Automatic optical detection method, automatic optical detection system and recording medium

Abstract

The present invention provides an automated optical inspection method, system, and recording medium. The method acquires an original image of an object under test via an optical lens, wherein the original image includes multiple first images; performs edge detection on the original image to obtain an edge image having an edge pattern, wherein the edge image includes multiple second images; calculates at least one of the maximum, minimum, and average values of pixel values in the second images; divides the edge image into multiple image blocks based on unit area, and calculates multiple eigenvalues based on at least one of the maximum, minimum, and average values corresponding to the multiple second images included in the image blocks; and trains a regression model corresponding to the defects of the object under test based on the eigenvalues and the object under test data to obtain an optimal regression model.

Description

Technical Field

[0001] The present invention relates to a defect detection technology, and in particular to an automatic optical detection method, an automatic optical detection system and a recording medium. Background Art

[0002] Before electronic components leave the factory, they are typically visually inspected by experienced visual inspectors to determine if they have defects or are flat. For example, when determining the flatness of silicon carbide (SiC) wafers, the haze value is often used as an indicator of flatness. Generally speaking, haze value testing involves manual interpretation of the SiC wafer using the human eye.

[0003] However, visual inspection by human inspectors often leads to misjudgment due to their subjective interpretation. Therefore, how to avoid the problem of overly subjective inspection results caused by relying on visual inspection is indeed a topic of concern to those skilled in the art. Summary of the Invention

[0004] The present invention provides an automatic optical inspection method, an automatic optical inspection system and a recording medium, which can improve the training efficiency of a classification model and the detection accuracy of defects by an optical inspection device.

[0005] An automatic optical inspection method, applicable to an optical inspection system including an optical lens and a processor, comprises: acquiring an original image of an object to be inspected via the optical lens, wherein the original image includes multiple first images; performing edge detection on the original image to obtain an edge image having an edge pattern, wherein the edge image includes multiple second images having edge patterns; calculating at least one of a maximum value, a minimum value, and an average value of pixel values in the multiple second images; dividing the edge image into multiple image blocks according to a unit area, and calculating multiple eigenvalues based on at least one of the maximum value, the minimum value, and the average value corresponding to the multiple second images included in the multiple image blocks; and training a regression model corresponding to the defects of the object to be inspected based on the multiple eigenvalues and the object to be inspected data to obtain an optimal regression model.

[0006] In an exemplary embodiment of the present invention, the plurality of characteristic values include at least one of a standard deviation, a coefficient of variation, a mean, a range, and a mean absolute deviation.

[0007] In an exemplary embodiment of the present invention, the above-mentioned test object data includes expected output, and the step of training the regression model corresponding to the test object defect based on the multiple feature values and the test object data to obtain the optimal regression model includes: inputting the multiple feature values and the expected output as input data into the regression model to train the regression model, and generating multiple weights corresponding to the multiple feature values.

[0008] In an exemplary embodiment of the present invention, the above-mentioned automatic optical inspection method further includes: classifying the multiple feature values according to the multiple weights based on the trained regression model to generate classification results corresponding to the multiple feature values; and comparing the classification results with the expected output to determine whether the classification results meet expectations.

[0009] In an exemplary embodiment of the present invention, the above-mentioned automatic optical inspection method further includes: if the classification result meets expectations, setting the trained regression model as the optimal regression model; and if the classification result does not meet expectations, reselecting multiple feature values used to train the regression model and retraining the regression model.

[0010] In an exemplary embodiment of the present invention, the automatic optical inspection method further includes: classifying the third image based on the optimal regression model to generate a classification result corresponding to the third image.

[0011] An automatic optical inspection system includes an optical lens and a processor. The optical lens is configured to acquire an original image of an object to be inspected, wherein the original image includes a plurality of first images. The processor is coupled to the optical lens and configured to: perform edge detection on the original image to acquire an edge image having an edge pattern, wherein the edge image includes a plurality of second images having edge patterns; calculate at least one of the maximum value, minimum value, and average value of pixel values in the plurality of second images; divide the edge image into a plurality of image blocks according to a unit area, and calculate a plurality of eigenvalues according to at least one of the maximum value, minimum value, and average value corresponding to the plurality of second images included in the plurality of image blocks; and train a regression model corresponding to the defects of the object to be inspected based on the plurality of eigenvalues and the object to be inspected data to obtain an optimal regression model.

[0012] In an exemplary embodiment of the present invention, the plurality of characteristic values include at least one of a standard deviation, a coefficient of variation, a mean, a range, and a mean absolute deviation.

[0013] In an exemplary embodiment of the present invention, the above-mentioned test object data includes expected output, and the processor is configured to: input the multiple feature values and the expected output as input data into the regression model to train the regression model, and generate multiple weights corresponding to the multiple feature values.

[0014] In an exemplary embodiment of the present invention, the processor is further configured to: classify the multiple feature values according to the multiple weights based on the trained regression model to generate classification results corresponding to the multiple feature values; and compare the classification results with the expected output to determine whether the classification results meet expectations.

[0015] In an exemplary embodiment of the present invention, if the classification result meets expectations, the processor sets the trained regression model as the optimal regression model, and if the classification result does not meet expectations, the processor reselects multiple feature values used to train the regression model and retrains the regression model.

[0016] In an exemplary embodiment of the present invention, the processor is configured to: classify the third image based on the optimal regression model to generate a classification result corresponding to the third image.

[0017] In an exemplary embodiment of the present invention, the processor is configured in a remote server.

[0018] A non-transitory computer-readable recording medium records a program that is loaded by a processor to execute the following steps: acquiring an original image of an object to be tested through an optical lens, wherein the original image includes multiple first images; performing edge detection on the original image to obtain an edge image having an edge pattern, wherein the edge image includes multiple second images having edge patterns; calculating at least one of the maximum value, the minimum value, and the average value of pixel values in the multiple second images; dividing the edge image into multiple image blocks based on unit area, and calculating multiple eigenvalues based on at least one of the maximum value, the minimum value, and the average value corresponding to the multiple second images included in the multiple image blocks; and training a regression model corresponding to the defects of the object to be tested based on the multiple eigenvalues and the object to be tested data to obtain an optimal regression model.

[0019] Based on the above, after an image of the object under test is captured by the optical lens of an automated optical inspection device, the image can be segmented into multiple image blocks, and the feature values corresponding to each image block can be used to train a classification model. This effectively improves the training efficiency of the classification model and enhances the optical inspection device's defect detection accuracy.

[0020] In order to make the above features and advantages of the present invention more clearly understood, embodiments are given below with reference to the accompanying drawings for detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a schematic diagram of an automatic optical inspection system according to an embodiment of the present invention;

[0022] Figure 2 is a block diagram of an automatic optical inspection system according to an embodiment of the present invention;

[0023] Figure 3 is a flow chart of an automatic optical inspection method according to an embodiment of the present invention;

[0024] Figure 4 is a schematic diagram of segmenting an edge image according to an embodiment of the present invention;

[0025] Figure 5 is a schematic diagram of a characteristic value calculation result according to an embodiment of the present invention;

[0026] Figure 6 FIG. 4 is a schematic diagram of a regression model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] Reference will now be made in detail to exemplary embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Whenever possible, the same reference numerals are used in the drawings and the description to refer to the same or like parts.

[0028] Figure 1 FIG. 1 is a schematic diagram of an automatic optical inspection system according to an embodiment of the present invention. Figure 2 FIG is a block diagram of an automatic optical inspection system according to an embodiment of the present invention. Figure 1 and Figure 2 The automated optical inspection system 100 can be used in automated optical inspection (AOI) equipment to perform surface defect inspection on semiconductor chips, wafers, circuit boards, panels, and other objects to be inspected. In other words, the automated optical inspection system 100 can be used to detect surface defects on the objects to be inspected.

[0029] The automatic optical inspection system 100 may include an optical inspection device 110, a transfer device 120, and a light source device 130. In one embodiment, the optical inspection device 110 may send a control signal in a wired or wireless manner to control at least one of the optical lens 111, the transfer device 120, and the light source device 130. The optical inspection device 110 has an optical lens 111. The optical lens 111 may adopt an area scan camera and / or a line scan camera. Line scan cameras are more often used in conjunction with dynamic scanning detection to take pictures while the object to be tested 101 moves. In this way, the continuity of the inspection process can be ensured. The transfer device 120 is used to achieve fully automated inspection. For example, the transfer device 120 can transfer the object to be tested 101 to the inspection area, and take pictures through the optical lens 111 arranged on one side of the inspection area to obtain an image of the object to be tested 101 and perform subsequent image analysis.

[0030] The light source device 130 is used to provide a light source for auxiliary illumination of the object under test 101. The type of light source device 130 may be, for example, a parallel light fixture, a diffuse light fixture, or a dome light fixture, but the present invention is not limited thereto. The light source device 130 may emit various types of light, such as white light, red light, green light, blue light, ultraviolet light, and infrared light. Furthermore, the type of light source device 130 may be changed to correspond to different types of objects under test 101. It should be noted that the present invention does not limit the number of optical inspection devices 110, transfer devices 120, and light source devices 130.

[0031] In optical inspection, local defects on the object under test 101 can generally be detected microscopically, and the surface flatness of the object under test 101 can be checked macroscopically. Taking wafer inspection as an example, early methods typically involved machine detection of local defects, followed by manual inspection to confirm the wafer's flatness. If zebra stripes, sunburst patterns, wavy patterns, or other patterns appear in the wafer image, it can be determined that the wafer corresponding to this image is not flat. However, under certain conditions, the presence of a few patterns in the wafer image can still be used to determine that the wafer corresponding to this image is not defective. Therefore, visual inspectors may interpret the flatness judgment criteria in their own subjective way, leading to misjudgments. Therefore, the efficiency and reliability of manual inspection are often lower than those of machine vision. Furthermore, manual inspection often takes longer than machine vision inspection. In one embodiment, the optical inspection device 110 can automatically complete the entire inspection process, thereby improving inspection efficiency and reliability.

[0032] The optical detection device 110 includes an optical lens 111, a processor 112, and a storage device 113. The optical lens 111 may include one or more optical lenses, and the present invention does not limit the type of the optical lens 111.

[0033] The processor 112 may be, for example, a central processing unit (CPU), or other programmable general-purpose or special-purpose microcontroller unit (MCU), microprocessor, digital signal processor (DSP), programmable controller, application-specific integrated circuit (ASIC), graphics processing unit (GPU), arithmetic logic unit (ALU), complex programmable logic device (CPLD), field programmable gate array (FPGA), or other similar components or combinations thereof. Furthermore, the processor 112 may also be a hardware circuit designed using a hardware description language (HDL) or any other digital circuit design method known to those skilled in the art. The processor 112 may be responsible for all or part of the operation of the optical detection device 110.

[0034] The storage device 113 may include volatile storage media and / or non-volatile storage media and may be used to store data. For example, the volatile storage media may be a random access memory (RAM), while the non-volatile storage media may be a read-only memory (ROM), a solid-state drive (SSD), or a hard disk drive (HDD).

[0035] Figure 3 This is a flow chart of an automatic optical inspection method according to an embodiment of the present invention. Figures 1 to 3 The method of this embodiment is applicable to the above-mentioned automatic optical inspection system 100, and the following is used in conjunction with Figure 1 and Figure 2 The various components of the automatic optical inspection system 100 are used to illustrate the detailed process of the method of this embodiment.

[0036] It should be noted that Figure 3 Each step can be implemented as multiple codes or circuits, and the present invention is not limited thereto. Figure 3 The method can be used in conjunction with the following exemplary embodiments or can be used alone, and the present invention is not limited thereto.

[0037] In step S302, the processor 112 captures an original image of the object under test 101 via the optical lens 111, where the original image includes multiple sub-images (also referred to as first images). In step S304, the processor 112 performs edge detection on the original image to obtain an edge image having an edge pattern, where the edge image includes multiple edge sub-images having the edge pattern (also referred to as second images). In other words, the processor 112 performs edge detection on the first image to obtain a second image having an edge pattern. For example, the edge detection may detect the edges (also referred to as contours) of at least a portion of the pattern in the original image and present the detected edges in the edge image as additional edge patterns. From another perspective, the processor 112 may convert an original image obtained by the optical lens 111 without an edge pattern and including multiple first images into an edge image having an edge pattern and including multiple second images. The present invention does not limit the image analysis technology used by the processor 112 to perform edge detection. Furthermore, the processor 112 may de-emphasize, simplify, or remove image data in the edge image that does not represent the edge pattern to highlight the edge pattern in the edge image.

[0038] In one embodiment, the edge detection technology employed by processor 112 is, for example, the Sobel operator, although the present invention is not limited thereto. Edge detection can enhance defective portions of an image, thereby improving the accuracy of processor 112 in identifying defective lines. For example, compared to directly performing defect detection on the original image, detecting defective lines on an edge image can achieve higher accuracy.

[0039] In one embodiment, the processor 112 may perform other image pre-processing operations on the original image before performing edge detection. Generally speaking, the original image captured by the optical lens 111 includes multiple first images that may present a pattern of the captured electronic component. The processor 112 may analyze the original image captured by the optical lens 111 to detect defects on the surface of the object under test 101 through the image. For example, after the original image is captured by the optical lens 111, the processor 112 may perform image pre-processing operations on the original image (e.g., performing image enhancement, contrast enhancement, edge enhancement, noise removal, feature extraction, image conversion, image compression, etc.). In one embodiment, the processor 112 performs image pre-processing operations to convert the original image into a haze map before performing subsequent image analysis. The processor 112 also performs edge detection on the original image converted into the haze map to obtain multiple edge images having edge patterns. In this embodiment, the multiple first images included in the original image are also haze maps.

[0040] Back to Figure 3Flowchart of . In step S306, the processor 112 calculates at least one of the maximum value, minimum value and average value of the pixel values (pixelvalue) in the plurality of second images. Specifically, the processor 112 obtains the pixel value of each pixel included in the second image, and calculates at least one of the maximum value, minimum value and average value of the pixel value corresponding to the second image based on the pixel value of each pixel obtained. Pixel value is a technical name well known to those skilled in the art, such as grayscale value, brightness value, etc., but the present invention is not limited thereto. After calculating the aforementioned maximum value, minimum value and / or average value, the processor 112 generates a matrix based on the calculation results. In other words, through step S306, the second images included in the edge image are digitized into a matrix to record the values corresponding to each second image.

[0041] In step S308, the processor 112 divides the edge image into a plurality of image blocks based on a unit area, and calculates a plurality of feature values based on at least one of the maximum value, minimum value, and average value corresponding to the plurality of second images included in the plurality of image blocks. Specifically, the processor 112 may group the second images included in the edge image into a plurality of image blocks based on a unit area. In different embodiments, the size of the image blocks per unit area may be set based on user needs. For example, the processor 112 may use 1 / 25, 1 / 36, or 1 / 64 of the edge image area as a unit area to divide the edge image into 5x5, 6x6, or 8x8 image blocks. Thus, by dividing the edge image into a plurality of image blocks for subsequent calculations, it is possible to avoid calculating the values of the entire image at once, which may dilute the maximum or minimum values of a single pixel and cause misjudgment.

[0042] Next, the processor 112 obtains multiple maximum values, multiple minimum values, and / or multiple average values corresponding to a set of second images included in each image block, and calculates multiple feature values corresponding to the set of second images based on these maximum values, minimum values, and / or average values. Feature values may be statistical values such as standard deviation (Std), coefficient of variation (Cv), average (Ave), range (R), and mean absolute deviation (MAD). For example, the processor 112 may calculate the standard deviation, coefficient of variation, and average value corresponding to each image block based on at least one of the maximum value, minimum value, and / or average value of the pixel values of the multiple second images included in each image block. In this way, multiple feature values corresponding to each image block can be calculated. These feature values may represent the degree of dispersion or concentration between the multiple second images included in the image block.

[0043] For example, the edge image is divided into 5x5 image blocks as an example for description. In this embodiment, the pixel value is a grayscale value. Figure 4 is a schematic diagram of segmenting edge images according to an embodiment of the present invention. Figure 4 , edge image 401 includes multiple second images. Processor 112 obtains the pixel value of each pixel of the second image and calculates at least one of the maximum, minimum, and average grayscale values corresponding to the second image. The calculation results of the maximum, minimum, and average grayscale values of each second image in this embodiment can be referred to in Table 1 below.

[0044] Table 1

[0045] Image Block Second image Minimum Maximum average value ST11 1 2.95 7.20 4.26 … .. … … N 2.93 7.22 4.25 ST12 1 2.95 6.95 4.23 … … … … N 2.94 6.92 4.24 … … … … … ST55 1 2.95 10.70 4.25 … … … … N 2.91 10.90 4.35

[0046] The processor 112 uses 1 / 25 of the area of the edge image 401 as a unit area and divides the edge image 401 into 25 image blocks ST11 to ST55 according to the unit area. Figure 4 Segmented edge image 402 and Table 1. In this embodiment, each image block includes N second images, where N is an integer greater than 0 and the value of N is determined by the number of sub-images included when acquiring the original image. Next, processor 112 calculates the standard deviation, coefficient of variation, and mean of the maximum grayscale values corresponding to each image block ST11-ST55 based on the maximum grayscale values of all second images included in each image block ST11-ST55 (see Table 1). Figure 5 is a schematic diagram showing the result of feature value calculation according to an embodiment of the present invention. The calculation results of the standard deviation, coefficient of variation and mean of the maximum value of the grayscale value of each image block ST11 to ST55 can be referred to Figure 5 And refer to Table 2 below for the corresponding information.

[0047] Table 2

[0048]

[0049] In step S310, the processor 112 trains a regression model corresponding to the defects of the object to be tested based on multiple eigenvalues and the object to be tested data to obtain an optimal regression model. Among them, the inspection personnel can make a judgment on the original image of the object to be tested in advance, and label the judgment result as the expected output of the original image. The expected output may include good products or NG products, etc. Based on this, the object to be tested data training may include the wafer ID and expected output corresponding to the original image. The processor 112 can input the eigenvalues and expected output corresponding to each image block (i.e., each group of second images) calculated from step S308 as input data into the regression model to train and generate the optimal regression model. In one embodiment, the regression model may be, for example, a classification algorithm model such as a logistic regression model (LogisticRegression), a K-nearest neighbor classification algorithm (KNN), a support vector machine (SVM), a decision tree classification, etc., and the present invention is not limited thereto.

[0050] Figure 6 is a schematic diagram of a regression model according to an embodiment of the present invention. Figure 6 The regression model 61 may be a logistic regression model, which mainly includes a net input function 620, an activation function 630, a threshold function 640, and an error 650. The net input function is a linear combination of the calculation weights and the input data. The activation function may be a sigmoid activation function. The error may be a logarithmic loss function.

[0051] When training the regression model 61, at least one set of input data may be input to the regression model 61. After setting the error, the weights of the input data in the net input function are trained by gradient descent or maximum likelihood. A set of input data corresponds to input values X1 to X2 included in an image block. m (i.e., the characteristic values corresponding to each image block, including standard deviation, coefficient of variation, and mean, etc.) and expected output I (i.e., good product or NG product, etc.). After training, the regression model 61 can generate each input value X1~X m The weights W1~W m And the weight W0 of the expected output I. After the training of the regression model 61 is completed, the regression model 61 is trained according to the weights W0~W m For at least one set of input values X1~X mClassification is performed to generate corresponding at least one set of input values X1~X m In other words, a classification result corresponds to an image block in an original image, and the classification result LB1 may include good quality or bad quality. In one embodiment, if the classification result LB1 of at least one image block in an original image is bad quality, the final classification result of the original image is bad quality.

[0052] After obtaining the final classification result, the processor 112 may compare the final classification result with the expected output I to determine whether the final classification result meets expectations. If it meets expectations, the regression model trained to correspond to the defect of the object to be tested may be set as the optimal regression model. If it does not meet expectations, the processor 112 may reselect the feature values used to train the regression model 61 and retrain the regression model 61. In one embodiment, the processor 112 may use a Chi-Squared Test of Independence to determine whether the final classification result meets expectations. For example, the processor 112 may use a Chi-Squared Test of Independence to determine whether there is a significant relationship between the final classification result and the expected output I. The present invention is not limited to the method of testing. By repeating the above operation, the optimal regression model corresponding to the defect of the object to be tested can be gradually trained, thereby improving the detection accuracy of the regression model 61 for the defect of the object to be tested. In one embodiment, the Chi-Squared Test of Independence is, for example, a Pearson's Chi-Squared test, but the present invention is not limited to this.

[0053] For example, the comparison between the expected output I and the final classification result generated by the trained regression model 61 is shown in Table 3. The processor 112 can use a chi-square independence test to determine whether there is a significant relationship between the expected output I and the final classification result.

[0054] Table 3

[0055] Wafer ID Expected Output I Final classification results Whether it meets expectations 1 Good product Good product yes 2 NG products NG products yes 3 Good product Good product yes 4 Good product NG products no 5 NG products NG products yes … … … … N Good product Good product yes

[0056] When performing defect detection on an unlabeled image (also referred to as the third image) based on the optimal regression model, processor 112 calculates multiple feature values corresponding to each image block in the third image. Through operations on an excitation function 630 and a threshold function 640, a final classification result corresponding to the third image is generated. This final classification result may indicate whether the object under test corresponding to the third image is a good product or a bad product.

[0057] In another embodiment, processor 112 may further color-code image blocks identified as NG products in the third image, and determine the defect pattern in the third image based on the distribution of the color-coded image blocks to determine the type of defect. For example, the distribution of image blocks identified as NG products may be irregularly scattered or regularly arranged to form a specific pattern (e.g., zebra stripes, sunburst patterns, wavy patterns, etc.), and processor 112 may determine the type of defect corresponding to the third image based on the distribution. In one embodiment, a visual inspector may determine the defect pattern based on the distribution of image blocks identified as NG products, or processor 112 may determine the defect pattern, but the present invention is not limited thereto.

[0058] In summary, the automated optical inspection method provided by the embodiments of the present invention, by segmenting the original image into multiple image blocks, not only prevents the dilution of extreme values of individual pixels, which can lead to the neglect of point-like defects, but also utilizes the classification results of each image block to determine whether the defective image block has formed specific defective lines. Therefore, by segmenting the wafer image into multiple image blocks and classifying each image block, the method provided by the present invention facilitates macroscopic determination of the presence of specific lines in the wafer image, thereby confirming the wafer's flatness.

[0059] In one embodiment, the regression model 61 is implemented in the optical detection device 110. For example, the regression model 61 can be implemented in software, firmware, or hardware. Figure 2 Alternatively, in another embodiment, the regression model 61 is implemented in another remote server (not shown, also referred to as a server network). This remote server can perform the automatic optical inspection.

[0060] The optical inspection device 110 can be connected to a remote server via a wired or wireless network. In this embodiment, the remote server may include a processor and a storage device to coordinate the operations required to perform the steps. In one embodiment, the optical inspection device 110 can provide the original image to the remote server via the network to perform the training in steps S304-S310. The relevant operational details have been described above and will not be repeated here.

[0061] The present invention further provides a non-transitory computer-readable recording medium containing a computer program for executing the steps of the automated optical inspection method. The computer program comprises a plurality of code snippets (e.g., a code snippet for creating an organizational chart, a code snippet for creating a sign-off form, a code snippet for configuring settings, and a code snippet for deploying the program). When loaded into a processor and executed, these code snippets complete the steps of the automated optical inspection method.

[0062] In summary, the automated optical inspection method, automated optical inspection system, and recording medium provided by embodiments of the present invention can effectively improve the accuracy of defect detection by automated optical inspection devices or systems. By segmenting an image into multiple image blocks and training a classification model using the feature values corresponding to each image block, the embodiments of the present invention can effectively improve the training efficiency and accuracy of the classification model. In one embodiment, the automated optical inspection method, automated optical inspection system, and recording medium provided by embodiments of the present invention can replace the microscopic inspection of wafer defect patterns followed by manual macroscopic interpretation of wafer flatness, thereby improving inspection efficiency and reliability and reducing labor costs.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An automatic optical inspection method, applicable to an optical inspection system including an optical lens and a processor, comprising: Acquiring an original image of the surface of the object to be measured through the optical lens, wherein the original image includes a plurality of first images; performing edge detection on the original image to obtain an edge image having an edge pattern, wherein the edge image includes a plurality of second images having edge patterns corresponding to the plurality of first images; Calculating at least one of a maximum value, a minimum value, and an average value of pixel values in each of the plurality of second images; grouping the plurality of second images into a plurality of image blocks according to unit area, and calculating a plurality of characteristic values according to at least one of the maximum value, the minimum value, and the average value corresponding to a group of second images included in each of the plurality of image blocks; as well as A regression model corresponding to the defects of the object to be tested is trained according to the multiple characteristic values and the object to be tested data to obtain an optimal regression model. 2 . The automated optical inspection method according to claim 1 , wherein the plurality of characteristic values comprises at least one of a standard deviation, a coefficient of variation, a mean, a range, and a mean absolute deviation.

3. The automated optical inspection method according to claim 1 , wherein the DUT data includes expected outputs, and the step of training the regression model corresponding to the DUT defect based on the plurality of feature values and the DUT data to obtain the optimal regression model comprises: The plurality of feature values and the expected output are input as input data to the regression model to train the regression model, and a plurality of weights corresponding to the plurality of feature values are generated.

4. The automated optical inspection method according to claim 3, further comprising: Classifying the plurality of feature values according to the plurality of weights based on the trained regression model to generate classification results corresponding to the plurality of feature values; as well as The classification result is compared with the expected output to determine whether the classification result meets the expectations.

5. The automated optical inspection method according to claim 4, further comprising: If the classification result meets the expectation, the trained regression model is set as the optimal regression model, and If the classification result does not meet expectations, a plurality of feature values for training the regression model are reselected, and the regression model is retrained.

6. The automated optical inspection method according to claim 1 , further comprising: The third image is classified based on the optimal regression model to generate a classification result corresponding to the third image.

7. An automatic optical inspection system comprising: an optical lens configured to obtain an original image of a surface of an object to be measured, wherein the original image includes a plurality of first images; as well as A processor, coupled to the optical lens, is configured to: performing edge detection on the original image to obtain an edge image having an edge pattern, wherein the edge image includes a plurality of second images having edge patterns corresponding to the plurality of first images; Calculating at least one of a maximum value, a minimum value, and an average value of pixel values in each of the plurality of second images; grouping the plurality of second images into a plurality of image blocks according to unit area, and calculating a plurality of characteristic values according to at least one of the maximum value, the minimum value, and the average value corresponding to a group of second images included in each of the plurality of image blocks; as well as A regression model corresponding to the defects of the object to be tested is trained according to the multiple characteristic values and the object to be tested data to obtain an optimal regression model. 8 . The automated optical inspection system according to claim 7 , wherein the plurality of characteristic values comprises at least one of a standard deviation, a coefficient of variation, a mean, a range, and a mean absolute deviation.

9. The automated optical inspection system of claim 7, wherein the object data comprises an expected output, and the processor is configured to: The plurality of feature values and the expected output are input as input data to the regression model to train the regression model, and a plurality of weights corresponding to the plurality of feature values are generated.

10. The automated optical inspection system of claim 9, wherein the processor is further configured to: classifying the plurality of feature values according to the plurality of weights based on the trained regression model to generate classification results corresponding to the plurality of feature values; and The classification result is compared with the expected output to determine whether the classification result meets the expectations.

11. An automatic optical inspection system according to claim 10, wherein if the classification result meets expectations, the processor sets the trained regression model as the optimal regression model, and if the classification result does not meet expectations, the processor reselects multiple feature values used to train the regression model and retrains the regression model.

12. The automated optical inspection system of claim 7, wherein the processor is configured to: The third image is classified based on the optimal regression model to generate a classification result corresponding to the third image.

13. The automated optical inspection system of claim 7, wherein the processor is configured in a remote server.

14. A non-transitory computer-readable recording medium recording a program loaded by a processor to execute the following steps: Acquire an original image of the surface of the object to be measured through an optical lens, wherein the original image includes a plurality of first images; performing edge detection on the original image to obtain an edge image having an edge pattern, wherein the edge image includes a plurality of second images having edge patterns corresponding to the plurality of first images; Calculating at least one of a maximum value, a minimum value, and an average value of pixel values in each of the plurality of second images; grouping the plurality of second images into a plurality of image blocks according to unit area, and calculating a plurality of characteristic values according to at least one of the maximum value, the minimum value, and the average value corresponding to a group of second images included in each of the plurality of image blocks; as well as A regression model corresponding to the defects of the object to be tested is trained according to the multiple characteristic values and the object to be tested data to obtain an optimal regression model.

Images