CN117094909B - Nanometer stamping wafer image acquisition processing method - Google Patents
Nanometer stamping wafer image acquisition processing method Download PDFInfo
- Publication number
- CN117094909B CN117094909B CN202311114562.4A CN202311114562A CN117094909B CN 117094909 B CN117094909 B CN 117094909B CN 202311114562 A CN202311114562 A CN 202311114562A CN 117094909 B CN117094909 B CN 117094909B
- Authority
- CN
- China
- Prior art keywords
- pixel
- pixel value
- blocks
- value
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 10
- 238000000605 extraction Methods 0.000 claims abstract description 40
- 238000011176 pooling Methods 0.000 claims description 44
- 230000002159 abnormal effect Effects 0.000 claims description 43
- 230000003213 activating effect Effects 0.000 claims description 3
- 238000000034 method Methods 0.000 abstract description 17
- 235000012431 wafers Nutrition 0.000 description 26
- 230000009286 beneficial effect Effects 0.000 description 10
- 230000007547 defect Effects 0.000 description 5
- 238000001514 detection method Methods 0.000 description 2
- 238000001127 nanoimprint lithography Methods 0.000 description 2
- 238000005192 partition Methods 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 239000003292 glue Substances 0.000 description 1
- 239000007921 spray Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G03—PHOTOGRAPHY; CINEMATOGRAPHY; ANALOGOUS TECHNIQUES USING WAVES OTHER THAN OPTICAL WAVES; ELECTROGRAPHY; HOLOGRAPHY
- G03F—PHOTOMECHANICAL PRODUCTION OF TEXTURED OR PATTERNED SURFACES, e.g. FOR PRINTING, FOR PROCESSING OF SEMICONDUCTOR DEVICES; MATERIALS THEREFOR; ORIGINALS THEREFOR; APPARATUS SPECIALLY ADAPTED THEREFOR
- G03F7/00—Photomechanical, e.g. photolithographic, production of textured or patterned surfaces, e.g. printing surfaces; Materials therefor, e.g. comprising photoresists; Apparatus specially adapted therefor
- G03F7/0002—Lithographic processes using patterning methods other than those involving the exposure to radiation, e.g. by stamping
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30148—Semiconductor; IC; Wafer
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a method for acquiring and processing images of a nanoimprint wafer, which belongs to the technical field of image processing, and comprises the steps of respectively segmenting a standard nanoimprint wafer image and an acquired nanoimprint wafer image to respectively obtain standard image sub-blocks and acquired image sub-blocks, so that the two images are segmented, region comparison is convenient, feature extraction is carried out by adopting a feature extraction channel to respectively obtain second feature data and first feature data, the noise distribution degree of each acquired image sub-block is obtained through the comparison of the second feature data and the first feature data, denoising processing is carried out on the corresponding acquired image sub-blocks to obtain a denoised nanoimprint wafer image, and the problem that the image edge is unclear and the blurring degree of the image is increased due to the existing image denoising processing method is solved.
Description
Technical Field
The invention relates to the technical field of image processing, and provides a method for acquiring and processing an image of a nanoimprint wafer.
Background
Nanoimprint lithography uses nanoimprint techniques to spray a droplet of nanoimprint resist onto a wafer at a location where an electronic circuit pattern is to be imprinted. Therefore, the quality of all circuits and devices of the chip is determined by the sprayed position of the nano imprinting glue, so that the defects in the nano imprinting lithography process are detected by machine vision in the prior art, and the quality of the circuit formed in the nano imprinting process is ensured. However, in the image acquisition process, a large amount of noise exists, and the noise can interfere with the quality of the nanoimprint wafer image, so that the defect detection precision is not high.
The existing image denoising processing method comprises the following steps: mean denoising, median denoising, gaussian denoising and the like, but the conventional image denoising processing method can cause unclear image edges, increase the blurring degree of the image and are not beneficial to defect detection.
Disclosure of Invention
Aiming at the defects in the prior art, the method for detecting the defects of the nanoimprint wafer solves the problems that the edges of the image are unclear and the blurring degree of the image is increased in the existing image denoising processing method.
In order to achieve the aim of the invention, the invention adopts the following technical scheme: the nano-imprint wafer image acquisition processing method comprises the following steps:
s1, respectively segmenting a standard nano-imprint wafer image and an acquired nano-imprint wafer image to obtain standard image sub-blocks and acquired image sub-blocks;
s2, carrying out feature extraction on the standard image sub-blocks by adopting a first feature extraction channel to obtain first feature data;
s3, performing feature extraction on the acquired image sub-blocks by adopting a second feature extraction channel to obtain second feature data;
s4, obtaining noise distribution degree according to the second characteristic data and the first characteristic data;
s5, denoising the acquired nanoimprint wafer image according to the noise distribution degree corresponding to each acquired image sub-block to obtain a denoised nanoimprint wafer image.
Further, the structures of the first feature extraction channel and the second feature extraction channel in S2 and S3 are the same, and each of them includes: a first convolution layer, a second convolution layer, a first maximum pooling layer, a second maximum pooling layer, a first average pooling layer, a second average pooling layer, a first Concat layer, and a second Concat layer;
the input end of the first convolution layer is used as the input end of the first characteristic extraction channel or the second characteristic extraction channel, and the output end of the first convolution layer is respectively connected with the input end of the first maximum pooling layer and the input end of the first average pooling layer; the input end of the first Concat layer is respectively connected with the output end of the first maximum pooling layer and the output end of the first average pooling layer, and the output end of the first Concat layer is connected with the input end of the second convolution layer; the output end of the second convolution layer is respectively connected with the input end of the second maximum pooling layer and the input end of the second average pooling layer; the input end of the second Concat layer is respectively connected with the output end of the second maximum pooling layer and the output end of the second average pooling layer, and the output end of the second Concat layer is used as the output end of the first characteristic extraction channel or the second characteristic extraction channel.
The beneficial effects of the above further scheme are: the invention adopts the twice maximum pooling layer and the average pooling layer to realize the maximum degree reduction of data quantity, simultaneously reserves data characteristics, and can characterize the difference through the data processed by the maximum pooling layer and the average pooling layer after the standard image sub-block and the collected image sub-block are processed by the same characteristic extraction channel.
Further, the step S4 includes the following sub-steps:
s41, carrying out phase-wise subtraction on the second characteristic data and the first characteristic data to obtain a difference sequence;
s42, calculating the noise distribution degree according to the difference value sequence.
The beneficial effects of the above further scheme are: the second characteristic data and the first characteristic data are subtracted according to the same position, so that a difference value of each position is obtained, the difference value is constructed into a difference value sequence, and the noise distribution degree is calculated according to the situation of the difference value sequence.
Further, the formula for calculating the noise distribution degree in S42 is:
wherein->For the degree of noise distribution->As the weight of the mean value of the weight,is the maximum weight, x i For the ith difference in the sequence of differences, max { x } i And the value is maximum, N is the number of differences in the difference sequence, and the value is absolute value operation.
The beneficial effects of the above further scheme are: the invention takes the average value of the absolute values of the differences in the difference sequence and the absolute value of the difference of the maximum value, and represents the overall distribution condition of the differences in the difference sequence, thereby representing the distribution condition of noise on the sub-blocks of the acquired image.
Further, the step S5 includes the following sub-steps:
s51, denoising the acquired image sub-blocks at the same position according to the noise distribution degree corresponding to each acquired image sub-block to obtain denoised image sub-blocks;
and S52, splicing the denoising image sub-blocks to obtain the denoising nanoimprint wafer image.
The beneficial effects of the above further scheme are: according to the method and the device, the collected image sub-blocks at the same position are subjected to denoising processing according to the noise distribution degree corresponding to the collected image sub-blocks, so that partition denoising according to different noise conditions is realized, and denoising precision is improved.
Further, the step S51 includes the following sub-steps:
s511, finding out abnormal pixel points on the sub-blocks of the acquired image;
s512, inputting the pixel value of the abnormal pixel point, the pixel value of the pixel points in the adjacent area and the noise distribution degree of the corresponding acquired image sub-block into a pixel value prediction model to obtain a predicted pixel value of the abnormal pixel point;
s513, replacing the original pixel value of the abnormal pixel point with the predicted pixel value of the abnormal pixel point to obtain a denoising image sub-block.
The beneficial effects of the above further scheme are: in the invention, firstly, abnormal pixel points on the acquired image sub-block are found, so that the problems that if all the pixel points are denoised, the image edge is unclear, the blurring degree of the image is increased and the non-noise points are denoised are avoided. According to the pixel value of the abnormal pixel point, the pixel value of the pixel point in the adjacent area of the abnormal pixel point and the noise distribution degree of the acquired image sub-block, the real pixel value of the abnormal pixel point is predicted, and the denoising precision is improved.
Further, when the pixel point on the acquired image sub-block in S511 satisfies the following conditional formula, the pixel point is an abnormal pixel point, where the conditional formula is:
wherein f (x,y) Is the pixel value of the pixel point at (x, y), f (x,y+1) Is the pixel value of the pixel point at (x, y+1), f (x,y-1) Is the pixel value of the pixel point at (x, y-1), f (x+1,y) Is the pixel value of the pixel point at (x+1, y), f (x-1,y) Is the pixel value of the pixel point at (x-1, y), the absolute value is calculated, f th For the anomaly threshold, x is the abscissa of the pixel, and y is the ordinate of the pixel.
The beneficial effects of the above further scheme are: in the invention, four pixel points in the adjacent area of the abnormal pixel point are taken, the pixel value condition of the whole range is reflected by the pixel average value of the four pixel points, and if the difference between the pixel value of the central pixel point and the peripheral pixel value is overlarge, the point is the abnormal pixel point.
Further, the pixel value prediction model in S512 is:
wherein->Predicted pixel value for kth outlier pixel, for>Activating a function for sigmoid->To the degree of noise distribution, f k Is the original pixel value of the kth abnormal pixel point, f k,j For the pixel value of the jth pixel point in the adjacent area of the kth abnormal pixel point, M is the number of the pixel points in the adjacent area, θ is the correction coefficient, +.>Is an offset coefficient.
The beneficial effects of the above further scheme are: according to the method, the original pixel value of the abnormal pixel point is denoised through the noise distribution degree, and then the denoising process is further corrected through the correction coefficient and the offset coefficient, so that the prediction accuracy is improved.
Further, the correction coefficient θ and the offset coefficientThe specific values of the pixel value prediction model are obtained through training, the obtained pixel value of the abnormal pixel point, the pixel value of the pixel point in the adjacent area and the noise distribution degree of the corresponding collected image sub-block are constructed to be training samples, the pixel value prediction model is trained by adopting the training samples, and the training is ended until the absolute value of the difference value between the predicted pixel value output by the pixel value prediction model and the marked real pixel value is smaller than an error threshold value;
the loss function of the training pixel value prediction model is:
wherein L is n For the loss function of the nth training, +.>Predicted pixel values output for the nth trained pixel value prediction model, e being a natural constant, f n And (3) the true pixel value of the label corresponding to the nth training, and ln is a logarithmic function.
The beneficial effects of the above further scheme are: according to the invention, the distance between the predicted pixel value and the marked real pixel value is enhanced by adopting an exponential function, then the ratio condition of the predicted pixel value and the marked real pixel value is calculated by adopting a logarithmic function, and the difference between the predicted pixel value and the marked real pixel value is reflected by the ratio condition, so that the real difference between the predicted pixel value and the marked real pixel value is represented to the greatest extent, and the prediction precision of a pixel value prediction model is improved.
The beneficial effects of the invention are as follows: according to the method, the standard nanoimprint wafer image and the acquired nanoimprint wafer image are segmented respectively to obtain the standard image sub-blocks and the acquired image sub-blocks, so that the segmentation of the two images is realized, regional comparison is convenient, feature extraction is carried out by adopting a feature extraction channel, second feature data and first feature data are obtained respectively, the noise distribution degree of each acquired image sub-block is obtained through the comparison of the second feature data and the first feature data, the denoising processing is carried out on the corresponding acquired image sub-blocks, and the denoising nanoimprint wafer image is obtained, and the problems that the image edge is unclear and the blurring degree of the image is increased due to the existing image denoising processing method are solved.
Drawings
FIG. 1 is a flow chart of a method for image acquisition and processing of nanoimprint wafers;
fig. 2 is a schematic structural view of the first feature extraction channel and the second feature extraction channel.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and all the inventions which make use of the inventive concept are protected by the spirit and scope of the present invention as defined and defined in the appended claims to those skilled in the art.
As shown in fig. 1, a method for acquiring and processing an image of a nanoimprint wafer includes the following steps:
s1, respectively segmenting a standard nano-imprint wafer image and an acquired nano-imprint wafer image to obtain standard image sub-blocks and acquired image sub-blocks;
in the invention, the standard nano-imprint wafer image is an image obtained by manually denoising the acquired nano-imprint wafer image.
S2, carrying out feature extraction on the standard image sub-blocks by adopting a first feature extraction channel to obtain first feature data;
s3, performing feature extraction on the acquired image sub-blocks by adopting a second feature extraction channel to obtain second feature data;
as shown in fig. 2, the first feature extraction channel and the second feature extraction channel in S2 and S3 have the same structure, and each include: a first convolution layer, a second convolution layer, a first maximum pooling layer, a second maximum pooling layer, a first average pooling layer, a second average pooling layer, a first Concat layer, and a second Concat layer;
the input end of the first convolution layer is used as the input end of the first characteristic extraction channel or the second characteristic extraction channel, and the output end of the first convolution layer is respectively connected with the input end of the first maximum pooling layer and the input end of the first average pooling layer; the input end of the first Concat layer is respectively connected with the output end of the first maximum pooling layer and the output end of the first average pooling layer, and the output end of the first Concat layer is connected with the input end of the second convolution layer; the output end of the second convolution layer is respectively connected with the input end of the second maximum pooling layer and the input end of the second average pooling layer; the input end of the second Concat layer is respectively connected with the output end of the second maximum pooling layer and the output end of the second average pooling layer, and the output end of the second Concat layer is used as the output end of the first characteristic extraction channel or the second characteristic extraction channel.
The invention adopts the twice maximum pooling layer and the average pooling layer to realize the maximum degree reduction of data quantity, simultaneously reserves data characteristics, and can characterize the difference through the data processed by the maximum pooling layer and the average pooling layer after the standard image sub-block and the collected image sub-block are processed by the same characteristic extraction channel.
S4, obtaining noise distribution degree according to the second characteristic data and the first characteristic data;
the step S4 comprises the following substeps:
s41, carrying out phase-wise subtraction on the second characteristic data and the first characteristic data to obtain a difference sequence;
s42, calculating the noise distribution degree according to the difference value sequence.
The second characteristic data and the first characteristic data are subtracted according to the same position, so that a difference value of each position is obtained, the difference value is constructed into a difference value sequence, and the noise distribution degree is calculated according to the situation of the difference value sequence.
The formula for calculating the noise distribution degree in S42 is as follows:
wherein->For the degree of noise distribution->Is the mean weight>Is the maximum weight, x i For the ith difference in the sequence of differences, max { x } i And the value is maximum, N is the number of differences in the difference sequence, and the value is absolute value operation.
The invention takes the average value of the absolute values of the differences in the difference sequence and the absolute value of the difference of the maximum value, and represents the overall distribution condition of the differences in the difference sequence, thereby representing the distribution condition of noise on the sub-blocks of the acquired image.
S5, denoising the acquired nanoimprint wafer image according to the noise distribution degree corresponding to each acquired image sub-block to obtain a denoised nanoimprint wafer image.
The step S5 comprises the following substeps:
s51, denoising the acquired image sub-blocks at the same position according to the noise distribution degree corresponding to each acquired image sub-block to obtain denoised image sub-blocks;
and S52, splicing the denoising image sub-blocks to obtain the denoising nanoimprint wafer image.
According to the method and the device, the collected image sub-blocks at the same position are subjected to denoising processing according to the noise distribution degree corresponding to the collected image sub-blocks, so that partition denoising according to different noise conditions is realized, and denoising precision is improved.
The step S51 includes the following sub-steps:
s511, finding out abnormal pixel points on the sub-blocks of the acquired image;
s512, inputting the pixel value of the abnormal pixel point, the pixel value of the pixel points in the adjacent area and the noise distribution degree of the corresponding acquired image sub-block into a pixel value prediction model to obtain a predicted pixel value of the abnormal pixel point;
s513, replacing the original pixel value of the abnormal pixel point with the predicted pixel value of the abnormal pixel point to obtain a denoising image sub-block.
In the invention, firstly, abnormal pixel points on the acquired image sub-block are found, so that the problems that if all the pixel points are denoised, the image edge is unclear, the blurring degree of the image is increased and the non-noise points are denoised are avoided. According to the pixel value of the abnormal pixel point, the pixel value of the pixel point in the adjacent area of the abnormal pixel point and the noise distribution degree of the acquired image sub-block, the real pixel value of the abnormal pixel point is predicted, and the denoising precision is improved.
When the pixel point on the acquired image sub-block in S511 satisfies the following conditional formula, the pixel point is an abnormal pixel point, where the conditional formula is:
wherein f (x,y) Is the pixel value of the pixel point at (x, y), f (x,y+1) Is the pixel value of the pixel point at (x, y+1), f (x,y-1) Is the pixel value of the pixel point at (x, y-1), f (x+1,y) Is the pixel value of the pixel point at (x+1, y), f (x-1,y) Is the pixel value of the pixel point at (x-1, y), the absolute value is calculated, f th Is an abnormal threshold value, x is the transverse sitting of the pixel pointAnd y is the ordinate of the pixel point.
In the invention, four pixel points in the adjacent area of the abnormal pixel point are taken, the pixel value condition of the whole range is reflected by the pixel average value of the four pixel points, and if the difference between the pixel value of the central pixel point and the peripheral pixel value is overlarge, the point is the abnormal pixel point.
The pixel value prediction model in S512 is:
wherein->Predicted pixel value for kth outlier pixel, for>Activating a function for sigmoid->To the degree of noise distribution, f k Is the original pixel value of the kth abnormal pixel point, f k,j For the pixel value of the jth pixel point in the adjacent area of the kth abnormal pixel point, M is the number of the pixel points in the adjacent area, θ is the correction coefficient, +.>Is an offset coefficient.
According to the method, the original pixel value of the abnormal pixel point is denoised through the noise distribution degree, and then the denoising process is further corrected through the correction coefficient and the offset coefficient, so that the prediction accuracy is improved.
The correction coefficient theta and the offset coefficientThe specific values of the image are obtained through training, the obtained pixel values of the abnormal pixel points, the pixel values of the pixel points in the adjacent areas and the noise distribution degree of the corresponding collected image sub-blocks are constructed as training samples, and the training samples are adopted to train the pixel value prediction model until the pixel value is predictedEnding training when the absolute value of the difference value between the predicted pixel value output by the model and the marked real pixel value is smaller than an error threshold value;
the loss function of the training pixel value prediction model is:
wherein L is n For the loss function of the nth training, +.>Predicted pixel values output for the nth trained pixel value prediction model, e being a natural constant, f n And (3) the true pixel value of the label corresponding to the nth training, and ln is a logarithmic function.
According to the invention, the distance between the predicted pixel value and the marked real pixel value is enhanced by adopting an exponential function, then the ratio condition of the predicted pixel value and the marked real pixel value is calculated by adopting a logarithmic function, and the difference between the predicted pixel value and the marked real pixel value is reflected by the ratio condition, so that the real difference between the predicted pixel value and the marked real pixel value is represented to the greatest extent, and the prediction precision of a pixel value prediction model is improved.
According to the method, the standard nanoimprint wafer image and the acquired nanoimprint wafer image are segmented respectively to obtain the standard image sub-blocks and the acquired image sub-blocks, so that the segmentation of the two images is realized, regional comparison is convenient, feature extraction is carried out by adopting a feature extraction channel, second feature data and first feature data are obtained respectively, the noise distribution degree of each acquired image sub-block is obtained through the comparison of the second feature data and the first feature data, the denoising processing is carried out on the corresponding acquired image sub-blocks, and the denoising nanoimprint wafer image is obtained, and the problems that the image edge is unclear and the blurring degree of the image is increased due to the existing image denoising processing method are solved.
Claims (1)
1. The nano-imprint wafer image acquisition processing method is characterized by comprising the following steps of:
s1, respectively segmenting a standard nano-imprint wafer image and an acquired nano-imprint wafer image to obtain standard image sub-blocks and acquired image sub-blocks;
s2, carrying out feature extraction on the standard image sub-blocks by adopting a first feature extraction channel to obtain first feature data;
s3, performing feature extraction on the acquired image sub-blocks by adopting a second feature extraction channel to obtain second feature data;
s4, obtaining noise distribution degree according to the second characteristic data and the first characteristic data;
s5, denoising the acquired nanoimprint wafer image according to the noise distribution degree corresponding to each acquired image sub-block to obtain a denoised nanoimprint wafer image;
the first feature extraction channel and the second feature extraction channel in S2 and S3 have the same structure, and each of them includes: a first convolution layer, a second convolution layer, a first maximum pooling layer, a second maximum pooling layer, a first average pooling layer, a second average pooling layer, a first Concat layer, and a second Concat layer;
the input end of the first convolution layer is used as the input end of the first characteristic extraction channel or the second characteristic extraction channel, and the output end of the first convolution layer is respectively connected with the input end of the first maximum pooling layer and the input end of the first average pooling layer; the input end of the first Concat layer is respectively connected with the output end of the first maximum pooling layer and the output end of the first average pooling layer, and the output end of the first Concat layer is connected with the input end of the second convolution layer; the output end of the second convolution layer is respectively connected with the input end of the second maximum pooling layer and the input end of the second average pooling layer; the input end of the second Concat layer is respectively connected with the output end of the second maximum pooling layer and the output end of the second average pooling layer, and the output end of the second Concat layer is used as the output end of the first characteristic extraction channel or the second characteristic extraction channel;
the step S4 comprises the following substeps:
s41, carrying out phase-wise subtraction on the second characteristic data and the first characteristic data to obtain a difference sequence;
s42, calculating noise distribution degree according to the difference value sequence;
the formula for calculating the noise distribution degree in S42 is as follows:
,
wherein,for the degree of noise distribution->Is the mean weight>Is the maximum weight, x i For the ith difference in the sequence of differences, max { x } i The value is maximum, N is the number of differences in the difference sequence, and I is absolute value operation;
the step S5 comprises the following substeps:
s51, denoising the acquired image sub-blocks at the same position according to the noise distribution degree corresponding to each acquired image sub-block to obtain denoised image sub-blocks;
s52, splicing the denoising image sub-blocks to obtain a denoising nanoimprint wafer image;
the step S51 includes the following sub-steps:
s511, finding out abnormal pixel points on the sub-blocks of the acquired image;
s512, inputting the pixel value of the abnormal pixel point, the pixel value of the pixel points in the adjacent area and the noise distribution degree of the corresponding acquired image sub-block into a pixel value prediction model to obtain a predicted pixel value of the abnormal pixel point;
s513, replacing original pixel values of the abnormal pixel points with predicted pixel values of the abnormal pixel points to obtain denoising image sub-blocks;
when the pixel point on the acquired image sub-block in S511 satisfies the following conditional formula, the pixel point is an abnormal pixel point, where the conditional formula is:
,
wherein f (x,y) Is the pixel value of the pixel point at (x, y), f (x,y+1) Is the pixel value of the pixel point at (x, y+1), f (x,y-1) Is the pixel value of the pixel point at (x, y-1), f (x+1,y) Is the pixel value of the pixel point at (x+1, y), f (x-1,y) Is the pixel value of the pixel point at (x-1, y), the absolute value is calculated, f th As an abnormal threshold, x is the abscissa of the pixel point, and y is the ordinate of the pixel point;
the pixel value prediction model in S512 is:
,
wherein,predicted pixel value for kth outlier pixel, for>Activating a function for sigmoid->To the degree of noise distribution, f k Is the original pixel value of the kth abnormal pixel point, f k,j For the pixel value of the jth pixel point in the adjacent area of the kth abnormal pixel point, M is the number of the pixel points in the adjacent area, θ is the correction coefficient, +.>Is an offset coefficient;
the correction coefficient theta and the offset coefficientThe specific values of the image are obtained through training, the obtained pixel values of the abnormal pixel points, the pixel values of the pixel points in the adjacent area and the noise distribution degree of the corresponding acquired image sub-blocks are constructed as training samples, and the training samples are adopted to train a pixel value prediction modelTraining is finished until the absolute value of the difference value between the predicted pixel value output by the pixel value prediction model and the marked real pixel value is smaller than an error threshold value;
the loss function of the training pixel value prediction model is:
,
wherein L is n As a loss function of the nth training,predicted pixel values output for the nth trained pixel value prediction model, e being a natural constant, f n And (3) the true pixel value of the label corresponding to the nth training, and ln is a logarithmic function.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311114562.4A CN117094909B (en) | 2023-08-31 | 2023-08-31 | Nanometer stamping wafer image acquisition processing method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311114562.4A CN117094909B (en) | 2023-08-31 | 2023-08-31 | Nanometer stamping wafer image acquisition processing method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117094909A CN117094909A (en) | 2023-11-21 |
CN117094909B true CN117094909B (en) | 2024-04-02 |
Family
ID=88780104
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311114562.4A Active CN117094909B (en) | 2023-08-31 | 2023-08-31 | Nanometer stamping wafer image acquisition processing method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117094909B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117372431B (en) * | 2023-12-07 | 2024-02-20 | 青岛天仁微纳科技有限责任公司 | Image detection method of nano-imprint mold |
CN117423113B (en) * | 2023-12-18 | 2024-03-05 | 青岛华正信息技术股份有限公司 | Adaptive denoising method for archive OCR (optical character recognition) image |
CN117495722B (en) * | 2023-12-25 | 2024-03-29 | 青岛天仁微纳科技有限责任公司 | Image processing method for nanoimprint lithography |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105069758A (en) * | 2015-08-21 | 2015-11-18 | 武汉大学 | Hyperspectral image denoising method based on robust low-rank tensor |
CN109754376A (en) * | 2018-12-28 | 2019-05-14 | 深圳美图创新科技有限公司 | Image de-noising method and device |
CN112508810A (en) * | 2020-11-30 | 2021-03-16 | 上海云从汇临人工智能科技有限公司 | Non-local mean blind image denoising method, system and device |
CN114049342A (en) * | 2021-11-19 | 2022-02-15 | 上海集成电路装备材料产业创新中心有限公司 | Denoising model generation method, system, device and medium |
CN114697574A (en) * | 2020-12-30 | 2022-07-01 | 菲力尔商业系统公司 | Abnormal pixel detection system and method |
CN115063413A (en) * | 2022-08-04 | 2022-09-16 | 宁波鑫芯微电子科技有限公司 | Feature extraction method for abnormal data of super-large-scale wafer |
CN115829883A (en) * | 2023-02-16 | 2023-03-21 | 汶上县恒安钢结构有限公司 | Surface image denoising method for dissimilar metal structural member |
CN116152079A (en) * | 2022-09-06 | 2023-05-23 | 马上消费金融股份有限公司 | Image processing method and image processing model training method |
CN116188325A (en) * | 2023-03-31 | 2023-05-30 | 东北大学 | Image denoising method based on deep learning and image color space characteristics |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8345130B2 (en) * | 2010-01-29 | 2013-01-01 | Eastman Kodak Company | Denoising CFA images using weighted pixel differences |
-
2023
- 2023-08-31 CN CN202311114562.4A patent/CN117094909B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105069758A (en) * | 2015-08-21 | 2015-11-18 | 武汉大学 | Hyperspectral image denoising method based on robust low-rank tensor |
CN109754376A (en) * | 2018-12-28 | 2019-05-14 | 深圳美图创新科技有限公司 | Image de-noising method and device |
CN112508810A (en) * | 2020-11-30 | 2021-03-16 | 上海云从汇临人工智能科技有限公司 | Non-local mean blind image denoising method, system and device |
CN114697574A (en) * | 2020-12-30 | 2022-07-01 | 菲力尔商业系统公司 | Abnormal pixel detection system and method |
CN114049342A (en) * | 2021-11-19 | 2022-02-15 | 上海集成电路装备材料产业创新中心有限公司 | Denoising model generation method, system, device and medium |
CN115063413A (en) * | 2022-08-04 | 2022-09-16 | 宁波鑫芯微电子科技有限公司 | Feature extraction method for abnormal data of super-large-scale wafer |
CN116152079A (en) * | 2022-09-06 | 2023-05-23 | 马上消费金融股份有限公司 | Image processing method and image processing model training method |
CN115829883A (en) * | 2023-02-16 | 2023-03-21 | 汶上县恒安钢结构有限公司 | Surface image denoising method for dissimilar metal structural member |
CN116188325A (en) * | 2023-03-31 | 2023-05-30 | 东北大学 | Image denoising method based on deep learning and image color space characteristics |
Non-Patent Citations (3)
Title |
---|
Zhaoming Kong 等.Color Image and Multispectral Image Denoising Using Block Diagonal Representation.《IEEE TRANSACTIONS ON IMAGE PROCESSING》.2019,第28卷(第9期),4247-4259. * |
兰红 等.面向全景拼接的图像配准技术研究及应用.《计算机工程与科学》.2016,第38卷(第2期),317-324. * |
李杭 等.基于噪声方差估计的伪造图像盲检测方法.《计算机应用研究》.2017,第34卷(第1期),314-316. * |
Also Published As
Publication number | Publication date |
---|---|
CN117094909A (en) | 2023-11-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN117094909B (en) | Nanometer stamping wafer image acquisition processing method | |
CN109829903B (en) | Chip surface defect detection method based on convolution denoising autoencoder | |
CN105976330B (en) | A kind of embedded greasy weather real time video image stabilization | |
CN112651968B (en) | Wood board deformation and pit detection method based on depth information | |
CN115035122B (en) | Artificial intelligence-based integrated circuit wafer surface defect detection method | |
CN106097256B (en) | A kind of video image fuzziness detection method based on Image Blind deblurring | |
CN105913396A (en) | Noise estimation-based image edge preservation mixed de-noising method | |
CN102156996A (en) | Image edge detection method | |
CN106485182A (en) | A kind of fuzzy Q R code restored method based on affine transformation | |
CN110598613B (en) | Expressway agglomerate fog monitoring method | |
CN116630813B (en) | Highway road surface construction quality intelligent detection system | |
CN115311282B (en) | Wafer defect detection method based on image enhancement | |
CN116660286A (en) | Wire harness head peeling measurement and defect detection method and system based on image segmentation | |
CN117152129B (en) | Visual detection method and system for surface defects of battery cover plate | |
CN116779465B (en) | Nano-imprinting wafer defect detection method | |
Ma et al. | An automatic detection method of Mura defects for liquid crystal display | |
CN114170165A (en) | Chip surface defect detection method and device | |
CN109919150A (en) | A kind of non-division recognition sequence method and system of 3D pressed characters | |
CN113888536A (en) | Printed matter double image detection method and system based on computer vision | |
Yu et al. | SEM image quality enhancement: an unsupervised deep learning approach | |
CN104616266B (en) | A kind of noise variance estimation method based on broad sense autoregression heteroscedastic model | |
CN106023097A (en) | Iterative-method-based flow field image preprocessing algorithm | |
Wu et al. | Research on crack detection algorithm of asphalt pavement | |
CN117152214A (en) | Defect identification method based on improved optical flow detection | |
CN102682434B (en) | Image denoising method on basis of edge prior and NSCT (Non-sampling Contourlet Transform)-domain GSM (gaussian scale mixture model) |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |