Optical waveguide alignment method based on glass substrate

Abstract

The invention relates to the technical field of photoelectric precision alignment, in particular to an optical waveguide alignment method based on a glass substrate, which comprises the following steps: acquiring a global optical image of a target glass substrate through a high-resolution area array image sensor, identifying and extracting a waveguide structure pattern to generate contour point cloud, and analyzing geometric characteristics to obtain a waveguide theoretical central axis coordinate expression; a waveguide port high-contrast local image is separated, a port edge sub-pixel coordinate set is extracted through edge sharpening and sub-pixel positioning, the port edge sub-pixel coordinate set is matched with a theoretical central axis space to solve a position deviation vector, a bearing table calibration motion parameter is combined to generate a driving control quantity, and the bearing table plane displacement is driven to complete alignment correction. According to the method, a reference is constructed based on a global structure, accurate deviation calculation is achieved through sub-pixel positioning and space matching, and glass substrate optical waveguide fine alignment is completed.

Description

Technical Field

[0001] This invention relates to the field of optoelectronic precision alignment technology, and in particular to an alignment method based on a glass substrate optical waveguide. Background Technology

[0002] Existing glass substrate optical waveguide alignment technologies mostly employ local image acquisition methods, relying on local features of the waveguide port to complete the benchmark positioning. The port position information is obtained through conventional pixel-level edge extraction methods. After simple alignment matching between the port position and the preset benchmark, the alignment operation is performed by a single displacement of the support stage.

[0003] Local image acquisition methods cannot obtain information about the overall waveguide structure. Reliance positioning based solely on local features is prone to reference offset issues. Conventional edge extraction can only achieve pixel-level positioning accuracy. The accuracy of waveguide port edge coordinate extraction is limited. The spatial matching accuracy between the local port position and the overall waveguide axis is insufficient. The position deviation vector calculation results contain errors. The matching degree between the drive control quantity of the platform motion compensation and the actual deviation is low.

[0004] It is necessary to construct a theoretical central axis reference based on the global structural contour of the waveguide, obtain the sub-pixel precision edge coordinates of the waveguide port and complete the precise spatial matching with the theoretical central axis, and generate the appropriate carrier platform planar compensation motion drive control quantity based on the precise position deviation vector. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and propose a glass substrate optical waveguide alignment method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a glass substrate optical waveguide alignment method, comprising:

[0007] A global optical image of the target glass substrate on the support platform is acquired by a high-resolution area array image sensor, and the global optical image includes a waveguide structure pattern.

[0008] Waveguide structure pattern region identification and boundary extraction are performed on the global optical image to generate waveguide structure contour point cloud;

[0009] Based on the waveguide structure contour point cloud, the coordinate expression of the theoretical central axis of the waveguide structure in the sensor coordinate system is calculated through geometric feature analysis.

[0010] From the imaging results of the high-resolution area array image sensor, a high-contrast local image of the target waveguide port under specific illumination conditions is extracted;

[0011] Edge sharpening and subpixel localization processing are performed on the high-contrast local image to extract the subpixel precision coordinate set of the actual edge of the target waveguide port;

[0012] The sub-pixel precision coordinate set of the actual edge of the target waveguide port is spatially matched with the theoretical central axis of the waveguide structure to solve for the positional deviation vector between the actual port center and the theoretical axis.

[0013] Based on the position deviation vector and combined with the pre-calibrated motion parameters of the bearing platform, a drive control quantity for compensating motion in the plane of the bearing platform is generated;

[0014] The drive control quantity is sent to the motion controller of the support platform, which drives the support platform to perform planar displacement and completes one alignment correction.

[0015] As a further aspect of the present invention, waveguide structure pattern region identification and boundary extraction are performed on the global optical image to generate a waveguide structure contour point cloud, including:

[0016] The global optical image is processed by an adaptive threshold segmentation algorithm to separate the waveguide structure pattern region from the glass substrate background region, thereby obtaining a binary image of the waveguide structure.

[0017] A morphological opening operation is performed on the binarized image of the waveguide structure to remove noise and smooth the edges of the waveguide structure.

[0018] An edge tracking algorithm is performed on the smoothed binary image of the waveguide structure to obtain the pixel-level coordinate sequence of the waveguide structure edge;

[0019] The pixel-level coordinate sequence is smoothed and resampled using cubic spline interpolation to obtain a continuous discrete point set of edge curves, forming the waveguide structure contour point cloud.

[0020] As a further aspect of the present invention, based on the waveguide structure contour point cloud, the coordinate expression of the theoretical central axis of the waveguide structure in the sensor coordinate system is calculated through geometric feature analysis, including:

[0021] Select a subset of contour points located on the long sides of the waveguide structure from the contour point cloud of the waveguide structure.

[0022] Least square line fitting is performed on the subsets of contour points on the two long sides respectively to obtain two line equations representing the two edges of the waveguide.

[0023] Based on the equations of two straight lines representing the two edges of the waveguide, the midline of the two lines is calculated, which is the theoretical central axis of the waveguide structure.

[0024] The mathematical expression of the theoretical central axis in the sensor coordinate system is recorded in the form of point-slope form or parametric equation to obtain the coordinate expression.

[0025] As a further aspect of the present invention, edge sharpening and sub-pixel localization processing are performed on the high-contrast local image to extract a set of sub-pixel precision coordinates of the actual edge of the target waveguide port, including:

[0026] The Sobel operator is applied to the high-contrast local image to calculate gradients, thereby enhancing the gradient response at the waveguide port edges.

[0027] On the gradient image, a grayscale profile is scanned along the normal direction perpendicular to the expected edge direction to obtain the grayscale distribution curve;

[0028] On the grayscale value distribution curve, locate the interval with the steepest grayscale value change, and perform Gaussian fitting on the data points within the interval;

[0029] Using the center point of the Gaussian fitted curve as the subpixel precision position of the edge, all test points on the edge of the waveguide port are traversed to obtain a series of subpixel precision position coordinates, which constitute the subpixel precision coordinate set.

[0030] As a further aspect of the present invention, the sub-pixel precision coordinate set of the actual edge of the target waveguide port is spatially matched with the theoretical central axis of the waveguide structure to solve for the position deviation vector between the actual port center and the theoretical axis, including:

[0031] Based on the sub-pixel precision coordinate set of the actual edge of the target waveguide port, the actual geometric center coordinates of the target waveguide port are calculated by ellipse fitting or least squares circle fitting algorithm.

[0032] Calculate the coordinates of the axis point closest to the actual geometric center from the coordinate expression of the theoretical central axis of the waveguide structure in the sensor coordinate system;

[0033] Calculate the difference between the actual geometric center coordinates and the coordinates of the nearest axis point. The difference is decomposed into a longitudinal deviation component along the axis direction and a lateral deviation component perpendicular to the axis direction in the XY plane of the sensor coordinate system.

[0034] The longitudinal deviation component and the lateral deviation component together constitute the position deviation vector.

[0035] As a further aspect of the present invention, based on the position deviation vector and combined with the pre-calibrated motion parameters of the support platform, a drive control quantity for compensating motion in the plane of the support platform is generated, including:

[0036] Obtain motion parameters of the support platform, including calibration conversion coefficients between image sensor pixel size and actual movement distance of the support platform;

[0037] The lateral and longitudinal deviation components of the position deviation vector are multiplied by the calibration conversion coefficient to convert them into the lateral and longitudinal displacements that the bearing platform needs to compensate for in the actual physical space.

[0038] Based on the current position feedback value of the bearing platform in the motor driver, calculate the number of transverse motor drive pulses and the number of longitudinal motor drive pulses required to move the bearing platform to the target position.

[0039] The number of horizontal motor drive pulses and the number of vertical motor drive pulses are used as the drive control quantities.

[0040] As a further aspect of the present invention, obtaining the motion parameters of the support platform includes:

[0041] A calibration plate with a standard scale is fixed to the support platform;

[0042] The high-resolution area array image sensor is controlled to image the calibration board, and an image of the calibration board is acquired.

[0043] Identify at least two non-collinear feature points in the calibration plate image and record the pixel coordinates of the feature points in the image sensor pixel coordinate system;

[0044] Based on the actual physical distance between known feature points on the calibration board, the actual physical size corresponding to a single pixel of the image sensor is calculated to obtain the pixel equivalent calibration value.

[0045] The control platform moves a known physical displacement along two mutually orthogonal axes in the plane, and the high-resolution area array image sensor acquires images before and after the movement to identify the coordinate changes of the same feature point in the pixel coordinate system.

[0046] Based on the physical displacement and the corresponding pixel coordinate change, the proportional relationship between the actual physical displacement of the carrier platform under a unit pulse drive and the pixel movement of the image sensor is calculated when the carrier platform moves in each axial direction, thus obtaining the calibration conversion coefficient.

[0047] As a further aspect of the present invention, after the drive control quantity is sent to the motion controller of the support platform, the invention further includes:

[0048] After the support stage completes the planar displacement, the high-resolution area array image sensor again acquires a global optical image of the target glass substrate after alignment in the same area;

[0049] In the aligned global optical image, the steps of generating the waveguide structure contour point cloud, calculating the theoretical central axis of the waveguide structure, extracting the sub-pixel precision coordinate set of the actual edge of the target waveguide port, and solving the position deviation vector are repeated.

[0050] Determine whether the magnitude of the newly solved position deviation vector is less than the preset alignment accuracy threshold.

[0051] As a further aspect of the present invention, the step of determining whether the magnitude of the newly solved position deviation vector is less than a preset alignment accuracy threshold includes:

[0052] Calculate the sum of the squares of the lateral deviation component and the longitudinal deviation component of the newly solved position deviation vector;

[0053] The magnitude of the position deviation vector is obtained by taking the square root of the sum of the squares of the lateral deviation component and the squares of the longitudinal deviation component.

[0054] The magnitude of the position deviation vector is compared with the alignment accuracy threshold pre-stored in the system.

[0055] As a further aspect of the present invention, when the magnitude of the position deviation vector is not less than the alignment accuracy threshold, it further includes:

[0056] Using the newly solved position deviation vector as input, the step of generating the drive control quantity for the compensated motion in the plane of the bearing platform is repeated.

[0057] The newly generated drive control quantity is sent to the motion controller of the platform, driving the platform to perform a new round of planar displacement correction.

[0058] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0059] A high-resolution area array image sensor is used to acquire a global optical image containing a waveguide structure pattern. The waveguide structure pattern region is identified and the boundary is extracted from the global optical image to generate a waveguide structure contour point cloud. Based on the waveguide structure contour point cloud, the coordinate expression of the theoretical central axis of the waveguide structure in the sensor coordinate system is calculated through geometric feature analysis. The generation of the reference parameters relies on the complete structural features of the waveguide, avoiding the reference offset caused by local feature positioning. The coordinate expression of the theoretical central axis can be directly adapted to the sensor coordinate system, and the degree of adaptation between the reference and the coordinate system is improved.

[0060] A high-contrast local image of the target waveguide port is extracted from the imaging results of a high-resolution area array image sensor. Edge sharpening and sub-pixel localization processing are performed on the high-contrast local image to extract the sub-pixel precision coordinate set of the actual edge of the target waveguide port. The sub-pixel precision coordinate set is spatially matched with the theoretical central axis of the waveguide structure to solve for the position deviation vector between the actual port center and the theoretical axis. Based on the position deviation vector and the pre-calibrated motion parameters of the carrier platform, a drive control quantity for in-plane compensation motion of the carrier platform is generated. The drive control quantity is sent to the motion controller to drive the carrier platform to perform planar displacement. The extraction accuracy of the port edge coordinates breaks through the conventional pixel-level limitation. The solution result of the deviation vector closely matches the actual position difference. The drive control quantity can directly form an adaptive compensation motion command. The alignment correction execution process is consistent with the actual deviation state. Attached Figure Description

[0061] Figure 1 This is a flowchart of a glass substrate optical waveguide alignment method according to the present invention;

[0062] Figure 2 A flowchart for generating point clouds of waveguide structure contours;

[0063] Figure 3 A flowchart for calculating the theoretical central axis of a waveguide structure;

[0064] Figure 4 A convergence curve of alignment correction accuracy in multiple rounds;

[0065] Figure 5 A visualization of the waveguide structure outline and its theoretical central axis. Detailed Implementation

[0066] 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.

[0067] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0068] See Figure 1A global optical image of the target glass substrate on a support platform is acquired using a high-resolution area array image sensor. This global optical image contains a waveguide structure pattern. Waveguide structure pattern region identification and boundary extraction are performed on the global optical image to generate a waveguide structure contour point cloud. Based on the waveguide structure contour point cloud, the coordinate expression of the theoretical central axis of the waveguide structure in the sensor coordinate system is calculated through geometric feature analysis. A high-contrast local image of the target waveguide port under specific illumination conditions is separated from the imaging results of the high-contrast area array image sensor. Edge sharpening and sub-pixel localization processing are performed on the high-contrast local image to extract the sub-pixel precision coordinate set of the actual edge of the target waveguide port. The sub-pixel precision coordinate set of the actual edge of the target waveguide port is spatially matched with the theoretical central axis of the waveguide structure to solve for the position deviation vector between the actual port center and the theoretical axis. Based on the position deviation vector and combined with pre-calibrated platform motion parameters, a drive control quantity for in-plane compensation motion of the platform is generated. The drive control quantity is sent to the motion controller of the platform to drive the platform to perform planar displacement, completing one alignment correction.

[0069] See Figure 2 In one embodiment of the present invention, an adaptive threshold segmentation algorithm is used to process the global optical image. The adaptive threshold segmentation algorithm compares the grayscale value of each pixel in the global optical image with a calculated threshold to complete region separation. The adaptive threshold segmentation algorithm automatically determines the segmentation threshold based on the image grayscale histogram distribution characteristics. For example, when the grayscale level of the waveguide structure pattern region in the global optical image differs significantly from the grayscale level of the glass substrate background region, the adaptive threshold segmentation algorithm selects the optimal threshold by maximizing the inter-class variance. One implementation of the adaptive threshold segmentation algorithm uses the Otsu method, and the discrimination criterion formula of the Otsu method is expressed as:

[0070]

[0071] Where: symbol Indicates the candidate threshold, symbol Indicates below the threshold The proportion of pixels, symbol Indicates that it is higher than or equal to the threshold. The proportion of pixels, symbol Indicates below the threshold The average gray value of the pixels, symbol Indicates that it is higher than or equal to the threshold. The average gray value of the pixels, symbol Indicates at the threshold The inter-class variance is calculated. The adaptive threshold segmentation algorithm iterates through all possible thresholds. And calculate the corresponding inter-class variance. Selecting classes with different variances The highest threshold is used as the final segmentation threshold. In some embodiments, the adaptive thresholding algorithm can be dynamically adjusted by incorporating local grayscale characteristics of the image, for example, by calculating the threshold independently for each image block after dividing the global optical image into blocks. Optionally, the adaptive thresholding algorithm can perform Gaussian filtering on the global optical image before processing to suppress noise. It can be understood that the application of the adaptive thresholding algorithm enables the waveguide structure pattern region and the glass substrate background region to be clearly separated in the binary image, resulting in a binary image of the waveguide structure.

[0072] In specific implementation, a morphological opening operation is performed on the binarized image of the waveguide structure. This morphological opening operation is a sequential combination of morphological erosion and morphological dilation operations. The morphological opening operation uses structuring elements of a preset shape and size, typically circular or rectangular. First, the morphological opening operation performs morphological erosion on the binarized waveguide structure image to remove small noise points and shrink the edge regions of the waveguide structure. Then, it performs morphological dilation on the result of the morphological erosion operation to restore the main shape of the waveguide structure. The morphological opening operation can smooth out small irregular protrusions at the edges of the waveguide structure while maintaining the overall topology of the waveguide structure. In some embodiments, the size of the structuring element in the morphological opening operation can be selected based on the typical linewidth of the waveguide structure pattern and the image resolution. Optionally, the morphological opening operation can be iterated multiple times to enhance the smoothing effect. It can be understood that in the binarized image of the waveguide structure processed by the morphological opening operation, the waveguide structure region has better connectivity and smoother edges, creating favorable conditions for subsequent edge extraction.

[0073] In the specific implementation, an edge tracking algorithm is executed on the smoothed binary image of the waveguide structure. The edge tracking algorithm searches for the contour boundaries of the waveguide structure region within the binary image. Starting from the first white pixel encountered in the waveguide structure binary image, the algorithm tracks adjacent white pixels according to a preset neighborhood search order. The algorithm records the row and column coordinates of each boundary pixel, forming a pixel-level coordinate sequence of the waveguide structure edge. The edge tracking algorithm needs to handle the possible inner and outer contours of the waveguide structure. By distinguishing contour levels, the algorithm ensures the acquisition of a complete outer edge coordinate sequence of the waveguide structure. In the specific implementation, cubic spline interpolation is used to smooth and resample the pixel-level coordinate sequence. Cubic spline interpolation fits the discrete pixel-level coordinate sequence into a smooth curve. Cubic spline interpolation defines a cubic polynomial curve segment between every two adjacent pixel coordinate points. The cubic spline interpolation requires the curve to have continuous first and second derivatives at the connection points to ensure overall smoothness. After fitting the pixel-level coordinate sequence using cubic spline interpolation, a set of discrete points is resampled on the curve at fixed arc length intervals or fixed parameter intervals to obtain a continuous set of discrete points on the edge curve. This continuous set of discrete points on the edge curve forms the waveguide structure contour point cloud. Each point in the waveguide structure contour point cloud contains coordinate information with sub-pixel precision, and the waveguide structure contour point cloud accurately represents the geometry and position of the waveguide structure.

[0074] See Figure 3 In one embodiment of the present invention, a subset of contour points located on the two long sides of the waveguide structure is selected from the waveguide structure contour point cloud. The selection process is based on the spatial distribution geometric features of the points in the waveguide structure contour point cloud. The waveguide structure contour point cloud represents the closed boundary of the waveguide structure, and the two long sides of the waveguide structure correspond to the approximately straight and parallel parts of the contour. In some embodiments, the orientation and long side orientation of the waveguide structure can be initially determined by calculating the convex hull or the minimum bounding rectangle of the waveguide structure contour point cloud. Optionally, principal component analysis can be performed on the waveguide structure contour point cloud to obtain its principal axis direction, and contour points whose angle with the principal axis direction is within a certain range can be classified as the long side contour point subset. After selecting the subset of contour points located on the two long sides of the waveguide structure from the waveguide structure contour point cloud, least squares line fitting is performed on the contour point subsets on the two long sides respectively. The goal of least squares line fitting is to find a straight line such that the sum of the squares of the perpendicular distances from the straight line to all points in the contour point subsets on the two long sides is minimized. For a set of contour point subsets Let the equation of the line to be fitted be... Then, least squares linear fitting minimizes the error function. To solve for the parameters and Error function The expression is:

[0075]

[0076] Where: symbol and These represent the first point in the contour point set. The x and y coordinates of a point in the sensor coordinate system, with symbols... This indicates the total number of points where the contour points of the longer side are concentrated, denoted by the symbol. The intercept parameter of the fitted line is represented by the symbol. The slope parameter of the fitted line is represented by the symbol. This represents the sum of squares of the longitudinal deviations of all points from the line. This is achieved through the error function. Regarding the parameters respectively and parameters By taking the partial derivatives and setting them to zero, we can solve for the parameters. and parameters The optimal value is obtained, thus yielding a straight line equation representing one edge of the waveguide. The above least-squares line fitting process is performed on the subset of contour points on both sides of the waveguide structure to obtain two line equations representing the edges of the waveguide.

[0077] In practical implementation, the midline of the two lines is calculated based on the equations of the two lines representing the edges of the waveguide. Let the equation of the first edge line be... The equation of the second edge line is Because the waveguide structure is designed with parallel edges on both sides, theoretically the slope is... With slope They should be equal; the slopes in the actual fitting results are... With slope There may be slight differences; the slope can be taken. With slope The average value is used as the slope of the midline. ,Right now Intercept parameter of the centerline Intercept parameters of the two edge lines and The decision is made, and the calculation formula is as follows: It's understandable that the straight line obtained through this calculation... This is the theoretical central axis of the waveguide structure. In some embodiments, if the equations of the two edge lines are expressed in general form... and If we express that the equation of the midline can be obtained by averaging the coefficients, then... However, it is important to normalize the equations of the lines to ensure the comparability of the coefficients. Alternatively, the median can be geometrically constructed by first finding the locus of the midpoints of corresponding points (e.g., points with the same x-coordinate) on the two edge lines.

[0078] In practical implementation, the mathematical expression of the theoretical central axis in the sensor coordinate system is recorded in point-slope form or parametric equation form to obtain the coordinate expression. The point-slope form utilizes the coordinates of a known point on the centerline. and slope The expression is: , midpoint It can be any point on the median, for example, obtained by solving for the intersection of the median and a coordinate axis. The parametric equation form introduces a parameter. The coordinates of the points on the midline Represented as Functions, for example: , ,in It is the angle between the centerline and the positive X-axis direction of the sensor coordinate system, parameter The distance of the point from the reference point along the centerline direction. The directed distance. It's understandable that point-slope or parametric equation forms of coordinate expressions facilitate subsequent calculations of spatial geometric relationships.

[0079] In one embodiment of the present invention, the Sobel operator is applied to a high-contrast local image for gradient calculation. The Sobel operator includes two sets of 3x3 convolution kernels: one set of kernels detects gradient changes in the horizontal direction of the image, and the other set detects gradient changes in the vertical direction. The horizontal gradient component is obtained by convolving each pixel of the high-contrast local image with the horizontal convolution kernels of the Sobel operator. The vertical gradient component is obtained by convolving the vertical convolution kernel of the Sobel operator. Gradient magnitude of a pixel Through horizontal gradient components and vertical gradient components The calculation results show that the calculation relationship is as follows: gradient magnitude This characterizes the edge strength of the image at that pixel. The application of the Sobel operator can enhance the gradient response at waveguide port edges because these edges are regions in the image where grayscale values ​​undergo abrupt changes; convolving these regions with the Sobel operator will produce larger gradient magnitudes. In some embodiments, the high-contrast local image may be preprocessed with Gaussian filtering before applying the Sobel operator to reduce noise interference. Optionally, other edge detection operators such as the Prewitt operator or the Roberts operator may also be used for gradient calculation.

[0080] In specific implementation, a grayscale profile scan is performed on the gradient image along the normal direction perpendicular to the expected edge direction. The expected edge direction can be preset based on the approximate orientation of the waveguide port. For each predetermined checkpoint of the waveguide port edge to be tested on the gradient image, a scan line is set along its normal direction, centered on that checkpoint. The grayscale values ​​of the corresponding pixel positions in the original high-contrast local image are extracted along this scan line. Arranging these grayscale values ​​in the scan order yields a grayscale distribution curve. The grayscale distribution curve exhibits a step change from one grayscale plateau to another at the ideal edge; in the actual image, this change is a ramp transition. It can be understood that scanning along the normal direction can obtain the most significant grayscale change information perpendicular to the edge, which is crucial for subsequent accurate edge location. In some embodiments, the extracted original grayscale values ​​can be smoothed to suppress the influence of noise on the grayscale distribution curve.

[0081] In practice, the steepest interval of grayscale value change is located on the grayscale value distribution curve. This interval corresponds to the continuous interval with the largest absolute value of the first derivative of the grayscale value distribution curve. The grayscale difference between adjacent sampling points on the grayscale value distribution curve is calculated, and the region with the largest absolute value of the difference is identified as the steepest interval of grayscale value change. Optionally, the first difference of the grayscale value distribution curve can be directly calculated, or the derivative operator can be used to locate the steepest interval. Gaussian fitting is then performed on the data points within the steepest interval of grayscale value change. Gaussian fitting uses a Gaussian function model to approximate the shape of the grayscale value distribution curve within this interval. The expression for the Gaussian function used for fitting is:

[0082]

[0083] Where: symbol Indicates the position coordinates (e.g., pixel index) along the scan line, symbol Indicates the location grayscale value at, symbol The magnitude parameter of the Gaussian function is represented by the symbol. The central position parameter of the Gaussian function is represented by the symbol. The standard deviation parameter of the Gaussian function is represented by the symbol. This represents the background grayscale shift parameter. The parameter is adjusted using fitting algorithms such as the least squares method. , , and Make the Gaussian function The error between the actual grayscale data points and the data points within the steepest grayscale value range is minimized. It can be understood that Gaussian fitting can utilize information from multiple data points to smooth noise and obtain sub-pixel accuracy edge position estimation.

[0084] In practical implementation, the center point of the Gaussian fitted curve is used as the sub-pixel accuracy position of the edge, i.e., the Gaussian function parameter. The corresponding position. Parameters Typically not an integer, this indicates that the edge lies at a sub-pixel location between two pixels. By iterating through all test points on the waveguide port edge and repeating the steps of scanning along the normal, locating the steepest interval, and Gaussian fitting for each test point, sub-pixel precision position coordinates are obtained for each test point. By combining the row and column indices of each point to be measured in the image with its normal direction, the sub-pixel position coordinates along the scan line can be determined. Convert back to two-dimensional subpixel coordinates in the high-contrast local image coordinate system This series of two-dimensional sub-pixel coordinates This constitutes a set of sub-pixel precision coordinates of the actual edge of the target waveguide port.

[0085] In one embodiment of the present invention, the actual geometric center coordinates of the target waveguide port are calculated based on the sub-pixel precision coordinate set of the actual edge of the target waveguide port. The calculation process employs an ellipse fitting or least squares circle fitting algorithm. The sub-pixel precision coordinate set of the actual edge of the target waveguide port contains a series of two-dimensional point coordinates. , where the symbol Indicates the first The x-coordinates of the edge points, symbol Indicates the first The ordinates of the edge points, with symbols This represents the total number of edge points. The least squares circle fitting algorithm aims to find a circle that minimizes the sum of the squared radial distances from the circle's boundary to all edge points. Let the coordinates of the circle's center be... , radius is The optimization objective is to minimize the error function. Error function The expression is:

[0086]

[0087] Where: symbol The x-coordinate of the fitted circle's center is represented by the symbol. The ordinate of the fitted circle's center is represented by the symbol. This represents the radius of the fitted circle. The error function is solved by... Minimize parameters , and The actual geometric center coordinates of the target waveguide port can then be obtained. In some embodiments, if the waveguide port shape is closer to an ellipse, an ellipse fitting algorithm is used. This algorithm obtains the coefficients of the ellipse equation by solving a generalized eigenvalue problem, and then calculates the coordinates of the ellipse center as the actual geometric center coordinates. Optionally, before fitting, outlier removal can be performed on the sub-pixel precision coordinate set of the actual edge of the target waveguide port to improve fitting robustness.

[0088] In practical implementation, the coordinates of the axis point closest to the actual geometric center are calculated from the coordinate expression of the theoretical central axis of the waveguide structure in the sensor coordinate system. The theoretical central axis of the waveguide structure is represented by a point-slope form. Let the actual geometric center coordinates be... The slope of the straight line passing through that point and perpendicular to the theoretical central axis is... The equation of the perpendicular line is By simultaneously solving the equations of the central axis and the perpendicular line, the coordinates of the intersection point of the two lines can be obtained. Intersection coordinates That is, the coordinates of the distance from the actual geometric center on the theoretical central axis. The coordinates of the nearest point. In some embodiments, if the theoretical central axis is given in the form of a parametric equation, the coordinates of the nearest point can be calculated directly using the vector projection formula.

[0089] In practical implementation, the difference between the actual geometric center coordinates and the coordinates of the nearest axis point is calculated. This difference is decomposed in the XY plane of the sensor coordinate system into a longitudinal deviation component along the axis and a lateral deviation component perpendicular to the axis. (Longitudinal deviation component...) The lateral deviation component represents the distance between the actual geometric center and the nearest axis point along the direction of the theoretical central axis. This represents the distance by which the actual geometric center deviates from the theoretical central axis in the direction perpendicular to the theoretical central axis. This can be understood as the longitudinal deviation component. and lateral deviation components Together they constitute the position deviation vector In some embodiments, a local coordinate system based on the theoretical central axis can be established. The lateral and longitudinal coordinate values ​​corresponding to the actual geometric center coordinates can be directly calculated within this local coordinate system. These two coordinate values ​​represent the lateral deviation components, respectively. and longitudinal deviation components .

[0090] In practical implementation, the motion parameters of the support platform are acquired, including calibration conversion coefficients between the image sensor pixel size and the actual movement distance of the support platform. Acquiring these parameters involves fixing a calibration plate with a standard scale onto the support platform. The calibration plate has regularly arranged feature patterns with known physical dimensions, such as a checkerboard or dot matrix. A high-resolution area array image sensor is controlled to image the calibration plate, acquiring an image of the calibration plate. At least two non-collinear feature points are identified in the calibration plate image, and their pixel coordinates in the image sensor pixel coordinate system are recorded. Based on the known physical distance between the feature points on the calibration plate, the actual physical size corresponding to a single pixel of the image sensor is calculated, yielding the pixel equivalent calibration value. (Unit: micrometers / pixel). The control platform is moved by a known physical displacement along two mutually orthogonal axes (e.g., the X-axis and Y-axis) within a plane. and It also uses a high-resolution area array image sensor to acquire images before and after movement, and identifies the coordinate changes of the same feature point in the pixel coordinate system. and Based on the physical displacement and the corresponding pixel coordinate change, the proportional relationship between the actual physical displacement of the platform under a unit pulse drive and the pixel movement of the image sensor is calculated for each axial movement of the platform, thus obtaining the calibration conversion coefficient. Referring to Table 1, the calibration process of the platform motion parameters can generate example data as shown in Table 1.

[0091] Table 1: Calibration Table of Motion Parameters for Loading Platform

[0092] Motion axis Known physical displacement Image pixel coordinate change Calculated pixel equivalent Relationship between motor pulse count and displacement X-axis Y-axis

[0093] It is understandable that the pixel equivalent in Table 1 and Converting pixel deviations in an image into physical spatial deviations, and the relationship between motor pulse count and displacement. and The physical displacement is converted into the number of driving pulses, and the two are combined to form a complete calibration conversion coefficient.

[0094] In practical implementation, the lateral deviation component of the position deviation vector is... and longitudinal deviation components Multiply by the calibration conversion factor to convert the lateral displacement that the bearing platform needs to compensate for in the actual physical space. and longitudinal displacement The conversion formula is: and ,in and These are the pixel equivalent calibration values ​​corresponding to the horizontal and vertical directions, respectively. Since the horizontal and vertical directions may correspond to different directions in the sensor coordinate system, their calibration values ​​may be taken from Table 1. and Alternatively, it can be obtained through coordinate rotation conversion. Combining the current position feedback value of the platform in the motor driver, calculate the number of transverse motor drive pulses required to move the platform to the target position. and longitudinal motor drive pulse count The calculation formula is: and ,in and This represents the actual displacement (micrometers / pulse) corresponding to each motor pulse. (Number of pulses for lateral motor drive) and longitudinal motor drive pulse count It is sent to the motion controller as a drive control quantity.

[0095] See Figure 4 This is a convergence curve of alignment correction accuracy over multiple rounds, showing the trend of the position deviation vector magnitude with each correction round during the alignment of the glass substrate optical waveguide, intuitively reflecting the convergence process of alignment accuracy. The curve clearly shows that the deviation magnitude converges rapidly with each correction round, directly proving the feasibility and effectiveness of the glass substrate optical waveguide alignment method. The deviation magnitude decreases rapidly with the increase of correction rounds, with the effect being particularly significant in the first three rounds, decreasing from 8.5 pixels to 2.1 pixels, a reduction of over 75%. From the fifth round onwards, the deviation magnitude is below the common alignment accuracy threshold, and the system enters a stable high-precision alignment state. By using the deviation convergence speed and final accuracy, the minimum requirements for hardware such as image sensor resolution and stage motion accuracy can be deduced. It can serve as a visual indicator for production line quality control, used to monitor the alignment process of each substrate and promptly identify abnormal batches.

[0096] In one embodiment of the present invention, after the support stage completes planar displacement, a high-resolution area array image sensor is used to re-acquire a global optical image of the target glass substrate in the same region after alignment. The acquired global optical image after alignment is used to evaluate the effect of this alignment correction. In specific implementation, the steps of generating waveguide structure contour point cloud, calculating the theoretical central axis of the waveguide structure, extracting the sub-pixel precision coordinate set of the actual edge of the target waveguide port, and solving the position deviation vector are repeatedly performed in the global optical image after alignment. The methods used in the repeated steps are completely consistent with the methods used before the first alignment correction. The waveguide structure contour point cloud generation is based on the global optical image after alignment and performs adaptive threshold segmentation, morphological opening operation, edge tracking, and cubic spline interpolation processing again. Based on the newly generated waveguide structure contour point cloud, geometric feature analysis is performed again to calculate the theoretical central axis coordinate expression of the aligned waveguide structure in the current state. From the imaging results of the same target waveguide port by the high-resolution area array image sensor, a high-contrast local image is separated again and edge sharpening and sub-pixel positioning processing are performed to extract the sub-pixel precision coordinate set of the actual edge of the aligned target waveguide port. The sub-pixel precision coordinate set of the actual edge of the aligned target waveguide port is spatially matched with the theoretical center axis calculated after alignment. The position deviation vector between the new actual port center and the theoretical axis is then solved. This position deviation vector reflects the residual alignment error after one correction.

[0097] In practical implementation, determining whether the magnitude of the newly solved position deviation vector is less than a preset alignment accuracy threshold involves calculating the sum of the squares of the lateral deviation component and the longitudinal deviation component of the newly solved position deviation vector. The position deviation vector can be represented as... , where the symbol Represents the newly solved lateral deviation component, with the sign... This represents the newly solved longitudinal deviation component. The sum of the squares of the lateral deviation component and the longitudinal deviation component is calculated, i.e., the calculation... The magnitude of the position deviation vector is obtained by taking the square root of the sum of the squares of the lateral deviation component and the squares of the longitudinal deviation component. The calculation formula is:

[0098]

[0099] Where: the magnitude of the position deviation vector It is a scalar that represents the overall distance of the actual port center from the theoretical central axis. In some embodiments, other equivalent norm calculation methods may also be used when calculating the modulus. The modulus of the calculated position deviation vector is then expressed as... Alignment accuracy thresholds pre-stored in the system Comparison, alignment accuracy threshold It is a positive number preset according to process requirements, for example... Pixels. Comparison operations to determine inequalities. Is this true? In some embodiments, the alignment accuracy threshold... The threshold values ​​for the horizontal and vertical components can be set independently; in this case, they need to be judged separately. and Whether they are both true.

[0100] In practical implementation, when the magnitude of the position deviation vector is not less than the alignment accuracy threshold, the step of generating the drive control quantity for compensating motion within the plane of the bearing platform is repeatedly executed using the newly solved position deviation vector as input. This step involves calculation based on the position deviation vector and pre-calibrated bearing platform motion parameters. The newly solved position deviation vector... The position deviation vector used in the previous iteration will be replaced as the input for the new round of compensation motion. The process of generating the drive control quantity based on the position deviation vector includes: taking the lateral deviation component of the new position deviation vector... and longitudinal deviation components Multiply by the calibration conversion factor to convert the new lateral displacement that the bearing platform needs to compensate for in the actual physical space. and longitudinal displacement The number of transverse motor drive pulses for the next round is calculated based on the current position feedback value of the bearing platform. and longitudinal motor drive pulse count Optionally, a decay factor can be introduced in the new round of calculations to scale the drive control quantity to prevent system overshoot. It can be understood that when the magnitude of the position deviation vector is not less than the alignment accuracy threshold, it indicates that a single correction has failed to bring the system into the allowable error range, and iterative correction is required.

[0101] In practice, the newly generated drive control quantity is sent to the motion controller of the platform. The drive control quantity includes the number of drive pulses of the new round of transverse motor. and longitudinal motor drive pulse count The motion controller receives the drive control input and drives the platform to perform a new round of planar displacement correction. The platform moves according to the new pulse command to further reduce the position deviation. Optionally, the motion controller can adopt a closed-loop control mode, reading the position feedback of the grating ruler or encoder in real time during the movement of the platform to ensure that the platform eventually stops at the target position required by the command. After the platform completes a new round of planar displacement correction, it can trigger the high-resolution area array image sensor to acquire images again and repeat the judgment process, forming a closed-loop feedback control cycle of "measurement-calculation-drive-remeasurement" until the magnitude of the calculated position deviation vector is less than the preset alignment accuracy threshold, at which point the alignment process terminates.

[0102] See Figure 5 This is a visualization of the waveguide structure outline and its theoretical central axis, intuitively presenting the spatial relationship between the waveguide structure outline point cloud and the theoretical central axis during the alignment process of the optical waveguide on the glass substrate. The left edge point cloud represents the outline points of the left long side of the waveguide structure, obtained through image edge extraction and resampling, and is used for subsequent straight line fitting. The right edge point cloud represents the outline points of the right long side of the waveguide structure, which, together with the left point cloud, constitute the complete outline of the waveguide. The theoretical central axis is obtained by taking the midline after least-squares straight line fitting of the left and right edge point clouds, and is the ideal geometric center reference of the waveguide. The waveguide port edge represents the sub-pixel precision edge point set of the target waveguide port, used for subsequent deviation calculation with the theoretical axis. By transforming the abstract processes of "outline point cloud generation" and "axis calculation" into an intuitive spatial distribution, the geometric characteristics of the waveguide structure are clearly displayed.

[0103] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for aligning optical waveguides based on a glass substrate, characterized in that, The method includes: A global optical image of the target glass substrate on the support platform is acquired by a high-resolution area array image sensor, and the global optical image includes a waveguide structure pattern. Waveguide structure pattern region identification and boundary extraction are performed on the global optical image to generate waveguide structure contour point cloud; Based on the waveguide structure contour point cloud, the coordinate expression of the theoretical central axis of the waveguide structure in the sensor coordinate system is calculated through geometric feature analysis. From the imaging results of the high-resolution area array image sensor, a high-contrast local image of the target waveguide port under specific illumination conditions is extracted; Edge sharpening and subpixel localization processing are performed on the high-contrast local image to extract the subpixel precision coordinate set of the actual edge of the target waveguide port; The sub-pixel precision coordinate set of the actual edge of the target waveguide port is spatially matched with the theoretical central axis of the waveguide structure to solve for the positional deviation vector between the actual port center and the theoretical axis. Based on the position deviation vector and combined with the pre-calibrated motion parameters of the bearing platform, a drive control quantity for compensating motion in the plane of the bearing platform is generated; The drive control quantity is sent to the motion controller of the support platform, which drives the support platform to perform planar displacement and completes one alignment correction.

2. The method for aligning optical waveguides based on a glass substrate according to claim 1, characterized in that, The global optical image is subjected to waveguide structure pattern region identification and boundary extraction to generate a waveguide structure contour point cloud, including: The global optical image is processed by an adaptive threshold segmentation algorithm to separate the waveguide structure pattern region from the glass substrate background region, thereby obtaining a binary image of the waveguide structure. A morphological opening operation is performed on the binarized image of the waveguide structure to remove noise and smooth the edges of the waveguide structure. An edge tracking algorithm is performed on the smoothed binary image of the waveguide structure to obtain the pixel-level coordinate sequence of the waveguide structure edge; The pixel-level coordinate sequence is smoothed and resampled using cubic spline interpolation to obtain a continuous discrete point set of edge curves, forming the waveguide structure contour point cloud.

3. The method for aligning optical waveguides based on a glass substrate according to claim 2, characterized in that, Based on the waveguide structure contour point cloud, the coordinate expression of the theoretical central axis of the waveguide structure in the sensor coordinate system is calculated through geometric feature analysis, including: Select a subset of contour points located on the long sides of the waveguide structure from the contour point cloud of the waveguide structure. Least square line fitting is performed on the subsets of contour points on the two long sides respectively to obtain two line equations representing the two edges of the waveguide. Based on the equations of two straight lines representing the two edges of the waveguide, the midline of the two lines is calculated, which is the theoretical central axis of the waveguide structure. The mathematical expression of the theoretical central axis in the sensor coordinate system is recorded in the form of point-slope form or parametric equation to obtain the coordinate expression.

4. The method for aligning optical waveguides based on a glass substrate according to claim 3, characterized in that, Edge sharpening and subpixel localization processing are performed on the high-contrast local image to extract the subpixel precision coordinate set of the actual edge of the target waveguide port, including: The Sobel operator is applied to the high-contrast local image to calculate gradients, thereby enhancing the gradient response at the waveguide port edges. On the gradient image, a grayscale profile is scanned along the normal direction perpendicular to the expected edge direction to obtain the grayscale distribution curve; On the grayscale value distribution curve, locate the interval with the steepest grayscale value change, and perform Gaussian fitting on the data points within the interval; Using the center point of the Gaussian fitted curve as the subpixel precision position of the edge, all test points on the edge of the waveguide port are traversed to obtain a series of subpixel precision position coordinates, which constitute the subpixel precision coordinate set.

5. The method for aligning optical waveguides based on a glass substrate according to claim 4, characterized in that, The sub-pixel precision coordinate set of the actual edge of the target waveguide port is spatially matched with the theoretical central axis of the waveguide structure to solve for the position deviation vector between the actual port center and the theoretical axis, including: Based on the sub-pixel precision coordinate set of the actual edge of the target waveguide port, the actual geometric center coordinates of the target waveguide port are calculated by ellipse fitting or least squares circle fitting algorithm. Calculate the coordinates of the axis point closest to the actual geometric center from the coordinate expression of the theoretical central axis of the waveguide structure in the sensor coordinate system; Calculate the difference between the actual geometric center coordinates and the coordinates of the nearest axis point. The difference is decomposed into a longitudinal deviation component along the axis direction and a lateral deviation component perpendicular to the axis direction in the XY plane of the sensor coordinate system. The longitudinal deviation component and the lateral deviation component together constitute the position deviation vector.

6. The method for aligning optical waveguides based on a glass substrate according to claim 5, characterized in that, Based on the position deviation vector and combined with the pre-calibrated motion parameters of the bearing platform, a drive control quantity for compensating the motion in the plane of the bearing platform is generated, including: Obtain motion parameters of the support platform, including calibration conversion coefficients between image sensor pixel size and actual movement distance of the support platform; The lateral and longitudinal deviation components of the position deviation vector are multiplied by the calibration conversion coefficient to convert them into the lateral and longitudinal displacements that the bearing platform needs to compensate for in the actual physical space. Based on the current position feedback value of the bearing platform in the motor driver, calculate the number of transverse motor drive pulses and the number of longitudinal motor drive pulses required to move the bearing platform to the target position. The number of horizontal motor drive pulses and the number of vertical motor drive pulses are used as the drive control quantities.

7. The method for aligning optical waveguides based on a glass substrate according to claim 6, characterized in that, The acquisition of the motion parameters of the support platform includes: A calibration plate with a standard scale is fixed to the support platform; The high-resolution area array image sensor is controlled to image the calibration board, and an image of the calibration board is acquired. Identify at least two non-collinear feature points in the calibration plate image and record the pixel coordinates of the feature points in the image sensor pixel coordinate system; Based on the actual physical distance between known feature points on the calibration board, the actual physical size corresponding to a single pixel of the image sensor is calculated to obtain the pixel equivalent calibration value. The control platform moves a known physical displacement along two mutually orthogonal axes in the plane, and the high-resolution area array image sensor acquires images before and after the movement to identify the coordinate changes of the same feature point in the pixel coordinate system. Based on the physical displacement and the corresponding pixel coordinate change, the proportional relationship between the actual physical displacement of the carrier platform under a unit pulse drive and the pixel movement of the image sensor is calculated when the carrier platform moves in each axial direction, thus obtaining the calibration conversion coefficient.

8. The method for aligning optical waveguides based on a glass substrate according to claim 7, characterized in that, After sending the drive control quantity to the motion controller of the platform, the system further includes: After the support stage completes the planar displacement, the high-resolution area array image sensor again acquires a global optical image of the target glass substrate after alignment in the same area; In the aligned global optical image, the steps of generating the waveguide structure contour point cloud, calculating the theoretical central axis of the waveguide structure, extracting the sub-pixel precision coordinate set of the actual edge of the target waveguide port, and solving the position deviation vector are repeated. Determine whether the magnitude of the newly solved position deviation vector is less than the preset alignment accuracy threshold.

9. The method for aligning optical waveguides based on a glass substrate according to claim 8, characterized in that, The step of determining whether the magnitude of the newly solved position deviation vector is less than a preset alignment accuracy threshold includes: Calculate the sum of the squares of the lateral deviation component and the longitudinal deviation component of the newly solved position deviation vector; The magnitude of the position deviation vector is obtained by taking the square root of the sum of the squares of the lateral deviation component and the squares of the longitudinal deviation component. The magnitude of the position deviation vector is compared with the alignment accuracy threshold pre-stored in the system.

10. The method for aligning optical waveguides based on a glass substrate according to claim 9, characterized in that, When the magnitude of the position deviation vector is not less than the alignment accuracy threshold, the method further includes: Using the newly solved position deviation vector as input, the step of generating the drive control quantity for the compensated motion in the plane of the bearing platform is repeated. The newly generated drive control quantity is sent to the motion controller of the platform, driving the platform to perform a new round of planar displacement correction.

Images