Step S02: Enhance the wafer reference point image, label the reference points, and generate a training dataset. The enhancement process includes at least one of image translation, rotation, brightness transformation, and image distortion. Step S03: Input the training dataset into the initial large model for training to obtain the large model.

[0058] Understandably, by acquiring wafer reference point images from multiple imaging units at different locations and performing enhancement processing, the resulting training set enables the large model to quickly learn the reference point features under different imaging angles and image states with fewer samples, thus enhancing the model's adaptability. The training dataset covers various image variation scenarios, giving the large model stronger generalization ability. Even when images fluctuate or differ, it can still stably output accurate coarse localization results, improving the reliability of coarse localization and providing a reliable foundation for subsequent precise localization of reference points.

[0059] Please see Figure 10 The second embodiment of the present invention provides a wafer reference point positioning method, which differs from the wafer reference point positioning method of the first embodiment of the present invention in that: step S2 includes: Step S21a: Based on the auxiliary graphic information corresponding to the reference point, determine the geometric orientation of the line segments in the reference point region image that correspond to the auxiliary graphic. Step S22a: Along the geometric direction, traverse each pixel in the preset neighborhood of the reference line corresponding to the geometric direction. During the traversal of each pixel, for the currently traversed pixel, confirm the pixel in its preset neighborhood that conforms to the preset grayscale comparison rule as the target point. Step S23a: Construct a set of pixels that conform to the preset grayscale comparison rules based on each target point.

[0060] Understandably, traversing along the geometric direction corresponding to the line segments of the auxiliary graphic reduces interference from irrelevant pixels, improving filtering efficiency and the targeting of the pixel set. Determining the geometric direction corresponding to the auxiliary graphic in the reference point region image based on the auxiliary graphic information corresponding to the reference point means determining the direction along which its line segments are distributed based on the known shape of the auxiliary graphic (e.g., a 45° and 135° intersection line). The reference line can be understood as a virtual main line along the geometric direction. Traversing the reference line along this direction, each pixel within a preset neighborhood is traversed. This preset neighborhood can be a specific area around the reference line, or a specific area corresponding to the reference line after dividing the reference point region image into different regions. During the traversal of each pixel, for the currently traversed pixel, pixels within its preset neighborhood that conform to preset grayscale comparison rules are identified as target points, ultimately accurately extracting the pixel set that conforms to the preset grayscale comparison rules.

[0061] Please see Figure 11 The third embodiment of the present invention provides a testing device 100 for executing the wafer reference point positioning method of the first embodiment of the present invention. The testing device 100 includes: Coarse positioning module 1: Used to acquire wafer images and generate a reference point region image based on the wafer images according to a preset method. The reference point region image includes reference points. Pixel set extraction module 2: Used to obtain the set of pixels in the reference point region image that conforms to the preset grayscale comparison rules; Candidate point set generation module 3: used to determine the vertical candidate point set and the horizontal candidate point set based on the pixel point set; Fitting module 4 is used to fit based on the vertical candidate point set and the horizontal candidate point set respectively to obtain vertical lines and horizontal lines; Positioning module 5: Used to calculate the final position of the intersection of the vertical and horizontal lines as the reference point.

[0062] Understandably, the detection device 100 has the same beneficial effects as the wafer reference point positioning method of the first or second embodiment of the present invention, and will not be described again here.

[0063] The fourth embodiment of the present invention provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the wafer reference point positioning method of the first embodiment of the present invention.

[0064] Understandably, the computer program product has the same beneficial effects as the wafer reference point positioning method of the first or second embodiment of the present invention, and will not be described again here.

[0065] In the embodiments provided by this invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0066] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Those skilled in the art should also recognize that the embodiments described in the specification are optional embodiments, and the actions and modules involved are not necessarily essential to the invention.

[0067] In various embodiments of the present invention, it should be understood that the sequence number of each process does not necessarily imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0068] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It is particularly important to note that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0069] Compared with the prior art, the wafer reference point positioning method, detection equipment, and computer program product provided by the present invention have the following beneficial effects: 1. This invention provides a wafer reference point positioning method. The method involves acquiring a reference point region image including the reference point, obtaining a set of pixels conforming to a preset grayscale comparison rule from the reference point region image, identifying a vertical candidate point set and a horizontal candidate point set based on this pixel set, fitting vertical and horizontal lines based on the vertical and horizontal candidate point sets respectively, and finally calculating the intersection of the vertical and horizontal lines as the reference point position. This solves the problem of inaccurate reference point positioning caused by image blurring and distortion due to positional differences among multiple imaging units, significantly improving the accuracy and reliability of reference point positioning. Generating a reference point region image containing the reference point based on the wafer image allows for obtaining the approximate position of the reference point. The reference point falling within the reference point region image provides a focused area for subsequent more precise positioning, improving processing efficiency. By acquiring a set of pixels conforming to a preset grayscale comparison rule in the reference point region image, key pixels of the auxiliary graphic line segments corresponding to the reference point can be captured in blurred images, unaffected by line segment distortion or blurring caused by inconsistencies in TDI imaging. By identifying vertical and horizontal candidate point sets based on pixel sets and fitting straight lines to each set, the influence of distortion in the auxiliary graphic line segments corresponding to the reference point is offset. Using the intersection of the vertical and horizontal lines as the reference point position, the final precise coordinates of the reference point are obtained through geometric calculation. Since point fitting does not depend on the overall shape of the auxiliary graphic corresponding to the reference point or the continuity of line segments within the auxiliary graphic, but rather on the precise analysis of the pixel set to determine the reference point's position, even if the auxiliary graphic line segments corresponding to the reference point are distorted or blurred, point fitting based on the obtained vertical and horizontal candidate point sets can still yield accurate and reliable positioning results, unaffected by the distortion or blurring of the auxiliary graphic line segments corresponding to the reference point, effectively improving the accuracy and reliability of reference point positioning.

[0070] 2. The method for determining the vertical and horizontal candidate point sets based on a set of pixels provided in this embodiment includes the following steps: establishing a reference coordinate system based on a reference point region image, wherein the reference coordinate system is one of a pixel coordinate system, an image coordinate system, a camera coordinate system, or a world coordinate system; calculating the midpoint of paired points with the same vertical coordinate in the pixel set based on the reference coordinate system to obtain the vertical candidate point set; and calculating the midpoint of paired points with the same horizontal coordinate in the pixel set based on the reference coordinate system to obtain the horizontal candidate point set. By establishing a reference coordinate system based on a reference point region image, each pixel in the reference point region image has a corresponding coordinate value. This can be achieved regardless of whether a pixel coordinate system, or an image coordinate system, camera coordinate system, or world coordinate system obtained by pixel coordinate system transformation, is used. Calculating the midpoint of paired points with the same vertical coordinate and paired points with the same horizontal coordinate respectively utilizes the geometric symmetry of midpoint calculation to reduce the positional deviation of individual pixels caused by image blurring or noise, making the vertical and horizontal candidate point sets closer to the true distribution of the auxiliary graphic line segments corresponding to the reference point, further improving the fitting accuracy of vertical and horizontal lines.

[0071] 3. The method of obtaining a set of pixels conforming to a preset grayscale comparison rule in a reference point region image provided in this embodiment of the invention includes: traversing each pixel in the reference point region image row by row or column by column, and for each pixel, identifying pixels within its preset neighborhood that conform to the preset grayscale comparison rule as target points and constructing a pixel set. The row-by-row traversal method is highly versatile and applicable to reference points of different auxiliary graphics, comprehensively covering the reference point region image and avoiding omission of potentially conforming pixels. Based on the auxiliary graphic information corresponding to the reference point, the geometric direction corresponding to the line segment of the auxiliary graphic in the reference point region image is determined; along the geometric direction, each pixel within the preset neighborhood of the reference line corresponding to the geometric direction is traversed, and for each pixel, identifying pixels within its preset neighborhood that conform to the preset grayscale comparison rule as target points and constructing a pixel set. The method of traversing along the geometric direction corresponding to the line segment of the auxiliary graphic reduces interference from irrelevant pixels, improves screening efficiency, and enhances the targeting of the pixel set. Both of the above methods can improve the accuracy of the pixel set conforming to the preset grayscale comparison rule.

[0072] 4. The preset grayscale comparison rules provided in this embodiment of the invention include: obtaining the grayscale values ​​of each pixel within a preset neighborhood of the pixel; comparing the magnitude of the grayscale values ​​of each pixel and selecting the grayscale maximum point as the target point; or, obtaining the grayscale values ​​of each pixel within a preset neighborhood of the pixel; calculating the gradient magnitude of each pixel and selecting the point with the largest gradient magnitude as the target point. The grayscale maximum point is the pixel with the highest grayscale value within the preset neighborhood, which can highlight the grayscale difference between the auxiliary graphic line segment corresponding to the reference point and the background. When the grayscale maximum point is used as the target point, the core pixel corresponding to the auxiliary graphic line segment can be accurately located. The gradient magnitude reflects the degree of drastic change in pixel grayscale. The maximum gradient magnitude point usually corresponds to the edge of the line segment. When the maximum gradient magnitude point is used as the target point, even when the auxiliary graphic line segment corresponding to the reference point is blurred or distorted, its edge features can still be effectively captured, overcoming the problem of unclear line segment boundaries caused by the difference in image clarity in the reference point area. Both grayscale comparison rules can accurately extract key pixels of the auxiliary graphic line segments corresponding to the reference point from the reference point region image, ensuring that the obtained pixel set can truly reflect the distribution of the auxiliary graphic line segments corresponding to the reference point, and providing a reliable data foundation for subsequent steps.

[0073] 5. The method provided in this embodiment of the invention, which involves fitting vertical and horizontal candidate point sets to obtain vertical and horizontal lines respectively, includes the following steps: removing outliers from the vertical and horizontal candidate point sets to obtain a set of denoised vertical and horizontal lines; fitting the set of denoised vertical lines using the least squares method, RANSAC algorithm, or Hough transform to obtain a vertical line; and fitting the set of denoised horizontal lines to obtain a horizontal line. By first removing outliers and then using algorithms such as least squares, RANSAC, and Hough transform to fit line segments, the problem of candidate point position deviation caused by noise and pixel fluctuations in the image is specifically addressed. The point fitting algorithm, combined with the previous outlier denoising processing, can further improve the accuracy of the horizontal and vertical lines generated by point fitting, avoid interference from outliers on the fitting results, and ensure accurate positioning of the reference point's coordinates using the vertical and horizontal lines.

[0074] 6. In this embodiment of the invention, a first direction line is fitted based on a set of vertical candidate points. Points in the set of vertical candidate points whose distance to the first direction line exceeds a threshold are removed, and the remaining points are used as the set of denoised points for the vertical line. Similarly, a second direction line is fitted based on a set of horizontal candidate points. Points in the set of horizontal candidate points whose distance to the second direction line exceeds a threshold are removed, and the remaining points are used as the set of denoised points for the horizontal line. Since the first direction line is fitted based on the set of vertical candidate points, theoretically, the points in the set of vertical candidate points will not deviate too far from the first direction line. Similarly, since the second direction line is fitted based on the set of horizontal candidate points, theoretically, the points in the set of horizontal candidate points will not deviate too far from the second direction line. If the distance from a point in the set of vertical candidate points to the first direction line exceeds a threshold, or the distance from a point in the set of horizontal candidate points to the second direction line exceeds a threshold, it means that the point is an anomaly. This allows for the rapid screening of anomalies, purification of the point set, reduction of computational complexity, and improvement of the accuracy of the reference point positioning. Compared to complex denoising algorithms, this method is simpler, improving processing efficiency while ensuring denoising effect, and quickly obtaining reliable sets of denoised points for both vertical and horizontal lines as the basic data for fitting.

[0075] 7. The wafer image acquisition method provided in this embodiment of the invention generates a reference point region image based on the wafer image according to a preset method. The reference point region image includes reference points and includes the following steps: acquiring the wafer image; inputting the wafer image into a pre-trained large model to obtain coarse localization information; obtaining the region of interest (ROI) of the reference points based on the coarse localization information; and cropping the ROI to obtain the reference point region image. The reference point region image obtained by the above operation focuses on the local area where the reference points are located, eliminating irrelevant information interference from other non-reference point areas in the wafer image, reducing the complexity and computational load of subsequent processing, and improving the localization speed and accuracy.

[0076] 8. The training of the large model provided in this embodiment of the invention includes the following steps: acquiring wafer reference point images through multiple sets of imaging units at different locations; enhancing the wafer reference point images, labeling the reference points, and generating a training dataset, wherein the enhancement processing includes at least one of image translation, rotation, brightness transformation, and image distortion; inputting the training dataset into the initial large model for training to obtain the large model. Acquiring wafer reference point images through multiple sets of imaging units at different locations and performing enhancement processing generates a training set that enables the large model to quickly learn the reference point features under different imaging angles and image states based on fewer samples, enhancing the adaptability of the large model. The training dataset covers various image variation scenarios, giving the large model stronger generalization ability. Even when there are fluctuations or differences in the image, it can still stably output accurate coarse localization results, improving the reliability of coarse localization and providing a reliable foundation for subsequent accurate localization of reference points.

[0077] 9. The detection device and computer program product provided in the embodiments of the present invention have the same beneficial effects as the wafer reference point positioning method described above, and will not be repeated here.

[0078] The foregoing has provided a detailed description of a wafer reference point positioning method, testing equipment, and computer program product disclosed in the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention. Any modifications, equivalent substitutions, and improvements made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for locating wafer reference points, characterized in that: Includes the following steps: Acquire a wafer image, and generate a reference point region image based on the wafer image according to a preset method, wherein the reference point region image includes the reference point; Obtain the set of pixels in the reference point region image that conform to the preset grayscale comparison rules; Based on the set of pixels, a set of vertical candidate points and a set of horizontal candidate points are determined; The vertical line and the horizontal line are obtained by fitting the vertical candidate point set and the horizontal candidate point set respectively. The intersection of the vertical line and the horizontal line is calculated as the final position of the reference point; The step of determining the vertical candidate point set and the horizontal candidate point set based on the pixel set includes the following steps: A reference coordinate system is established based on the reference point area image. The reference coordinate system is one of the pixel coordinate system, image coordinate system, camera coordinate system, and world coordinate system. Based on the reference coordinate system, the midpoint is calculated for pairs of points with the same ordinate in the pixel set to obtain the set of vertical candidate points; Based on the reference coordinate system, the midpoint is calculated for pairs of points with the same horizontal coordinate in the pixel set to obtain the horizontal candidate point set.

2. The wafer reference point positioning method as described in claim 1, characterized in that: The set of pixels in the reference point region image that conforms to the preset grayscale comparison rules includes: The pixel in the reference point region image is traversed row by row or column by column. During the traversal of each pixel, for the currently traversed pixel, the pixel in its preset neighborhood that conforms to the preset grayscale comparison rule is identified as the target point. A set of pixels conforming to a preset grayscale comparison rule is formed based on each of the target points; or, Based on the auxiliary graphic information corresponding to the reference point, the geometric orientation of the line segment in the reference point region image corresponding to the auxiliary graphic is determined; Along the geometric direction, traverse each pixel in the preset neighborhood of the reference line corresponding to the geometric direction. During the traversal of each pixel, for the currently traversed pixel, identify the pixel in its preset neighborhood that conforms to the preset grayscale comparison rule as the target point. Each target point constitutes a set of pixels that conforms to a preset grayscale comparison rule.

3. The wafer reference point positioning method as described in claim 1, characterized in that: The preset grayscale comparison rule includes: obtaining the grayscale values ​​of each pixel within a preset neighborhood of the pixel; comparing the magnitudes of the grayscale values ​​of each pixel, and selecting the point with the maximum grayscale value as the target point; Alternatively, obtain the grayscale values ​​of each pixel within a preset neighborhood of the pixel; calculate the gradient magnitude of each pixel, and select the point with the largest gradient magnitude as the target point.

4. The wafer reference point positioning method as described in claim 1, characterized in that: The process of fitting the vertical and horizontal lines based on the vertical candidate point set and the horizontal candidate point set, respectively, includes the following steps: By removing outliers from the vertical candidate point set and the horizontal candidate point set, we obtain the vertical line noise reduction point set and the horizontal line noise reduction point set. The set of denoised points for the vertical line is fitted using the least squares method, RANSAC algorithm, or Hough transform to obtain the vertical line; the set of denoised points for the horizontal line is fitted to obtain the horizontal line.

5. The wafer reference point positioning method as described in claim 4, characterized in that: The process of removing outliers from the vertical candidate point set and the horizontal candidate point set to obtain the vertical line noise reduction point set and the horizontal line noise reduction point set includes the following steps: Based on the set of vertical candidate points, fit a first direction line, determine whether the distance from each point in the set of vertical candidate points to the first direction line exceeds a threshold. If so, remove the point and use the remaining points as the set of vertical line noise reduction points. Based on the set of horizontal candidate points, fit a second direction line, determine whether the distance from each point in the set of horizontal candidate points to the second direction line exceeds a threshold. If so, remove the point and use the remaining points as the set of noise reduction points for the horizontal line.

6. The wafer reference point positioning method as described in claim 1, characterized in that: The process of acquiring a wafer image and generating a reference point region image based on the wafer image according to a preset method, wherein the reference point region image includes the reference points, includes the following steps: The wafer image is acquired and input into a pre-trained large model to obtain coarse localization information; Based on the coarse positioning information, the region of interest of the reference point is obtained, and the region of interest is cropped to obtain the image of the reference point region.

7. The wafer reference point positioning method as described in claim 6, characterized in that: The training of the large model includes the following steps: Images of wafer reference points are acquired using multiple imaging units at different locations; The wafer reference point image is enhanced, the reference points are labeled, and a training dataset is generated. The enhancement process includes at least one of image translation, rotation, brightness transformation, and image distortion. The training dataset is input into the initial large model for training to obtain the large model.

8. A testing device for performing the wafer reference point positioning method as described in any one of claims 1 to 7, characterized in that: Includes a coarse positioning module: used to acquire a wafer image, and generate a reference point region image based on the wafer image according to a preset method, wherein the reference point region image includes the reference point; Pixel set extraction module: used to obtain the set of pixels in the reference point region image that conforms to the preset grayscale comparison rules; Candidate point set generation module: used to determine a vertical candidate point set and a horizontal candidate point set based on the pixel set; wherein, determining the vertical and horizontal candidate point sets based on the pixel set includes establishing a reference coordinate system based on the reference point region image, wherein the reference coordinate system is one of pixel coordinate system, image coordinate system, camera coordinate system, and world coordinate system; calculating the midpoint of pairs of points with the same ordinate in the pixel set based on the reference coordinate system to obtain the vertical candidate point set; and calculating the midpoint of pairs of points with the same abscissa in the pixel set based on the reference coordinate system to obtain the horizontal candidate point set. The fitting module is used to perform fitting based on the set of vertical candidate points and the set of horizontal candidate points respectively, to obtain vertical lines and horizontal lines; Positioning module: used to calculate the intersection of the vertical line and the horizontal line as the final position of the reference point.

9. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the wafer reference point positioning method as described in any one of claims 1 to 7.

Images