Wafer image processing method, electronic equipment and storage medium

Abstract

The invention provides a wafer image processing method, electronic equipment and a storage medium, and the method comprises the steps: obtaining multiple frames of original images of a to-be-detected wafer and multi-modal scanning information bound with each original image, the multi-mode scanning information comprises the Z-axis height when the original image is shot, an illumination mode and a two-dimensional coordinate of a two-dimensional platform where the to-be-detected wafer is located; according to the multi-modal scanning information, performing denoising processing on each original image to obtain each processed image; and according to the multi-modal scanning information, splicing the processed images to obtain a spliced image, the spliced image being used as an input image of a defect identification model, so that the defect identification model performs defect detection on the to-be-detected wafer based on the spliced image, and the processing effect of the wafer image is improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and more specifically, to a wafer image processing method, electronic device, and storage medium. Background Technology

[0002] In semiconductor manufacturing, wafer defect detection is a crucial step in ensuring chip yield. High-resolution cameras are typically used to scan and image the wafer surface, and image preprocessing steps such as noise reduction, enhancement, and stitching are employed to improve the accuracy of subsequent defect identification.

[0003] However, in existing technologies, the image preprocessing module operates independently, resulting in a lack of targeted preprocessing strategies and difficulty in resolving the contradiction between noise suppression and the preservation of minor defects. For example, general noise reduction algorithms may smooth out minor scratches and defects, while over-enhancement can amplify noise and produce false defects. In addition, existing technologies rely heavily on feature-point-based matching algorithms during image stitching, which can easily produce blurred feature points, leading to stitching misalignment or ghosting phenomena.

[0004] Therefore, the preprocessing of wafer images in the existing technology for wafer defect detection has certain limitations. Summary of the Invention

[0005] The purpose of this application is to address the shortcomings of the prior art by providing a wafer image processing method, electronic device, and storage medium to solve the practical problem of limitations in the preprocessing of wafer images during wafer defect detection.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, embodiments of this application provide a wafer image processing method, the method comprising: Acquire multiple frames of raw images of the wafer under test and multimodal scanning information bound to each raw image. The multimodal scanning information includes the Z-axis height, illumination mode, and two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located when the raw images are captured. Based on the multimodal scanning information, each of the original images is denoised to obtain a processed image. Based on the multimodal scanning information, the processed images are stitched together to obtain a stitched image. The stitched image is used as the input image for the defect recognition model, so that the defect recognition model can perform defect detection on the wafer under test based on the stitched image.

[0007] As an optional implementation, the step of denoising each of the original images according to the multimodal scanning information to obtain each processed image includes: Based on the lighting pattern, determine the target denoising algorithm; Based on the multimodal scanning information, the target denoising algorithm is used to denoise the original image to obtain the processed image.

[0008] As an optional implementation, determining the target denoising algorithm based on the lighting pattern includes: If the lighting mode is a bright field lighting mode, then the target denoising algorithm is determined to be a guided filtering algorithm that introduces Z-axis height. If the lighting mode is a dark field lighting mode, then the target denoising algorithm is determined to be the CLAHE algorithm.

[0009] As an optional implementation, the step of denoising the original image using the target denoising algorithm based on the multimodal scanning information to obtain the processed image includes: In the bright field illumination mode, a template image of the wafer under test is acquired; Based on the template image and the Z-axis height when the original image was captured, the original image is linearly filtered using the guided filtering algorithm that introduces the Z-axis height to obtain the processed image.

[0010] As an optional implementation, the step of performing linear filtering on the original image using the guided filtering algorithm that incorporates the Z-axis height, based on the template image and the Z-axis height when the original image was captured, includes: Based on the template image, determine the guide image for the original image; The defocus height is determined based on the Z-axis height and the preset target focal plane height; Within the local window corresponding to the original image, based on the guiding image, the defocus height, and the original image, the minimum guided filter cost function corresponding to the local window is determined with the goal of minimizing the error between the processed image and the original image within the local window; Based on the minimized guided filtering cost function corresponding to the local window, the first target linear coefficient and the second target linear coefficient corresponding to the local window are determined, and the original image is subjected to linear filtering processing based on the first target linear coefficient, the second target linear coefficient and the guided image.

[0011] As an optional implementation, the step of denoising the original image using the target denoising algorithm based on the multimodal scanning information to obtain the processed image further includes: In the dark field illumination mode, the CLAHE algorithm is used to divide the original image into multiple sub-regions, and histogram equalization is performed on each sub-region to obtain the histogram of each sub-region. Based on a preset amplitude limiting threshold, the histograms of each sub-region are cropped and redistributed to obtain the processed histograms of each sub-region. Based on the processed histograms of each sub-region, the CDF of the original image is determined, and the original image is then equalized according to the CDF to obtain the processed image.

[0012] As an optional implementation, the step of stitching together the processed images based on the multimodal scanning information to obtain a stitched image includes: Based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, the processed images are initially stitched together to obtain the initial stitched image. Based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, a sub-pixel registration algorithm is used to adjust the overlapping areas of each adjacent processed image in the preliminary stitched image to obtain the stitched image.

[0013] As an optional implementation, the step of adjusting the overlapping areas of adjacent processed images in the preliminary stitched image by using a sub-pixel registration algorithm based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured includes: Determine whether there is an overlapping region between the first processed image and the adjacent second processed image; If so, the overlapping region is extracted from the first processed image as the first image, and the overlapping region is extracted from the second processed image as the second image; Using the subpixel registration algorithm, Fourier transform and inverse Fourier transform are performed on the first image and the second image respectively to obtain the subpixel-level displacement deviation between the first processed image and the second processed image. Based on the subpixel-level displacement deviation, the position of the second processed image is adjusted to obtain an adjusted second processed image, so that the first processed image and the adjacent adjusted second processed image do not overlap.

[0014] Secondly, embodiments of this application provide a wafer image processing apparatus, the apparatus comprising: The acquisition module is used to acquire multiple frames of original images of the wafer under test and multimodal scanning information bound to each of the original images. The multimodal scanning information includes the Z-axis height, illumination mode, and two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located when the original images are captured. The processing module is used to perform denoising processing on each of the original images according to the multimodal scanning information to obtain each processed image; The stitching module is used to stitch together the processed images according to the multimodal scanning information to obtain a stitched image. The stitched image is used as the input image of the defect recognition model so that the defect recognition model can perform defect detection on the wafer under test based on the stitched image.

[0015] As an optional implementation, the processing module is specifically used for: Based on the lighting pattern, determine the target denoising algorithm; Based on the multimodal scanning information, the target denoising algorithm is used to denoise the original image to obtain the processed image.

[0016] As an optional implementation, the processing module is specifically used for: If the lighting mode is a bright field lighting mode, then the target denoising algorithm is determined to be a guided filtering algorithm that introduces Z-axis height. If the lighting mode is a dark field lighting mode, then the target denoising algorithm is determined to be the CLAHE algorithm.

[0017] As an optional implementation, the processing module is specifically used for: In the bright field illumination mode, a template image of the wafer under test is acquired; Based on the template image and the Z-axis height when the original image was captured, the original image is linearly filtered using the guided filtering algorithm that introduces the Z-axis height to obtain the processed image.

[0018] As an optional implementation, the processing module is specifically used for: Based on the template image, determine the guide image for the original image; The defocus height is determined based on the Z-axis height and the preset target focal plane height; Within the local window corresponding to the original image, based on the guiding image, the defocus height, and the original image, the minimum guided filter cost function corresponding to the local window is determined with the goal of minimizing the error between the processed image and the original image within the local window; Based on the minimized guided filtering cost function corresponding to the local window, the first target linear coefficient and the second target linear coefficient corresponding to the local window are determined, and the original image is subjected to linear filtering processing based on the first target linear coefficient, the second target linear coefficient and the guided image.

[0019] As an optional implementation, the processing module is further used for: In the dark field illumination mode, the CLAHE algorithm is used to divide the original image into multiple sub-regions, and histogram equalization is performed on each sub-region to obtain the histogram of each sub-region. Based on a preset amplitude limiting threshold, the histograms of each sub-region are cropped and redistributed to obtain the processed histograms of each sub-region. Based on the processed histograms of each sub-region, the CDF of the original image is determined, and the original image is then equalized according to the CDF to obtain the processed image.

[0020] As an optional implementation, the splicing module is specifically used for: Based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, the processed images are initially stitched together to obtain the initial stitched image. Based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, a sub-pixel registration algorithm is used to adjust the overlapping areas of each adjacent processed image in the preliminary stitched image to obtain the stitched image.

[0021] As an optional implementation, the splicing module is specifically used for: Determine whether there is an overlapping region between the first processed image and the adjacent second processed image; If so, the overlapping region is extracted from the first processed image as the first image, and the overlapping region is extracted from the second processed image as the second image; Using the subpixel registration algorithm, Fourier transform and inverse Fourier transform are performed on the first image and the second image respectively to obtain the subpixel-level displacement deviation between the first processed image and the second processed image. Based on the subpixel-level displacement deviation, the position of the second processed image is adjusted to obtain an adjusted second processed image, so that the first processed image and the adjacent adjusted second processed image do not overlap.

[0022] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the wafer image processing method as described in the first aspect above.

[0023] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the wafer image processing method described in the first aspect above.

[0024] The beneficial effects of this application are: This application provides a wafer image processing method, electronic device, and storage medium. The method acquires multiple frames of original images of a wafer under test and multimodal scanning information bound to each original image. The multimodal scanning information includes the Z-axis height, illumination mode, and two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located when the original images are captured. Based on the multimodal scanning information, each original image is denoised to obtain processed images. Then, based on the multimodal scanning information, the processed images are stitched together to obtain a stitched image. The stitched image is used as the input image for a defect recognition model, enabling the defect recognition model to perform defect detection on the wafer under test based on the stitched image. Since wafer image processing is based on multimodal scanning information bound to each original image of the wafer under test, differential denoising can be achieved, preserving scratch defects in each original image and filtering out noise. Furthermore, by using the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured as the position reference, a stitched image is generated, which makes the stitched image contain clear textures and scratch defects across the entire wafer under test, greatly improving the processing effect of wafer images and thus improving the accuracy and efficiency of wafer defect detection. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 1 ; Figure 2 This is a schematic diagram of the structure of the wafer defect detection device provided in the embodiments of this application; Figure 3 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 2 ; Figure 4 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 3 ; Figure 5 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 4 ; Figure 6 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 5 ; Figure 7 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 6 ; Figure 8 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 7 ; Figure 9 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 8 ; Figure 10 A module structure diagram of the wafer image processing apparatus provided in the embodiments of this application; Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0027] Figure reference numerals: Imaging unit: 10; Robotic arm: 20; Two-dimensional platform: 30; Wafer under test: 40; Industrial control computer: 50; Height adjustment unit: 60. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0029] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0030] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0031] In semiconductor manufacturing, wafer defect detection is a crucial step in ensuring chip yield. Current technologies employ independent image preprocessing modules, resulting in a lack of targeted preprocessing strategies and difficulty in resolving the conflict between noise suppression and the preservation of subtle defects. Furthermore, existing technologies often rely on feature-point matching algorithms for image stitching, which can easily produce blurred feature points, leading to stitching misalignments or ghosting phenomena. In other words, current technologies have limitations in preprocessing wafer images during wafer defect detection.

[0032] Based on the above-mentioned problems, this application provides a wafer image processing method that performs denoising and stitching processing on each original image based on the multimodal scanning information bound to each original image, thereby improving the processing effect of wafer images.

[0033] Figure 1 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 1 The subject executing this method can be any electronic device with computing power. For example... Figure 1 As shown, the method includes: S101. Acquire multiple frames of original images of the wafer under test and multimodal scanning information bound to each original image. The multimodal scanning information includes the Z-axis height, illumination mode, and two-dimensional coordinates of the two-dimensional platform where the wafer under test is located when the original images are captured.

[0034] Optionally, Figure 2 This is a schematic diagram of the wafer defect detection device provided in the embodiments of this application, as shown below. Figure 2 As shown, the wafer defect detection device includes an imaging unit 10, a robotic arm 20, a two-dimensional platform 30, an industrial control computer 50, and a height adjustment unit 60.

[0035] During the full-area scanning inspection of the wafer under test, the robotic arm 20 picks up the wafer 40 from the wafer cassette, precisely transfers it, and places it on the initial position of the two-dimensional platform 30. The two-dimensional platform 30 supports the wafer 40 and performs high-precision two-dimensional motion under the control of the industrial computer 50. During the two-dimensional motion, a grating ruler is used for closed-loop position feedback to ensure positioning accuracy. The height adjustment unit 60 is used to precisely adjust the relative distance between the imaging unit 10 and the surface of the wafer 40 under test, so that the imaging unit 10 can capture a clear image of the wafer 40.

[0036] The wafer defect detection device moves the wafer 40 under test through the two-dimensional platform 30, and precisely adjusts the relative distance between the imaging unit 10 and the surface of the wafer 40 under test through the height adjustment unit 60. It acquires multiple frames of original images of the wafer 40 under test at different physical coordinate positions, with each frame of original image corresponding to a local area of ​​the wafer 40 under test.

[0037] Multiple frames of raw images of the wafer under test are bound to the multimodal scanning information captured at the time of shooting. Before preprocessing the multiple frames of raw images of the wafer under test, the multiple frames of raw images of the wafer under test and the multimodal scanning information bound to each raw image are obtained.

[0038] The first wafer under test Taking the original image of frame 1 as an example, the first frame The multimodal scanning information bound to the original frame image specifically includes: the image captured at the beginning of the frame. Z-axis height of the original image frame Lighting modes Two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located Among them, the first shot Z-axis height of the original image frame For filming the first The relative distance between imaging unit 10 and wafer 40 under test when the original image is captured.

[0039] S102. Based on the multimodal scanning information, denoise each original image is processed to obtain each processed image.

[0040] Optionally, based on the Z-axis height and illumination mode in the multimodal scanning information bound to each original image, differential denoising processing is performed on each original image to obtain each processed image.

[0041] Specifically, in the case of the first When denoising the original frame image, it is necessary to consider the number of frames captured. Z-axis height of the original image frame and lighting modes Determine the height relative to the Z-axis and lighting modes An adaptive denoising algorithm is used to preserve the first denoising value to the greatest extent possible. Remove scratches and defects from the original image and filter out noise.

[0042] S103. Based on the multimodal scanning information, the processed images are stitched together to obtain a stitched image. The stitched image is used as the input image of the defect recognition model so that the defect recognition model can perform defect detection on the wafer under test based on the stitched image.

[0043] Optionally, based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test is located in the multimodal scanning information bound to each original image, the processed images are globally stitched together to obtain the stitched image.

[0044] Specifically, using the two-dimensional coordinates of the two-dimensional platform where the wafer under test is located when each original image is captured as the position reference, the discrete local processed images are stitched together to form a panoramic image covering the entire area of ​​the wafer under test, generating a stitched image. The stitched image contains clear textures (such as lattice lines) and scratches and defects of the entire area of ​​the wafer under test.

[0045] The stitched image is used as the input image to the defect recognition model so that the defect recognition model can detect defects in the wafer under test based on the stitched image, locate the defect location, determine the cause of the defect, and thus improve the wafer manufacturing process.

[0046] In this embodiment, multiple frames of original images of the wafer under test and multimodal scanning information bound to each original image are acquired. The multimodal scanning information includes the Z-axis height, illumination mode, and two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located when the original images are captured. Based on the multimodal scanning information, each original image is denoised to obtain processed images. Then, based on the multimodal scanning information, the processed images are stitched together to obtain a stitched image. The stitched image is used as the input image for a defect recognition model, enabling the model to detect defects in the wafer under test based on the stitched image. Since wafer image processing is based on multimodal scanning information bound to each original image of the wafer under test, differentiated denoising can be achieved, preserving scratch defects in each original image while filtering out noise. Furthermore, by using the two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located when each original image is captured as the position reference, the stitched image contains clear textures and scratch defects across the entire wafer under test, greatly improving the wafer image processing effect and thus enhancing the accuracy and efficiency of wafer defect detection.

[0047] Figure 3 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 2 ,like Figure 3 As shown, in step S102 above, each original image is denoised based on the multimodal scanning information to obtain each processed image, including: S201. Determine the target denoising algorithm based on the lighting pattern.

[0048] Optionally, different image denoising algorithms are used for different lighting modes of the original images to perform differentiated denoising processing on each original image. The lighting modes include bright field lighting mode and dark field lighting mode.

[0049] Specifically, according to the shooting of the first Illumination mode when the original image is framed Determine the lighting mode Adapted target denoising processing algorithm.

[0050] S202. Based on the multimodal scanning information, a target denoising algorithm is used to denoise the original image to obtain the processed image.

[0051] Optionally, a target denoising algorithm is used to perform differential denoising on the original image based on multimodal scanning information to obtain the processed image.

[0052] Among them, the target denoising processing algorithm can preserve the scratches and defects of the original image captured under the corresponding lighting mode and filter out noise, avoiding the flattening of minor scratches and defects during denoising processing, and preventing the generation of false defects due to excessive enhancement of image noise.

[0053] In this embodiment, a target denoising algorithm is determined based on the illumination mode, and the algorithm is applied to the original image based on multimodal scanning information to obtain the processed image. This process preserves the scratches and defects of the original image while filtering out noise, preventing the smoothing of minor scratches and defects during denoising and avoiding the over-enhancing of the image to amplify noise and create false defects, thus improving the denoising effect.

[0054] Figure 4 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 3 ,like Figure 4 As shown, in step S201 above, determining the target denoising algorithm based on the lighting pattern includes: S301. If the lighting mode is bright field lighting mode, then the target denoising algorithm is determined to be a guided filtering algorithm that introduces Z-axis height.

[0055] Optionally, if the first shot is taken Illumination mode when the original image is framed If it is a bright field lighting mode, then determine the first... The target denoising algorithm for the original frame image is a guided filtering algorithm that introduces Z-axis height.

[0056] A guided filtering algorithm incorporating Z-axis height is used to process the images captured under bright field illumination mode. The original image is denoised, which can smooth out noise while preserving lattice lines and scratches similar to lattice lines, based on the Z-axis height. Confirm the shooting of the first The defocusing degree of the wafer under test is determined by the original image frame. The denoising strategy for original images taken when the wafer is not at its optimal focal plane is dynamically adjusted to compensate for the image quality degradation caused by defocusing. This allows the guided filtering algorithm incorporating Z-axis height to adaptively adjust noise smoothing intensity and brightness, thereby improving the image quality of images taken under bright field illumination. The denoising effect of the original frame image.

[0057] S302. If the lighting mode is dark field lighting mode, then the target denoising algorithm is determined to be the CLAHE algorithm.

[0058] Optionally, if the first shot is taken Illumination mode when the original image is framed If it is a dark field lighting mode, then determine the first... The target denoising algorithm for the original frame image is the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm.

[0059] The CLAHE algorithm was used to analyze the images taken in dark lighting mode. Denoising the original image frame can enhance image contrast and highlight defects while suppressing background noise.

[0060] In this embodiment, if the lighting mode is bright field lighting, the target denoising algorithm is determined to be a guided filtering algorithm that introduces Z-axis height; if the lighting mode is dark field lighting, the target denoising algorithm is determined to be the CLAHE algorithm. Differentiated denoising processing is performed on the original images under different lighting modes to improve the image denoising effect and provide high-quality image data for subsequent defect detection and identification.

[0061] Figure 5 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 4 ,like Figure 5 As shown, in step S202 above, based on the multimodal scanning information, a target denoising algorithm is used to denoise the original image to obtain the processed image, including: S401. In bright field illumination mode, acquire the template image of the wafer to be tested.

[0062] Optionally, if the first shot is taken Illumination mode when the original image is framed In bright-field illumination mode, a template image of the wafer under test is acquired. The template image is a pre-scanned image of a defect-free standard wafer from the same batch as the wafer under test, taken at the optimal focal plane.

[0063] The template image is used as a reference for guided filtering. The template image contains the standard lattice lines of the wafer and is used to guide the guided filtering algorithm that introduces Z-axis height. This ensures that the guided filtering algorithm that introduces Z-axis height only smooths noise and blurred areas, while retaining the lattice lines and scratches in the original image that are similar to the standard lattice lines.

[0064] S402. Based on the template image and the Z-axis height when the original image was captured, a guided filtering algorithm that incorporates the Z-axis height is used to perform linear filtering on the original image to obtain the processed image.

[0065] Optionally, based on the template image and the photographed first... Z-axis height of the original image frame The guided filtering algorithm that incorporates Z-axis height is used to determine the first... The linear relationship between the original image and the template image in the first frame, adaptively adjusting the noise smoothing intensity and brightness, for the first frame... The original image of frame 1 is linearly filtered to obtain the first frame. The processed image enhances the performance of images captured in bright field illumination mode. The denoising effect of the original frame image.

[0066] Among them, the Z-axis height It is used to dynamically control the filter strength, maximize the compensation for image blur caused by defocus, enhance smoothing when there is slight defocus, and further enhance smoothing when there is severe defocus, and suppress false brightness changes caused by defocus.

[0067] In this embodiment, under bright-field illumination mode, a template image of the wafer under test is acquired. Based on the template image and the Z-axis height when the original image was captured, a guided filtering algorithm incorporating Z-axis height is used to perform linear filtering on the original image to obtain the processed image. This guided filtering algorithm incorporating Z-axis height achieves defocus blur compensation and avoids image distortion.

[0068] Figure 6 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 5 ,like Figure 6 As shown, in step S402 above, based on the template image and the Z-axis height when the original image was captured, a guided filtering algorithm incorporating Z-axis height is used to perform linear filtering on the original image, including: S501. Determine the guide image of the original image based on the template image.

[0069] Optionally, the first Original image of the frame Align with the template image, and align the elements in the template image with the first... Original image of the frame The aligned region is the first Guide image of the original frame Guide image The Middle The value of each pixel is , No. Original image of the frame The Middle The value of each pixel is .

[0070] In local window In the middle, the first output after filtering Image after frame processing It is the first Guide image of the original frame A linear transformation. Where, the local window... Represented by pixels The radius of the center is A two-dimensional window.

[0071] Specifically, the filtered output of the first Image after frame processing The Middle The value of each pixel With the Guide image of the original frame The Middle The value of each pixel The linear transformation relationship is expressed by the following formula:

[0072] in, For the first Image after frame processing The Middle The value of each pixel. and Local windows The corresponding first target linear coefficients and second target linear coefficients, For the first Guide image of the original frame The Middle The value of each pixel.

[0073] S502. Determine the defocus height based on the Z-axis height and the preset target focal plane height.

[0074] Alternatively, in an optical system, when the objective lens deviates from the optimal focal plane, i.e., when the first... The original image frame is at a non-optimal focal plane height (i.e., the preset target focal plane height). During filming, the first The original image of the first frame will be blurry when the objective lens of imaging unit 10 deviates from the optimal focal plane. The attenuation of lattice edges in the original frame image will be more obvious, and it will introduce additional noise and blur, as well as reduce the contrast between defects and background.

[0075] According to the filming Z-axis height of the original image frame and target focal plane height The number of shots is determined based on the following formula. Defocus height at frame original image :

[0076] in, For filming the first Defocus height of the original frame image. For filming the first Z-axis height of the original image frame. The target focal plane height.

[0077] S503. Within the local window corresponding to the original image, based on the guiding image, defocus height, and the original image, determine the minimum guiding filter cost function corresponding to the local window with the goal of minimizing the error between the processed image and the original image within the local window.

[0078] Optionally, in a local window In the middle, according to the first Guide image of the original frame Defocus height Passing the exam Original image of the frame , in local window The first internal output Image after frame processing With the input of the first Original image of the frame The local window is determined based on the following formula, with the goal of minimizing the error. The corresponding minimized guided filter cost function:

[0079]

[0080] in, It is the output of the first Image after frame processing At pixel value With the input of the first Original image of the frame At pixel value The cost function that minimizes the error between them. and Local windows The corresponding first linear coefficient and second linear coefficient, For the first Guide image of the original frame At pixel The value, For a highly adaptive regularization function, prevent the first linear coefficient Too large For the target focal plane height The regularization parameter at time, The preset adjustment coefficient, Indicates the defocus height The degree of image blur caused Used for adjustment right The extent of the impact. These are preset weighting coefficients used to adjust the Z-axis height constraint. The weights are adjusted to balance the filtering error with the suppression strength of brightness shift. The constraint function for the Z-axis height. This is the reference offset, used to constrain the second linear coefficient when out of focus. To avoid excessive brightness shifts.

[0081] S504. Based on the minimized guided filtering cost function corresponding to the local window, determine the first target linear coefficient and the second target linear coefficient corresponding to the local window, and perform linear filtering on the original image based on the first target linear coefficient, the second target linear coefficient, and the guided image.

[0082] Optionally, in a local window Within, the cost function of minimizing the guided filter is given by the first linear coefficients. Second linear coefficient Find the partial derivatives, and ensure the zero derivative is zero. Determine the local window based on the following formula. The corresponding first objective linear coefficient Second target linear coefficients :

[0083]

[0084]

[0085]

[0086] in, and Local windows The corresponding first target linear coefficients and second target linear coefficients, For local windows Inner Guide image of the original frame With the Original image of the frame covariance, For the defocus brightness constraint function, For local windows Inner Original image of the frame The pixel mean, representing the local window. Inner Original image of the frame average brightness For local windows Inner Guide image of the original frame The pixel mean, representing the local window. Inner Guide image of the original frame average brightness For local windows Inner Guide image of the original frame The pixel variance represents the first... Guide image of the original frame In local window Brightness fluctuations within, As the reference offset, For defocus smoothing constraint function, For preset weighting coefficients, The constraint function for the Z-axis height. It is a highly adaptive regularization function. For local windows The number of pixels present in the memory.

[0087] Based on minimizing the guided filter cost function, if the first shot is taken... Z-axis height of the original image frame For the target focal plane height Then the defocus height The constraint function for Z-axis height is 0. Also equal to 0, a highly adaptive regularization function And the defocus brightness constraint function If the result is 0, substitute these values ​​into the above method for determining the local window. The corresponding first objective linear coefficient Second target linear coefficients The formula determines the local window. The corresponding first objective linear coefficient Second target linear coefficients At this point, there is no need to introduce the Z-axis height; only the first... The original image of the frame is processed based on the first frame. Guide image of the original frame Guided filtering.

[0088] Accordingly, if the first shot is taken Slightly out of focus when viewing the original image frame, Z-axis height At the height of the target focal plane The absolute value of the difference is the defocus height. If the value is greater than or equal to the first preset threshold, then a highly adaptive regularization function is applied. Slightly larger Defocus smoothing constraint function Increase these values ​​and substitute them into the above method for determining the local window. The corresponding first objective linear coefficient Second target linear coefficients The formula determines the local window. The corresponding first objective linear coefficient Second target linear coefficients At this point, the Z-axis height is introduced, for the first... The original image of the frame is processed based on the first frame. Guide image of the original frame The guided filter balances the filtering error with the suppression strength of brightness shift, compensates for slight blur caused by defocus by enhancing smoothness, and prevents brightness shift.

[0089] If the first The original image was severely out of focus, with a Z-axis height of [missing information]. At the height of the target focal plane The absolute value of the difference is the defocus height. If the value is greater than or equal to the second preset threshold, then a highly adaptive regularization function is applied. Much larger Defocus smoothing constraint function Increase these values ​​and substitute them into the above method for determining the local window. The corresponding first objective linear coefficient Second target linear coefficients The formula determines the local window. The corresponding first objective linear coefficient Second target linear coefficients At this point, the Z-axis height is introduced, because Increasing both enhances the linearity coefficients for the first objective. The constraints further enhance the smoothing effect, and Make the second objective linear coefficient offset to reference Contraction is used to suppress false brightness changes caused by severe defocusing. This is based on the first target linear coefficient. Second target linear coefficients For the The original image of the frame is processed based on the first frame. Guide image of the original frame Guided filtering can further balance the suppression strength of filtering error and brightness shift, and improve the linear filtering effect.

[0090] In this embodiment, a guide image for the original image is determined based on the template image, and a defocus height is determined based on the Z-axis height and a preset target focal plane height. Within a local window corresponding to the original image, a minimum guided filtering cost function is determined based on the guide image, the defocus height, and the original image, with the goal of minimizing the error between the processed image and the original image within the local window. Based on this minimum guided filtering cost function, a first target linear coefficient and a second target linear coefficient are determined for the local window. Then, linear filtering is performed on the original image based on the first target linear coefficient, the second target linear coefficient, and the guide image. This balances the filtering error with the suppression strength of brightness shift, compensates for image blur caused by defocus, and improves the linear filtering effect.

[0091] Figure 7 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 6 ,like Figure 7 As shown, step S402 above, based on multimodal scanning information, employs a target denoising algorithm to denoise the original image to obtain the processed image, and further includes: S601. In dark field illumination mode, the CLAHE algorithm is used to divide the original image into multiple sub-regions, and histogram equalization is performed on each sub-region to obtain the histogram of each sub-region.

[0092] Optionally, if the first shot is taken Illumination mode when the original image is framed For dark field lighting mode, then the first... The original image frame is divided into multiple non-overlapping sub-regions.

[0093] The CLAHE algorithm is used to perform histogram equalization on each sub-region, that is, to perform HE transformation on each sub-region to obtain the histogram of each sub-region. The histogram of the sub-region is used to characterize the gray-level distribution characteristics of the sub-region.

[0094] Among them, the The histogram of subregions is the first The number of original pixels for each gray level is represented as .

[0095] S602. Based on the preset amplitude limiting threshold, the histograms of each sub-region are cropped and redistributed to obtain the processed histograms of each sub-region.

[0096] Optionally, based on a preset limiting threshold The histograms of each sub-region are cropped and redistributed to obtain the processed histograms of each sub-region.

[0097] With the first Taking the histogram of the nth sub-region as an example, based on the following formula, the... Crop the histogram of each sub-region:

[0098] in, For the first After cropping the histogram of each sub-region The number of pixels at each gray level. For the first The number of original pixels for each gray level. This is the preset amplitude limit threshold.

[0099] In other words, by setting a preset amplitude limit threshold The number of pixels at each gray level in the histogram of each sub-region is limited to be less than or equal to a preset limiting threshold. .

[0100] The number is determined based on the following formula. The total number of pixels clipped in the histogram of each sub-region :

[0101] in, For the first The total number of pixels clipped in the histogram of each sub-region. For the first The number of original pixels for each gray level. This is the preset amplitude limit threshold.

[0102] Based on the following formula, the first The total number of pixels clipped in the histogram of each sub-region Distribute evenly across all gray levels:

[0103] in, For the first The histogram of the processed sub-regions is shown in section 1. The number of pixels at each gray level. For the first After cropping the histogram of each sub-region The number of pixels at each gray level. For the first The total number of pixels clipped in the histogram of each sub-region. For the first The total gray level of the histogram of each sub-region.

[0104] S603. Based on the processed histogram of each sub-region, determine the CDF of the original image, and perform equalization processing on the original image according to the CDF to obtain the processed image.

[0105] Optionally, the first sub-region is determined based on the processed histogram of each sub-region. Histogram of the original frame image after processing And determine the first based on the following formula The cumulative distribution function (CDF) of the original frame image:

[0106] in, For the first CDF of the original frame image, For target gray level, For the first Histogram of the original frame image after processing.

[0107] For example, Indicates the first The sum of the number of pixels with gray levels 0 to 5 in the original image frame.

[0108] Based on the following formula, for the th The original image of frame 1 is subjected to equalization processing to obtain the first frame. Image after frame processing :

[0109] in, For the first Image after frame processing For the floor function, For the first CDF of the original frame image, For the first The minimum CDF value of the original frame image can be... , Indicates the first The total number of pixels in the original image of the frame. For the first The maximum gray level of the histogram of the original image frame.

[0110] In this embodiment, under dark lighting mode, the CLAHE algorithm is used to divide the original image into multiple sub-regions, and histogram equalization is performed on each sub-region to obtain its histogram. Based on a preset amplitude limiting threshold, the histograms of each sub-region are cropped and redistributed to obtain processed histograms. The CDF of the original image is then determined based on the processed histograms of each sub-region, and the original image is equalized according to the CDF to obtain the processed image. This enhances image contrast, highlights defects, and suppresses background noise.

[0111] Figure 8 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 7 ,like Figure 8 As shown, in step S103 above, the processed images are stitched together based on the multimodal scanning information to obtain a stitched image, including: S701. Based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test is located when each original image is captured, perform preliminary stitching processing on each processed image to obtain a preliminary stitched image.

[0112] Optionally, based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test is located when each original image is captured, each processed image is placed at the corresponding position on the global canvas to perform preliminary stitching processing on each processed image to obtain a preliminary stitched image.

[0113] In other words, the initial stitching process relies on the two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located when each original image is captured. The two-dimensional platform has high precision, and the relative positions of each processed image in the initial stitched image are basically correct. However, there may be slight mechanical errors in the movement of the two-dimensional platform, which may cause pixel-level misalignment in the overlapping areas of some adjacent processed images, resulting in blurring and ghosting. Fine-tuning of the overlapping areas of each adjacent processed image is required.

[0114] S702. Based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test is located when each original image is captured, a sub-pixel registration algorithm is used to adjust the overlapping areas of adjacent processed images in the preliminary stitched image to obtain the stitched image.

[0115] Optionally, based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test is located when each original image is captured, a phase-related subpixel registration algorithm is used to fine-tune the overlapping areas of adjacent processed images in the preliminary stitched image at the subpixel level, so as to eliminate the overlapping areas of adjacent processed images and obtain the stitched image.

[0116] In this embodiment, based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, the processed images are initially stitched together to obtain a preliminary stitched image. Then, based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, a sub-pixel registration algorithm is used to adjust the overlapping areas of adjacent processed images in the preliminary stitched image to obtain the final stitched image. This achieves high-precision and high-reliability image stitching, eliminating image stitching misalignment and ghosting.

[0117] Figure 9 Flowchart of the wafer image processing method provided in the embodiments of this application Figure 8 ,like Figure 9 As shown, in step S702 above, based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test is located when each original image is captured, a sub-pixel registration algorithm is used to adjust the overlapping areas of adjacent processed images in the initially stitched image, including: S801. Determine whether there is an overlapping region between the first processed image and the adjacent second processed image.

[0118] Optionally, based on the captured first processed image The corresponding original image is the two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located, and the second processed image is captured adjacent to the first processed image. The corresponding original image is the two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located, which determines the first processed image. With the adjacent second processed image Are there any overlapping areas between them?

[0119] Among them, the first processed image It can be any of the processed images from multiple frames.

[0120] S802. If so, then the overlapping region is extracted from the first processed image as the first image, and the overlapping region is extracted from the second processed image as the second image.

[0121] Optionally, if the first processed image is determined With the adjacent second processed image If there are overlapping regions, then from the first processed image... Extract the overlapping region as the first image. and from the second processed image Extract the overlapping region as the second image. .

[0122] Among them, the second image It can be seen as the first image Obtained by translation, i.e. ,in, and The assumed offset is used to determine the degree of offset through subsequent steps.

[0123] S803. Using a subpixel registration algorithm, perform Fourier transform and inverse Fourier transform on the first image and the second image respectively to obtain the subpixel-level displacement deviation between the first processed image and the second processed image.

[0124] Optionally, a sub-pixel registration algorithm is used for the first image. Second image Perform Fourier transforms on each image and then use the following formula to transform the first image. Second image Convert to frequency domain:

[0125]

[0126] in, For the first image The expression in the frequency domain, For the second image The expression in the frequency domain, and The first image is shown below. Two-dimensional coordinates and second image Two-dimensional coordinates.

[0127] According to the first image Expression in the frequency domain Second image Expression in the frequency domain The first image is determined based on the following formula. With the second image Cross power spectrum :

[0128] in, For the first image With the second image The cross power spectrum between them and This is a hypothetical offset.

[0129] Based on the following formula, the cross power spectrum Performing the inverse Fourier transform yields the two-dimensional impulse function:

[0130] If the two-dimensional impulse function peaks at the positional deviation, then the image after the first processing... Compared with the second processed image Subpixel-level displacement deviation .

[0131] S804. Based on the subpixel-level displacement deviation, adjust the position of the second processed image to obtain the adjusted second processed image, so that there is no overlapping area between the first processed image and the adjacent adjusted second processed image.

[0132] Optionally, based on the calculated sub-pixel level displacement deviation Adjust the image after the second processing. The adjusted, second-processed image is obtained at its position on the global canvas. So that the image after the first processing is Compared with the adjacent adjusted second processed image There are no overlapping regions.

[0133] Specifically, the image after the second processing is adjusted based on the following formula. The position of the second processed image on the global canvas was adjusted. :

[0134] in, This is the adjusted, second-processed image. This is the image after the second processing. The first processed image Compared with the second processed image Subpixel-level displacement deviation.

[0135] Specifically, after adjusting the image after the second processing... When positioning the image on the global canvas, the second processed image can be... Move left on the global canvas 1 pixel, and move upwards Pixels, to obtain the adjusted second-processed image. .

[0136] In this embodiment, it is determined whether there is an overlapping region between the first processed image and the adjacent second processed image. If so, the overlapping region is extracted from the first processed image as the first image, and the overlapping region is extracted from the second processed image as the second image. A sub-pixel registration algorithm is used to perform Fourier transform and inverse Fourier transform on the first and second images respectively, obtaining the sub-pixel-level displacement deviation between the first and second processed images. Based on the sub-pixel-level displacement deviation, the position of the second processed image is adjusted to obtain an adjusted second processed image, ensuring that there is no overlapping region between the first processed image and the adjacent adjusted second processed image. This improves the accuracy of the sub-pixel-level displacement deviation, thereby improving the image stitching effect and eliminating overlapping regions.

[0137] Based on the same inventive concept, this application also provides a wafer image processing apparatus corresponding to the wafer image processing method. Since the principle of the apparatus in this application is similar to that of the wafer image processing method described above in this application, the implementation of the apparatus can refer to the implementation of the method, and the repeated parts will not be described again.

[0138] Figure 10 A module structure diagram of the wafer image processing apparatus provided in the embodiments of this application is shown below. Figure 10 As shown, the device includes: The acquisition module 901 is used to acquire multiple frames of original images of the wafer under test and multimodal scanning information bound to each original image. The multimodal scanning information includes the Z-axis height, illumination mode, and two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located when the original images are captured. The processing module 902 is used to perform denoising processing on each original image according to the multimodal scanning information to obtain each processed image; The stitching module 903 is used to stitch together the processed images according to the multimodal scanning information to obtain a stitched image. The stitched image is used as the input image of the defect recognition model so that the defect recognition model can perform defect detection on the wafer under test based on the stitched image.

[0139] As an optional implementation, the processing module 902 is specifically used for: Determine the target denoising algorithm based on the lighting pattern; Based on the multimodal scanning information, a target denoising algorithm is used to denoise the original image, resulting in a processed image.

[0140] As an optional implementation, the processing module 902 is specifically used for: If the lighting mode is bright field lighting mode, then the target denoising algorithm is determined to be a guided filtering algorithm that introduces Z-axis height. If the lighting mode is dark field lighting mode, then the target denoising algorithm is determined to be the CLAHE algorithm.

[0141] As an optional implementation, the processing module 902 is specifically used for: In bright field illumination mode, acquire a template image of the wafer under test; Based on the template image and the Z-axis height when the original image was captured, a guided filtering algorithm that incorporates the Z-axis height is used to perform linear filtering on the original image to obtain the processed image.

[0142] As an optional implementation, the processing module 902 is specifically used for: Based on the template image, determine the guide image for the original image; Determine the defocus height based on the Z-axis height and the preset target focal plane height; Within the local window corresponding to the original image, based on the guiding image, defocus height, and the original image, the minimum guided filter cost function corresponding to the local window is determined with the goal of minimizing the error between the processed image and the original image within the local window. Based on the minimized guided filtering cost function corresponding to the local window, the first target linear coefficient and the second target linear coefficient corresponding to the local window are determined, and the original image is subjected to linear filtering processing based on the first target linear coefficient, the second target linear coefficient and the guided image.

[0143] As an optional implementation, the processing module 902 is further configured to: In dark field illumination mode, the CLAHE algorithm is used to divide the original image into multiple sub-regions, and histogram equalization is performed on each sub-region to obtain the histogram of each sub-region. Based on the preset amplitude limit threshold, the histograms of each sub-region are cropped and redistributed to obtain the processed histograms of each sub-region. Based on the processed histograms of each sub-region, the CDF of the original image is determined, and the original image is then equalized according to the CDF to obtain the processed image.

[0144] As an optional implementation, the splicing module 903 is specifically used for: Based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, the processed images are initially stitched together to obtain the initial stitched image. Based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, a sub-pixel registration algorithm is used to adjust the overlapping areas of each adjacent processed image in the initial stitched image to obtain the stitched image.

[0145] As an optional implementation, the splicing module 903 is specifically used for: Determine whether there is an overlapping region between the first processed image and the adjacent second processed image; If so, the overlapping region is extracted from the first processed image as the first image, and the overlapping region is extracted from the second processed image as the second image; A subpixel registration algorithm is used to perform Fourier transform and inverse Fourier transform on the first image and the second image respectively, to obtain the subpixel-level displacement deviation between the first processed image and the second processed image. Based on the subpixel-level displacement deviation, the position of the second processed image is adjusted to obtain the adjusted second processed image, so that there is no overlapping area between the first processed image and the adjacent adjusted second processed image.

[0146] This application also provides an electronic device, such as... Figure 11 The diagram shown is a schematic representation of the structure of an electronic device provided in an embodiment of this application, including a processor 111, a memory 112, and a bus 113. The memory 112 stores machine-readable instructions executable by the processor 111. When the electronic device is running, the processor 111 communicates with the memory 112 via the bus 113, and the processor 111 executes the machine-readable instructions to perform the steps of the wafer image processing method in the aforementioned embodiment.

[0147] This application also provides a computer-readable storage medium storing a computer program, which, when run by a processor, performs the steps of the wafer image processing method described in the foregoing embodiments.

[0148] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0149] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0150] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A wafer image processing method, characterized in that, include: Acquire multiple frames of raw images of the wafer under test and multimodal scanning information bound to each raw image. The multimodal scanning information includes the Z-axis height, illumination mode, and two-dimensional coordinates of the two-dimensional platform on which the wafer under test is located when the raw images are captured. Based on the multimodal scanning information, each of the original images is denoised to obtain a processed image. Based on the multimodal scanning information, the processed images are stitched together to obtain a stitched image. The stitched image is used as the input image for the defect recognition model, so that the defect recognition model can perform defect detection on the wafer under test based on the stitched image.

2. The method according to claim 1, characterized in that, The step of denoising each of the original images according to the multimodal scanning information to obtain each processed image includes: Based on the lighting pattern, determine the target denoising algorithm; Based on the multimodal scanning information, the target denoising algorithm is used to denoise the original image to obtain the processed image.

3. The method according to claim 2, characterized in that, The step of determining the target denoising algorithm based on the lighting pattern includes: If the lighting mode is a bright field lighting mode, then the target denoising algorithm is determined to be a guided filtering algorithm that introduces Z-axis height. If the lighting mode is a dark field lighting mode, then the target denoising algorithm is determined to be the Limiting Contrast Adaptive Histogram Equalization (CLAHE) algorithm.

4. The method according to claim 3, characterized in that, The step of denoising the original image using the target denoising algorithm based on the multimodal scanning information to obtain the processed image includes: In the bright field illumination mode, a template image of the wafer under test is acquired; Based on the template image and the Z-axis height when the original image was captured, the original image is linearly filtered using the guided filtering algorithm that introduces the Z-axis height to obtain the processed image.

5. The method according to claim 4, characterized in that, The step of performing linear filtering on the original image based on the template image and the Z-axis height when the original image was captured, using the guided filtering algorithm that incorporates the Z-axis height, includes: Based on the template image, determine the guide image for the original image; The defocus height is determined based on the Z-axis height and the preset target focal plane height; Within the local window corresponding to the original image, based on the guiding image, the defocus height, and the original image, the minimum guided filter cost function corresponding to the local window is determined with the goal of minimizing the error between the processed image and the original image within the local window; Based on the minimized guided filtering cost function corresponding to the local window, the first target linear coefficient and the second target linear coefficient corresponding to the local window are determined, and the original image is subjected to linear filtering processing based on the first target linear coefficient, the second target linear coefficient and the guided image.

6. The method according to claim 3, characterized in that, The step of denoising the original image using the target denoising algorithm based on the multimodal scanning information to obtain the processed image further includes: In the dark field illumination mode, the CLAHE algorithm is used to divide the original image into multiple sub-regions, and histogram equalization is performed on each sub-region to obtain the histogram of each sub-region. Based on a preset amplitude limiting threshold, the histograms of each sub-region are cropped and redistributed to obtain the processed histograms of each sub-region. Based on the processed histograms of each sub-region, the cumulative distribution function (CDF) of the original image is determined, and the original image is then subjected to equalization processing according to the CDF to obtain the processed image.

7. The method according to claim 1, characterized in that, The step of stitching together the processed images based on the multimodal scanning information to obtain a stitched image includes: Based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, the processed images are initially stitched together to obtain the initial stitched image. Based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, a sub-pixel registration algorithm is used to adjust the overlapping areas of each adjacent processed image in the preliminary stitched image to obtain the stitched image.

8. The method according to claim 7, characterized in that, The step of adjusting the overlapping areas of adjacent processed images in the preliminary stitched image, based on the two-dimensional coordinates of the two-dimensional platform where the wafer under test was located when each original image was captured, using a sub-pixel registration algorithm, includes: Determine whether there is an overlapping region between the first processed image and the adjacent second processed image; If so, the overlapping region is extracted from the first processed image as the first image, and the overlapping region is extracted from the second processed image as the second image; Using the subpixel registration algorithm, Fourier transform and inverse Fourier transform are performed on the first image and the second image respectively to obtain the subpixel-level displacement deviation between the first processed image and the second processed image. Based on the subpixel-level displacement deviation, the position of the second processed image is adjusted to obtain an adjusted second processed image, so that the first processed image and the adjacent adjusted second processed image do not overlap.

9. An electronic device, characterized in that, include: The electronic device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the image processing method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, performs the steps of the image processing method as described in any one of claims 1 to 8.

Images