Wafer image gray interval segmentation method and device, electronic equipment and storage medium
By integrating intensity and semantic segmentation thresholds in wafer defect detection, the inaccuracy caused by ignoring structural information in existing technologies is solved, achieving higher precision and reliability in grayscale segmentation and defect detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-19
- Publication Date
- 2026-04-10
AI Technical Summary
Existing intensity-based grayscale segmentation methods ignore the structural information of images in wafer defect detection, resulting in segmentation thresholds that do not match the actual physical condition of the wafer, thus affecting the accuracy and reliability of detection.
By fusing the intensity segmentation threshold from intensity segmentation processing and the semantic segmentation threshold from semantic segmentation processing, and considering both grayscale and semantic information of the image, the boundaries and ranges of the grayscale segmentation intervals are dynamically adjusted to better adapt to the structure and grayscale of the image.
It improves the rationality and accuracy of grayscale segmentation in wafer images, enhances the accuracy and reliability of defect detection, reduces false positives and false negatives, and provides more reliable quality assurance.
Smart Images
Figure CN121837291A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of optical detection, and in particular to a wafer image gray interval segmentation method and device, an electronic device, and a storage medium. BACKGROUND
[0002] In optical-based wafer surface defect detection, the detection image obtained by the optical imaging equipment of the wafer defect detection system generally has a large gray value range, and in different gray value image regions, generally has different noise intensity and geometric characteristics, so it is usually necessary to segment the image to be detected based on the gray value range.
[0003] The related art mainly uses an intensity-based gray interval segmentation method (such as a median intensity-based segmentation method MBS, a projection-based segmentation method PBS, etc.) to group pixels with similar intensities in the same interval to form relatively independent image regions, so that different detection methods and sensitivity configurations can be used in different image regions to improve detection efficiency and accuracy.
[0004] However, when using the intensity-based gray interval segmentation method, the structural information of the image is ignored, which may cause the gray segmentation interval to be inconsistent with the true physical condition of the wafer, thereby affecting the accuracy and reliability of wafer defect detection and adversely affecting the quality control of wafer production. SUMMARY
[0005] The present application provides a wafer image gray interval segmentation method, device, electronic device, and storage medium to improve the accuracy of wafer image gray interval segmentation, thereby further improving the accuracy and reliability of wafer defect detection.
[0006] According to a first aspect of the present application, the present application provides a wafer image gray interval segmentation method, comprising:
[0007] obtaining a wafer gray image of a wafer to be detected;
[0008] in response to a selection operation of a user, obtaining a point image corresponding to at least one reference point of at least three dies on the wafer gray image;
[0009] performing image fusion processing on the point images of the same reference point of the at least three dies to obtain a point fusion image;
[0010] performing intensity segmentation processing and semantic segmentation processing on the point fusion image, respectively, to obtain an intensity segmentation threshold and a semantic segmentation threshold of the point fusion image;
[0011] performing threshold fusion processing on the intensity segmentation threshold and the semantic segmentation threshold of the point fusion image to obtain a fusion segmentation threshold of the reference point.
[0012] In the present scheme, the intensity segmentation threshold of the fusion intensity segmentation processing and the semantic segmentation threshold of the semantic segmentation processing are fused, while considering the gray information and semantic information of the image, so that the segmentation process is more comprehensive and accurate. Under this fusion strategy, the gray scale segmentation interval can better adapt to the structure and gray scale of the image, so as to dynamically adjust the boundary and range of the gray scale segmentation interval according to the gray scale distribution and semantic structure of different regions in the image, so that the segmentation result is more in line with the actual characteristics of the image, thereby effectively improving the rationality and accuracy of the wafer image gray scale interval segmentation. Accurate gray scale interval segmentation provides more accurate basic data for subsequent wafer defect detection, so that the defect detection algorithm can flexibly configure the corresponding defect detection method and parameter according to the divided image region, and more accurately identify various defects on the wafer. This not only improves the accuracy of wafer defect detection, reduces the misjudgment and omission, enhances the reliability of the defect detection result, and provides more reliable quality guarantee for wafer manufacturing.
[0013] In a feasible implementation manner, the intensity segmentation threshold comprises at least one first gray scale threshold, the semantic segmentation threshold comprises at least one second gray scale threshold, and the intensity segmentation threshold and the semantic segmentation threshold of the point position fusion image are subjected to threshold fusion processing to obtain the fusion segmentation threshold of the reference point position, comprising:
[0014] Obtaining the minimum absolute distance relationship between the at least one second gray scale threshold and the at least one first gray scale threshold;
[0015] Comparing the minimum absolute distance relationship with the numerical value of the at least one threshold stored locally, and fusing the at least one first gray scale threshold and the at least one second gray scale threshold.
[0016] In the present embodiment, the intensity segmentation threshold is taken as a reference, and the semantic segmentation threshold is taken as an auxiliary. The intensity segmentation threshold is fused and updated by the semantic segmentation threshold to obtain a fusion segmentation threshold considering the gray information and semantic information of the image, so that the segmentation threshold is more in line with the actual characteristics of the image.
[0017] In a feasible implementation manner, the at least one threshold comprises a new threshold and a merged threshold, the new threshold is greater than the merged threshold, and the at least one first gray scale threshold and the at least one second gray scale threshold are fused by comparing the minimum absolute distance relationship with the numerical value of the at least one threshold stored locally, comprising:
[0018] Obtaining the minimum absolute distance between the second gray scale threshold and the first gray scale threshold in the intensity segmentation threshold;
[0019] if the minimum absolute distance is greater than the new threshold, the second gray threshold is inserted into the intensity segmentation threshold;
[0020] if the minimum absolute distance is less than the merging threshold, the second gray threshold and the first gray threshold corresponding to the minimum absolute distance are fused to obtain a weighted fusion threshold, and the first gray threshold corresponding to the minimum absolute distance is replaced by the weighted fusion threshold;
[0021] if the minimum absolute distance is greater than or equal to the merging threshold and less than or equal to the new threshold, the first gray threshold corresponding to the minimum absolute distance is retained;
[0022] the updated intensity segmentation threshold is obtained.
[0023] In the embodiment, the intensity segmentation threshold is updated according to the absolute distance between the second gray threshold and the intensity segmentation threshold, so as to ensure the fusion accuracy of the intensity segmentation threshold and the language segmentation threshold.
[0024] In an available embodiment, the point image of the same reference point of the at least three bare dies is subjected to image fusion processing to obtain a point fusion image, including:
[0025] The point fusion image is obtained based on the point image of the same reference point of the at least three bare dies and an image fusion algorithm, wherein the image fusion algorithm is a median filter fusion algorithm, a mean filter fusion algorithm or a weighted filter fusion algorithm.
[0026] In the embodiment, the point image of the same reference point is subjected to image fusion processing, so as to enhance the applicability of the subsequent segmentation result on different bare dies.
[0027] In an available embodiment, the intensity segmentation processing is realized based on the following method:
[0028] The intensity segmentation threshold is obtained based on the point fusion image and an intensity segmentation algorithm, wherein the intensity segmentation algorithm is a median-based segmentation algorithm MBS or a projection-based segmentation algorithm PBS.
[0029] In the embodiment, MBS realizes accurate segmentation of a target region by using the median characteristic of pixel intensity in an image and combining an adaptive deformation algorithm, and PBS realizes efficient segmentation of an image by projecting image data to a specific feature space and using the data distribution after projection.
[0030] In an available embodiment, the semantic segmentation processing is realized based on the following method:
[0031] segment the point position fusion graph based on the point position fusion graph and the semantic segmentation model to obtain at least one semantic image region;
[0032] obtain the semantic segmentation threshold based on a threshold value of a gray interval corresponding to each semantic image region.
[0033] In this embodiment, the semantic segmentation model obtained through training of the deep learning model can accurately identify the boundaries and spatial distribution of different structures in an image, thereby converting an input image into a segmentation graph with clear semantics, and has high image analysis capability and generalization.
[0034] According to a second aspect of the present application, the present application provides a wafer image gray interval segmentation device, comprising:
[0035] An image acquisition module is configured to acquire a wafer gray image of a wafer to be detected.
[0036] An image extraction module is configured to acquire, in response to a selection operation of a user, a point position image corresponding to at least one reference point position of at least three dies on the wafer gray image.
[0037] An image fusion module is configured to perform image fusion processing on the point position images of the same reference point position of the at least three dies to obtain a point position fusion graph.
[0038] An image segmentation module is configured to perform intensity segmentation processing and semantic segmentation processing on the point position fusion graph respectively to obtain an intensity segmentation threshold and a semantic segmentation threshold of the point position fusion graph.
[0039] A threshold fusion module is configured to perform threshold fusion processing on the intensity segmentation threshold and the semantic segmentation threshold of the point position fusion graph to obtain a fusion segmentation threshold of the reference point position.
[0040] According to a third aspect of the present application, the present application provides an electronic device comprising a processor and a memory in communication connection with the processor;
[0041] The memory stores computer execution instructions.
[0042] The processor executes the computer execution instructions stored in the memory to implement the method of any one of the first aspect.
[0043] According to a fourth aspect of the present application, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed to implement the method of any one of the first aspect.
[0044] According to a fifth aspect of the present application, the present application provides a computer program product comprising a computer program for implementing the method of any one of the first aspect when executed.
[0045] According to a sixth aspect of the present application, the present application provides a semiconductor device comprising the electronic device of the third aspect. BRIEF DESCRIPTION OF DRAWINGS
[0046] The accompanying drawings, which are incorporated herein and constitute part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the present application.
[0047] Figure 1 A system framework diagram of a wafer defect detection system provided for an embodiment of the present application;
[0048] Figure 2 A structure schematic diagram of a wafer provided for an embodiment of the present application;
[0049] Figure 3 A flow schematic diagram of a wafer image gray scale interval segmentation method provided for an embodiment of the present application;
[0050] Figure 4 A structure schematic diagram of a wafer image gray scale interval segmentation device provided for an embodiment of the present application;
[0051] Figure 5 A structure schematic diagram of an electronic device provided for an embodiment of the present application.
[0052] BRIEF DESCRIPTION OF DRAWINGS
[0053] 100 - wafer defect detection system;
[0054] 101 - optical imaging device;
[0055] 102 - cluster processing device;
[0056] 103 - host computer;
[0057] 200 - wafer;
[0058] 201 - bare chip.
[0059] The specific embodiments of the present application have been shown through the above drawings, and will be described in more detail hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0060] The exemplary embodiments will be described in detail herein with reference to the attached drawings. In the following description, like reference numerals refer to like elements throughout the description. The following exemplary embodiments described herein represent illustrations only. That is, those of ordinary skill in the art will recognize that the applications described herein can be with various modifications not specified herein. The exemplary embodiments described herein are meant to be illustrative only and are not meant to be limiting in any way.
[0061] In optical-based wafer surface defect detection, it is necessary to acquire wafer grayscale images by optical imaging devices of a wafer defect detection system for wafer defect detection. Since the wafer grayscale images acquired by the optical imaging devices often have a large range of grayscale values, image regions with different grayscale values have different noise intensities and geometric characteristics. Therefore, in order to more effectively detect wafer surface defects, it is usually necessary to perform grayscale interval segmentation processing on the images to be detected according to the range of grayscale values. After the images are segmented into different image regions according to the grayscale intervals, different defect detection methods and sensitivity configurations can be used for defect detection on different image regions according to the characteristics of different grayscale intervals, so that the defect detection process can more accurately fit the actual situation of different image regions, thereby improving the efficiency and accuracy of wafer defect detection.
[0062] At present, the related art mainly uses intensity-based grayscale interval segmentation methods, such as a median-based segmentation (MBS) method, a project-based segmentation (PBS) method, etc. These grayscale interval segmentation methods can divide pixels with similar grayscale values in a detection image into the same interval. In this way, pixels with similar grayscale characteristics can be gathered together to form relatively independent image regions.
[0063] For example, the segmentation threshold of the current intensity-based grayscale interval segmentation method includes at least one grayscale threshold. The grayscale value range of pixels in a known image is (0, 255). The grayscale threshold can be used to divide the grayscale value into multiple grayscale intervals. Assuming that the grayscale threshold is {100, 200}, the grayscale value is divided into intervals with 100 and 200 as the thresholds, respectively. The divided grayscale intervals are (0, 100), (101, 200), and (201, 255).
[0064] On this basis, when subsequently performing optical detection on the wafer grayscale image, only the grayscale interval needs to be divided to extract the image region with the corresponding grayscale value from the wafer grayscale image, and the corresponding defect detection method and sensitivity configuration are used for defect detection on the image region according to the characteristics of the grayscale interval.
[0065] However, when using the intensity-based interval segmentation method, the structural information of the image is ignored. The structural information of the image contains important contents such as the spatial relationship between pixels, edge features, and the overall shape of the object. For example, in some cases, although the gray levels of two image regions are not similar, they are adjacent in space and may belong to similar physical structure parts on the wafer. If the segmentation is only based on the gray level, the regions that belong to the similar physical structure may be divided into different intervals; conversely, the regions with different physical structures but slightly different gray levels may be divided into the same interval. Therefore, the current gray interval segmentation method which ignores the structural information may cause the gray interval segmentation threshold to be inconsistent with the real physical condition of the wafer, thereby affecting the accuracy and reliability of wafer defect detection, and adversely affecting the quality control of wafer production.
[0066] To solve the above technical problems, the present application provides a wafer image gray interval segmentation method, device, electronic equipment and storage medium. By fusing the intensity segmentation threshold of the intensity segmentation processing and the semantic segmentation threshold of the semantic segmentation processing, the gray information and semantic information of the image are considered at the same time, so that the gray segmentation interval of the wafer image can better adapt to the structure and gray of the image, effectively improving the rationality and precision of the wafer image gray interval segmentation, and further improving the accuracy and reliability of wafer defect detection.
[0067] The technical scheme of the wafer image gray interval segmentation method provided by the present application will be described in detail below through specific embodiments. It should be noted that the following embodiments can exist independently or be combined with each other. For the same or similar content, it may not be repeated in different embodiments.
[0068] Figure 1 A system architecture diagram of a wafer defect detection system 100 provided by an embodiment of the present application is shown in FIG. 1. Figure 1 The system architecture includes an optical imaging device 101, a cluster processing device 102 and a host computer 103.
[0069] The optical imaging device 101 usually includes an image sensor, a light source and a collection system, etc. The image sensor (such as a photomultiplier tube PMT, a charge-coupled device CCD, etc.) is used to acquire 2D image information of the wafer to be detected; the light source is used to provide appropriate wavelength for imaging illumination; the collection system is used to receive light and transmit it to the image sensor.
[0070] The cluster processing device 102 usually includes CPU, GPU, NPU, memory, etc., and is mainly used for system optimization, image preprocessing (including gray interval segmentation of the image) and running of the defect detection algorithm.
[0071] The host computer 103 mainly includes a display, a CPU, a memory, an external storage and the like, and is mainly used for controlling the operation and parameter configuration of the cluster processing device and the optical imaging device, and generating and displaying corresponding operation interface, data interface and the like.
[0072] It should be noted that the execution subject of the wafer image gray scale interval segmentation method provided in the embodiments of the present application is the cluster processing device 102, and correspondingly, the wafer image gray scale interval segmentation device is also arranged in the cluster processing device 102.
[0073] Figure 3 A flowchart of a wafer image gray scale interval segmentation method provided in the embodiments of the present application is shown in FIG. 2. Figure 3 In some embodiments, the flow of the wafer image gray scale interval segmentation method includes the following steps:
[0074] S301, obtaining a wafer gray scale image of a wafer to be detected.
[0075] The wafer gray scale image of the wafer to be detected is collected by the optical imaging device to provide data support for subsequent gray scale interval segmentation.
[0076] S302, in response to a selection operation of a user, obtaining a point image corresponding to at least one reference point position of at least three dies on the wafer gray scale image.
[0077] Since the wafer surface defect detection is after the die processing and before the packaging, referring to FIG. 1, Figure 2 It can be seen that the dies 201 are still arranged in an array on the wafer 200 at this time. The optical detection of the wafer 200 surface defect is essentially the detection of the dies 201 on the wafer 200 which have been processed but not yet packaged, to detect whether the processed dies 201 have defects. The wafer 200 surface defect detection is arranged at this stage because the wafer gray scale image containing multiple die gray scale images can be collected at this time, which is convenient for batch detection of all dies 201 on the same wafer 200, to improve the detection efficiency. Specifically, the die is a micro semiconductor chip unit cut from the wafer, which has not been packaged and contains complete integrated circuit functions. When the wafer is processed, multiple dies with the same structure will be made on a wafer.
[0078] Referring to FIG. 1, Figure 2As can be seen, there are multiple dies 201 on the wafer 200 to be detected, and the subsequent defect detection is essentially detecting each die 201. In order to further reduce the amount of data to be processed during detection and improve detection efficiency, the detection is generally performed on the point image extracted from the key detection area of the die 201 where defects are prone to occur, so as to reflect the overall defect condition of the die 201 by local sampling detection. Therefore, the gray interval segmentation of the wafer 201 is essentially the segmentation of the point image of the region of interest in the die gray image.
[0079] Meanwhile, since the die structure on the wafer is consistent, the gray interval segmentation result of the reference die can be extended to other dies by selecting part of the dies as the reference, and the gray interval segmentation result of all the dies is used to reflect the gray interval segmentation result of the entire wafer. Therefore, through the selection operation of the user, the user selects at least three dies on the wafer gray image, and obtains the point image at the reference point position in the region of interest where defects are prone to occur and need to be focused on, so as to reflect the gray interval segmentation of the entire wafer based on the gray interval segmentation result of the point image, thereby further reducing the amount of data to be processed for wafer gray interval segmentation and improving the efficiency of wafer gray interval segmentation.
[0080] Specifically, the region of interest is a specific local range in an image or scene that is particularly concerned, has key information or unique features, and needs to be focused on for analysis, processing or research. The region of interest of the reference die is the key detection area of the die on the wafer where defects are prone to occur, and the reference point is the key detection position in the region of interest.
[0081] Specifically, the point image is an image taken at the reference point position in the region of interest of the reference die with a fixed window size. If multiple reference dies are selected, a corresponding point image will be extracted at the same reference point position of each reference die.
[0082] Specifically, in order to determine that the reference die can represent the gray interval segmentation effect of all dies on the wafer, at least three die gray images on the wafer gray image are selected as reference images.
[0083] Optionally, before extracting the point image corresponding to each reference point, the method further comprises:
[0084] Step 1, generating a visual detection page based on the point images of the reference points; wherein the visual detection page comprises a detection feedback control.
[0085] Wherein, the visual detection interface is generated and displayed through the point image, and the detection feedback control is set on the visual detection page, so as to facilitate the user to visually detect the point image,
[0086] Specifically, the user checks whether the point position image is abnormal through visual observation.
[0087] Specifically, the visual detection interface is displayed through the display of the host computer in the wafer defect detection system.
[0088] Specifically, the detection feedback control can adopt a selection control, and the user triggers the detection feedback operation through a selection operation, and determines the visual detection result through the feedback content. For example, the selection control can be set to be able to feed back two results of “normal” and “abnormal”, so that when the user selects “normal”, it is confirmed that the point position image is normal, and when the user selects “abnormal”, it is confirmed that the point position image is abnormal.
[0089] Step 2, in response to the detection feedback operation of the user on the detection feedback control, it is determined whether the point position image of the reference point position is abnormal.
[0090] Among them, through the operation of the user on the detection feedback control, it is judged whether the point position image extracted by the reference point position is abnormal.
[0091] Specifically, the abnormality includes image abnormal conditions such as overexposure, underexposure, and texture abnormality.
[0092] If the point position image of the reference point position is abnormal, the user reselects the reference point position.
[0093] Among them, since the abnormal point position image will affect the accuracy of subsequent gray interval segmentation, if it is found that the point position image of the reference point position is abnormal, the user reselects a new reference point position to ensure the accuracy of subsequent gray interval segmentation.
[0094] Specifically, different reference dies will extract multiple point position images at the same reference point position, and when there is at least one abnormal image in the point position image, it indicates that the point position image of the reference point position is abnormal.
[0095] If the point position image of the reference point position is not abnormal, S303 is executed.
[0096] S303, image fusion processing is performed on the point position images of the same reference point position of at least three dies to obtain a point position fusion image.
[0097] Among them, the point position images of the same reference point position are integrated to enhance the applicability of the subsequent segmentation threshold on different dies.
[0098] Optionally, based on the point position images of the same reference point position of at least three dies and an image fusion algorithm, a point position fusion image is obtained; wherein the image fusion algorithm is a median filter fusion algorithm, a mean filter fusion algorithm or a weighted filter fusion algorithm.
[0099] The image fusion algorithm changes the gray value of the pixel by mathematical transformation or spatial operation to fuse the point image to obtain a point fusion image. Meanwhile, the image fusion algorithm can also adjust the image gray distribution, enhance the contrast or suppress the noise.
[0100] Specifically, the median filter fusion algorithm replaces the original pixel value with the median value of the gray values in the neighborhood window of the pixel point to realize image smoothing and noise suppression. In the fusion of multiple gray images, for each pixel point of each image, a fixed size neighborhood window (such as 3x3, 5x5) is selected with the point as the center, and the middle value of the sorted gray values of all pixels in the window is taken as the new value of the center pixel. Since the median filter can effectively eliminate isolated noise points (such as salt and pepper noise, impulse noise), while retaining the edge features of the image, the fused image can better maintain the edge information and detailed structure of the original image while removing noise.
[0101] The mean filter fusion algorithm replaces the pixel value with the arithmetic mean of all pixels in the neighborhood to smooth the image and suppress noise. In the fusion of multiple gray images, for each pixel point of each image, a fixed size neighborhood window (such as 3x3, 5x5) is selected, the average value of all pixels in the window is calculated, and the average value is taken as the new value of the center pixel. The mean filter can effectively reduce sharp changes in the image, smooth image details, and suppress high-frequency noise while retaining low-frequency information.
[0102] The weighted filter fusion algorithm introduces a weight coefficient when calculating the average value of the pixels in the neighborhood, so that some pixels have a greater contribution when calculating the average value. In the fusion of multiple gray images, the algorithm can assign different weight coefficients to each pixel point according to the signal-to-noise ratio, stability or importance of the characteristics of each image. When fusing, for each pixel point of each image, a fixed size neighborhood window is selected, the weighted average value of all pixels in the window is calculated, and the weighted average value is taken as the new value of the center pixel. The weighted filter fusion algorithm can more flexibly adapt to different application scenarios and requirements, highlight important areas, suppress noise or outliers, and better preserve the structural information of the image by adjusting the weight coefficients and considering the spatial correlation between pixels.
[0103] S304, the point fusion image is subjected to intensity segmentation processing and semantic segmentation processing respectively to obtain an intensity segmentation threshold and a semantic segmentation threshold of the point fusion image.
[0104] The intensity segmentation processing considers the gray scale size information of the point fusion image to segment the point fusion image, and the semantic segmentation processing considers the semantic information of the point fusion image to segment the point fusion image. After the intensity segmentation processing and the semantic segmentation processing are respectively performed, at least one intensity segmentation threshold based on the gray scale size of the point fusion image is segmented to form a first gray scale threshold, and at least one semantic segmentation threshold based on the semantic structure of the point fusion image is segmented to form a second gray scale threshold.
[0105] Optionally, the intensity segmentation processing is implemented based on the following method: obtaining an intensity segmentation threshold based on the point fusion image and an intensity segmentation algorithm; wherein the intensity segmentation algorithm is a median intensity-based segmentation algorithm MBS or a projection-based segmentation algorithm PBS.
[0106] Wherein, MBS uses the median value characteristics of the pixel intensity in the image, combined with an adaptive deformation algorithm to accurately segment the target region; PBS projects the image data to a specific feature space, and uses the distribution of the projected data to achieve efficient segmentation of the image.
[0107] Optionally, the semantic segmentation processing is implemented based on the following method:
[0108] Step 1, based on the point fusion image and the semantic segmentation model, segmenting the point fusion image to obtain at least one semantic image region.
[0109] Wherein, the semantic segmentation model obtained by training the deep learning model can accurately identify the boundaries and spatial distribution of different structures in the image, thereby converting the input image into a segmented image with clear semantics, and has high image analysis ability and generalization.
[0110] Step 2, based on the threshold of the gray scale interval corresponding to each semantic image region, obtaining a semantic segmentation threshold.
[0111] Wherein, in the semantic image region formed by the semantic segmentation, each semantic image region has a gray value interval, so that the semantic segmentation threshold is obtained according to the gray value interval of the semantic image region.
[0112] For example, assuming that the gray value interval of semantic image region A is (0, 100), and the gray value interval of semantic image region B is (101, 200), then the gray scale threshold can be obtained as 100.
[0113] For regions with similar gray scale intervals in the semantic image region, interval fusion processing can be used to reduce the data processing amount of subsequent threshold fusion. For example, set the difference threshold to 10, then the gray value difference of the gray scale intervals (80, 150) and (85, 145) is within the difference threshold range, then the two gray scale intervals can be merged to obtain (80, 150).
[0114] Specifically, the semantic segmentation model is obtained based on the following method:
[0115] Step 1, obtaining a training set; wherein the training set includes a plurality of training sample images and real semantic segmentation images corresponding to the training sample images.
[0116] Wherein, the image data collected by wafer optical detection is used as the training sample image, and the corresponding real semantic segmentation image is labeled for each training sample image to provide training data for the deep learning model.
[0117] Step 2, input the training set into the deep learning model to obtain a predicted semantic segmentation image output by the deep learning model.
[0118] Wherein, after the training sample image is input into the deep learning model, it is subjected to convolution layer, activation function (such as ReLU), normalization (such as BatchNorm) and other operations, and finally outputs the predicted semantic segmentation image.
[0119] Specifically, the deep learning model can select an encoder-decoder architecture (such as U-Net, DeepLab, PSPNet, etc.).
[0120] Step 3, constructing a loss function based on the difference between the real semantic segmentation image and the predicted semantic segmentation image.
[0121] Wherein, the loss function can quantify the error between the prediction result and the real label, and provide a direction for the parameter optimization of the deep learning model.
[0122] Specifically, the loss function can adopt cross-entropy loss, intersection over union (IoU) loss, etc.
[0123] Step 4, iteratively training the deep learning model based on the loss function to obtain a semantic segmentation model.
[0124] Wherein, the deep learning model gradually learns the mapping relationship from the image to the semantic segmentation by optimizing the loss function, and the iterative training ensures that the deep learning model parameters approach the optimal solution and improve the model generalization ability.
[0125] S305, threshold fusion processing is performed on the intensity segmentation threshold and the semantic segmentation threshold of the point position fusion image to obtain a fusion segmentation threshold of the reference point position.
[0126] The intensity segmentation result includes at least one gray threshold obtained based on the intensity size, the semantic segmentation result includes at least one gray threshold obtained based on the semantic structure, and the gray thresholds obtained by fusing the two segmentation results are as close as possible to obtain the fused gray segmentation threshold. The intensity segmentation threshold and the semantic segmentation threshold of the point fusion image are fused, the gray information and the semantic information of the image are considered at the same time, the gray segmentation interval of the wafer image can better adapt to the structure and the gray of the image, and the rationality and the precision of the segmentation interval are effectively improved.
[0127] Specifically, by fusing the segmentation threshold to divide the gray value into intervals, the corresponding gray division interval can be obtained.
[0128] Optionally, if there are multiple different reference points, the fusion segmentation threshold of each reference point in the region of interest is subjected to threshold fusion processing to obtain the gray interval segmentation threshold of the region of interest.
[0129] Wherein, when there are multiple different reference points, because the point image of multiple reference points in the region of interest is used, the fusion segmentation result of each reference point is fused again, and the gray interval segmentation result after fusion can better represent the overall image information of the region of interest, further improving the precision and efficiency of the gray interval segmentation.
[0130] Optionally, the intensity segmentation threshold includes at least one first gray threshold, the semantic segmentation threshold includes at least one second gray threshold, the intensity segmentation threshold and the semantic segmentation threshold of the point fusion image are subjected to threshold fusion processing to obtain the fusion segmentation threshold of the reference point, and specifically include:
[0131] Step 1, obtaining the minimum absolute distance relationship between at least one second gray threshold and at least one first gray threshold.
[0132] Step 2, comparing the minimum absolute distance relationship with the numerical size of the at least one threshold stored locally, and fusing the at least one first gray threshold and the at least one second gray threshold.
[0133] Wherein, according to the minimum absolute distance relationship between the second gray threshold and the first gray threshold, the minimum absolute distance between the second gray threshold and the second gray threshold is determined, and the size relationship between the minimum absolute distance and the at least one threshold stored locally is compared to determine how to fuse and update the intensity segmentation threshold according to the second gray threshold.
[0134] In some embodiments, the intensity segmentation threshold is updated by the semantic segmentation threshold, and a fusion segmentation threshold considering the gray information and the semantic information of the image is obtained, so that the segmentation threshold is more consistent with the actual characteristics of the image.
[0135] Optionally, the at least one threshold includes an added threshold and a merged threshold, the added threshold is greater than the merged threshold, the value of the minimum absolute distance relationship is compared with the value of the at least one threshold stored locally, and the at least one first gray threshold and the at least one second gray threshold are fused, including:
[0136] Step 1: If the minimum absolute distance is greater than the added threshold, the second gray threshold is inserted into the intensity segmentation threshold.
[0137] If the minimum absolute distance is greater than the added threshold, it indicates that the second gray threshold is too far away from the first gray threshold corresponding to the minimum absolute distance, and the semantic information is not considered here, so the second gray threshold is inserted into the intensity segmentation threshold as a new threshold.
[0138] Step 2: If the minimum absolute distance is less than the merged threshold, the second interval threshold and the first gray threshold corresponding to the minimum absolute distance are weighted and fused to obtain a weighted fusion threshold, and the first gray threshold corresponding to the minimum absolute distance is replaced by the weighted fusion threshold; wherein the added threshold is greater than the merged threshold.
[0139] If the minimum absolute distance is less than the merged threshold, it indicates that the second gray threshold is close to the first gray threshold corresponding to the minimum absolute distance, and a compromise adjustment is made to the position, so the first gray threshold and the second gray threshold corresponding to the minimum absolute distance are weighted and fused, and the first gray threshold corresponding to the minimum absolute distance is replaced by the weighted fusion threshold.
[0140] Specifically, the first gray threshold with the minimum absolute distance to the second gray threshold in the intensity segmentation threshold is the first gray threshold with the minimum absolute distance to the second gray threshold.
[0141] Step 3: If the minimum absolute distance is greater than or equal to the merged threshold and the minimum absolute distance is less than or equal to the added threshold, the first gray threshold corresponding to the minimum absolute distance is retained.
[0142] Correspondingly, if the minimum absolute distance is not greater than the added threshold and not less than the merged threshold, it indicates that the distance is moderate, and the intensity segmentation threshold is not updated.
[0143] Steps 1, 2 and 3 are executed in no particular order and can be executed simultaneously.
[0144] Step 4: Obtain the updated intensity segmentation threshold.
[0145] The intensity segmentation threshold after the fusion update is taken as the fusion segmentation threshold.
[0146] The intensity segmentation threshold is updated by the absolute distance between the second gray threshold and the intensity segmentation threshold to ensure the fusion accuracy of the intensity segmentation threshold and the language segmentation threshold.
[0147] In the embodiment, after the point position fusion map is respectively subjected to the intensity segmentation processing and the semantic segmentation processing, the intensity segmentation threshold and the semantic segmentation threshold are obtained, and then the two groups of thresholds are fused to obtain the final fusion segmentation threshold. For example, the threshold fusion is further described in combination with a specific case:
[0148] Firstly, the related parameters are assumed as follows:
[0149] GapThreshold represents the new threshold, specifically, GapThreshold=60;
[0150] MergeThreshold represents the merging threshold, specifically, MergeThreshol=20;
[0151] alpha represents the weight of the intensity segmentation threshold in the weighted fusion, and the value range is (0, 1). Correspondingly, the weight of the semantic segmentation threshold is 1-alpha. Specifically, alpha=0.7;
[0152] A represents the intensity segmentation threshold, specifically, A={100, 200}, wherein 100 and 200 respectively represent the corresponding first gray threshold;
[0153] B represents the semantic segmentation threshold, specifically, B={10, 104, 150, 230}, wherein 10, 104, 150 and 230 respectively represent the corresponding second gray threshold.
[0154] Then, each second gray threshold b in the semantic segmentation threshold B is taken out in sequence, and each second gray threshold b is fused with the first gray threshold a in the semantic segmentation threshold A, and the specific threshold fusion process includes:
[0155] The absolute distance d=|b-a| between b and each a is calculated, and the minimum absolute distance d_min between b and a and the first gray threshold a_min corresponding to the minimum absolute distance d_min are determined. If there are two a with the same absolute distance from b, a_min takes the smaller value.
[0156] If d_min>GapThreshold, b is inserted into the sequence A.
[0157] If d_min < MergeThreshold, then get a weighted merge threshold c = alpha * a + (1 - alpha) * b, and replace a_min in A with c;
[0158] If MergeThreshold < d_min < GapThreshold, then keep a_min in A.
[0159] Specifically, in combination with the above specific parameters, the threshold merging process includes:
[0160] When the second gray threshold b = 10, the first gray threshold a_min corresponding to the minimum absolute distance is 100, and d_min = 90 > 60, b = 10 is inserted into A, and A is updated to {10, 100, 200};
[0161] When the second gray threshold b = 104, the first gray threshold a_min corresponding to the minimum absolute distance is 100, and d_min = 4 < 20, a weighted merge obtains a weighted merge threshold c = 0.7 * 100 + 0.3 * 104 = 101, and a_min = 100 in A is replaced by the weighted merge threshold c = 101, and A is updated to {10, 101, 200};
[0162] When the second gray threshold b = 150, the first gray threshold a_min corresponding to the minimum absolute distance is 101, and d_min = 20 < 49 < 60, a_min = 101 in A is kept;
[0163] When the second gray threshold b = 230, the first gray threshold a_min corresponding to the minimum absolute distance is 200, and d_min = 20 < 30 < 60, a_min = 200 in A is kept.
[0164] Finally, the updated intensity segmentation threshold A = {10, 101, 200} is obtained as the merged segmentation threshold.
[0165] In addition, the intensity range of the image gray is [min, max] = [0, 255], and it is assumed that the gray interval formed based on the gray threshold segmentation is denoted as (start, end), and the gray interval is required to satisfy the following rules:
[0166] The start of the first gray interval is min;
[0167] The end + 1 of the last gray interval is the start of the next gray interval;
[0168] The end of the last gray interval is max.
[0169] The fusion threshold value of the above-mentioned segmentation is {10, 101, 200}, and the gray interval formed by the segmentation is (0, 10), (11, 101), (102, 200), and (201, 255).
[0170] It can be seen that the gray threshold value is the segmentation line of the gray interval, and the gray threshold value is composed of the end values (not including max) of all gray intervals. By determining the segmentation threshold value, the gray interval segmentation can be realized.
[0171] In some embodiments, the semantic segmentation threshold value can also be taken as a reference, and the intensity segmentation threshold value can be taken as an auxiliary to update the semantic segmentation threshold value.
[0172] In this embodiment, the intensity segmentation threshold value of the intensity segmentation processing and the semantic segmentation threshold value of the semantic segmentation processing are fused, and the gray information and the semantic information of the image are considered at the same time, so that the segmentation process is more comprehensive and accurate. Under this fusion strategy, the gray segmentation interval can better adapt to the structure and gray scale of the image, so as to dynamically adjust the boundary and range of the gray segmentation interval according to the gray distribution and semantic structure of different regions in the image, so that the segmentation result is more in line with the actual characteristics of the image, thereby effectively improving the rationality and precision of the wafer image gray interval segmentation. Accurate gray interval segmentation provides more accurate basic data for subsequent wafer defect detection, so that the defect detection algorithm can flexibly configure the corresponding defect detection method and parameter according to the divided image region, and more accurately identify various defects on the wafer. This not only improves the accuracy of wafer defect detection, reduces the misjudgment and omission, enhances the reliability of the defect detection result, and provides more reliable quality guarantee for wafer manufacturing.
[0173] Figure 4 is a structural schematic diagram of a wafer image gray interval segmentation device provided by an embodiment of the present application, referring to Figure 4 The wafer image gray interval segmentation device includes various functional modules for implementing the foregoing wafer image gray interval segmentation method, and any functional module can be realized by software and / or hardware.
[0174] In some embodiments, the wafer image gray interval segmentation device includes an image acquisition module 401, an image extraction module 402, an image fusion module 403, an image segmentation module 404, and a threshold value fusion module 405. Among them:
[0175] The image acquisition module 401 is configured to acquire a wafer gray image of a wafer to be detected.
[0176] The image extraction module 402 is configured to acquire a point image corresponding to at least one reference point of at least three dies on the wafer gray image in response to a selection operation of a user.
[0177] The image fusion module 403 is configured to perform image fusion processing on the point image of the same reference point of the at least three dies to obtain a point fusion image;
[0178] The image segmentation module 404 is configured to perform intensity segmentation processing and semantic segmentation processing on the point fusion image respectively to obtain an intensity segmentation threshold and a semantic segmentation threshold of the point fusion image;
[0179] The threshold fusion module 405 is configured to perform region fusion processing on the intensity segmentation threshold and the semantic segmentation threshold of the point fusion image to obtain a fusion segmentation threshold of the reference point.
[0180] In some embodiments, the intensity segmentation threshold includes at least one first gray threshold, and the semantic segmentation threshold includes at least one second gray threshold, and the threshold fusion module 405 is specifically configured to:
[0181] obtain a minimum absolute distance relationship between the at least one second gray threshold and the at least one first gray threshold;
[0182] compare the minimum absolute distance relationship with a numerical value of at least one threshold stored locally, and fuse the at least one first gray threshold and the at least one second gray threshold.
[0183] In some embodiments, the threshold fusion module 405 is specifically configured to:
[0184] if the minimum absolute distance is greater than a new threshold, the second gray threshold is inserted into the intensity segmentation threshold;
[0185] if the minimum absolute distance is less than a merging threshold, the second gray threshold and the first gray threshold corresponding to the minimum absolute distance are fused by weighting to obtain a weighted fusion threshold, and the first gray threshold corresponding to the minimum absolute distance is replaced by the weighted fusion threshold; wherein the new threshold is greater than the merging threshold;
[0186] if the minimum absolute distance is greater than or equal to the merging threshold and the minimum absolute distance is less than or equal to the new threshold, the first gray threshold corresponding to the minimum absolute distance is retained;
[0187] obtain an updated intensity segmentation threshold.
[0188] In some embodiments, the image fusion module 403 is specifically configured to:
[0189] obtain the point fusion image based on the point image of the same reference point of the at least three dies and an image fusion algorithm; wherein the image fusion algorithm is a median filter fusion algorithm, a mean filter fusion algorithm, or a weighted filter fusion algorithm.
[0190] In some embodiments, the image segmentation module 404 is specifically configured to:
[0191] obtain the intensity segmentation threshold based on the point fusion image and an intensity segmentation algorithm; the intensity segmentation algorithm is a median intensity-based segmentation algorithm MBS or a projection-based segmentation algorithm PBS.
[0192] In some embodiments, the image segmentation module 404 is specifically configured to:
[0193] segment the point fusion image based on the point fusion image and a semantic segmentation model to obtain at least one semantic image region;
[0194] obtain a semantic segmentation threshold based on a threshold of a gray interval corresponding to each semantic image region.
[0195] The wafer image gray interval segmentation device 300 provided by the embodiments of the present application is used to execute the technical solutions provided by the foregoing wafer image gray interval segmentation method embodiments, and has similar implementation principles and technical effects to those in the foregoing method embodiments, which will not be described here again.
[0196] It should be noted that the division of each module of the above device is only a logical functional division, and all or part of the actual implementation can be integrated into one physical entity, or can be physically separated. And these modules can all be implemented in the form of software called by a processing element, or all be implemented in the form of hardware, or part of the modules are implemented in the form of software called by a processing element, and part of the modules are implemented in the form of hardware. For example, the image acquisition module 401 can be a separately established processing element, or can be integrated in a chip of the above device, and in addition, the image acquisition module 401 can be stored in the form of program code in the memory of the above device, and the functions of the image acquisition module 401 are called and executed by a processing element of the above device. The implementation of other modules is similar. In addition, all or part of the modules can be integrated together, or can be independently implemented. The processing element here can be an integrated circuit with signal processing capability. In the implementation process, each step of the above method or each module can be completed by the integrated logic circuit of hardware or the instruction of software in the processing element.
[0197] Figure 5 For the structure schematic diagram of an electronic device provided by the embodiments of the present application, refer to Figure 5 The electronic device 500 includes a processor 501 and a memory 502 in communication connection with the processor 501.
[0198] The memory 502 stores computer execution instructions.
[0199] The processor 501 executes the computer execution instructions stored in the memory 502 to realize the technical solutions of the foregoing wafer image gray interval segmentation method.
[0200] In the electronic device 500 described above, the memory 502 and the processor 501 are directly or indirectly electrically connected to realize the transmission or interaction of data. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines, such as through bus connection. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc., but it does not mean that there is only one bus or only one type of bus. The memory 502 stores computer execution instructions for implementing the wafer image gray interval segmentation method described above, including at least one software function module stored in the memory 502 in the form of software or firmware, and the processor 501 executes various function applications and data processing by running the software program and module stored in the memory 502.
[0201] The memory 502 at least includes one type of readable storage medium, not limited to a Random Access Memory (RAM), a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), an Electric Erasable Programmable Read-Only Memory (EEPROM), etc. Among them, the memory 502 is used to store programs, and the processor 501 executes the programs after receiving the execution instructions. Further, the software program and module in the memory 502 described above can also include an operating system, which can include various software components and / or drivers for managing system tasks (such as memory management, storage device control, power management, etc.), and can communicate with various hardware or software components to provide a running environment for other software components.
[0202] The processor 501 can be an integrated circuit chip having a processing capability of signals. The processor 501 can be a general processor, including a central processing unit (CPU), a network processor (NP), a digital signal processor (DSP), an application specific integrated circuit (ASIC), etc. The processor 501 can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the present application. The general processor can be a microprocessor, or the processor 501 can also be any conventional processor or the like.
[0203] The electronic device 500 is configured to implement the technical solutions provided by the wafer image gray interval segmentation method embodiments, and the implementation principle and technical effects are similar to those of the method embodiments, which will not be described herein.
[0204] The embodiments of the present application further provide a computer readable storage medium, and the computer readable storage medium stores computer execution instructions. When the computer execution instructions are executed, the technical solutions of the wafer image gray interval segmentation method are implemented.
[0205] The computer readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk. The computer readable storage medium can be any available medium that can be accessed by a general or special purpose computer.
[0206] An exemplary computer readable storage medium is coupled to the processor, so that the processor can read information from the computer readable storage medium and write information to the computer readable storage medium. Of course, the computer readable storage medium can also be an integral part of the processor. The processor and the computer readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the computer readable storage medium can also exist as discrete components in the control device of the wafer image gray interval segmentation device.
[0207] The embodiments of the present application further provide a computer program product, including a computer program. When the computer program is executed, the technical solutions of the wafer image gray interval segmentation method are implemented.
[0208] The embodiment of the present application further provides a semiconductor device, which comprises the electronic device.
[0209] In the above embodiments, those skilled in the art can understand that the implementation of the above-mentioned method embodiments can be realized by software, hardware, firmware or any combination thereof, entirely or partially. When software is used for implementation, the software can be realized in the form of a computer program product, entirely or partially. The computer program product comprises one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present application are entirely or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through a wired (for example, coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (for example, infrared, wireless network, microwave, etc.) manner. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available media. The available media can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD) or a semiconductor medium (for example, solid state disk (SSD)) and the like.
[0210] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments. The technical features of the above embodiments can be combined arbitrarily, and in order to make the description concise, not all possible combinations of the technical features in the above embodiments are described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.
[0211] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. The specification and examples given are exemplary only, and the true scope and spirit of the application is indicated by the claims which follow. The claims are intended to cover any and all alternatives, modifications, equivalents, and variations of the present application which follow the principles of the application and include known or customary practice in the art to which the application pertains. The specification and examples are to be considered exemplary only, with the true scope and spirit of the application indicated by the following claims.
[0212] It is to be understood that the application is not limited to the precise construction already described above and shown in the drawings, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the application. The scope of the application should only be limited by the claims appended hereto.
Claims
1. A method for grayscale region segmentation of a wafer image, characterized in that, include: Obtain the grayscale image of the wafer to be inspected; In response to the user's selection operation, acquire point images corresponding to at least one reference point of at least three bare dies on the wafer grayscale image; Image fusion processing is performed on the point images of the same reference point of the at least three bare crystals to obtain a point fusion map; The point fusion map is subjected to intensity segmentation and semantic segmentation respectively to obtain the intensity segmentation threshold and semantic segmentation threshold of the point fusion map; The intensity segmentation threshold and semantic segmentation threshold of the fused point map are subjected to threshold fusion processing to obtain the fusion segmentation threshold of the reference point.
2. The method according to claim 1, characterized in that, The intensity segmentation threshold includes at least one first grayscale threshold, and the semantic segmentation threshold includes at least one second grayscale threshold. Threshold fusion processing is performed on the intensity segmentation threshold and semantic segmentation threshold of the point fusion map to obtain the fusion segmentation threshold for the reference point, including: Obtain the minimum absolute distance relationship between the at least one second grayscale threshold and the at least one first grayscale threshold; The minimum absolute distance relationship is compared with the numerical value of at least one threshold stored locally, and the at least one first grayscale threshold is fused with the at least one second grayscale threshold.
3. The method according to claim 2, characterized in that, The at least one threshold includes a new threshold and a merge threshold, wherein the new threshold is greater than the merge threshold. The minimum absolute distance relationship is compared with the numerical value of at least one locally stored threshold. The at least one first grayscale threshold is then merged with the at least one second grayscale threshold, including: If the minimum absolute distance is greater than the newly added threshold, then the second grayscale threshold is inserted into the intensity segmentation threshold; If the minimum absolute distance is less than the merging threshold, then the second grayscale threshold and the first grayscale threshold corresponding to the minimum absolute distance are weighted and fused to obtain a weighted fusion threshold, and the weighted fusion threshold is used to replace the first grayscale threshold corresponding to the minimum absolute distance. If the minimum absolute distance is greater than or equal to the merging threshold and the minimum absolute distance is less than or equal to the new threshold, then the first grayscale threshold corresponding to the minimum absolute distance is retained. Obtain the updated intensity segmentation threshold.
4. The method according to any one of claims 1-3, characterized in that, Image fusion processing is performed on the point images of the same reference point of the at least three bare crystals to obtain a point fusion map, including: The point fusion image is obtained based on the point images of the same reference point of the at least three bare dies and the image fusion algorithm; wherein the image fusion algorithm is a median filtering fusion algorithm, a mean filtering fusion algorithm, or a weighted filtering fusion algorithm.
5. The method according to any one of claims 1-3, characterized in that, The intensity segmentation process is implemented based on the following method: Based on the point fusion map and the intensity segmentation algorithm, the intensity segmentation threshold is obtained; wherein the intensity segmentation algorithm is either the median intensity-based segmentation algorithm MBS or the projection-based segmentation algorithm PBS.
6. The method according to any one of claims 1-3, characterized in that, The semantic segmentation process is implemented based on the following method: Based on the point fusion map and the semantic segmentation model, the point fusion map is segmented to obtain at least one semantic image region; The semantic segmentation threshold is obtained based on the threshold of the grayscale interval corresponding to each semantic image region.
7. A wafer image grayscale range segmentation device, characterized in that, include: The image acquisition module is used to acquire the grayscale image of the wafer to be inspected; The image extraction module is used to, in response to a user's selection operation, acquire point images corresponding to at least one reference point of at least three bare dies on the wafer grayscale image; The image fusion module is used to perform image fusion processing on the point images of the same reference point of the at least three bare crystals to obtain a point fusion map; The image segmentation module is used to perform intensity segmentation and semantic segmentation on the point-location fusion image to obtain the intensity segmentation threshold and semantic segmentation threshold of the point-location fusion image. The threshold fusion module is used to perform threshold fusion processing on the intensity segmentation threshold and semantic segmentation threshold of the point fusion map to obtain the fusion segmentation threshold of the reference point.
8. An electronic device, characterized in that, Includes a processor and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed, are used to implement the method as described in any one of claims 1 to 6.
10. A computer program product comprising a computer program, which, when executed, performs the method as described in any one of claims 1 to 6.
11. A semiconductor device, characterized in that, The semiconductor device includes the electronic device as described in claim 8.