A wafer character quality detection method, device, computer equipment and storage medium

Abstract

The present invention discloses a wafer character quality detection method, device, computer equipment and storage medium, belonging to the field of semiconductor detection technology. By acquiring the template feature point coordinates in the template image, the detection feature point coordinates of the detection image are obtained according to the template feature point coordinates; the detection image is registered by the template feature point coordinates and the detection feature point coordinates to obtain the registered image; the registered image and the template image are XOR-processed to obtain the difference point image, the bad points are found in the difference point image, and the scores of all the bad points are calculated; the registered image and the template image are processed for over-printing statistics and under-printing statistics to obtain the over-printing and under-printing scores; the final similarity score is obtained by calculating the bad point score and the over-printing and under-printing score; and the character quality in the detection image is judged according to the final similarity score. As a result, the efficiency of character quality detection is significantly improved, which helps to promote the equipment to realize automated detection and improve product quality, and has strong practicality and wide applicability.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor detection technology, and more specifically, to a wafer character quality detection method, device, computer equipment and storage medium. Background Art

[0002] Each character on the wafer reflects the unique sample serial number of each wafer, which is composed of numbers and letters. The serial number is made through a special process, which can achieve the purpose of product anti-counterfeiting on the one hand, and on the other hand, when a product has defects, the production time and production process can be determined through the serial number to achieve the purpose of product quality traceability.

[0003] Currently, the printing of product serial numbers is carried out on the product production line. Although this can improve chip production efficiency, the complex environment on the production line will reduce the quality of character printing, which may lead to character defects, such as dirty characters, missing characters, blurred character printing, etc. In order to improve the quality and stability of chips and prevent poor quality chips from entering the market, product quality inspections are carried out before the chips leave the factory. Therefore, using a wafer character quality inspection system based on machine vision to inspect the quality of sample serial number characters before the chips leave the factory is of great significance to improving chip quality and helping production lines improve production processes.

[0004] In the prior art, quality inspection of wafer characters usually involves extracting each character on the wafer and matching them one by one with the characters on the template image, and then calculating the character similarity. This requires a large amount of data calculation and has low inspection efficiency. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] In view of the problem of low efficiency of wafer character quality detection in the prior art, the present invention provides a wafer character quality detection method, device, computer equipment and storage medium, which judge the chip character quality by calculating the similarity score of wafer character contour points based on machine vision.

[0007] 2. Technical solution

[0008] The purpose of the present invention is achieved through the following technical solutions.

[0009] A wafer character quality detection method comprises the following steps:

[0010] Collecting and inputting a wafer character image, wherein the wafer character image includes a template image and a detection image;

[0011] Obtaining the coordinates of the template feature points in the template image, and obtaining the coordinates of the detection feature points of the detection image according to the coordinates of the template feature points;

[0012] The detection image is registered by using the template feature point coordinates and the detection feature point coordinates to obtain a registered image;

[0013] Perform XOR processing on the registration image and the template image to obtain a difference point image, find bad pixels in the difference point image, and calculate the scores of all bad pixels;

[0014] Perform over-printing and under-printing statistics processing on the registered image and the template image to obtain over-printing and under-printing scores;

[0015] The final similarity score is obtained by calculating the bad pixel score and the overprinting and underprinting score, and whether the quality of the characters in the inspection image is qualified is judged according to the final similarity score.

[0016] Furthermore, the obtaining of the coordinates of the template feature points in the template image includes: converting the template image into a grayscale image; obtaining the contour edge points of the grayscale image; screening the contour edge points to obtain the template feature points, and obtaining the coordinates of the template feature points.

[0017] Further, obtaining the detection feature point coordinates of the detection image according to the template feature point coordinates includes:

[0018] Define function I (x, y, t), where x represents the horizontal coordinate of the feature point pixel, y represents the vertical coordinate of the feature point pixel, t represents the frame number, and function I (x, y, t) represents the pixel intensity at frame t;

[0019] Set the template image to the image at frame t0 and the detection image to the image at frame t1;

[0020] If lim(dt→0), the displacement does not change the pixel intensity of the grayscale image, then the pixel intensity of the detection feature point coordinates of the detection image is expressed as:

[0021]

[0022] Among them, dx represents the displacement in the horizontal direction, dy represents the displacement in the vertical direction, and dt represents the interval time.

[0023] Furthermore, the method of registering the detection image by using the template feature point coordinates and the detection feature point coordinates to obtain the registered image includes: obtaining the affine transformation matrix by the correspondence between the template feature point coordinates and the detection feature point coordinates, and multiplying each detection feature point in the detection image by the affine transformation matrix to obtain the registered image.

[0024] Furthermore, a dedicated threshold is set to perform binarization on the detected image and then extract the characters in the registered image. The calculation formula is:

[0025]

[0026] Among them, Tfinal represents the exclusive threshold, T otsu represents the global threshold, T part represents the local threshold, α and β represent weight factors, and T1 and T2 represent constraints.

[0027] Furthermore, the step of searching for bad pixels in the difference point image and calculating the scores of all bad pixels includes:

[0028] Take the template image feature point as the anchor point, traverse all feature points, take each anchor point as the center, select the outermost layer of pixels in the window within the n×n local window range of the position on the difference point image for continuity judgment, set the threshold, if the number of continuous difference points counted is greater than the threshold, the difference point at the current position is a bad point, count all bad points, calculate all bad point scores, the calculation formula is:

[0029]

[0030] Among them, Blod score Indicates the bad pixel score , represents the statistics of the number of non-zero points in a binary image, Indicates the number of bad pixels. Indicates the number of feature points.

[0031] Furthermore, the processing of over-printing statistics and under-printing statistics on the registered image and the template image includes: performing XOR operation on the registered image and the template image to obtain an XOR image, and then performing OR operation on the XOR image and the template image to obtain an OR image; performing AND operation on the XOR image and the template image to obtain an AND image;

[0032] Calculate the over-printing and under-printing scores using the following formula:

[0033]

[0034] Among them, Print score Indicates the value of printing more or less, Or ing Representation or image, And ing Representation and image, Temp ing Represents the template binary image;

[0035] The final similarity score is obtained by the bad pixel score and the over-printing or under-printing score. The calculation formula is:

[0036]

[0037] Among them, Final score Represents the final similarity score.

[0038] A wafer character quality detection device, comprising:

[0039] An input module collects and inputs a wafer character image, wherein the wafer character image includes a template image and a detection image;

[0040] An acquisition module is used to acquire the coordinates of the template feature points in the template image, and obtain the coordinates of the detection feature points of the detection image according to the coordinates of the template feature points;

[0041] A registration module, which registers the detection image by using the template feature point coordinates and the detection feature point coordinates to obtain a registered image;

[0042] The calculation module performs XOR processing on the registration image and the template image to obtain a difference point image, finds bad pixels in the difference point image, and calculates the scores of all bad pixels; performs over-printing statistics and under-printing statistics processing on the registration image and the template image to obtain over-printing and under-printing scores;

[0043] The judgment module obtains a final similarity score by calculating the bad pixel score and the overprinting / underprinting score; and judges whether the quality of the characters in the detection image is qualified according to the final similarity score.

[0044] A computer device comprises a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor implements any of the above-mentioned methods when executing the computer program.

[0045] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any of the above methods is executed.

[0046] 3. Beneficial effects

[0047] Compared with the prior art, the advantages of the present invention are:

[0048] The present invention provides a wafer character quality detection method, device, computer equipment and storage medium. By selecting each character contour point in the wafer and uniformly sampling each character contour point, the amount of similarity calculation data can be effectively reduced; the detection image is registered by the template feature point coordinates and the detection feature point coordinates to obtain the registered image, so that the registered image and the template image are in a one-to-one correspondence, so that the characters on the detection image can be directly extracted by the position of the characters on the template image and then compared with the characters at the corresponding positions of the template image for similarity, which significantly reduces the amount of calculated data and effectively improves the detection efficiency; at the same time, in the registered image, the local threshold and the global threshold are combined for multi-angle judgment, so that an exclusive threshold can be calculated for each pixel point, so as to extract more complete characters; the character quality is judged by the final similarity score, so that the product quality can be quickly screened out, which is helpful to promote the equipment to realize automatic detection and improve product quality, and has strong practicality and wide applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flow chart of a method according to an embodiment of the present invention;

[0050] Figure 2 A random selection graph of feature points in an embodiment of the present invention;

[0051] Figure 3 It is a uniform selection map of feature points in an embodiment of the present invention;

[0052] Figure 4 This is a distribution diagram of feature points of a template image according to an embodiment of the present invention;

[0053] Figure 5 A distribution map of feature points of an image to be registered in accordance with an embodiment of the present invention;

[0054] Figure 6 The present invention is to extract wafer character effect diagrams by different methods in the embodiments;

[0055] Figure 7 This is a bad pixel image according to an embodiment of the present invention;

[0056] Figure 8 The wafer character image of the embodiment of the present invention;

[0057] Fig. 9 This is the first test result image of the embodiment of the present invention;

[0058] Fig.10 This is the second test result image of the embodiment of the present invention. DETAILED DESCRIPTION

[0059] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0060] Example

[0061] like Figure 1 As shown, a wafer character quality detection method provided by the present embodiment includes the following steps: collecting and inputting a wafer character image, the wafer character image includes a template image and a detection image; obtaining the coordinates of the template feature points in the template image, and obtaining the detection feature point coordinates of the detection image according to the template feature point coordinates; aligning the detection image through the template feature point coordinates and the detection feature point coordinates to obtain a registered image; performing XOR processing on the registered image and the template image to obtain a difference point image, searching for bad points in the difference point image, and calculating the scores of all bad points; performing over-printing statistical processing and under-printing statistical processing on the registered image and the template image to obtain over-printing and under-printing scores; obtaining a final similarity score by calculating the bad point score and the over-printing and under-printing score; and judging whether the character quality in the detection image is qualified according to the final similarity score.

[0062] Specifically in this embodiment, first, a wafer character image is collected and input, and the wafer character image includes a template image and a detection image.

[0063] Furthermore, the coordinates of the template feature points are obtained in the template image. In this embodiment, the template feature points need to be uniformly selected in the template image so that the shape of the characters in the template image is more representative. Figure 2 As shown in , if the feature points are randomly selected, the extracted character shape will be incomplete. Figure 3 As shown, after the feature points are uniformly selected, the extracted character shape is more complete and more representative. In this embodiment, specifically, after the template image is converted into a grayscale image, the outermost contour edge points of each character in the grayscale image are obtained, and the outermost contour edge points of the characters are used as character feature points. The character feature points are then screened, and the screened character feature points are used as template feature points, and then the template feature point coordinates are obtained. It should be noted that in this embodiment, the number of selected template feature points is set to m, and the initial sampling step length is the total number of character feature points / m. Therefore, in order to make the obtained template feature points as evenly distributed as possible, the sampling step length decreases or increases as the number of character feature points finally screened is greater than or less than the set number m. Therefore, in this embodiment, by selecting each character contour point in the wafer and uniformly sampling each character contour point, the amount of similarity calculation data can be effectively reduced.

[0064] Furthermore, the detection feature point coordinates of the detection image are obtained according to the template feature point coordinates. Specifically, the wafer character image is converted into a grayscale image, and the function I (x, y, t) is defined, where x represents the horizontal coordinate of the feature point pixel, y represents the vertical coordinate of the feature point pixel, and t represents the number of frames. Thus, the function I (x, y, t) represents the pixel intensity at the t frame. In this embodiment, the template image is set to the image at the t0 frame, and the detection image is set to the image at the t1 frame. In a very short time, that is, lim(dt→0), the displacement does not change the pixel intensity of the grayscale image, and the pixel intensity of the detection feature point coordinates of the detection image is expressed as:

[0065]

[0066] Among them, dx represents the displacement in the horizontal direction, dy represents the displacement in the vertical direction, and dt represents the interval time.

[0067] Will Performing Taylor expansion at I (x, y, t) yields:

[0068]

[0069]

[0070] Among them, ε represents the second-order infinitesimal term, which can be ignored. , , whereby u and v represent the motion velocity vectors along the x-axis and y-axis directions. In this embodiment, the constraint condition-based gradient method (Horn-Schunck algorithm) can be used to approximately calculate the velocity vectors u and v of the pixel, and then we can obtain: dx=u×dt, dy=v×dt. Therefore, in this embodiment, given the template feature coordinates in the template image I (x, y, t), we can obtain the detection feature point coordinates I of the detection image. t (x+dx,y+dy,t+dt).

[0071] Furthermore, the detected image is registered by the template feature point coordinates and the detected feature point coordinates to obtain the registered image. Specifically, the affine transformation matrix is ​​obtained by the correspondence between the template feature point coordinates and the detected feature point coordinates. Specifically, the template feature point coordinates are used as the dependent variable, and the detected image feature point coordinates are used as the independent variable. Then, a set of equations is established according to the number of corresponding points, and finally, the multi-equation system equation is obtained to obtain the affine transformation matrix. Each detected feature point on the detected image is multiplied by the affine transformation matrix to obtain the registered image. Therefore, in this embodiment, Figure 4 As shown in, it is the distribution diagram of the feature points of the template image; Figure 5 As shown, the registration image and the template image are in one-to-one correspondence, and the characters on the detection image can be directly extracted by the position of the characters on the template image. After obtaining the registration image, a dedicated threshold is set to perform binarization processing on the registration image, and then the characters in the registration image are extracted. In the prior art, if the brightness of the characters is uneven, extracting the characters by using a fixed threshold or OTSU will result in incomplete character extraction. Figure 6 As shown in the figure, the effect diagram of wafer character extraction by different methods is shown in the figure. Figure 6 As shown in a in , select a wafer character original image; Figure 6 As shown in b in FIG, the wafer character original image is binarized by a fixed threshold, and the extracted characters are incomplete; Figure 6 As shown in c in the figure, the wafer character original image is binarized by OTSU, and the extracted characters are incomplete; Figure 6 As shown in d, in this embodiment, after the registration image is binarized by setting a dedicated threshold, a more complete character can be extracted. Specifically, the gray value at the pixel point (x, y) of the registration image is set to g(x, y), and a local threshold is set. The local threshold is determined by the gray value of the pixel in an r×r window. The local threshold calculation formula is as follows:

[0072]

[0073] Among them, Tpart represents the local threshold, m(x,y) represents the mean grayscale value of pixels in the r×r window, s(x,y) represents the variance of the grayscale value of pixels in the r×r window, k represents the correction factor, R represents the fixed value of the pixel. In this embodiment, R is 128 pixels, and g(i,j) represents the pixel intensity value at any point in the image.

[0074] In this embodiment, in order to reduce the generation of pseudo noise in the background area, the global threshold T is obtained by using the existing maximum between-class variance (OTSU) otsu , assisting in distinguishing targets from backgrounds.

[0075] In this embodiment, in order to further improve the algorithm's anti-noise ability, constraint parameters T1 and T2 are introduced. Specifically, the grayscale mean m1 of the entire grayscale image is calculated, and grayscale values ​​less than m1 are defined as m1 class, and grayscale values ​​greater than or equal to m1 are defined as m2 class. The m1 class mean is recorded as T1, and the m2 class mean is recorded as T2.

[0076] Therefore, the global threshold T otsu With the local threshold T part Weighted, and under the constraints T1 and T2, get the exclusive threshold for each feature point, and the final threshold is expressed as:

[0077]

[0078] Among them, T final represents the exclusive threshold, T otsu represents the global threshold, T part Represents the local threshold, α and β both represent weight factors, where α+β=1, and the larger the α value, the greater the contribution of the global threshold to the exclusive threshold, and vice versa. T1 and T2 both represent constraint conditions. Therefore, in the present embodiment, in the registered image, the local threshold and the global threshold are combined for multi-angle judgment, so that an exclusive threshold can be calculated for each pixel point, thereby extracting more complete characters.

[0079] Furthermore, the registration image and the template image are XORed to obtain a difference point image, bad points are found in the difference point image, and the scores of all bad points are calculated. Specifically, the feature points of the template image are used as anchor points, and all difference points are traversed. It should be noted that in this embodiment, Figure 7As shown, the difference point refers to a white point with a grayscale value of 255 pixels. Then, taking each anchor point as the center, within the n×n local window range of the position on the difference point image, select the outermost layer of pixels of the window for continuity judgment, set a threshold, and if the number of consecutive difference points counted is greater than the threshold, the difference point at the current position is a bad point, count all the bad points, and calculate the bad point score. For example, at the 5×5 local window range on the difference point image, in the outermost 25 difference points, set the threshold to 4, if the number of consecutive difference points counted is greater than 4, the difference point at the current position is a bad point, and count all the bad points. In this embodiment, the bad point score calculation formula is:

[0080]

[0081] Among them, Blod score Indicates the bad pixel score, Represents the statistics of the number of non-zero points in a binary image. Indicates the number of bad pixels. Indicates the number of feature points.

[0082] Furthermore, the registration image and the template image are processed for over-printing statistics and under-printing statistics to obtain over-printing and under-printing scores. Specifically, the detection image and the template image are XORed to obtain an XOR image, and then the XOR image and the template image are processed by OR to obtain an OR image; the XOR image and the template image are processed by AND to obtain an AND image; the over-printing and under-printing scores are calculated, and the calculation formula is:

[0083]

[0084] Among them, Print score Indicates the value of printing more or less. ing Representation or image, And ing Representation and image, Temp ing Represents the template binary image;

[0085] The final similarity score is calculated by the bad pixel score and the over-printing or under-printing score. The calculation formula is:

[0086]

[0087] Among them, Final score Represents the final similarity score. Finally, the quality of the characters in the detected image is judged based on the final score.

[0088] It should be noted that, in this embodiment, in order to limit the final score to the range of [0,1], the restriction formula is designed as follows:

[0089]

[0090] Among them, (a,b) represents the range of the value range limit interval, max(a,b) represents the upper limit of the specified range, and the upper limit of the specified range is set to α; min(a,b) represents the lower limit of the specified range, and the lower limit of the specified range is set to β.

[0091] If min(num,α), it means that only when the specified value exceeds the upper limit of the range, the return value is α, otherwise it returns num. If min(a,b), it means the specified lower limit β is returned. If num exceeds the upper limit of the range, the return value is max(α,β). At this time, the upper limit α of the range is returned, which is the value closest to num in the entire value range. If num does not exceed the upper limit of the range, the return value is max(num,β). At this time, if num is within the value range, it will be greater than the lower limit β of the range, and num will be returned. If num is less than the lower limit β of the range, β is returned, which is the value closest to num in the entire value range.

[0092] Furthermore, in this embodiment, the restriction of Blod score The value range is [0.5,1], Print score The value range is [0,0.5].

[0093] like Figure 8 As shown, a wafer character image is selected as a template image, and a wafer character quality detection method provided in this embodiment is used to perform character quality detection. Fig. 9 As shown, after the test image and the template image are registered, each character in the test image corresponds to each character in the template image one by one. When there are no defects in the characters in the test image, it can be concluded through detection that the similarity score between each character in the test image and each character in the template image is high. Fig.10 As shown, when a character in the test image has a defect or stain, it can be concluded through detection that the similarity score between the defect or stain character in the test image and the corresponding character in the template image is low.

[0094] Therefore, the wafer character quality detection method provided in this embodiment can quickly screen whether the product quality is qualified by judging the character quality through the final similarity score, which helps to promote the equipment to realize automated detection and improve product quality. It has strong practicality and wide applicability.

[0095] The present embodiment also provides a wafer character quality detection device, including an input module, an acquisition module, a registration module, a calculation module and a judgment module. The input module is used to collect and input a wafer character image, and the wafer character image includes a template image and a detection image; the acquisition module is used to obtain the coordinates of the template feature points in the template image, and obtain the detection feature point coordinates of the detection image according to the template feature point coordinates; the registration module registers the detection image through the template feature point coordinates and the detection feature point coordinates to obtain the registration image; the calculation module is used to perform XOR processing on the registration image and the template image to obtain a difference point image, find bad points in the difference point image, and calculate all bad point scores; perform over-printing statistics and under-printing statistics processing on the registration image and the template image to obtain over-printing and under-printing scores; the judgment module obtains the final similarity score by calculating the bad point score and the over-printing and under-printing score, and judges whether the character quality in the detection image is qualified according to the final similarity score. The wafer character quality detection device provided in the present embodiment can implement any of the wafer character quality detection methods, and the specific working process of a wafer character quality detection device can refer to the corresponding process in the embodiment of the wafer character quality detection method. The method and device provided in this embodiment can be implemented in other ways. For example, the device embodiment described above is only schematic; for example, the division of a module is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual connection or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or it can be an electrical, mechanical or other form of connection.

[0096] This embodiment also provides a computer device. A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the wafer character quality detection method when executing the computer program.

[0097] This embodiment also provides a computer-readable storage medium. A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, a wafer character quality detection method described in this embodiment is executed. Among them, the computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, device or device; the program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0098] The above schematically describes the invention and its implementation methods, which is not restrictive. Without departing from the spirit or basic features of the invention, the invention can be implemented in other specific forms. What is shown in the accompanying drawings is only one of the implementation methods of the invention. The actual structure is not limited thereto, and any figure mark in the claims should not limit the claims involved. Therefore, if a person of ordinary skill in the art is inspired by it, without departing from the purpose of the invention, a structural method and an embodiment similar to the technical solution are designed without creativity, which should all belong to the protection scope of the present invention. In addition, the word "including" does not exclude other elements or steps, and the word "one" before the element does not exclude the inclusion of "multiple" elements. The multiple elements stated in the product claim can also be implemented by one element through software or hardware. The words first, second, etc. are used to indicate the name, and do not indicate any specific order.

Claims

1. A wafer character quality detection method, comprising the following steps: Collecting and inputting a wafer character image, wherein the wafer character image includes a template image and a detection image; Obtaining the coordinates of the template feature points in the template image, and obtaining the coordinates of the detection feature points of the detection image according to the coordinates of the template feature points; The detected image is registered by the template feature point coordinates and the detected feature point coordinates to obtain the registered image; the characters in the registered image are extracted after the detection image is binarized by setting a dedicated threshold. The calculation formula is: Among them, T final represents the exclusive threshold, T otsu represents the global threshold, T part represents the local threshold, α and β represent weight factors, and T1 and T2 represent constraint conditions; Perform XOR processing on the registration image and the template image to obtain a difference point image, find bad points in the difference point image, and calculate the scores of all bad points; including: taking the feature point of the template image as the anchor point, traversing all feature points, taking each anchor point as the center, within the n×n local window range of the position on the difference point image, select the outermost layer of pixels of the window for continuity judgment, set a threshold, if the number of continuous difference points counted is greater than the threshold, the difference point at the current position is a bad point, count all bad points, and calculate the scores of all bad points. The calculation formula is: Among them, Blod score Indicates the bad pixel score, Represents the statistics of the number of non-zero points in a binary image. Indicates the number of bad pixels. Indicates the number of feature points; Performing over-printing statistics and under-printing statistics processing on the registered image and the template image to obtain over-printing and under-printing scores; including: performing XOR processing on the registered image and the template image to obtain an XOR image, then performing an OR processing on the XOR image and the template image to obtain an OR image; performing an AND operation on the XOR image and the template image to obtain an AND image; calculating the over-printing and under-printing scores, the calculation formula is: Among them, Print score Indicates the value of printing more or less, Or ing Representation or image, And ing Representation and image, Temp ing Represents the template binary image; The final similarity score is obtained by calculating the bad pixel score and the over-printing and under-printing score. The calculation formula is: Among them, Final score Represents the final similarity score; The final similarity score is used to determine whether the quality of the characters in the detected image is qualified.

2. A wafer character quality detection method according to claim 1, characterized in that: The method of obtaining the coordinates of the template feature points in the template image includes: converting the template image into a grayscale image; obtaining the contour edge points of the grayscale image; screening the contour edge points to obtain the template feature points, and obtaining the coordinates of the template feature points.

3. A wafer character quality detection method according to claim 2, characterized in that: The step of obtaining the detection feature point coordinates of the detection image according to the template feature point coordinates includes: Define function I (x, y, t), where x represents the horizontal coordinate of the feature point pixel, y represents the vertical coordinate of the feature point pixel, t represents the frame number, and function I (x, y, t) represents the pixel intensity at frame t; Set the template image to the image at frame t0 and the detection image to the image at frame t1; If lim(dt→0), the displacement does not change the pixel intensity of the grayscale image, then the pixel intensity of the detection feature point coordinates of the detection image is expressed as: Among them, dx represents the displacement in the horizontal direction, dy represents the displacement in the vertical direction, and dt represents the interval time.

4. A wafer character quality detection method according to claim 3, characterized in that: The method of registering the detection image by using the template feature point coordinates and the detection feature point coordinates to obtain the registered image includes: obtaining an affine transformation matrix by the correspondence between the template feature point coordinates and the detection feature point coordinates, and multiplying each detection feature point in the detection image by the affine transformation matrix to obtain the registered image.

5. A wafer character quality detection device, characterized in that: include: An input module collects and inputs a wafer character image, wherein the wafer character image includes a template image and a detection image; An acquisition module is used to acquire the coordinates of the template feature points in the template image, and obtain the coordinates of the detection feature points of the detection image according to the coordinates of the template feature points; The registration module registers the detection image through the template feature point coordinates and the detection feature point coordinates to obtain the registered image; sets a dedicated threshold to perform binarization on the detection image and then extracts the characters in the registered image. The calculation formula is: Among them, T final represents the exclusive threshold, T otsu represents the global threshold, T part represents the local threshold, α and β represent weight factors, and T1 and T2 represent constraint conditions; The calculation module performs XOR processing on the registration image and the template image to obtain a difference point image, searches for bad points in the difference point image, and calculates the scores of all bad points; including: taking the feature point of the template image as the anchor point, traversing all feature points, taking each anchor point as the center, selecting the outermost layer of pixels in the window within the n×n local window range at the position on the difference point image for continuity judgment, setting a threshold, if the number of continuous difference points counted is greater than the threshold, the difference point at the current position is a bad point, counting the number of all bad points, and calculating the scores of all bad points. The calculation formula is: Among them, Blod score Indicates the bad pixel score, Represents the statistics of the number of non-zero points in a binary image. Indicates the number of bad pixels. Indicates the number of feature points; Performing over-printing statistics and under-printing statistics processing on the registered image and the template image to obtain over-printing and under-printing scores; including: performing XOR processing on the registered image and the template image to obtain an XOR image, then performing an OR processing on the XOR image and the template image to obtain an OR image; performing an AND operation on the XOR image and the template image to obtain an AND image; calculating the over-printing and under-printing scores, the calculation formula is: Among them, Print score Indicates the value of printing more or less, Or ing Representation or image, And ing Representation and image, Temp ing Represents the template binary image; The judgment module calculates the bad pixel score and the over-printing or under-printing score to obtain the final similarity score; the calculation formula is: Among them, Final score Represents the final similarity score; The final similarity score is used to determine whether the quality of the characters in the detected image is qualified.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 4 is executed.

Images