An optical detection method without standard sample and its application

Abstract

The present invention provides an optical detection method without standard samples, including: S1. Continuously obtain the graphic information of the target product; S2. Perform graying processing on the obtained graphic information of the target product; S3. Obtain the reference point of each grayed graphic by reference point calculation; S4. Obtain the similarities and differences among the grayed graphics of at least N consecutive target products after superposition based on the reference points; S5. Construct a virtual standard sample based on the similarities in step S4, and use the virtual standard sample as the standard sample for subsequent target product grayed graphic comparison detection, and record the results of each target product compared with the virtual standard sample. The present invention does not need to make standard samples, but directly uses the virtual standard sample as a reference sample through the method of forming a virtual standard sample, saving the cost of making standard samples.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical detection, and in particular relates to an optical detection method without the need for a standard sample and an application thereof. Background Art

[0002] Traditional AOI (Automated Optical Inspection) optical inspection technology is widely used for the inspection of lines, blocks, and electronic components. During the operation, standard samples must be pre-loaded and compared with the photos to be tested. The difference is the abnormal point (Defect) of the tested photo. However, due to the continuous innovation of technology, the production of small quantities of diverse and variable products is becoming more and more popular. If the traditional AOI optical inspection technology is continued to be used, a standard film will be required for each group of products, which will lead to the continuous increase in the cost of traditional AOI. Summary of the invention

[0003] The present invention aims to solve the problems existing in the prior art and provides an optical detection method without the need for a standard sample, comprising:

[0004] S1. Continuously obtain graphic information of the target product;

[0005] S2. graying the acquired graphic information of the target product;

[0006] S3. Obtain the reference point of each grayed-out graphic by calculating the reference point;

[0007] S4. After superimposing at least N consecutive gray images of the target product based on the reference point, similarities and differences are obtained;

[0008] S5. Construct a virtual standard sample based on the similar points in step S4, and use the virtual standard sample as a standard sample for subsequent target product graying pattern comparison detection, and record the comparison results of each target product with the virtual standard sample.

[0009] Furthermore, the ashing method in step S2 includes:

[0010] First, obtain the three primary color values ​​of each pixel in the graphic data, and use the XY composition system to express it as follows:

[0011] [[B11,G11,R11][B12,G12,R12][B13,G13,R13]...[B1n,G1n,R1n]]

[0012] [[B21,G21,R21][B22,G22,R22][B23,G23,R23]...[B2n,G2n,R2n]]

[0013] [[B31,G31,R31][B32,G32,R32][B33,G33,R33]...[B3n,G3n,R3n]] ...

[0015] [[Bm1,Gm1,Rm1][Bm2,Gm2,Rm2][Bm3,Gm3,Rm3]...[Bmn,Gmn,Rmn]];

[0016] Then, the following numerical conversion is performed on the three primary color values ​​of each pixel:

[0017] [Max(B11,G11,R11)-Min(B11,G11,R11)][Max(B12,G12,R12)-Min(B12,G12,R12)]...[Max(B1n,G1n,R1n)-Min(B1n,G1n,R1n)]

[0018] [Max(B21,G21,R21)-Min(B21,G21,R21)][Max(B22,G22,R22)-Min(B22,G22,R22)]...[Max(B2n,G2n,R2n)-Min(B2n,G2n,R2n)] ...

[0020] [Max(Bm1,Gm1,Rm1)-Min(Bm1,Gm1,Rm1)][Max(Bm2,Gm2,Rm2)-Min(Bm2,Gm2,Rm2)]...[Max(Bmn,Gmn,Rmn)-Min(Bmn,Gmn,Rmn)];

[0021] After the three primary color values ​​of all the pixels are converted, the graying process is completed; the three primary color values ​​of each pixel are taken as integers and classified into a value range of 0-255.

[0022] Furthermore, the method for obtaining the reference point of each grayed figure in step S3 includes: first, taking the median of each horizontal row in the dot matrix after the three primary color values ​​are converted to form a longitudinal reference point, and taking the median of each vertical row in the dot matrix after the three primary color values ​​are converted to form a transverse reference point; then, taking the intersection of the longitudinal reference point and the transverse reference point, and marking it as the reference point of the grayed figure.

[0023] Furthermore, the method for obtaining similarities and differences in step S4 includes: first, accumulating the X-axes of N consecutive target products into a y-axis transformation curve, and using FFD Fourier transform to find the high-frequency repetition pattern and the low-frequency repetition pattern of the y-axis; similarly, accumulating the Y-axes of N consecutive target products into an x-axis transformation curve, and using FFD Fourier transform to find the high-frequency repetition pattern and the low-frequency repetition pattern of the x-axis; then defining the high-frequency repetition area of ​​the x-axis and the y-axis as similarities, and defining the low-frequency repetition area of ​​the x-axis and the y-axis as differences.

[0024] Furthermore, the method for comparing and detecting the graying pattern of the target product in step S5 includes:

[0025] (1) Calculate the numerical value A of each pixel point in the data column of the grayed image of the target product and the numerical value B of each pixel point in the virtual standard sample: C=AB. When the calculated value of C is within the preset threshold range, the numerical value A is considered to be a normal point; when the calculated value of C is outside the preset threshold range, the numerical value A is considered to be a suspected abnormal point and is marked with a characteristic color; when the number of suspected abnormal points exceeds the preset threshold, all suspected abnormal points are marked with characteristic colors and the product is eliminated;

[0026] (2) The data of the target product graying pattern is superimposed on the virtual standard sample and then linear Fourier transform is performed, where the target product graying pattern area below the preset frequency is the suspected abnormal point area.

[0027] Furthermore, after forming the virtual standard sample described in step S5, firstly, the difference points in the N consecutive target products that are referenced when the virtual standard sample is formed are marked in a color difference standard manner; then, based on a secondary judgment method, it is determined whether these difference points are retained as part of the virtual standard sample.

[0028] Furthermore, the secondary judgment adopts at least one of: manual judgment and judgment after deep learning of AI with deep learning function.

[0029] Furthermore, for products with an area less than or equal to Km 2 For products with an area larger than Km2, the overall comparison method is adopted, that is, the grayed pattern of the product as a whole is compared with the virtual standard sample; for products with an area larger than Km2, the local comparison method is adopted, that is, the grayed pattern of the product and the virtual standard sample are divided into at least two parts, and then the reference points of each part are obtained in turn for superposition comparison.

[0030] In addition, the present invention also provides an application of an optical detection method without the need for a standard sample in the detection of surface circuit anomalies of a semiconductor wafer, and the surface circuit of a semiconductor wafer is detected for anomalies using the optical detection method without the need for a standard sample.

[0031] In addition, the present invention also provides an application of an optical detection method without the need for a standard sample in abnormal detection of printed circuits on a printed circuit board, and the optical detection method without the need for a standard sample is used to perform abnormal detection on the printed circuits of the printed circuit board.

[0032] The present invention has at least one of the following advantages:

[0033] 1. The present invention does not need to produce standard samples, but directly uses virtual standard samples as reference samples by forming virtual standard samples, thereby saving the cost of producing standard samples.

[0034] 2. The present invention can accurately find the similarities and differences among samples, thereby achieving accurate product testing.

[0035] 3. The technology of the present invention can be combined with artificial intelligence technology to significantly reduce the sample size of deep learning of artificial intelligence in the field of optical detection, improve learning efficiency, and enhance the judgment accuracy of artificial intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Shown is a photographic original image of a printed circuit board product.

[0037] Figure 2 Shown Figure 1 The ash pattern of the product after ash treatment.

[0038] Figure 3 Shown Figure 2 The grayed-out graphic is a schematic diagram for obtaining the longitudinal reference point.

[0039] Figure 4 Shown Figure 2 The grayed-out graphic is a schematic diagram for obtaining the horizontal reference point.

[0040] Figure 5 Shown Figure 3 and Figure 4 Get after superposition Figure 2 Schematic diagram of the reference points of the grayed-out graphic.

[0041] Figure 6 Shown Figure 1 The product shown is a virtual standard sample image obtained by the method of the present invention.

[0042] Figure 7 Shown Figure 1 Grayed-out graphics of successors to the product shown.

[0043] Figure 8 Shown Figure 7 Products and Figure 6 Comparison results of the virtual standard sample.

[0044] Fig. 9 Shown is an original photograph of the surface circuit of a semiconductor wafer.

[0045] Fig.10 Shown Fig. 9 Schematic diagram of the reference points of a product obtained through the method of the present invention.

[0046] Fig.11 Shown Fig. 9 The virtual standard sample image of the rest of the products in the same batch is obtained by the method of the present invention.

[0047] Fig.12 Shown Fig. 9 Products and Fig.11 Comparison results of the virtual standard sample. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0049] The present invention exemplarily provides an optical detection method without the need for a standard sample, comprising:

[0050] S1. Continuously obtain graphic information of the target product;

[0051] S2. graying the acquired graphic information of the target product;

[0052] S3. Obtain the reference point of each grayed-out graphic by calculating the reference point;

[0053] S4. After superimposing at least N consecutive gray images of the target product based on the reference point, similarities and differences are obtained;

[0054] S5. Construct a virtual standard sample based on the similar points in step S4, and use the virtual standard sample as a standard sample for subsequent target product graying pattern comparison detection, and record the comparison results of each target product with the virtual standard sample.

[0055] Traditional AOI optical inspection technology must produce an absolutely accurate standard sample. Due to the accuracy requirements of the standard sample, its production cost has always been high. For products that need to be produced in large quantities, the production cost of the standard sample will not appear very high after being apportioned, so it can be accepted by the manufacturer. However, with the innovation of modern technology, more and more small-batch and variable products are produced. If a standard sample is required for each small batch of products and each deformed product, the production cost will increase significantly. The optical inspection method of the present invention does not require a standard sample. By forming a virtual standard sample, it circumvents the traditional method of comparative detection by standard samples, effectively reducing the production cost of small-batch and variable products.

[0056] The present invention exemplarily provides a method for ashing in step S2, comprising:

[0057] First, obtain the three primary color values ​​of each pixel in the graphic data, and use the XY composition system to express it as follows:

[0058] [[B11,G11,R11][B12,G12,R12][B13,G13,R13]...[B1n,G1n,R1n]]

[0059] [[B21,G21,R21][B22,G22,R22][B23,G23,R23]...[B2n,G2n,R2n]]

[0060] [[B31,G31,R31][B32,G32,R32][B33,G33,R33]...[B3n,G3n,R3n]] ...

[0062] [[Bm1,Gm1,Rm1][Bm2,Gm2,Rm2][Bm3,Gm3,Rm3]...[Bmn,Gmn,Rmn]];

[0063] Then, the following numerical conversion is performed on the three primary color values ​​of each pixel:

[0064] [Max(B11,G11,R11)-Min(B11,G11,R11)][Max(B12,G12,R12)-Min(B12,G12,R12)]...[Max(B1n,G1n,R1n)-Min(B1n,G1n,R1n)]

[0065] [Max(B21,G21,R21)-Min(B21,G21,R21)][Max(B22,G22,R22)-Min(B22,G22,R22)]...[Max(B2n,G2n,R2n)-Min(B2n,G2n,R2n)] ...

[0067] [Max(Bm1,Gm1,Rm1)-Min(Bm1,Gm1,Rm1)][Max(Bm2,Gm2,Rm2)-Min(Bm2,Gm2,Rm2)]...[Max(Bmn,Gmn,Rmn)-Min(Bmn,Gmn,Rmn)];

[0068] After the three primary color values ​​of all the pixels are converted, the graying process is completed; the three primary color values ​​of each pixel are taken as integers and classified into a value range of 0-255.

[0069] The existing AOI optical inspection technology is generally based on color image comparison. This method has a relatively accurate detection accuracy rate when comparing the whole, but when comparing the details, it is often difficult to identify because the color changes are too shallow. After research, the applicant found that in the actual production process, the defect points of the product after graying are often close to neutral gray. After the pixel point value conversion is performed using the method of the present invention, the defect point value will be minimized (even if it is a white point, gray point, or black point defect, it will be converted to the minimum value), so it will be very significantly reflected in the grayed graphics, and there will be significant differences in the values, thereby significantly improving the accuracy of subsequent inspections.

[0070] The present invention exemplarily provides a method for obtaining the reference point of each grayed graphic, including: first, taking the median of each horizontal row in the dot matrix after the three primary color values ​​are converted to form a longitudinal reference point, and taking the median of each vertical row in the dot matrix after the three primary color values ​​are converted to form a transverse reference point; then, taking the intersection of the longitudinal reference point and the transverse reference point, and marking it as the reference point of the grayed graphic.

[0071] The base point is the basis for graphic comparison. If the base point is not found correctly, the accuracy of subsequent detection will be seriously affected. The reason why the present invention takes the median of the X-axis and Y-axis instead of the average is that the median can better exclude the deviation value compared with the average, so that the base point tends to be more uniform, avoiding comparison errors caused by base point errors.

[0072] The present invention exemplarily provides a method for obtaining similarities and differences therein, including: first, the X-axis is accumulated into a y-axis transformation curve, and the high-frequency repetition law and the low-frequency repetition law of the y-axis are found by using FFD Fourier transform; similarly, the Y-axis is accumulated into an x-axis transformation curve, and the high-frequency repetition law and the low-frequency repetition law of the x-axis are found by using FFD Fourier transform; then the high-frequency repetition area of ​​the x-axis and the y-axis is defined as the similarity point, and the low-frequency repetition area of ​​the x-axis and the y-axis is defined as the difference point.

[0073] Based on the Fourier transformation after superposition of a certain number of product graphics, the frequency of occurrence of the pixels of these product graphics can be obtained. For industrial products, under modern manufacturing process conditions, their repeatability is generally very high. Therefore, the high-frequency repeated parts of several consecutive products can be used as the determined non-abnormal parts. Based on this principle, the present invention analyzes the frequency of occurrence of the repeated parts of several consecutive products, selects the high-frequency repeated parts to construct a virtual standard sample, and can obtain a virtual standard sample with extremely high accuracy, thereby avoiding the cost of making actual standard samples.

[0074] The present invention exemplarily provides a method for comparative detection of graying patterns of a target product, comprising:

[0075] (1) Calculate the numerical value A of each pixel point in the data column of the grayed image of the target product and the numerical value B of each pixel point in the virtual standard sample: C=AB. When the calculated value of C is within the preset threshold range, the numerical value A is considered to be a normal point; when the calculated value of C is outside the preset threshold range, the numerical value A is considered to be a suspected abnormal point and is marked with a characteristic color; when the number of suspected abnormal points exceeds the preset threshold, all suspected abnormal points are marked with characteristic colors and the product is eliminated;

[0076] (2) The data of the target product graying pattern is superimposed on the virtual standard sample and then linear Fourier transform is performed, where the target product graying pattern area below the preset frequency is the suspected abnormal point area.

[0077] The present invention simultaneously performs point analysis and surface analysis on the product graphics, that is, by changing the color of the pixel points, the pixel points with significantly different colors are found. This is because the graphics color of the abnormal points of the product is often light white or dark gray after the graying treatment of the present invention, which is significantly different from the median grayscale of other areas. Therefore, these abnormal points can be found by comparing the color values ​​of the pixels. Sometimes the abnormal points appear in a regional way, but it is difficult to identify from the pixel color, so it is necessary to introduce Fourier transform analysis to mark the areas with low frequency of occurrence as suspected abnormal points. Through the comparative analysis of the present invention, the detection accuracy of abnormal points can be increased to more than 95%, while if AI is used for learning and then detection and judgment, it will take hundreds of deep learning times to achieve an accuracy of more than 90%.

[0078] The present invention exemplarily provides a supplementary method for comparative detection of graying patterns of target products, comprising: for a product area less than or equal to Km 2 For products with an area larger than Km, the overall comparison method is used, that is, the grayed pattern of the entire product is compared with the virtual standard sample; for products with an area larger than Km 2 For products, the local comparison method is adopted, that is, the product gray pattern and the virtual standard sample are divided into at least two parts, and then the reference points are obtained for each part in turn and then superimposed and compared.

[0079] This method can choose whether to divide the block analysis according to the area of ​​the product to be tested. Its purpose is to avoid the problem of inaccurate identification due to the small area of ​​abnormal points when testing large-area products. After regional division, the large-area product can be divided into several small areas that can be accurately identified, thereby improving the accuracy of detection.

[0080] The present invention provides an exemplary method for supplementary judgment after forming a virtual standard sample, including: firstly, marking the difference points of N consecutive target products that are referenced when the virtual standard sample is formed in a color difference standard manner; then judging whether these difference points are retained as part of the virtual standard sample in a secondary judgment manner. The secondary judgment adopts at least one of: manual judgment and judgment after deep learning of AI with deep learning function.

[0081] This method can further improve the accuracy of the virtual standard sample through subsequent supplementary judgment and reduce the deviation between the virtual standard sample and the actual standard sample.

[0082] The present invention exemplarily provides an application of an optical detection method without the need for a standard sample in the detection of surface circuit anomalies of a semiconductor wafer. The optical detection method without the need for a standard sample is used to perform abnormality detection on the surface circuit of a semiconductor wafer.

[0083] The present invention exemplarily provides an application of an optical detection method without the need for a standard sample in abnormal detection of printed circuits on a printed circuit board. The optical detection method without the need for a standard sample is used to perform abnormal detection on the printed circuits of the printed circuit board.

[0084] In order to explain the technical solution of the present application in more detail, the technology of the present invention is further specifically described below in combination with a specific embodiment of the present invention applied in abnormal detection of printed circuits on a printed circuit board.

[0085] Example 1

[0086] A printed circuit board printed circuit product is tested by using the optical testing method of the present invention without the need for a standard sample:

[0087] S1. Figure 1 The original image of a product in the printed circuit board printed circuit product is obtained after scanning / photographing (the original image is originally in color).

[0088] S2. Yes Figure 1 Graying treatment is carried out to turn the originally colored Figure 1 Processed Figure 2 Grayed out graphics shown.

[0089] S3. First, take the median of each horizontal row in the dot matrix after the three primary color values ​​are converted to form the vertical reference point, such as Figure 3 The median of each vertical row in the dot matrix after the three primary color values ​​are converted is used to form the horizontal reference point, as shown in Figure 4 Then, take the intersection of the longitudinal reference point and the transverse reference point and mark it as the reference point of the grayed-out figure, as shown in Figure 5 shown.

[0090] S4. First, the X-axes of N consecutive target products are accumulated into a y-axis transformation curve, and the high-frequency repetition law and low-frequency repetition law of the y-axis are found by using the FFD Fourier transform. Similarly, the Y-axes of N consecutive target products are accumulated into an x-axis transformation curve, and the high-frequency repetition law and low-frequency repetition law of the x-axis are found by using the FFD Fourier transform. Then, the high-frequency repetition areas of the x-axis and the y-axis are defined as similarities, and the low-frequency repetition areas of the x-axis and the y-axis are defined as differences.

[0091] At this time, after Fourier transformation, the high-frequency repetitive parts and low-frequency repetitive parts of these continuous products will be obtained. As a modern industrial product, the high-frequency repetitive part is generally the compliant part, and the low-frequency repetitive part is the suspected abnormal area.

[0092] S5. Construct a virtual standard sample based on the similar points in step S4, such as Figure 6 As shown in the figure, the virtual standard sample is used as the subsequent target product graying diagram, as shown in the figure Figure 7 As shown, the standard sample for shape comparison test is Figure 8 The white circle marks the part. Meanwhile, the comparison results between each target product and the virtual standard are recorded.

[0093] At this point, by comparing the virtual standard sample with the product's graphics, the inspection of the printed circuit board printed circuit product can be completed.

[0094] Example 2

[0095] A surface circuit diagram of a semiconductor wafer product is detected by using the optical detection method of the present invention without the need for a standard sample:

[0096] S1. The original image of a product in the printed circuit board product obtained after scanning / photographing. Fig. 9 As shown, Fig. 9 There are significant abnormalities on the circuit panel. Fig. 9 The rest of the products in the same batch do not have this abnormality.

[0097] S2. Performing an ashing process to obtain an ashing pattern.

[0098] S3. First, take the median of each horizontal row in the dot matrix after the three primary color values ​​are converted to form the vertical reference point. Take the median of each vertical row in the dot matrix after the three primary color values ​​are converted to form the horizontal reference point; then, take the intersection of the vertical reference point and the horizontal reference point and mark it as the reference point of the grayed figure. Fig.10 shown.

[0099] S4. First, the X-axes of N consecutive target products are accumulated into a y-axis transformation curve, and the high-frequency repetition law and low-frequency repetition law of the y-axis are found by using the FFD Fourier transform. Similarly, the Y-axes of N consecutive target products are accumulated into an x-axis transformation curve, and the high-frequency repetition law and low-frequency repetition law of the x-axis are found by using the FFD Fourier transform. Then, the high-frequency repetition areas of the x-axis and the y-axis are defined as similarities, and the low-frequency repetition areas of the x-axis and the y-axis are defined as differences.

[0100] At this time, after Fourier transformation, the high-frequency repetitive parts and low-frequency repetitive parts of these continuous products will be obtained. As a modern industrial product, the high-frequency repetitive part is generally the compliant part, and the low-frequency repetitive part is the suspected abnormal area.

[0101] S5. Construct a virtual standard sample based on the similar points in step S4, such as Fig.11 As shown ( Fig.11 The virtual standard shown is Fig. 9 The product shown is constructed from the other defect-free products in the same batch), and the virtual standard sample is used as the standard sample for subsequent target product graying diagram and shape comparison test, such as Fig.12 The light grey area marked by the white frame line in the middle. The comparison results of each target product and the virtual standard sample are recorded at the same time.

[0102] At this point, by comparing the virtual standard sample with the product's graphics, the inspection of the printed circuit board printed circuit product can be completed.

[0103] It should be noted and understood that various modifications and improvements can be made to the invention described in detail above without departing from the spirit and scope of the claims of the present invention. Therefore, the scope of the claimed technical solution is not limited by any specific exemplary teaching given.

Claims

1. An optical detection method without the need for a standard sample, characterized in that: include: S1. Continuously obtain the graphic information of the target product; S2. graying the acquired graphic information of the target product, specifically including: First, obtain the three primary color values ​​of each pixel in the graphic data, and use the XY composition system to express it as follows: [[B11,G11,R11][B12,G12,R12][B13,G13,R13]...[B1n,G1n,R1n]] [[B21,G21,R21][B22,G22,R22][B23,G23,R23]...[B2n,G2n,R2n]] [[B31,G31,R31][B32,G32,R32][B33,G33,R33]...[B3n,G3n,R3n]] ... [[Bm1,Gm1,Rm1][Bm2,Gm2,Rm2][Bm3,Gm3,Rm3]...[Bmn,Gmn,Rmn]]; Then, the following numerical conversion is performed on the three primary color values ​​of each pixel: [Max(B11,G11,R11)-Min(B11,G11,R11)][Max(B12,G12,R12)-Min(B12,G12,R12)]...[Max(B1n,G1n,R1n)-Min(B1n,G1n,R1n)] [Max(B21,G21,R21)-Min(B21,G21,R21)][Max(B22,G22,R22)-Min(B22,G22,R22)]...[Max(B2n,G2n,R2n)-Min(B2n,G2n,R2n)] ... [Max(Bm1,Gm1,Rm1)-Min(Bm1,Gm1,Rm1)][Max(Bm2,Gm2,Rm2)-Min(Bm2,Gm2,Rm2)]...[Max(Bmn,Gmn,Rmn)-Min(Bmn,Gmn,Rmn)]; After the three primary color values ​​of all pixels are converted, the graying process is completed; the three primary color values ​​of each pixel are taken as integers and classified into a value range of 0-255; S3. Obtain the reference point of each grayed image by calculating the reference point; S4. After superimposing the grayed images of at least N consecutive target products based on the reference points, similarities and differences therein are obtained, specifically including: First, the X-axis of N consecutive target products is accumulated into a y-axis transformation curve, and the high-frequency repetition law and low-frequency repetition law of the y-axis can be found by using FFD Fourier transform; similarly, the Y-axis of N consecutive target products is accumulated into an x-axis transformation curve, and the high-frequency repetition law and low-frequency repetition law of the x-axis can be found by using FFD Fourier transform; then the high-frequency repetition area of ​​the x-axis and the y-axis is defined as the similarity point, and the low-frequency repetition area of ​​the x-axis and the y-axis is defined as the difference point S5. construct a virtual standard sample based on the similar points in step S4, and use the virtual standard sample as the standard sample for subsequent target product graying pattern comparison detection, and record the comparison results of each target product with the virtual standard sample; The method for comparative detection of graying patterns of target products comprises: (1) Calculate the numerical value A of each pixel point in the data column of the grayed image of the target product and the numerical value B of each pixel point in the virtual standard sample: C = AB. When the calculated value of C is within the preset threshold range, the numerical value A is considered to be a normal point; when the calculated value of C is outside the preset threshold range, the numerical value A is considered to be a suspected abnormal point and is marked with a characteristic color; when the number of suspected abnormal points exceeds the preset threshold, all suspected abnormal points are marked with characteristic colors and the product is eliminated; (2) The data of the target product graying pattern is superimposed on the virtual standard sample and then linear Fourier transform is performed, wherein the target product graying pattern area below the preset frequency is the suspected abnormal point area.

2. The optical detection method without standard sample according to claim 1, characterized in that: The method for obtaining the reference point of each grayed figure described in step S3 includes: first, taking the median of each horizontal row in the dot matrix after the three primary color values ​​are converted to form a longitudinal reference point, and taking the median of each vertical row in the dot matrix after the three primary color values ​​are converted to form a transverse reference point; then, taking the intersection of the longitudinal reference point and the transverse reference point, and marking it as the reference point of the grayed figure.

3. The optical detection method without standard sample according to claim 1, characterized in that: After forming the virtual standard sample described in step S5, firstly, the difference points of the N consecutive target products used as reference when forming the virtual standard sample are marked in a color difference standard manner; then, it is determined whether these difference points are retained as part of the virtual standard sample in a secondary judgment manner.

4. The optical detection method without standard sample according to claim 3, characterized in that: The secondary judgment adopts at least one of: manual judgment and judgment after deep learning of AI with deep learning function.

5. The optical detection method without standard sample according to claim 1, characterized in that: For products with an area less than or equal to Km 2 For products with an area larger than Km, the overall comparison method is used, that is, the grayed pattern of the entire product is compared with the virtual standard sample; for products with an area larger than Km 2 For products, the local comparison method is adopted, that is, the product gray pattern and the virtual standard sample are divided into at least two parts, and then the reference points are obtained for each part in turn and then superimposed and compared.

6. An optical detection method without the need for a standard sample is used in the detection of surface circuit anomalies of a semiconductor wafer, characterized in that: The optical detection method without the need for a standard sample as described in any one of claims 1 to 5 is used to perform abnormality detection on the surface circuits of a semiconductor wafer.

7. An optical detection method without standard samples is used in the detection of abnormalities in printed circuit boards, characterized in that: The optical detection method without standard samples as described in any one of claims 1 to 5 is used to detect abnormalities in the printed circuits of the printed circuit board.

Images