Glass substrate rotating jig alignment method and system based on visual guidance

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a glass substrate rotating jig alignment method and system based on visual guidance, and the method comprises the steps: constructing a complex gradient vector field according to the gradient of each pixel point in a glass substrate grayscale image, carrying out the four-order phase mapping of the complex gradient vector field, constructing a harmonic transformation field, and carrying out the four-order phase mapping of the harmonic transformation field; calculating a square frame saliency map according to the harmonic transformation field, calculating a circular saliency map based on a complex gradient vector field by using gradient convergence characteristics of solid dots, performing dual fusion on the square frame saliency map and the circular saliency map, constructing a target probability potential energy field, and searching extreme points in the target probability potential energy field to determine true reference point coordinates. And the rotating jig is driven to execute alignment compensation. According to the invention, the interference of metal grid textures, linear scratches and circular stains can be resisted, the rotation invariance is realized, and the high-precision visual alignment is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a visual guidance-based glass substrate rotary jig alignment method and system. BACKGROUND

[0002] In the automated production process of glass substrates, the accuracy of the visual alignment system directly determines the yield of subsequent hole insertion, screen printing or lamination processes. In order to ensure the uniqueness and robustness of alignment, the reference point is usually designed as a composite geometric structure of concentrically nested solid black dots and hollow black squares.

[0003] However, as the process requirements improve, the measured objects gradually develop in the direction of ultra-thin and high light transmission. In actual engineering scenarios, due to the high transparency of the glass substrate, the surface features of the metal jig table below are clearly imaged by the industrial camera when shooting the reference point. These features are usually regular metal grid textures, vacuum suction hole arrays or random straight-line scratches. In addition, the circular stains formed by ink droplets falling in the production environment and the shadows caused by uneven edge lighting are highly similar to the reference point in grayscale features.

[0004] Traditional visual algorithms face serious challenges when dealing with such complex scenes. A single circle detection algorithm cannot distinguish between true reference points with square constraints and independent circular stains. A pure square detection algorithm not only has a large amount of calculation, but is also easily disturbed by straight-line scratches, which are misjudged as square edges. Existing feature extraction methods based on histogram of oriented gradients can distinguish between cluttered textures to some extent, but lack robustness when dealing with rotating targets, and cannot effectively distinguish between one-way distributed straight-line interference and four-way orthogonal distributed square structures, resulting in a decrease in the recognition accuracy of composite reference points and positioning failure, affecting the accuracy and stability of glass substrate rotary jig alignment. SUMMARY

[0005] To solve the technical problems of large background interference of glass substrates, reference point recognition easily affected by scratches and stains, and lack of rotational invariance in traditional algorithms in the prior art, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides a visual guidance-based glass substrate rotary jig alignment method, comprising: The gray scale image of the glass substrate is collected, a complex gradient vector field is constructed according to the gradient of each pixel point in the gray scale image of the glass substrate, the fourth-order phase mapping is performed on the complex gradient vector field, a harmonic transformation field is constructed, the box saliency value of the pixel point is calculated according to the fourth-order harmonic vector of the pixel point in the harmonic transformation field, and a box saliency map is formed; the gradient convergence characteristics of the solid circle point are utilized, and a circular saliency map is calculated based on the complex gradient vector field; the box saliency map and the circular saliency map are dual fused based on a topological concentric constraint mechanism, a target probability potential field is constructed, extreme points in the target probability potential field are searched to determine the true fiducial point coordinates, and a rotary tool is driven to perform alignment compensation according to the true fiducial point coordinates.

[0007] The application maps the image gradient field to the complex domain, constructs a fourth-order harmonic transformation field, utilizes the four-way orthogonal gradient of the box structure to cause constructive interference in the four times frequency domain, and utilizes the continuous gradient of the circular structure to cause destructive interference in the four times frequency domain, realizes the extraction of the box structure, effectively suppresses the circular interference and the single-direction straight line scratch, utilizes the all-directional gradient convergence characteristics of the solid circle point to construct a circular saliency map which is only sensitive to the center of the circle, and through the topological concentric constraint mechanism, the box saliency map and the circular saliency map are dual fused, the center of the circle is forced to fall within the tolerance range of the box center, so that the circular stain and the straight line scratch are completely filtered out, the recognition accuracy is greatly improved, the application is based on complex vector operation, has rotation invariance, does not need to pre-correct the attitude angle of the glass substrate, and has simple parameters and high calculation efficiency, and meets the real-time alignment demand on the industrial high-speed production line.

[0008] Preferably, the complex gradient vector field is constructed, including: calculating the gradient of the glass substrate gray scale image in the horizontal direction and the vertical direction respectively by using a Scharr operator; calculating the complex gradient vector of each pixel point in the glass substrate gray scale image to form a complex gradient vector field, and the complex gradient vector satisfies the expression: , wherein, represents the complex gradient vector at the pixel point ; represents the gradient amplitude at the pixel point ; represents the gradient direction angle at the pixel point ; represents the horizontal direction gradient at the pixel point ; represents the vertical direction gradient at the pixel point ; is an imaginary unit.

[0009] The application extracts edge gradient information of a glass substrate image by using a Scharr operator, and constructs the edge gradient information as a complex gradient vector field, which can effectively reduce angle error of a rotating target edge detection, and uses a complex form to simultaneously retain intensity and direction information of the gradient, converts traditional scalar image processing into vector field analysis, and provides a mathematical basis for processing symmetry of a geometric structure and performing frequency domain harmonic analysis by using a complex rotating operator.

[0010] Preferably, the fourth-order harmonic vector satisfies an expression: ; in the expression, represents the fourth-order harmonic vector at a pixel point ; represents a gradient amplitude at the pixel point ; represents a gradient direction angle at the pixel point ; is an imaginary unit; and 4 is a harmonic order.

[0011] The application maps the gradient field of the image to a four-fold frequency domain by constructing the fourth-order harmonic vector, uses the four-fold rotational symmetry of the square geometric structure to make the gradient directions corresponding to four edges that are originally orthogonal to each other in space achieve phase alignment in the frequency domain, thereby creating conditions for subsequent feature resonance enhancement by vector accumulation, and simultaneously makes interference features that do not have the symmetry achieve phase dispersion in the frequency domain, thereby achieving spectral separation of the target feature and background noise from a mathematical bottom layer.

[0012] Preferably, the method further includes: setting a first neighborhood with the current pixel point as a center, calculating a vector sum of the fourth-order harmonic vectors of all pixel points in the first neighborhood, and taking a module length of the vector sum as the box saliency value of the current pixel point.

[0013] The application calculates the vector sum of the fourth-order harmonic vectors in the first neighborhood and the module length thereof as the box saliency value, uses the vector superposition principle to make the gradients of the four edges of the box structure interfere constructively after phase alignment, and produce a high-intensity resonance response, while the gradients of the circular structure or the disordered texture interfere destructively in the superposition process due to continuous changes or random distribution of the directions, and the response value approaches zero, thereby realizing high signal-to-noise ratio extraction of the reference point box component without relying on complex geometric fitting, and the process naturally has rotation invariance.

[0014] Preferably, the method further includes: calculating a circular saliency value at each pixel point to form a circular saliency map, and the circular saliency value satisfies an expression: , in the expression, represents the circular saliency value at a pixel point ; is a pixel point The second neighborhood centered on; Represents the pixels in the second neighborhood. Gradient magnitude at; Represents the pixels in the second neighborhood. The gradient direction angle at that location; Represents the second neighboring pixel. Pointing to the center pixel The displacement vector angle in the image coordinate system; Pi is the mathematical constant of a circle.

[0015] This invention corrects the inverse gradient features in a white background with black dots scene by calculating the saliency value of the circle and introducing phase compensation, and statistically analyzes the gradient centripetal degree in the second neighborhood, which can enhance the response intensity of solid black circles. This invention utilizes the geometric convergence characteristics of the omnidirectional gradient of solid circles to complement the discrete gradient features of the box structure, thereby effectively extracting the circular component of the reference point and eliminating misjudgments of hollow boxes or non-circular interference objects.

[0016] Preferably, constructing the target probability potential field includes: calculating the probability potential energy of each pixel belonging to the center of the true reference point, constructing the target probability potential field, wherein the probability potential energy of a pixel belonging to the center of the true reference point satisfies the expression: , Represents pixels The probabilistic potential energy belonging to the true reference point center; Represents pixels The circular significant value at the location; The standard deviation is expressed as Gaussian kernel; This represents the convolution operation; Represents pixels The significance value of the box at the location; This represents the standard deviation parameter of the Gaussian kernel.

[0017] This invention constructs a target probability potential field, uses a Gaussian kernel to smoothly expand the salient values ​​of the bounding boxes to establish a tolerance region, and performs dual fusion of the circular salient values ​​and the expanded bounding box salient values ​​to implement topological concentricity constraints. This invention forces that the center of the high-response circle must fall within the tolerance range of the center of the high-response bounding box, thereby filtering out circular stains or straight scratches with only a single feature, ensuring that only true reference points with a nested square-circle topological structure can generate global extrema, greatly improving the accuracy of recognition and anti-interference ability.

[0018] Preferably, the standard deviation parameter of the Gaussian kernel is ,in To allow for physical concentricity deviation, This refers to the image resolution.

[0019] Preferably, determining the true reference point coordinates includes: searching for the global maximum pixel coordinates in the target probabilistic potential energy field; selecting a third neighborhood centered on the global maximum pixel coordinates; fitting the peak vertex of the response surface using a bivariate quadratic polynomial to obtain the sub-pixel level true reference point coordinates.

[0020] This invention breaks through the limitation of physical pixel resolution of image sensors by searching for the global maximum value in the target probabilistic potential energy field and selecting the third neighborhood for binary quadratic polynomial fitting, and using the surface fitting method to analyze the sub-pixel position of the response peak. It can further extract the high-precision true reference point center coordinates while retaining the robustness of global search, thus meeting the stringent requirements of precision manufacturing for micron-level alignment accuracy.

[0021] Preferably, the step of driving the rotary fixture to perform alignment compensation based on the true reference point coordinates includes: calculating the positional deviation between the true reference point coordinates and the preset target position; converting the positional deviation into pulse signals in the mechanical coordinate system and sending them to the motion control card to drive the X-axis, Y-axis, and... The shaft motor performs linkage compensation until the position deviation is less than the preset process threshold.

[0022] Secondly, the present invention provides a vision-guided glass substrate rotation fixture alignment system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned vision-guided glass substrate rotation fixture alignment method is implemented.

[0023] By adopting the above technical solution, a computer program for the above-mentioned vision-guided glass substrate rotation fixture alignment method is generated and stored in a memory for loading and execution by a processor. This allows for the creation of a terminal device based on the memory and processor, facilitating its use.

[0024] The beneficial effects of this invention are as follows: This invention maps image edge features to the complex domain by constructing a complex gradient vector field. It utilizes a fourth-order harmonic transform field to ensure that the four-directional orthogonal gradients of the reference point box component are phase-aligned and constructively interfered in the frequency domain, while the reference point circular component and linear interference undergo destructive interference due to phase discretization. This achieves the extraction of the box structure without angle pre-correction. Simultaneously, this invention utilizes the omnidirectional convergence characteristic of the solid black circle edge gradient to capture the center position. Based on this, this invention uses a topological concentricity constraint mechanism to perform dual fusion of the box saliency map and the circular saliency map, establishing an effective verification region with the box center as the reference. Mathematically, this forces the center of the high-response circle to be located inside the high-response box, thus filtering out circular stains, linear scratches, and background mesh texture interference with only a single geometric feature. Finally, combined with sub-pixel-level surface fitting and motion control feedback, it achieves high-precision and robust automatic alignment of the glass substrate under complex background interference. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a vision-guided glass substrate rotation fixture alignment method according to the present invention. Figure 2 This is a grayscale image of a glass substrate. Figure 3 This is a gradient amplitude image of the grayscale image of the glass substrate; Figure 4 A saliency map of the box; Figure 5 A circular saliency diagram; Figure 6 A visualization of the target probability potential field; Figure 7 This is a schematic diagram of the true reference point positioning results. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0028] This invention discloses a vision-guided glass substrate rotation fixture alignment method, referring to... Figure 1 This includes steps S1-S5: S1. Acquire a grayscale image of the glass substrate and construct a complex gradient vector field based on the gradient of each pixel in the grayscale image of the glass substrate.

[0029] It should be noted that the objective of this invention is to extract the black reference point under bright backlight. In order to extract the faint geometric edges from the background containing metallic mesh texture and random noise, and to provide a complete vector basis for subsequent frequency domain analysis, this invention no longer processes gradients in Euclidean space, but maps the gradient information of the image to the complex domain, processes the symmetry of the geometric structure through complex rotation operators, and preserves the magnitude and direction information of the gradient.

[0030] Specifically, the backlight unit located below the glass substrate is turned on, and an industrial camera mounted above the rotating fixture acquires a grayscale image of the glass substrate containing reference points. To obtain better rotational invariance, the Scharr operator is used to calculate the gradient of the grayscale image of the glass substrate in the horizontal direction. and the gradient in the vertical direction For example, Figure 2 This is a grayscale image of a glass substrate. Figure 3 This is the gradient magnitude image of the grayscale image of the glass substrate.

[0031] For each pixel in the grayscale image of the glass substrate, a complex gradient vector is calculated to form a complex gradient vector field, which satisfies the expression:

[0032] In the formula, Represents pixels The complex gradient vector at point; Represents pixels The gradient magnitude at a given point is calculated as follows: This reflects the intensity of the edge, when the pixel... When located at the edge of the reference point or the edge of the background texture, Larger, and vice versa; Represents pixels The gradient direction angle at that point is calculated as follows: The range of values ​​is ; Represents pixels The horizontal gradient at that location; Represents pixels The vertical gradient at that location; The imaginary unit; It is a two-parameter arctangent function.

[0033] The complex gradient vectors of all pixels are used to construct a complex gradient vector field.

[0034] Thus, this invention transforms a scalar grayscale image of a glass substrate into a dense complex gradient vector field, providing a mathematical basis for subsequent harmonic analysis.

[0035] S2. Perform fourth-order phase mapping on the complex gradient vector field to construct a harmonic transform field. Calculate the saliency value of the pixel based on the fourth-order harmonic vector of the pixel in the harmonic transform field to form a saliency map.

[0036] It should be noted that the reference point box component exhibits significant 4-fold rotational symmetry, meaning that the normal directions of its four sides differ by 90 degrees in space. In the conventional gradient domain, these four directions are mutually exclusive and difficult to directly accumulate for enhancement. However, in the fourth harmonic domain, the phases of all orthogonal directions will resonate. For example, suppose the overall rotation angle of the box is... Then the gradient directions of its four sides are respectively After multiplying by 4, these four phases become By utilizing the periodicity of complex phases, they point in the same direction in the complex plane. However, for circular or linear structures, this resonance phenomenon does not occur or is extremely weak. This invention utilizes this physical property to perform fourth-order phase mapping on the complex gradient field to construct a harmonic transformation field, thereby enhancing the rectangular structure and suppressing circular and linear structures.

[0037] Specifically, a fourth-order phase mapping is performed on the complex gradient field to construct a harmonic transform field:

[0038] In the formula, Represents pixels The fourth harmonic vector at that location; Represents pixels Gradient magnitude at; Represents pixels The gradient direction angle at that location; 4 is the imaginary unit; 4 is the harmonic order, corresponding to the fourfold symmetry of the square.

[0039] Furthermore, define a first neighborhood. The saliency value of the bounding box of the center pixel is calculated based on the fourth harmonic vectors of the neighboring pixels within the first neighborhood of the center pixel:

[0040] In the formula, Represents pixels The significance value of the box at the location; In pixels The first neighborhood centered at the center has a side length of . Since the saliency of the box depends on the phase resonance of the four sides of the box, The value of needs to be able to cover the components of the reference point box. In this embodiment, The value is set to 1.0 to 1.2 times the side length of the reference point box, for example, 100 pixels. In other embodiments, the implementer can set it according to the actual pixel size of the reference point in the image. ; Represented by pixels Pixels in the first neighborhood of the center The fourth harmonic vector at that location; Represents the accumulation of complex vectors; This indicates taking the modulus of a complex number.

[0041] When the first neighborhood When the covered window is located at the center of the box, the gradient vectors of the four sides of the box are phase-aligned in the fourth harmonic domain. After accumulation, the magnitude reaches a maximum value, resulting in constructive interference. Larger; when the first neighborhood When the covered window is located at the center of the circular component of the reference point or the circular stain, its gradient direction From 0 Continuous change, mapped to After 0 to Rapid rotation, after integration, results in vectors that connect end to end, with the magnitude approaching zero, leading to destructive interference. Minimal; when the first neighborhood When the covered window is located on a straight scratch, there is only unidirectional energy and a lack of multidirectional superposition, making... It is significantly lower than the center of the box.

[0042] Thus, a saliency map of the boxes that is only sensitive to the box structure was obtained. Figure 4 This is a saliency map of the box.

[0043] S3. Utilize the gradient convergence property of solid dots to calculate the circular saliency map based on the complex gradient vector field.

[0044] It should be noted that, in order to accurately locate the reference point and eliminate interference from empty boxes, an algorithm complementary to harmonic resonance is needed to detect solid circles. Since this invention targets a white background with black dots, the gradient direction at the edge of a solid black circle points outward, i.e., to the high grayscale region, while the displacement vector pointing towards the center points inward; the two are in opposite directions. Therefore, this invention constructs a circular structure saliency model incorporating a phase compensation mechanism to correctly capture the features of black circles.

[0045] Specifically, calculate the circular saliency value for each pixel:

[0046] In the formula, Represents pixels The circular significant value at the location; In pixels The second neighborhood centered at the center has a side length of . Since the circular component of the reference point is nested inside the box and its size is small, in order to avoid introducing unnecessary background gradient noise, It should be less than the side length of the first neighborhood. In this embodiment, The value is set to 1.2 to 1.5 times the diameter of the solid black circle inside the reference point, for example, 40 pixels. In other embodiments, the implementer may set the value according to the actual pixel size of the reference point in the image. ; Represents the pixels in the second neighborhood. Gradient magnitude at; Represents the pixels in the second neighborhood. The gradient direction angle at that location; Represents the pixels in the second neighborhood. Pointing to the center pixel The displacement vector angle in the image coordinate system is calculated using the following formula: ; Pi is used to achieve 180° phase compensation to adapt to the gradient inversion characteristics in a white background with a black circle scene. This term is used to measure the consistency between the gradient direction and the centripetal direction; The function is used to remove gradient components that deviate from the center and retain only the centripetal convergence components; Center pixel Coordinates in the image coordinate system For the second neighboring pixels Coordinates in the image coordinate system.

[0047] When pixel When located at the center of a solid black circle, its edge gradient points omnidirectionally outward from the circle, with phase compensation added. Afterwards, the equivalent direction points towards the center of the circle, maximizing the accumulated centripetal components, thus making... Extremely high; when pixel When located at the center of the box, only a small amount of gradient at the four corners points towards the center, making Weaker; when pixel When located on a straight scratch, there is no center of convergence, making Extremely low.

[0048] Thus, a circular saliency map that is only sensitive to circular structures was obtained. Figure 5 It is a circular saliency diagram.

[0049] S4. Based on the topological concentric constraint mechanism, the box saliency map and the circular saliency map are dually fused to construct the target probability potential field.

[0050] It should be noted that after steps S2 and S3, two feature maps with completely orthogonal properties are obtained: a saliency map of a response box and a suppression circle, and a saliency map of a response circle and a suppression square. The topological definition of a true reference point is a strong response circle surrounded and concentrically enclosed by a strong response box. Simple pixel-level multiplication may lead to signal attenuation due to printing concentricity errors. Therefore, this invention introduces Gaussian relaxation constraints to establish an effective verification region based on the center of the box. As long as the center of the circle falls within this region, it can be determined as a true reference point.

[0051] Specifically, the probability potential energy of each pixel belonging to the center of the true reference point is calculated to construct the final target probability potential energy field. The probability potential energy of a pixel belonging to the center of the true reference point satisfies the following expression:

[0052] In the formula, Represents pixels The probabilistic potential energy belonging to the true reference point center; Represents pixels The circular significant value at the location; The standard deviation is expressed as A Gaussian kernel is used for smooth expansion operations; This represents the convolution operation; Represents pixels The significance value of the box at the location; In this embodiment, the standard deviation parameter of the Gaussian kernel is represented. The value is set to 5 pixels. In other embodiments, the implementer can set it according to the concentricity error range allowed by the printing process. For example, if the allowable physical concentricity deviation is millimeters, image resolution is pixels / millimeter It should be set to .

[0053] When pixel When located at a circular stain, although It is relatively tall, but the lack of a surrounding frame structure leads to... Approaching 0, making Suppressed; when pixel When located at an empty box or grid intersection, although It is relatively tall, but the lack of a circular structure in the center leads to... Approaching 0, making Suppressed; only when pixel When located at the true reference point, both are high, and their product produces a global extremum.

[0054] For example, Figure 6 A visualization of the target probability potential field.

[0055] S5. Search for extreme points in the target probabilistic potential energy field to determine the coordinates of the true reference point, and drive the rotating fixture to perform alignment compensation based on the coordinates of the true reference point.

[0056] Specifically, the global maximum pixel coordinates are searched in the target probability potential field, and a third neighborhood is selected centered on these coordinates. The third neighborhood is fitted using a bivariate quadratic polynomial. The probability potential energy of all pixels belonging to the true reference point center is used to obtain the peak vertex of the fitted surface, thus obtaining sub-pixel level precise coordinates. It should be noted that this invention uses a bivariate quadratic polynomial for fitting to refine the already locked global extremum points to sub-pixel level, focusing only on local surface features near the peak. Therefore, the third neighborhood... The scale should be much smaller than the first and second neighbors, and is usually set to... or Pixel.

[0057] For example, Figure 7 This is a schematic diagram of the true reference point positioning results.

[0058] The precise coordinates are compared with the preset theoretical reference coordinates to calculate the positional deviation. This positional deviation is then converted into pulse signals in the mechanical coordinate system and sent to the motion control card to drive the X-axis, Y-axis, and... The shaft motor performs linkage compensation until the deviation is less than the preset process threshold.

[0059] This invention also discloses a vision-guided glass substrate rotation fixture alignment system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a vision-guided glass substrate rotation fixture alignment method according to the present invention.

[0060] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for aligning a vision-guided glass substrate spin chuck, the method comprising: The method comprises the following steps: Collecting a glass substrate gray image, and constructing a complex gradient vector field according to the gradient of each pixel point in the glass substrate gray image; Performing fourth-order phase mapping on the complex gradient vector field to construct a harmonic transformation field, calculating a box saliency value of the pixel point according to the fourth-order harmonic vector of the pixel point in the harmonic transformation field to form a box saliency map; Calculating a circular saliency map based on the complex gradient vector field by using the gradient convergence characteristics of the solid circle point; Based on the topological concentric constraint mechanism, the box saliency map and the circular saliency map are dual fused to construct a target probability potential field; Searching for an extreme point in the target probability potential field to determine the true fiducial point coordinates, and driving a rotating tool to perform alignment compensation according to the true fiducial point coordinates.

2. The vision-guided glass substrate spin chuck alignment method of claim 1, wherein, The complex gradient vector field is constructed, comprising: The Scharr operator is used to calculate the gradients in the horizontal and vertical directions of the grayscale image of the glass substrate. The complex gradient vector of each pixel in the grayscale image of the glass substrate is then calculated to form a complex gradient vector field, which satisfies the expression: ,in, Represents pixels The complex gradient vector at point; Represents pixels Gradient magnitude at; Represents pixels The gradient direction angle at that location; Represents pixels The horizontal gradient at that location; Represents pixels The vertical gradient at that location; It is the imaginary unit.

3. The vision-guided glass substrate spin chuck alignment method of claim 1, wherein, The fourth-order harmonic vector satisfies the expression: ； wherein represents the fourth order harmonic vector at the pixel point ; represents the gradient magnitude at the pixel point ; represents the gradient direction angle at the pixel point ; is the imaginary unit; and 4 is the harmonic order.

4. The vision-guided glass substrate spin chuck alignment method of claim 1, wherein, The box saliency value of the pixel point is calculated, comprising: A first neighborhood is set with the current pixel point as the center, the vector sum of the fourth-order harmonic vectors of all pixel points in the first neighborhood is calculated, and the module length of the vector sum is taken as the box saliency value of the current pixel point.

5. The vision-guided glass substrate spin chuck alignment method of claim 1, wherein, The circular saliency map is calculated, comprising: A circular saliency value at each pixel point is calculated to form a circular saliency map, and the circular saliency value satisfies an expression: wherein, denotes the circular saliency value at the pixel point ; is a second neighborhood with the pixel point as the center; denotes the gradient amplitude at the pixel point in the second neighborhood; denotes the gradient direction angle at the pixel point in the second neighborhood; denotes the displacement vector angle in an image coordinate system in which the second neighborhood pixel points point to the center pixel point ; is the constant pi.

6. The vision-guided glass substrate spin chuck alignment method of claim 1, wherein, The target probability potential field is constructed, comprising: A probability potential of each pixel point belonging to a center of a true fiducial point is calculated, a target probability potential field is constructed, and the probability potential of each pixel point belonging to the center of the true fiducial point satisfies an expression: , represents a probability potential of a pixel point belonging to a center of a true fiducial point; represents a circular saliency value at a pixel point ; represents a Gaussian kernel with a standard deviation of ; represents a convolution operation; represents a box saliency value at a pixel point ; represents a standard deviation parameter of the Gaussian kernel.

7. The vision-guided glass substrate spin chuck alignment method of claim 6, wherein, The standard deviation parameter of the Gaussian kernel is where is the allowed physical concentricity deviation, is the image resolution.

8. The vision-guided glass substrate spin chuck alignment method of claim 1, wherein, The true fiducial point coordinates are determined, comprising: A global maximum value pixel coordinate is searched in the target probability potential field; a third neighborhood is selected with the global maximum value pixel coordinate as the center, a peak vertex of a response surface is fitted by using a binary quadratic polynomial to obtain sub-pixel level true fiducial point coordinates.

9. The vision-guided glass substrate spin chuck alignment method of claim 1, wherein, The true fiducial point coordinates are used to drive the rotating tool to perform alignment compensation, comprising: Calculating the position deviation of the true fiducial point coordinates and a preset target position; The position deviation is converted into a pulse signal in a mechanical coordinate system, and is sent to a motion control card to drive X-axis, Y-axis and Z-axis of the rotary fixture The shaft motor performs linkage compensation until the position deviation is less than a preset process threshold.

10. A vision-guided glass substrate spin chuck alignment system, comprising: The method comprises the following steps: A processor and a memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, a visual guidance-based glass substrate rotating tool alignment method according to any one of claims 1-9 is realized.

Images