Position recognition and calibration method of vacuum printing plate based on machine vision

Abstract

The present invention relates to the technical field of image processing, and in particular to a method for identifying and calibrating the position of a vacuum printed plate based on machine vision, the method comprising: acquiring an image of a printed plate; taking any pixel point in the image as a target point and acquiring a gradient value of the target point; constructing a diffusion function of anisotropic filtering based on the gradient value to filter the target point and finally obtain a filtered image; and identifying and calibrating the position of the printed plate on the filtered image; that is, the scheme of the present invention can acquire a high-quality image of a printed plate and can accurately identify and calibrate the position of the printed plate.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more specifically, to a method for identifying and calibrating the position of a vacuum printed plate based on machine vision. Background Art

[0002] Vacuum printing is a precision manufacturing technology that is mainly used in electronic components, display panels, thin-film circuits and other products. During the vacuum printing process, accurate identification and calibration of the position of the printed plate is the key to ensuring the subsequent printing quality and efficiency.

[0003] In traditional manual or mechanical position recognition methods, the positioning of the plate is usually manually adjusted, which results in high error risk, cumbersome operation, low efficiency, etc. At present, by introducing a machine vision system, the position of the plate can be automatically recognized and accurately calibrated by analyzing the collected images, thus greatly improving the automation level of the production line and the accuracy of plate processing.

[0004] In the related technology, for example, the patent application document with publication number CN114998432A discloses a circuit board detection point positioning method based on YOLOv5, which locates the pixel position of the detection point in the captured circuit board image by using the YOLOv5 target detection algorithm; the average error of the positioning accuracy meets the requirements, and the efficiency is much higher than manual detection.

[0005] When performing positioning, the above scheme takes enough pictures under different ambient light and different circuit board angles, uses them as training sets and verification sets, and trains the network model to accurately locate the detection points.

[0006] However, it should be noted that due to the existence of noise, uneven lighting, reflection and scattering problems of the image capturing device itself, the influence of noise and lighting will still affect the position recognition and calibration of the plate, especially for the recognition and calibration of the position of the printed plate in the printed plate with texture, resulting in a certain deviation in the recognition of the position of the printed plate, making it difficult to calibrate accurately. Therefore, high-quality printed plate images are one of the necessary conditions to ensure the accuracy of printed plate position recognition and calibration. Therefore, it is necessary to filter the printed plate image in advance to obtain a high-quality printed plate image.

[0007] At present, since anisotropic filtering can retain the edge information of the image, and the retention of edge information on the printed plate is crucial for the subsequent position recognition and calibration, anisotropic filtering is usually used when filtering the printed plate image. Anisotropic filtering is an algorithm that obtains the diffusion function for filtering based on the gradient of the pixel point and the diffusion constant. When the lighting is uneven, the vacuum printed plate may have problems such as blurred edges and loss of texture details, which will lead to pixel gradient distortion. At this time, the gradient of the pixel point is inaccurate, which will lead to a decrease in the accuracy of the diffusion function, affect the filtering effect, and thus reduce the accuracy of the position recognition and calibration of the printed plate in the plate.

[0008] Therefore, it is particularly important to accurately filter the printed plate image in order to accurately identify and calibrate the position in the printed plate. Summary of the invention

[0009] The purpose of the present invention is to propose a vacuum printing plate position recognition and calibration method based on machine vision to solve the problem in the prior art that the filtering effect of the printing plate image is poor and the position of the printing plate in the printing plate cannot be accurately recognized and calibrated.

[0010] The present invention provides a method for identifying and calibrating a vacuum printing plate position based on machine vision, comprising:

[0011] acquiring an image of the printed plate;

[0012] Taking any pixel point in the image as the target point, obtaining the gradient value of the target point; constructing the diffusion function of anisotropic filtering based on the gradient value to filter the target point and obtain the filtered image;

[0013] Perform position recognition and calibration on the filtered image;

[0014] The gradient value is obtained by weighting the gradient of the target point in different neighborhood directions; wherein the gradient weight during weighting is the value after the contrast value is normalized; the ratio is the ratio of the texture expression degree and the influence degree of the target point in any neighborhood direction; the texture expression degree is the product of the distribution uniformity and the gradient consistency; the influence degree is the ratio of the local contrast to the distribution uniformity;

[0015] The distribution uniformity represents the distribution of gray values; the local contrast represents the fluctuation of gray values; the gradient consistency for: ; , are the sth pixel point in the window in the ath neighborhood direction of the ith pixel point, The gradient amplitude of the pixel point in the vertical direction of the ath neighborhood direction, the window starts from the ith pixel point and has a length of Window, The natural constant An exponential function with base .

[0016] In the above scheme, by first analyzing the pixel points in the window in different neighborhood directions of each pixel point in the image, the gradient weights in different neighborhood directions of each pixel point can be obtained, and the gradients in all neighborhood directions are weighted to obtain the gradient value of each pixel point, and the accurate gradient information of each pixel point is obtained, thereby reducing the gradient distortion caused by uneven illumination; then, the diffusion function in the anisotropic filtering algorithm is constructed based on the gradient value to realize the filtering processing of the pixel points in the image, thereby obtaining a high-quality image and ensuring the accuracy of the position recognition and calibration of the printed plate.

[0017] Optionally, the gradient value for:

[0018] ; is the gradient weight of the ith pixel in the direction of the ath neighborhood, is the number of neighborhood directions, is the gradient of the ith pixel in the direction of the ath neighborhood.

[0019] The above scheme can obtain accurate gradient values ​​of pixel points by weighting the gradients of different neighborhood directions of each pixel point.

[0020] Optionally, the neighborhood directions include 0°, 45°, 90° and 135°.

[0021] Optionally, the distribution uniformity for:

[0022] ; is the number of grayscale types in the ath neighborhood direction of the ith pixel, is the gray value of the pth pixel in the window in the ath neighborhood direction of the ith pixel, is the number of pixels in the window in the ath neighborhood direction of the i-th pixel The gray value of a pixel, is the length of the window in the ath neighborhood direction.

[0023] The above scheme can characterize the distribution of grayscale values ​​in the window by calculating the distribution uniformity of each pixel point in the window in different neighborhood directions.

[0024] Optionally, the local contrast is a grayscale standard deviation or grayscale variance of the target point within a window in any neighborhood direction.

[0025] The above calculation of local contrast can analyze the brightness of the pixels in the window.

[0026] Optionally, the gradient weight for: ;in, is the ratio of the ith pixel in the ath neighborhood direction, and n is the number of neighborhood directions.

[0027] The above calculation can be simplified by normalization.

[0028] Optionally, a diffusion function of anisotropic filtering is constructed based on the gradient value to filter the target point as follows: a diffusion function is constructed based on the gradient value of each pixel point, and the diffusion function of each pixel point is applied to iteratively update the grayscale value of the corresponding pixel point in the image until the set number of iterations is reached, at which time the iteration stops and the grayscale value of each pixel point after filtering is obtained.

[0029] Optionally, the diffusion function for: ;

[0030] In the formula, is the gradient value of the i-th pixel, is the diffusion constant.

[0031] The beneficial effects of the present invention are:

[0032] The solution of the present invention performs anisotropic filtering on the vacuum printed plate image to reduce the influence of uneven illumination, and in the filtering process, applies weighted gradient to reduce the gradient distortion caused by uneven illumination, so as to improve the image filtering effect, thereby improving the accuracy of the printed plate position recognition and calibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0034] Figure 1 The flowchart of the method for identifying and calibrating the position of a vacuum printing plate based on machine vision in this embodiment is schematically shown;

[0035] Figure 2 A schematic diagram schematically shows the gradient direction of a pixel point on a texture. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0037] The scenario targeted by the present invention is to accurately align the printed plate through the automatic alignment system during the printing process of the vacuum printer. Taking the automatic alignment vacuum printer as an example, the CCD automatic alignment system is used to identify and calibrate the position of the printed plate. Specifically, the system uses a camera to capture the image of the printed plate and performs position calculation on the image to obtain the actual position of the printed plate, realize the identification and calibration of the printed plate, and ensure accurate alignment during the printing process.

[0038] However, when identifying and locating the position of a printing plate, a high-quality printing plate image is particularly important. The existing image processing generally involves filtering the collected image of the printing plate first, and then identifying and calibrating the position of the printing plate based on the filtered image.

[0039] The existing anisotropic filtering can retain the edge information of the image. The retention of edge information on the printed plate is crucial for subsequent position recognition and calibration. Therefore, in the image filtering process, the anisotropic filtering applies different diffusion functions in different directions to maintain a higher gradient value in the edge area, thereby avoiding edge blur. However, when the captured image has problems such as edge blur and loss of texture details, it is difficult for the anisotropic filtering to filter effectively, and thus it is impossible to accurately identify the position of the printed plate in the image.

[0040] Based on the above problems, the present invention provides a vacuum printing plate position identification and calibration method based on machine vision, which can effectively filter low-quality images (such as blurred edges and lost texture details in the image), and provide data support for the subsequent accurate acquisition of the position of the printing plate.

[0041] Taking the identification and calibration of the positioning holes in the circuit board as an example, the vacuum printed board position identification and calibration method based on machine vision in this embodiment is as follows: Figure 1 As shown, the following steps are included:

[0042] Step S1, collecting an image of a printed board. Specifically, the printed board is a circuit board.

[0043] In this embodiment, when collecting images of a printed plate placed on the collection platform, it is necessary to ensure that the printed plate is flat and has no deformation; and according to the position and shape of the printed plate, the angle and focal length of the camera are adjusted to ensure that the collected image is complete.

[0044] Furthermore, in this embodiment, the image of the printing plate is preprocessed to obtain a preprocessed image; wherein the preprocessing is grayscale processing.

[0045] Step S2, taking any pixel point in the image as the target point, setting different neighborhood directions with the target point as the center, obtaining the gradient weight of the target point in the corresponding neighborhood direction according to the distribution characteristics of the target point in each neighborhood direction, and weighting the gradient of the target point in each neighborhood direction to obtain the gradient value of the target point.

[0046] When collecting images of printed plates, problems such as noise, uneven illumination, reflection and scattering of the shooting equipment itself still exist. For example, changes in illumination may cause asymmetric gradient changes in different directions of the lower pixel points in the image. This uneven illumination will cause gradient distortion. When using anisotropic filtering, it is easy to cause large differences in the filtering effects of pixels in different directions, which in turn affects the filtering effect of pixels, especially for edge pixels on printed plates. These edge pixels may become blurred and edge information may be lost after filtering.

[0047] Therefore, in this embodiment, the gradient of the pixel point in each neighborhood direction is weighted to obtain a relatively realistic gradient value of each pixel point in the image.

[0048] The process of obtaining the gradient weight during weighting includes steps S21 to S24, which are specifically as follows:

[0049] Step S21, obtaining the distribution uniformity and local contrast of the target point in any neighborhood direction.

[0050] In this embodiment, any pixel point is taken as a target point to obtain the neighborhood direction of the target point. Specifically, the neighborhood direction includes 0°, 45°, 90° and 135° directions.

[0051] After obtaining the neighborhood direction of the target point, the windows in different neighborhood directions are also obtained, that is, the window is a window of a set length along the neighborhood direction starting from the target point. It should be noted that the sizes of the windows in different neighborhood directions are the same, and the difference lies only in the pixels within the window.

[0052] The window size in this embodiment is 1×11, that is, the length of the window is 11; of course, as other implementations, the set length can also be determined according to actual conditions.

[0053] The above-mentioned acquisition of different neighborhood directions is due to the fact that uneven illumination will cause the gradient of the image to differ in different directions. Regions with large illumination changes may produce pseudo gradients, interfering with the representation of the real structure. Therefore, it is necessary to obtain different neighborhood directions of the target point.

[0054] Among them, taking the i-th pixel point as the target point, the relationship between the difference in grayscale values ​​of two adjacent pixels of the target point in any neighborhood direction and the grayscale type is obtained, and the distribution uniformity of the target point in any neighborhood direction is obtained.

[0055] Specifically, the distribution uniformity of the ath neighborhood direction of the i-th pixel point It is expressed as: ;

[0056] In the formula, is the number of grayscale types in the ath neighborhood direction of the ith pixel, is the gray value of the pth pixel in the window in the ath neighborhood direction of the ith pixel, is the number of pixels in the window in the ath neighborhood direction of the i-th pixel The gray value of a pixel, is the length of the window in the ath neighborhood direction.

[0057] The ath neighborhood direction can be 0°, 45°, 90° or 135°, and the total number of neighborhood directions is 4.

[0058] in, Indicates the distance between the pth pixel in the window in the ath neighborhood direction of the i-th pixel and the pth pixel in the window in the ath neighborhood direction. The grayscale difference of pixels, It is the cumulative sum of the absolute values ​​of the grayscale differences between adjacent pixels in the a-th neighborhood direction of the ith pixel. The smaller the cumulative sum of the grayscale differences between adjacent pixels in the window in the a-th neighborhood direction of the ith pixel, and the fewer the number of grayscale types, the more uniform the grayscale distribution of the pixels in the a-th neighborhood direction of the ith pixel.

[0059] In this embodiment, the local contrast of the target point in any neighborhood direction is the standard deviation or variance of the grayscale values ​​of the pixels in the window in the corresponding neighborhood direction.

[0060] The purpose of the above-obtained distribution uniformity and local contrast is that, since the local contrast reflects the range of variation of the grayscale values ​​of pixels within a certain area in the image, if the brightness in the neighborhood changes dramatically, it means that the area may be affected by strong illumination unevenness. The distribution uniformity characterizes whether the distribution of the grayscale values ​​of pixels in a certain area in the image is uniform. Under the condition of uniform illumination, the distribution of grayscale values ​​in different areas of the image should be relatively uniform. In the case of uneven illumination, the grayscale values ​​of certain areas in the image may be concentrated in a higher or lower range, resulting in uneven grayscale distribution. Therefore, when quantifying the degree of influence of illumination unevenness in any neighborhood direction of the target point, the distribution uniformity and local contrast in the neighborhood direction can be used for side description.

[0061] Step S22, calculating the influence of the target point in any neighborhood direction according to the local contrast and distribution uniformity of the target point in any neighborhood direction.

[0062] In this embodiment, the influence of the illumination non-uniformity on the neighborhood direction of the pixel point is described according to the ratio of the local contrast of the target point in any neighborhood direction to the distribution uniformity.

[0063] Specifically, the degree of influence of illumination non-uniformity on the ath neighborhood direction of the i-th pixel is It is expressed as: ;

[0064] In the formula, is the local contrast of the ath neighborhood direction of the ith pixel, is the distribution uniformity of the a-th neighborhood direction of the i-th pixel.

[0065] Among them, the higher the local contrast in the a-th neighborhood direction of the ith pixel point and the smaller the distribution uniformity, the higher the degree of influence of uneven illumination on the a-th neighborhood direction.

[0066] Step S23, calculating the gradient consistency of the target point; and obtaining the texture expression degree based on the gradient consistency and the distribution uniformity.

[0067] Since the texture direction and the gradient direction in the image are perpendicular, if there is texture in a certain direction in the image, the gradient of the pixels in the perpendicular direction to that direction should be the same, that is, they have gradient consistency, and the distribution of the grayscale values ​​of the pixels in the texture direction where the texture exists is relatively uniform. Therefore, in this embodiment, the texture expression in the neighborhood direction of the pixel can be quantified based on the texture distribution characteristics of the pixel. For example, Figure 2 As shown in the figure, assuming that there is a texture in the horizontal direction (0° direction), then the gradient direction of each pixel on the texture is perpendicular to the horizontal direction, that is, the gradient direction is 90° ( Figure 2 The more consistent the gradients of these pixels in the vertical direction are, the higher the degree of feature expression of the texture is and the more regular the texture is.

[0068] Therefore, in this embodiment, it is necessary to obtain the gradient consistency of the target point in any neighborhood direction; specifically, the difference between the gradient amplitudes of the target point in the vertical direction of any neighborhood direction is calculated to obtain the gradient consistency.

[0069] Specifically, gradient consistency for: ; , are the sth pixel point in the window in the ath neighborhood direction of the ith pixel point, The gradient amplitude of the pixel point in the vertical direction of the ath neighborhood direction, the window starts from the ith pixel point and has a length of Window, The natural constant An exponential function with base .

[0070] At this point, the gradient consistency of the target point in all neighborhood directions can be obtained.

[0071] In this embodiment, the texture expression degree in any neighborhood direction of the target point is obtained according to the distribution uniformity and gradient consistency of the target point in any neighborhood direction.

[0072] Specifically, when a=1, and the first neighborhood direction corresponds to the 0° neighborhood direction, the texture expression of the first neighborhood direction of the i-th pixel is It is expressed as: ;

[0073] In the formula, is the distribution uniformity of the first neighborhood direction of the i-th pixel, is the gradient consistency in the first neighborhood direction of the i-th pixel.

[0074] Among them, the higher the distribution uniformity of the 0° neighborhood direction of the i-th pixel point and the higher the gradient consistency in the neighborhood direction, the higher the texture expression of the local texture feature in the 0° neighborhood direction of the i-th pixel point.

[0075] In the above embodiment, the 0° neighborhood direction of the i-th pixel is taken as an example for analysis, and the analysis methods for other neighborhood directions are the same and will not be described in detail here.

[0076] It should be noted that a is merely the serial number of the neighborhood direction; as another implementation, when a=1, it may also correspond to 45°, 90° or 135°.

[0077] Step S24, calculating the gradient weight of the target point in any neighborhood direction.

[0078] The gradient weight in this embodiment is a normalized value of the ratio of the influence degree to the texture expression degree in any neighborhood direction of the target point.

[0079] Specifically, taking the i-th pixel as the target point, the gradient weight of the i-th pixel in any neighborhood direction is: ;

[0080] in, is the ratio of the ith pixel in the ath neighborhood direction, n is the number of neighborhood directions, = , is the texture representation of the uneven illumination in the ath neighborhood direction of the i-th pixel, is the influence degree of the ath neighborhood direction of the i-th pixel;

[0081] Among them, the higher the texture expression of the local texture features of the target point in any neighborhood direction and the lower the influence of illumination non-uniformity, the greater the gradient weight in the neighborhood direction.

[0082] in It is a normalization process of weights to ensure that the sum of weights is 1. As another implementation, the above normalization method may also adopt a maximum and minimum value normalization method.

[0083] The gradient value of the target point in this embodiment is obtained by weighting the gradients in all neighborhood directions using the gradient weights of the target point in each neighborhood direction.

[0084] Specifically, taking the i-th pixel as the target point, the gradient value of the target point is:

[0085] ;

[0086] In the formula, is the gradient value of the i-th pixel, is the gradient weight of the ith pixel in the direction of the ath neighborhood, is the number of neighborhood directions, is the gradient of the ith pixel in the direction of the ath neighborhood.

[0087] The calculation method of the gradient in this embodiment includes the Sobel operator and the Prewitt operator. Since the above-mentioned calculation methods of the gradient are all existing technologies, they will not be described here in detail.

[0088] Since the main purpose of filtering the printed plate image is to reduce the influence of gradient distortion caused by illumination non-uniformity on the filtering effect, when obtaining the gradient weights in each direction, the present embodiment mainly considers illumination non-uniformity. At the same time, based on the directional characteristics of the texture on the printed plate, the gradient is weighted according to the local texture features, which helps to retain the texture details and avoid excessive smoothing in the texture area, especially the texture area that has an important influence on the local details in the image. That is, the influence of illumination non-uniformity on the pixel points in different directions and the texture features are comprehensively considered to obtain. When the illumination non-uniformity on the pixel point in this direction is lower, the gradient weight in this direction is higher, so as to reduce the influence of gradient distortion caused by illumination non-uniformity on the final gradient of the pixel point. The more obvious the texture features of the pixel point in this direction are, the higher the gradient weight in this direction is, so as to capture these details more accurately, which is conducive to the accuracy of the diffusion function obtained based on the pixel point gradient, and further conducive to the improvement of the filtering effect.

[0089] Step S3, constructing a diffusion function of anisotropic filtering based on the gradient value to filter the target point, and finally obtaining a filtered image; and performing position recognition and calibration on the filtered image.

[0090] In this embodiment, the specific process of filtering the target point is: constructing a diffusion function based on the gradient value of each pixel point, applying the diffusion function of each pixel point to iteratively update the grayscale value of the corresponding pixel point in the image until the set number of iterations is reached, and the iteration stops. At this time, the final grayscale value of the pixel point is the grayscale value of the pixel point after filtering, so as to realize image filtering.

[0091] In this embodiment, only the gradient in the diffusion function in the anisotropic filtering is improved, and other steps are not improved. Therefore, since the specific steps of the anisotropic filtering are existing technologies, they will not be described in detail here.

[0092] The above iteration number can be 5, 10 or more times.

[0093] In this embodiment, the diffusion constant is set , according to the above steps, the gradient value of the pixel is obtained and the diffusion function of the pixel is calculated.

[0094] Among them, the diffusion function of the i-th pixel is for:

[0095] ; is the gradient value of the i-th pixel, is the diffusion constant.

[0096] The diffusion constant is Of course, it can also be obtained according to the actual situation, usually selected empirically or expressed by a function related to image noise. If the absolute value of the gradient value is much larger than the diffusion constant, that is , the value of the diffusion function tends to 0, and the diffusion is inhibited; conversely, the value of the diffusion function tends to 1, and the diffusion is enhanced.

[0097] After acquiring the filtered image of the printing plate, the position of the printing plate is identified and calibrated based on the filtered image.

[0098] Since the specific method for position recognition and calibration of the filtered image of the printed plate is a prior art, it will not be described in detail here.

[0099] In this embodiment, the diffusion function of the pixel point when filtering is obtained according to the weighted gradient value and the diffusion constant, instead of applying different diffusion functions in different directions for filtering in traditional anisotropic filtering. This can not only avoid gradient distortion caused by uneven illumination, but also improve the filtering efficiency of anisotropic filtering in processing printed images (because anisotropic filtering itself is an iterative operation, and each pixel point needs to apply different diffusion functions according to the gradient in different directions. Although the algorithm itself can better retain the image edge, the amount of calculation is large).

[0100] The solution of the present invention performs anisotropic filtering on the vacuum printed plate image to reduce the influence of uneven illumination, thereby improving the accuracy of subsequent position identification and calibration of the printed plate.

[0101] In the description of this specification, “plurality” means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0102] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A vacuum printing plate position recognition and calibration method based on machine vision, characterized in that: include: acquiring an image of the printed plate; Take any pixel point in the image as the target point and obtain the gradient value of the target point; A diffusion function of anisotropic filtering is constructed based on the gradient value to filter the target point and obtain a filtered image; Diffusion function for: ; In the formula, is the gradient value of the i-th pixel, is the diffusion constant; Performing position recognition and calibration of the printed plate on the filtered image; The gradient value is obtained by weighting the gradient of the target point in different neighborhood directions; wherein the gradient weight during weighting is the value after the contrast value is normalized; the ratio is the ratio of the texture expression degree and the influence degree of the target point in any neighborhood direction; the texture expression degree is the product of the distribution uniformity and the gradient consistency; the influence degree is the ratio of the local contrast to the distribution uniformity; The distribution uniformity represents the distribution of gray values; the local contrast represents the fluctuation of gray values; the gradient consistency for: ; , are the ath neighborhood direction of the ith pixel in the window. pixels, The gradient amplitude of the pixel point in the vertical direction of the ath neighborhood direction, the window is the starting point of the ith pixel point, and the length along the neighborhood direction is Window, The natural constant An exponential function with base .

2. The method for position identification and calibration of vacuum printed sheet material based on machine vision according to claim 1, characterized in that: The gradient value for: ; is the gradient weight of the ith pixel in the direction of the ath neighborhood, is the number of neighborhood directions, is the gradient of the ith pixel in the direction of the ath neighborhood.

3. The method for position identification and calibration of vacuum printed sheet material based on machine vision according to claim 1 or 2, characterized in that: The neighborhood directions include 0°, 45°, 90° and 135°.

4. The method for position identification and calibration of vacuum printed sheet material based on machine vision according to claim 3, characterized in that: The distribution uniformity for: ; is the number of grayscale types in the ath neighborhood direction of the ith pixel, is the gray value of the pth pixel in the window in the ath neighborhood direction of the ith pixel, is the number of pixels in the window in the ath neighborhood direction of the i-th pixel The gray value of a pixel, is the length of the window in the ath neighborhood direction.

5. The method for position identification and calibration of vacuum printed sheet material based on machine vision according to claim 4, characterized in that: The local contrast is the grayscale standard deviation or grayscale variance of the target point in a window in any neighborhood direction.

6. The method for position identification and calibration of vacuum printed sheet material based on machine vision according to claim 3, characterized in that: The gradient weight for: ;in, is the ratio of the ith pixel in the ath neighborhood direction, and n is the number of neighborhood directions.

7. The method for position identification and calibration of vacuum printed sheet material based on machine vision according to claim 1, characterized in that: The diffusion function of anisotropic filtering is constructed based on the gradient value to filter the target point: A diffusion function is constructed based on the gradient value of each pixel, and the diffusion function of each pixel is applied to iteratively update the grayscale value of the corresponding pixel in the image until the set number of iterations is reached, at which time the iteration stops and the filtered grayscale value of each pixel is obtained.

Images