CN106600600A - Wafer defect detection method based on characteristic matching - Google Patents

Wafer defect detection method based on characteristic matching Download PDF

Info

Publication number
CN106600600A
CN106600600A CN201611214743.4A CN201611214743A CN106600600A CN 106600600 A CN106600600 A CN 106600600A CN 201611214743 A CN201611214743 A CN 201611214743A CN 106600600 A CN106600600 A CN 106600600A
Authority
CN
China
Prior art keywords
image
point
chip
value
template
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.)
Pending
Application number
CN201611214743.4A
Other languages
Chinese (zh)
Inventor
杜娟
谭健胜
胡跃明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201611214743.4A priority Critical patent/CN106600600A/en
Publication of CN106600600A publication Critical patent/CN106600600A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image
    • 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
    • G06T2207/30141—Printed circuit board [PCB]

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a wafer defect detection method based on characteristic matching and relates to the image processing method technology field. The method comprises the following steps of collecting a wafer sample image, matching the wafer sample image with a chip template image and searching a single chip area to be detected in the wafer sample image; describing characteristic vectors of characteristic points of the chip template image and a chip image to be detected; matching the characteristic points of the chip template image and the chip image to be detected, rejecting a mismatching point, and according to a matched result, completing position adjusting and correction of the chip image to be detected; and carrying out binarization on the chip template image and the registered chip image to be detected so as to realize extraction of chip defect information. By using the method, small defects on a wafer can be effectively detected, detection efficiency is high and detection stability is good.

Description

The wafer defect detection method of feature based matching
Technical field
The present invention relates to image processing method technical field, more particularly to a kind of wafer defect detection of feature based matching Method.
Background technology
With the fast development of ic manufacturing technology, the characteristic size of wafer constantly reduces, and causes more small Defect.The defect of crystal column surface has become the major obstacle for affecting yield.Therefore, the research of defect becomes more and more important, The result of study of defect can be used to improve or modified technique, so as to improve the yield rate of product, while IC cores can be improved The reliability of piece.And the general method detected by naked eyes is detected to wafer defect in prior art, not only Detection results It is bad and detection efficiency is relatively low.
The content of the invention
The technical problem to be solved is how to provide one kind can effectively detect tiny flaw on wafer, and The wafer defect detection method of the high feature based matching of detection efficiency.
To solve above-mentioned technical problem, the technical solution used in the present invention is：A kind of wafer of feature based matching lacks Sunken detection method, it is characterised in that comprise the steps：
Collection wafer sample image, wafer sample image is matched with chip template image, finds wafer sample figure The single chip area to be detected as in；
The characteristic vector of the characteristic point of chip template image and chip image to be detected is described；
The characteristic point of chip template image and chip image to be detected is matched, then Mismatching point is rejected, root The position adjustment and correction of chip image to be detected are completed according to the result of matching；
Binaryzation is carried out respectively to the chip image to be detected after chip template image and registration, chip defect information is realized Extraction.
Preferably, described collection wafer sample image, wafer sample image is matched with chip template image, is sought The method for looking for single chip area to be detected in wafer sample image is as follows：
First by the similarity measure of pixel calculation template image and image to be searched, the similar of maximum or minimum is then found Gauge region is used as matched position.
Preferably, it is described first by pixel calculation template image and the similarity measure of image to be searched, then find most Big or minimum similarity measure region is as follows as the method for matched position：
Using the template matching method based on gray value, the template matching method based on gray value is by the gray scale of entire image Value as similarity measure, using the search strategy for defining according to order from top to bottom, from left to right in image to be searched Qualified region is searched for, by the search window for setting a specified size, scans for comparing in the search window；
The pose of object is described by translating in image to be searched, and template is represented by image t (r, c), therein Area-of-interest is appointed as T, and template matches are exactly to translate template area-of-interest T according to certain order in image to be matched, Then the similarity value s of the region and template area-of-interest in image to be matched is calculated, similarity measure is described by following formula:
S (r, c)=s { t (u, v), f (r+u, c+v)；(u,v)∈T}
Wherein s (r, c) represents the similarity measure calculated based on gray value, and t (u, v) represents the gray value of each point in template, f (r+u, c+v) represents that template area-of-interest moves on to the gray value of image current location；
The method for asking for similarity measure is absolute value sum SAD or all differences for calculating gray value difference between two images Quadratic sum SSD, SAD and SSD represents respectively with following two formula:
Wherein, n represents the quantity of pixel in the template interest region, i.e. n=| T |；It is similar for SAD and SSD The value of tolerance is bigger, and the difference between image to be searched and template is also bigger.
Preferably, what described chip template image and the characteristic vector of the characteristic point of chip image to be detected was described Method is as follows：
Extreme point in SIFT algorithms is replaced using Harris operator extractions characteristic point, and is accurately positioned extreme point, then Principal direction is defined for each characteristic point, feature point description is finally generated using PCA.
Preferably, described employing Harris operator extractions characteristic point replaces the method for the extreme point in SIFT algorithms such as Under：
The value of Harris operators is：
R=det (C)-ktr2(C)
In formula, det is determinant of a matrix；Tr is the mark of matrix；C is correlation matrix, and：
In formula, Iu(x)、Iv(x)、IuvX () is respectively local derviation and second order of the gray scale of picture point x in u and v directions and mixes Local derviation；K is empirical value；When the Harris operators R that certain is put is more than given threshold T, the point is angle point.
Preferably, the described method for being accurately positioned extreme point is as follows：
The accurate of characteristic point position and yardstick coordinate is obtained using second order Taylor expansions D (X) interpolation of DOG functions Value：
Vector X=(x, y, σ) in formula, represents position, the yardstick skew between sampled point and characteristic point；Make the single order of above formula Derivative is 0, and the offset vector that can obtain characteristic point exact position is：
WillThe coordinate X of former thick characteristic point is added to, that is, the sub-pix precise interpolation for obtaining characteristic point is estimated, through mathematics generation Can obtain after entering：
When | D (X) | values are less than certain threshold value, this feature point is cast out；
Additionally should cast out with unstable skirt response point, such extreme point of DOG functions is generally tangential at edge There is larger principal curvatures, and the vertical direction at edge has less principal curvatures；In order to detect principal curvatures whether in certain thresholding r Under, only need to detect whether to meet：
H is the Hessian matrixes of DOG functions in formula:.
Preferably, the described method that feature point description is generated using PCA is as follows：
The neighborhood that a size is 41 × 41 is determined to characteristic point, this neighborhood is rotated to principal direction；
Calculate the horizontal gradient and vertical gradient of pixel in neighborhood, each characteristic point determine a size for 39 × 39 × The Feature Descriptor of 2=3042 dimensions；
M characteristic point is gathered for characteristics of image, primitive character matrix M, a 3042 × m of matrix size is reached into, is calculated The covariance matrix N of matrix；
Calculate the characteristic vector of covariance matrix N, sorted according to the size of characteristic root, select corresponding front n feature to Amount, constitutes projection matrix T；
To new feature interpretation subvector, projection matrix T is multiplied by, obtains the characteristic vector that 3042 dimensions drop to n dimensions.
Preferably, Feature Points Matching is carried out using k-d tree, matching process is as follows：
Build k-d tree：
In image feature descriptor matching process, each node of k-d tree will at least include multinomial information, and have each Specific meanings；The building process of k-d tree is a process launched step by step, all data is progressively divided, from level to level Determine space segmentation domain with segmentation numerical value；During feature registration, it is by relatively more each to select for the segmentation domain of every one-level Data variance size on individual dimension selects most scattered that one-dimensional conduct segmentation domain, while during segmentation domain is located at come what is determined Between value as partition value；
The arest neighbors of k-d tree is searched：
K-d tree arest neighbors search procedure is by constantly by the node on point to be checked and k-d tree in its partition dimension It is compared with the keyValue of node, it is less than keyValue values, continue to search for left subtree, on the contrary sub-tree search to the right, one Until leaf node.
Preferably, the described method rejected to Mismatching point is as follows：
From template image calculate match point coordinate maximum and minimum of a value, and accordingly in template image comprising matching The part of point is divided into b × b blocks；
Wherein, there is match point in some blocks, do not have in some blocks, the block without match point is removed；
Then from template image choose 8 mutually different piece, from this 8 pieces per block in randomly select a point, obtain To 8 pairs of distributions than more uniform match point；
When random sample collection quantity is selected than asking for model parameter needed for match to quantity more than one, first profit Whether determine the parameter of model with n data in this n+1 sample, find temporary pattern, then detect that (n+1)th sample exists On temporary pattern, if it is not, reselecting a random sample collection；If it is, this temporary pattern is candidate family, algorithm Continually look for the supported collection of this candidate family；If supported amount is sufficiently large, candidate family is found object module；It is no Then, gravity treatment random sample collection.
Preferably, the described chip image to be detected to after chip template image and registration carries out respectively binaryzation, real The method of the extraction of existing chip defect information is as follows：
Gray-scale pixel values of the tentative standard image in coordinate (x, y) place are f (x, y), and defect image is in coordinate (x, y) Gray-scale pixel values g (x, y) at place, then value of its absolute difference figure at coordinate (x, y) place can be expressed as | f (x, y)-g (x, y) |；
It is subsequently using the dividing method based on fixed threshold, each grey scale pixel value in result difference shadow figure is pre- with this Gating limit value compares, it is stipulated that grey scale pixel value is entered as into 255 more than the pixel of this threshold value, otherwise is entered as 0, defect Information is individually extracted.
It is using the beneficial effect produced by above-mentioned technical proposal：The method of the invention utilizes Digital Image Processing skill Art carries out feature extraction and matching to wafer sample, you can realizes the automatic on-line detection of wafer defect, can effectively detect Tiny flaw on wafer, and detection efficiency is high, stability is high.
Additionally, methods described extracts the extreme point in characteristic point replacement SIFT algorithms using Harris, and adopt principal component Analytic approach is characterized generation characteristic vector description, reduces the dimension of Feature Descriptor, substantially increases the speed of feature extraction Degree, meets the requirement of real-time of defects detection.
Matching search is carried out using k-d tree, match time is effectively saved, the efficiency of detection is improve, using improved RANSAC algorithms are rejected to Mismatching point, it is ensured that the accuracy of matching, are easy to that the registration of image is better achieved.
Description of the drawings
With reference to the accompanying drawings and detailed description the present invention is further detailed explanation.
Fig. 1 is the flow chart of embodiment of the present invention methods described；
Fig. 2 is image to be searched in embodiment of the present invention methods described；
Fig. 3 is template image in embodiment of the present invention methods described；
Fig. 4 is the schematic diagram of k-d tree matching method in embodiment of the present invention methods described；
Fig. 5 is improved RANSAC algorithms schematic diagram in embodiment of the present invention methods described.
Specific embodiment
With reference to the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Ground description, it is clear that described embodiment a part of embodiment only of the invention, rather than the embodiment of whole.It is based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made Embodiment, belongs to the scope of protection of the invention.
Many details are elaborated in the following description in order to fully understand the present invention, but the present invention can be with It is different from alternate manner described here to implement using other, those skilled in the art can be without prejudice to intension of the present invention In the case of do similar popularization, therefore the present invention is not limited by following public specific embodiment.
It is overall, as shown in figure 1, the invention discloses a kind of wafer defect detection method of feature based matching, including Following steps：
S101：Collection wafer sample image, wafer sample image is matched with chip template image, finds wafer sample Single chip area to be detected in product image；
S102：The characteristic vector of the characteristic point of chip template image and chip image to be detected is described；
S103：The characteristic point of chip template image and chip image to be detected is matched, then Mismatching point is picked Remove, the position adjustment and correction of chip image to be detected are completed according to the result of matching；
S104：Binaryzation is carried out respectively to the chip image to be detected after chip template image and registration, realizes that chip lacks The extraction of sunken information.
The method of the invention carries out feature extraction and matching using digital image processing techniques to wafer sample, you can real The automatic on-line detection of existing wafer defect, can effectively detect the tiny flaw on wafer, and detection efficiency is high, stability It is high.
Specifically, S101：By image acquisition device wafer sample image, by wafer sample image transmitting to calculating Machine, wafer sample image is matched with chip template image, finds single chip area to be detected in wafer sample image.
Template matches are the similarity measures by calculation template image and image to be searched, so as to look in image to be searched To the process of template image.The process of template matches can be expressed as：First by pixel calculation template image and image to be searched Similarity measure, then find the similarity measure region of maximum or minimum as matched position, its principle is as Figure 2-3.
Because the grey value profile in each region of wafer sample image is uniform fixed, therefore present invention employs based on gray scale The template matching method of value.Template matching method based on gray value as similarity measure, utilizes the gray value of entire image The search strategy for defining searches for qualified region according to order from top to bottom, from left to right in image to be searched, By the search window for setting a specified size, scan for comparing in the search window.
The pose of object can be described by translating in image to be searched.Template represents by image t (r, c), its In area-of-interest be appointed as T, template matches are exactly to translate template region of interest according to certain order in image to be matched Domain T, then calculates the similarity value s of the region and template area-of-interest in image to be matched.Similarity measure is retouched by following formula State:
S (r, c)=s { t (u, v), f (r+u, c+v)；(u,v)∈T}
Wherein s (r, c) represents the similarity measure calculated based on gray value, and t (u, v) represents the gray value of each point in template, f (r+u, c+v) represents that template area-of-interest moves on to the gray value of image current location.
The most straightforward procedure for asking for similarity measure be calculate two images between gray value difference absolute value sum (SAD) or The quadratic sum (SSD) of all differences, SAD and SSD can be represented respectively with following two formula:
Wherein, n represents the quantity of pixel in the interest region in template, i.e. n=| T |.For SAD and SSD, phase The value of likelihood metric is bigger, and the difference between image to be searched and template is also bigger.Using the template matches side based on gray value Method can determine that one single chip region to be detected.
S102：The characteristic vector of the characteristic point of chip template image and chip image to be detected is described；
SIFT algorithms are a kind of to maintain the invariance based on metric space, to image scaling, rotation even affine transformation Image local feature describes algorithm, and characteristic extraction procedure mainly has 4 steps as follows：
1) yardstick spatial extrema point is detected
One width two dimensional image, the metric space under different scale is represented can be obtained by image with Gaussian kernel convolution：
L (x, y, σ)=G (x, y, σ) * I (x, y)
(x, y) is space coordinates in formula；σ is yardstick coordinate；G (x, y, σ) is changeable scale Gaussian function：
In order to effectively in metric space detect stable characteristic point, it is proposed that Gaussian difference scale space, i.e. DOG are calculated Son, be dimension normalization LOG operators it is approximate, generated using Gaussian difference pyrene and the image convolution of different different scales：
D (x, y, σ)=(G (x, y, k σ)-G (x, y, σ)) * I (x, y)=L (x, y, k σ)-L (x, y, σ)
In order to find the extreme point of metric space, each sampled point will compare with its all of consecutive points, whether see it Than it image area and scale domain consecutive points it is big or little.Middle test point and with it with 8 consecutive points of yardstick and Totally 26 points compare corresponding 9 × 2 points of neighbouring yardstick, if the pixel is all extreme value in this 26 neighborhood territory pixels, Then as the extreme point of candidate.
2) it is accurately positioned extreme point
By the position and the yardstick that fit three-dimensional quadratic function accurately to determine characteristic point, while removing the spy of low contrast Levy a little and unstable skirt response point (because DOG operators can produce stronger skirt response), to strengthen matching stability, carry High noise resisting ability.
The exact value of characteristic point position and yardstick coordinate is obtained using second order Taylor expansions D (X) interpolation of DOG functions (sub-pixel precision)：
Vector X=(x, y, σ) in formula, represents position, the yardstick skew between sampled point and characteristic point.Make the single order of above formula Derivative is 0, and the offset vector that can obtain characteristic point exact position is：
WillThe coordinate X of former thick characteristic point is added to, that is, the sub-pix precise interpolation for obtaining characteristic point is estimated.Through mathematics generation Can obtain after entering：
When | D (X) | value less than certain threshold value when, this feature point can be cast out, generally, this feature point pairs noise-sensitive so It is unstable.
Additionally should cast out with unstable skirt response point, such extreme point of DOG functions is generally tangential at edge There is larger principal curvatures, and the vertical direction at edge has less principal curvatures.In order to detect principal curvatures whether in certain thresholding r Under, only need to detect whether to meet：
H is the Hessian matrixes of DOG functions in formula:
3) it is each feature point selection principal direction
It is each characteristic point assigned direction parameter using the gradient direction distribution characteristic of characteristic point neighborhood territory pixel, makes operator Possesses rotational invariance.Calculate the gradient-norm of characteristic point and direction in Gaussian spatial first：
Then sample in feature neighborhood of a point in Gaussian spatial, create gradient orientation histogram.Histogram is made per 10 degree For a post, totally 36 posts.Then each sampled point in neighborhood is included into into appropriate post by gradient direction θ, with gradient-norm m work For the weight of contribution.Histogrammic main peak value is finally selected as the principal direction of characteristic point, value is chosen and is reached main peak value 80% Local peaking above is used as auxiliary direction.Such a characteristic point may be designated with multiple directions, can be strengthened The robustness matched somebody with somebody.
4) feature point description is generated
Reference axis is rotated to be into the direction of characteristic point, to guarantee rotational invariance.Next 8 are taken centered on characteristic point × 8 window.With window central point as the one of each little lattice representative feature vertex neighborhood place metric space of the position of current signature point Individual pixel, the direction of arrow represents the gradient direction of the pixel, and arrow length represents gradient modulus value, then on the fritter per 4 × 4 Calculate
The gradient orientation histogram in 8 directions, draws the accumulated value of each gradient direction, you can form a seed point.
The united thought of this neighborhood directivity information enhances the antimierophonic ability of algorithm, simultaneously for missing containing positioning Poor characteristic matching also provides preferable fault-tolerance.It is special to each in order to strengthen the robustness of matching during Practical Calculation Levy using 4 × 4 totally 16 seed points describing, for a characteristic point can just produce 128 data, that is, ultimately form The SIFT feature vector of 128 dimensions.Now SIFT feature vector has eliminated the shadow of the geometry deformation factor such as dimensional variation, rotation Ring, be further continued for by the length normalization method of characteristic vector, then can further removing the impact of illumination variation.
Because SIFT algorithms have been used for multiple times convolution smooth operation and weighted histogram statistics, algorithm complex is higher, meter Evaluation time is long, while generating, characteristic point is excessive, the dimension of feature point description is too high, affects matching and search speed, in order to full The real time problems of sufficient defect detecting system,
The present invention is improved SIFT feature extraction algorithm in terms of two, as follows：
The first employ Harris operators to replace SIFT in extreme points extraction algorithm.Harris operators are that one kind has The point feature extraction operator of effect, its value is：
R=det (C)-ktr2(C)
In formula, det is determinant of a matrix；Tr is the mark of matrix；C is correlation matrix, and：
In formula, Iu(x)、Iv(x)、IuvX () is respectively local derviation and second order of the gray scale of picture point x in u and v directions and mixes Local derviation；K is empirical value, generally takes 0.04～0.06.When the Harris operators R that certain is put is more than given threshold T, the point is angle Point.
Harris operators calculate simple, and the characteristic point of extraction uniformly rationally, therefore extracts characteristic point replacement from Harris Extreme point in SIFT algorithms, then defines principal direction for each characteristic point, is that each characteristic point generates characteristic vector description, The requirement of real-time of defects detection algorithm can well be met.
It two is that feature point description is generated using PCA, and its principle steps is as follows：
1) neighborhood that a size is 41 × 41 is determined to characteristic point, this neighborhood is rotated to principal direction.
2) horizontal gradient and vertical gradient of pixel in neighborhood are calculated, so each characteristic point determines a size and is The Feature Descriptor of 39 × 39 × 2=3042 dimensions.
3) m characteristic point is gathered for characteristics of image, has so reached into a primitive character matrix M, matrix size 3042 × m, the covariance matrix N of calculating matrix；
4) characteristic vector of covariance matrix N is calculated, is sorted according to the size of characteristic root, select corresponding front n feature Vector (present invention selects n=20), constitutes projection matrix T；
5) to new feature interpretation subvector, projection matrix T is multiplied by, obtains the characteristic vector that 3042 dimensions drop to n dimensions；
In fact, step 3) and step 4) be to calculate before, i.e., projection matrix is adopted by same class image set PCA principles are calculated in advance.Characteristic point is calculated the feature interpretation period of the day from 11 p.m. to 1 a.m of 3042 dimensions from image, it is only necessary to projection matrix It is multiplied, you can reach dimensionality reduction.With regard to the selection of the n of projection matrix, the value of n can be as needed fixed, also dependent on covariance The characteristic value energy value percentage of matrix, automatically determines the size of n.The present invention adopts n=20 best results.
Methods described extracts the extreme point in characteristic point replacement SIFT algorithms using Harris, and adopts PCA Generation characteristic vector description is characterized, the dimension of Feature Descriptor is reduced, the speed of feature extraction is substantially increased, is met The requirement of real-time of defects detection.
S103：The characteristic point of chip template image and chip image to be detected is matched, then Mismatching point is picked Remove, the position adjustment and correction of chip image to be detected are completed according to the result of matching.
K-d tree is a kind of space partition tree, whole space is divided into into specific several sub-spaces, then in search procedure Just can scan in specific subspace, constantly reduce spatial dimension.Mainly included based on the Feature Points Matching of k-d tree Two steps are as follows：
Build k-d tree：
In image feature descriptor matching process, each node of k-d tree will at least include multinomial information, such as Partition, keyValue etc., there is respective specific meanings.The building process of k-d tree is a process launched step by step, right All data are progressively divided, and determine space segmentation domain with segmentation numerical value from level to level.During feature registration, for each The segmentation domain selection of level is what is determined by comparing the data variance size on each dimension, selects most scattered that one-dimensional work To split domain, while splitting domain is located at middle value as partition value.
The arest neighbors of k-d tree is searched：
K-d tree arest neighbors search procedure is by constantly by the node on point to be checked and k-d tree in its partition dimension It is compared with the keyValue of node, it is less than keyValue values, continue to search for left subtree, on the contrary sub-tree search to the right, one Until leaf node.The leaf node for searching out is not necessarily nearest neighbor point, as shown in Fig. 4 is in two-dimensional space.Point S Jing k-d to be checked Tree search falls in D point places space, but actually A points are nearest neighbor points, therefore backtracking is also needed to after leaf node is searched.
Mismatching point is needed on very big impact being brought on the calculating of transformation parameter, therefore when transformation parameter estimation is carried out Carry out the purification of matching double points.Algorithm calculating process is reliable and stable, and high precision, right for RANSAC (random sampling uniformity) Mismatching point can preferably reject Mismatching point pair to there is very strong adaptability.
But the RANSAC algorithms of routine have two：When the first randomly selects sample set, there are two candidates Point hypotelorism is considered as that a point arranges inaccurate problem so as to try to achieve fundamental matrix.It two is to select one at random every time Random sample collection, will find the supported collection of its correspondence candidate family parameter.For the observation data set existed compared with multiple error, will Having many times is wasted on the corresponding supported collection point of searching.
A kind of improved RANSAC algorithms are proposed for the problems referred to above present invention, mainly in terms of two improving The effect of RANSAC algorithms：
1) randomly select in match point, take match point as shown in Figure 5 to randomly select method by block.It is specific as follows：From Calculate the maximum and minimum of a value of match point coordinate in template image, and accordingly putting down comprising the part of match point in template image It is divided into b × b blocks.In figure, b=4, wherein, there is match point in some blocks, do not have in some blocks, the block without match point Remove.
Then from template image choose 8 mutually different piece, from this 8 pieces per block in randomly select a point, can 8 pairs of distributions are obtained than more uniform match point.The fundamental matrix calculated with such 8 pairs of match points is more stable, accurate.
2) match needed for when random sample collection quantity is selected than asking for model parameter to quantity more than one, first Determine the parameter of model using n data in this n+1 sample, find temporary pattern, then detect that (n+1)th sample is On temporary pattern, if it is not, reselecting a random sample collection (n+1 sample)；If it is, this temporary pattern is Candidate family, algorithm continually looks for the supported collection of this candidate family.If supported amount is sufficiently large, candidate family is and is found Object module；Otherwise, gravity treatment random sample collection.So may not necessarily be when supported collection be determined every time all by all observation data Point detection one time, reduces amount of calculation.
Matching search is carried out using k-d tree, match time is effectively saved, the efficiency of detection is improve, using improved RANSAC algorithms are rejected to Mismatching point, it is ensured that the accuracy of matching, are easy to that the registration of image is better achieved.
S104：Binaryzation is carried out respectively to the chip image to be detected after chip template image and registration, realizes that chip lacks The extraction of sunken information.
After the matching work of template image and defect image to be detected is completed, the defect image after registration can be obtained. Subsequently binaryzation is carried out respectively to template image and defect image, then defect information is extracted using gray scale difference shadow method, its Algorithm is substantially the absolute difference of image, and main thought is：
Gray-scale pixel values of the tentative standard image in coordinate (x, y) place are f (x, y), and defect image is in coordinate (x, y) Gray-scale pixel values g (x, y) at place, then value of its absolute difference figure at coordinate (x, y) place can be expressed as | f (x, y)-g (x, y) |.Comprising defective entrained half-tone information in the difference map in strength for now obtaining.Subsequently use based on fixed threshold Dividing method, the method using a threshold value is manually set presets each grey scale pixel value in result difference shadow figure with this Threshold value compares, it is stipulated that grey scale pixel value is entered as into 255 (white pixel points) more than the pixel of this threshold value, on the contrary assignment For 0 (black pixel point).Then defect information can be individually extracted, finally realize the work of the online defects detection of wafer.

Claims (10)

1. the wafer defect detection method that a kind of feature based is matched, it is characterised in that comprise the steps：
Collection wafer sample image, wafer sample image is matched with chip template image, in finding wafer sample image Single chip area to be detected；
The characteristic vector of the characteristic point of chip template image and chip image to be detected is described；
The characteristic point of chip template image and chip image to be detected is matched, then Mismatching point is rejected, according to The result matched somebody with somebody completes the position adjustment and correction of chip image to be detected；
Binaryzation is carried out respectively to the chip image to be detected after chip template image and registration, carrying for chip defect information is realized Take.
2. the wafer defect detection method that feature based as claimed in claim 1 is matched, it is characterised in that described collection is brilliant Circle sample image, wafer sample image is matched with chip template image, finds single to be detected in wafer sample image The method of chip area is as follows：
First by pixel calculation template image and the similarity measure of image to be searched, the similarity measure of maximum or minimum is then found Region is used as matched position.
3. the wafer defect detection method that feature based as claimed in claim 2 is matched, it is characterised in that described presses first The similarity measure of pixel calculation template image and image to be searched, then find the similarity measure region of maximum or minimum as Method with position is as follows：
Using the template matching method based on gray value, the template matching method based on gray value makees the gray value of entire image For similarity measure, searched in image to be searched according to order from top to bottom, from left to right using the search strategy for defining Qualified region, by the search window for setting a specified size, scans for comparing in the search window；
The pose of object is described by translating in image to be searched, and template is represented by image t (r, c), and sense therein is emerging Interesting region is appointed as T, and template matches are exactly to translate template area-of-interest T according to certain order in image to be matched, then The similarity value s of the region and template area-of-interest in image to be matched is calculated, similarity measure is described by following formula:
S (r, c)=s { t (u, v), f (r+u, c+v)；(u,v)∈T}
Wherein s (r, c) represents the similarity measure calculated based on gray value, and t (u, v) represents the gray value of each point in template, f (r+ U, c+v) represent that template area-of-interest moves on to the gray value of image current location；
The method for asking for similarity measure is calculate absolute value sum SAD of gray value difference between two images or all differences flat Side and SSD, SAD and SSD are represented respectively with following two formula:
S A D ( r , c ) = 1 n Σ ( u , v ) ∈ T | t ( u , v ) - f ( r + u , c + v ) |
S S D ( r , c ) = 1 n Σ ( u , v ) ∈ T [ t ( u , v ) - f ( r + u , c + v ) ] 2
Wherein, n represents the quantity of pixel in the template interest region, i.e. n=| T |；For SAD and SSD, similarity measure Value it is bigger, the difference between image to be searched and template is also bigger.
4. the wafer defect detection method that feature based as claimed in claim 1 is matched, it is characterised in that described chip dies The method that the characteristic vector of the characteristic point of domain picture and chip image to be detected is described is as follows：
Extreme point in SIFT algorithms is replaced using Harris operator extractions characteristic point, and is accurately positioned extreme point, be then every Individual characteristic point defines principal direction, and feature point description is finally generated using PCA.
5. the wafer defect detection method that feature based as claimed in claim 4 is matched, it is characterised in that described employing Harris operator extractions characteristic point replaces the method for the extreme point in SIFT algorithms as follows：
The value of Harris operators is：
R=det (C)-ktr2(C)
In formula, det is determinant of a matrix；Tr is the mark of matrix；C is correlation matrix, and：
C ( x ) = I u 2 ( x ) I u v ( x ) I u v ( x ) I v 2 ( x )
In formula, Iu(x)、Iv(x)、IuvX () is respectively the local derviation and second order mixing local derviation of the gray scale in u and v directions of picture point x； K is empirical value；When the Harris operators R that certain is put is more than given threshold T, the point is angle point.
6. the wafer defect detection method that feature based as claimed in claim 4 is matched, it is characterised in that described is accurate fixed The method of position extreme point is as follows：
The exact value of characteristic point position and yardstick coordinate is obtained using second order Taylor expansions D (X) interpolation of DOG functions：
D ( x ) = D + ∂ D T ∂ X X + 1 2 X T ∂ D T ∂ X 2 X
Vector X=(x, y, σ) in formula, represents position, the yardstick skew between sampled point and characteristic point；Make the first derivative of above formula For 0, the offset vector that can obtain characteristic point exact position is：
X ^ = - ∂ D - 1 ∂ X 2 ∂ D ∂ X
WillBe added to the coordinate X of former thick characteristic point, that is, the sub-pix precise interpolation for obtaining characteristic point is estimated, through mathematics substitute into After can obtain：
D ( X ^ ) = D + 1 2 ∂ D ∂ X X ^
When | D (X) | values are less than certain threshold value, this feature point is cast out；
Additionally should cast out with unstable skirt response point, such extreme point of DOG functions generally edge tangentially have compared with Big principal curvatures, and the vertical direction at edge has less principal curvatures；In order to detect principal curvatures whether under certain thresholding r, only Need to detect whether to meet：
T r ( H ) 2 D e t ( H ) < ( r + 1 ) 2 r
H is the Hessian matrixes of DOG functions in formula:
7. the wafer defect detection method that feature based as claimed in claim 4 is matched, it is characterised in that described adopts master Componential analysis are as follows come the method for generating feature point description：
The neighborhood that a size is 41 × 41 is determined to characteristic point, this neighborhood is rotated to principal direction；
The horizontal gradient and vertical gradient of pixel in neighborhood are calculated, each characteristic point determines that a size is 39 × 39 × 2= The Feature Descriptor of 3042 dimensions；
M characteristic point is gathered for characteristics of image, a primitive character matrix M, 3042 × m of matrix size, calculating matrix are reached into Covariance matrix N；
The characteristic vector of covariance matrix N is calculated, is sorted according to the size of characteristic root, select corresponding front n characteristic vector, structure Into projection matrix T；
To new feature interpretation subvector, projection matrix T is multiplied by, obtains the characteristic vector that 3042 dimensions drop to n dimensions.
8. the wafer defect detection method that feature based as claimed in claim 1 is matched, it is characterised in that carried out using k-d tree Feature Points Matching, matching process is as follows：
Build k-d tree：
In image feature descriptor matching process, each node of k-d tree will at least include multinomial information, and have respective spy Determine implication；The building process of k-d tree is a process launched step by step, and all data are progressively divided, and is determined from level to level Split domain with segmentation numerical value in space；During feature registration, it is by comparing each dimension to select for the segmentation domain of every one-level Come what is determined, domain is split in most scattered that the one-dimensional conduct of selection to data variance size on number, while splitting domain is located at centre Value is used as partition value；
The arest neighbors of k-d tree is searched：
K-d tree arest neighbors search procedure is by constantly by the node on point to be checked and k-d tree in its partition dimension and section The keyValue of point is compared, less than keyValue values, continues to search for left subtree, on the contrary sub-tree search to the right, until Leaf node.
9. the wafer defect detection method of feature based as claimed in claim 1 matching, it is characterised in that described to by mistake The method rejected with point is as follows：
Calculate the maximum and minimum of a value of match point coordinate from template image, and accordingly in template image comprising match point Part is divided into b × b blocks；
Wherein, there is match point in some blocks, do not have in some blocks, the block without match point is removed；
Then from template image choose 8 mutually different piece, from this 8 pieces per block in randomly select a point, obtain 8 pairs Distribution is than more uniform match point；
Select random sample collection quantity when than asking for model parameter needed for match to quantity more than one, first with this N data in n+1 samples determine the parameter of model, find temporary pattern, then detect that (n+1)th sample is interim On model, if it is not, reselecting a random sample collection；If it is, this temporary pattern is candidate family, algorithm continues Find the supported collection of this candidate family；If supported amount is sufficiently large, candidate family is found object module；Otherwise, Gravity treatment random sample collection.
10. the wafer defect detection method of feature based as claimed in claim 1 matching, it is characterised in that described to chip Chip image to be detected after template image and registration carries out respectively binaryzation, realizes the method for extraction of chip defect information such as Under：
Gray-scale pixel values of the tentative standard image in coordinate (x, y) place are f (x, y), and defect image is in coordinate (x, y) place Gray-scale pixel values g (x, y), then value of its absolute difference figure at coordinate (x, y) place can be expressed as | f (x, y)-g (x, y) |；
Subsequently using the dividing method based on fixed threshold, by each grey scale pixel value and this pre- gating in result difference shadow figure Limit value compares, it is stipulated that grey scale pixel value is entered as into 255 more than the pixel of this threshold value, otherwise is entered as 0, defect information Individually extract.
CN201611214743.4A 2016-12-26 2016-12-26 Wafer defect detection method based on characteristic matching Pending CN106600600A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611214743.4A CN106600600A (en) 2016-12-26 2016-12-26 Wafer defect detection method based on characteristic matching

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611214743.4A CN106600600A (en) 2016-12-26 2016-12-26 Wafer defect detection method based on characteristic matching

Publications (1)

Publication Number Publication Date
CN106600600A true CN106600600A (en) 2017-04-26

Family

ID=58601658

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611214743.4A Pending CN106600600A (en) 2016-12-26 2016-12-26 Wafer defect detection method based on characteristic matching

Country Status (1)

Country Link
CN (1) CN106600600A (en)

Cited By (58)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107292901A (en) * 2017-07-24 2017-10-24 北京小米移动软件有限公司 Edge detection method and device
CN107389697A (en) * 2017-07-10 2017-11-24 北京交通大学 A kind of crack detection method based on half interactive mode
CN108122230A (en) * 2018-01-10 2018-06-05 广东工业大学 The soldered ball position-recognizing system of the recognition methods of image block, device and flip-chip
CN108256565A (en) * 2018-01-10 2018-07-06 广东工业大学 A kind of adaptive masterplate matching process and device based on cross entropy distinctiveness ratio
CN108257174A (en) * 2018-01-10 2018-07-06 广东工业大学 Image block matching method and soldered ball position-recognizing system based on sub-pixel level image
CN108256566A (en) * 2018-01-10 2018-07-06 广东工业大学 A kind of adaptive masterplate matching process and device based on cosine similarity
CN108279241A (en) * 2017-10-20 2018-07-13 同济大学 A kind of workpiece configurations detection method based on machine vision
CN108376655A (en) * 2018-01-30 2018-08-07 北京世纪金光半导体有限公司 The positioning of detection defect and tracking in a kind of wafer manufacturing process
CN108447070A (en) * 2018-03-15 2018-08-24 中国科学院沈阳自动化研究所 A kind of industrial part defect detection algorithm based on pixel vectors invariant relation feature
CN108446486A (en) * 2018-03-16 2018-08-24 珠海市睿晶聚源科技有限公司 Integrated circuit diagram Hot spots detection network training and hot spot detecting method
CN108648168A (en) * 2018-03-15 2018-10-12 北京京仪仪器仪表研究总院有限公司 IC wafer surface defects detection methods
CN108765389A (en) * 2018-05-18 2018-11-06 浙江大学 A kind of microcosmic wafer surface defects image detecting method
CN109211917A (en) * 2018-08-20 2019-01-15 苏州富鑫林光电科技有限公司 A kind of general complex surface defect inspection method
CN109308489A (en) * 2018-09-03 2019-02-05 中国人民解放军国防科技大学 Method for detecting welding quality of element arc welding
CN109325930A (en) * 2018-09-12 2019-02-12 苏州优纳科技有限公司 Detection method, device and the detection device of boundary defect
CN109360185A (en) * 2018-08-28 2019-02-19 中国科学院微电子研究所 A kind of domain resolution chart extracting method, device, equipment and medium
CN109543300A (en) * 2018-11-22 2019-03-29 北京华大九天软件有限公司 A kind of method of pixel device in extraction domain
CN109671078A (en) * 2018-12-24 2019-04-23 广东理致技术有限公司 A kind of product surface image abnormity detection method and device
CN110245674A (en) * 2018-11-23 2019-09-17 浙江大华技术股份有限公司 Template matching method, device, equipment and computer storage medium
CN110363756A (en) * 2019-07-18 2019-10-22 佛山市高明金石建材有限公司 A kind of wear detecting system and detection method for bistrique
CN110517265A (en) * 2019-09-04 2019-11-29 北京理工大学 A kind of detection method of surface defects of products, device and storage medium
CN110544238A (en) * 2019-08-12 2019-12-06 华南理工大学 Flexible circuit board line defect identification method based on geometric positioning
CN111161208A (en) * 2019-11-21 2020-05-15 深圳航天智控科技有限公司 Image detection method and device
CN111340788A (en) * 2020-02-28 2020-06-26 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） Hardware trojan layout detection method and device, electronic equipment and readable storage medium
CN111429408A (en) * 2020-03-11 2020-07-17 苏州杰锐思智能科技股份有限公司 Method for detecting gold wire of packaged chip
CN111462113A (en) * 2020-04-24 2020-07-28 上海精测半导体技术有限公司 Rechecking method of non-pattern wafer
CN111707677A (en) * 2020-06-24 2020-09-25 江西景旺精密电路有限公司 PCB appearance detection method and device
CN111724335A (en) * 2019-03-21 2020-09-29 深圳中科飞测科技有限公司 Detection method and detection system
CN111738247A (en) * 2020-05-15 2020-10-02 上海望友信息科技有限公司 Identification method and identification device of polarity identification, electronic equipment and storage medium
CN111862076A (en) * 2020-07-30 2020-10-30 上海华力集成电路制造有限公司 Method for improving bright field defect detection precision and noise caused by chromatic aberration in bright field defect detection process
CN111896539A (en) * 2019-05-06 2020-11-06 迪密机电(杭州)有限公司 High-precision workpiece quality detection equipment and detection method thereof
CN112213314A (en) * 2019-07-12 2021-01-12 长鑫存储技术有限公司 Detection method and detection system for wafer side surface defects
CN112288734A (en) * 2020-11-06 2021-01-29 西安工程大学 Printed fabric surface defect detection method based on image processing
CN112508846A (en) * 2020-10-30 2021-03-16 北京市商汤科技开发有限公司 Defect detection method and device, electronic equipment and storage medium
CN112598632A (en) * 2020-12-16 2021-04-02 北京卫星制造厂有限公司 Appearance detection method and device for contact element of crimp connector
CN112819799A (en) * 2021-02-09 2021-05-18 上海众壹云计算科技有限公司 Target defect detection method, device, system, electronic equipment and storage medium
CN113269743A (en) * 2021-05-20 2021-08-17 北京理工大学重庆创新中心 Chip quantity detection method based on iterative translation verification
CN113441842A (en) * 2021-09-01 2021-09-28 佛山市联动科技股份有限公司 Wafer laser marking method
CN113454445A (en) * 2018-10-12 2021-09-28 快速图像处理有限责任公司 Compensating for reference misalignment during part inspection
CN113486729A (en) * 2021-06-15 2021-10-08 北京道达天际科技有限公司 Unmanned aerial vehicle image feature point extraction method based on GPU
CN113658080A (en) * 2021-08-23 2021-11-16 宁波棱镜空间智能科技有限公司 Method and device for geometric correction of line-scanning cylinder based on feature point matching
CN113689397A (en) * 2021-08-23 2021-11-23 湖南视比特机器人有限公司 Workpiece circular hole feature detection method and workpiece circular hole feature detection device
WO2022057389A1 (en) * 2020-09-18 2022-03-24 长鑫存储技术有限公司 Wafer alignment method and apparatus
CN114419045A (en) * 2022-03-30 2022-04-29 武汉中导光电设备有限公司 Method, device and equipment for detecting defects of photoetching mask plate and readable storage medium
CN114526682A (en) * 2022-01-13 2022-05-24 华南理工大学 Deformation measurement method based on image feature enhanced digital volume image correlation method
CN114700941A (en) * 2022-03-28 2022-07-05 中科合肥智慧农业协同创新研究院 Strawberry picking method based on binocular vision and robot system
CN114757913A (en) * 2022-04-15 2022-07-15 电子科技大学 Display screen defect detection method
CN114926671A (en) * 2022-06-16 2022-08-19 福州大学 NLED/QLED/OLED pixel defect positioning method and system based on template matching
CN115049713A (en) * 2022-08-11 2022-09-13 武汉中导光电设备有限公司 Image registration method, device, equipment and readable storage medium
CN115184380A (en) * 2022-08-01 2022-10-14 国网上海市电力公司 Printed circuit board welding spot abnormity detection method based on machine vision
CN115205223A (en) * 2022-06-22 2022-10-18 锋睿领创(珠海)科技有限公司 Visual detection method and device for transparent object, computer equipment and medium
WO2023279558A1 (en) * 2021-07-09 2023-01-12 长鑫存储技术有限公司 Defect detection method and apparatus, device and storage medium
CN116486126A (en) * 2023-06-25 2023-07-25 合肥联宝信息技术有限公司 Template determination method, device, equipment and storage medium
CN116539638A (en) * 2023-07-04 2023-08-04 季华实验室 Doping concentration measuring method, device, electronic equipment and storage medium
TWI821506B (en) * 2019-01-17 2023-11-11 以色列商應用材料以色列公司 Computerized system and method for obtaining information about a region of an object
US11815470B2 (en) 2019-01-17 2023-11-14 Applied Materials Israel, Ltd. Multi-perspective wafer analysis
CN117250208A (en) * 2023-11-20 2023-12-19 青岛天仁微纳科技有限责任公司 Machine vision-based nano-imprint wafer defect accurate detection system and method
CN117392518A (en) * 2023-12-13 2024-01-12 南京耀宇视芯科技有限公司 Low-power-consumption visual positioning and mapping chip and method thereof

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103645284A (en) * 2013-12-13 2014-03-19 林丽君 Quick smell fingerprint detection method based on improved RANSAC (random sample consensus) theory
CN104156965A (en) * 2014-08-13 2014-11-19 徐州工程学院 Automatic fast mine monitoring image stitching method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103645284A (en) * 2013-12-13 2014-03-19 林丽君 Quick smell fingerprint detection method based on improved RANSAC (random sample consensus) theory
CN104156965A (en) * 2014-08-13 2014-11-19 徐州工程学院 Automatic fast mine monitoring image stitching method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
马敏: "基于OpenCV技术的毛巾标签缺陷检测算法研究与实现", 《中国优秀硕士学位论文全文数据库信息科技辑》 *

Cited By (82)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107389697A (en) * 2017-07-10 2017-11-24 北京交通大学 A kind of crack detection method based on half interactive mode
CN107292901A (en) * 2017-07-24 2017-10-24 北京小米移动软件有限公司 Edge detection method and device
CN107292901B (en) * 2017-07-24 2020-06-02 北京小米移动软件有限公司 Edge detection method and device
CN108279241A (en) * 2017-10-20 2018-07-13 同济大学 A kind of workpiece configurations detection method based on machine vision
CN108122230A (en) * 2018-01-10 2018-06-05 广东工业大学 The soldered ball position-recognizing system of the recognition methods of image block, device and flip-chip
CN108256565A (en) * 2018-01-10 2018-07-06 广东工业大学 A kind of adaptive masterplate matching process and device based on cross entropy distinctiveness ratio
CN108257174A (en) * 2018-01-10 2018-07-06 广东工业大学 Image block matching method and soldered ball position-recognizing system based on sub-pixel level image
CN108256566A (en) * 2018-01-10 2018-07-06 广东工业大学 A kind of adaptive masterplate matching process and device based on cosine similarity
CN108122230B (en) * 2018-01-10 2022-06-24 广东工业大学 Image block identification method and device and solder ball position identification system of flip chip
CN108376655A (en) * 2018-01-30 2018-08-07 北京世纪金光半导体有限公司 The positioning of detection defect and tracking in a kind of wafer manufacturing process
CN108376655B (en) * 2018-01-30 2021-05-11 北京世纪金光半导体有限公司 Method for positioning and tracking detection defects in wafer manufacturing process
CN108447070A (en) * 2018-03-15 2018-08-24 中国科学院沈阳自动化研究所 A kind of industrial part defect detection algorithm based on pixel vectors invariant relation feature
CN108648168A (en) * 2018-03-15 2018-10-12 北京京仪仪器仪表研究总院有限公司 IC wafer surface defects detection methods
CN108447070B (en) * 2018-03-15 2021-08-10 中国科学院沈阳自动化研究所 Industrial part defect detection algorithm based on pixel vector invariant relation features
CN108446486A (en) * 2018-03-16 2018-08-24 珠海市睿晶聚源科技有限公司 Integrated circuit diagram Hot spots detection network training and hot spot detecting method
CN108765389A (en) * 2018-05-18 2018-11-06 浙江大学 A kind of microcosmic wafer surface defects image detecting method
CN109211917A (en) * 2018-08-20 2019-01-15 苏州富鑫林光电科技有限公司 A kind of general complex surface defect inspection method
CN109360185A (en) * 2018-08-28 2019-02-19 中国科学院微电子研究所 A kind of domain resolution chart extracting method, device, equipment and medium
CN109360185B (en) * 2018-08-28 2022-07-26 中国科学院微电子研究所 Layout test pattern extraction method, device, equipment and medium
CN109308489A (en) * 2018-09-03 2019-02-05 中国人民解放军国防科技大学 Method for detecting welding quality of element arc welding
CN109308489B (en) * 2018-09-03 2021-09-10 中国人民解放军国防科技大学 Method for detecting welding quality of element arc welding
CN109325930A (en) * 2018-09-12 2019-02-12 苏州优纳科技有限公司 Detection method, device and the detection device of boundary defect
CN109325930B (en) * 2018-09-12 2021-09-28 苏州优纳科技有限公司 Boundary defect detection method, device and detection equipment
CN113454445A (en) * 2018-10-12 2021-09-28 快速图像处理有限责任公司 Compensating for reference misalignment during part inspection
CN109543300B (en) * 2018-11-22 2020-05-12 北京华大九天软件有限公司 Method for extracting pixel device in layout
CN109543300A (en) * 2018-11-22 2019-03-29 北京华大九天软件有限公司 A kind of method of pixel device in extraction domain
CN110245674B (en) * 2018-11-23 2023-09-15 浙江大华技术股份有限公司 Template matching method, device, equipment and computer storage medium
CN110245674A (en) * 2018-11-23 2019-09-17 浙江大华技术股份有限公司 Template matching method, device, equipment and computer storage medium
CN109671078B (en) * 2018-12-24 2022-11-01 广东理致技术有限公司 Method and device for detecting product surface image abnormity
CN109671078A (en) * 2018-12-24 2019-04-23 广东理致技术有限公司 A kind of product surface image abnormity detection method and device
TWI821506B (en) * 2019-01-17 2023-11-11 以色列商應用材料以色列公司 Computerized system and method for obtaining information about a region of an object
US11815470B2 (en) 2019-01-17 2023-11-14 Applied Materials Israel, Ltd. Multi-perspective wafer analysis
CN111724335A (en) * 2019-03-21 2020-09-29 深圳中科飞测科技有限公司 Detection method and detection system
CN111896539A (en) * 2019-05-06 2020-11-06 迪密机电(杭州)有限公司 High-precision workpiece quality detection equipment and detection method thereof
CN112213314B (en) * 2019-07-12 2022-11-29 长鑫存储技术有限公司 Detection method and detection system for wafer side surface defects
CN112213314A (en) * 2019-07-12 2021-01-12 长鑫存储技术有限公司 Detection method and detection system for wafer side surface defects
CN110363756A (en) * 2019-07-18 2019-10-22 佛山市高明金石建材有限公司 A kind of wear detecting system and detection method for bistrique
CN110544238B (en) * 2019-08-12 2023-01-06 华南理工大学 Flexible circuit board line defect identification method based on geometric positioning
CN110544238A (en) * 2019-08-12 2019-12-06 华南理工大学 Flexible circuit board line defect identification method based on geometric positioning
CN110517265A (en) * 2019-09-04 2019-11-29 北京理工大学 A kind of detection method of surface defects of products, device and storage medium
CN111161208A (en) * 2019-11-21 2020-05-15 深圳航天智控科技有限公司 Image detection method and device
CN111340788B (en) * 2020-02-28 2024-03-29 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） Hardware Trojan horse layout detection method and device, electronic equipment and readable storage medium
CN111340788A (en) * 2020-02-28 2020-06-26 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） Hardware trojan layout detection method and device, electronic equipment and readable storage medium
CN111429408A (en) * 2020-03-11 2020-07-17 苏州杰锐思智能科技股份有限公司 Method for detecting gold wire of packaged chip
CN111462113A (en) * 2020-04-24 2020-07-28 上海精测半导体技术有限公司 Rechecking method of non-pattern wafer
CN111738247A (en) * 2020-05-15 2020-10-02 上海望友信息科技有限公司 Identification method and identification device of polarity identification, electronic equipment and storage medium
CN111707677A (en) * 2020-06-24 2020-09-25 江西景旺精密电路有限公司 PCB appearance detection method and device
CN111862076B (en) * 2020-07-30 2023-11-14 上海华力集成电路制造有限公司 Method for improving bright field defect detection precision and noise caused by chromatic aberration in process of bright field defect detection precision
CN111862076A (en) * 2020-07-30 2020-10-30 上海华力集成电路制造有限公司 Method for improving bright field defect detection precision and noise caused by chromatic aberration in bright field defect detection process
WO2022057389A1 (en) * 2020-09-18 2022-03-24 长鑫存储技术有限公司 Wafer alignment method and apparatus
CN112508846A (en) * 2020-10-30 2021-03-16 北京市商汤科技开发有限公司 Defect detection method and device, electronic equipment and storage medium
CN112288734A (en) * 2020-11-06 2021-01-29 西安工程大学 Printed fabric surface defect detection method based on image processing
CN112598632A (en) * 2020-12-16 2021-04-02 北京卫星制造厂有限公司 Appearance detection method and device for contact element of crimp connector
CN112819799A (en) * 2021-02-09 2021-05-18 上海众壹云计算科技有限公司 Target defect detection method, device, system, electronic equipment and storage medium
CN112819799B (en) * 2021-02-09 2024-05-28 上海众壹云计算科技有限公司 Target defect detection method, device, system, electronic equipment and storage medium
CN113269743A (en) * 2021-05-20 2021-08-17 北京理工大学重庆创新中心 Chip quantity detection method based on iterative translation verification
CN113486729A (en) * 2021-06-15 2021-10-08 北京道达天际科技有限公司 Unmanned aerial vehicle image feature point extraction method based on GPU
WO2023279558A1 (en) * 2021-07-09 2023-01-12 长鑫存储技术有限公司 Defect detection method and apparatus, device and storage medium
CN113658080A (en) * 2021-08-23 2021-11-16 宁波棱镜空间智能科技有限公司 Method and device for geometric correction of line-scanning cylinder based on feature point matching
CN113689397A (en) * 2021-08-23 2021-11-23 湖南视比特机器人有限公司 Workpiece circular hole feature detection method and workpiece circular hole feature detection device
CN113658080B (en) * 2021-08-23 2023-12-22 宁波棱镜空间智能科技有限公司 Linear scanning cylinder geometric correction method and device based on characteristic point matching
CN113441842A (en) * 2021-09-01 2021-09-28 佛山市联动科技股份有限公司 Wafer laser marking method
CN114526682B (en) * 2022-01-13 2023-03-21 华南理工大学 Deformation measurement method based on image feature enhanced digital volume image correlation method
CN114526682A (en) * 2022-01-13 2022-05-24 华南理工大学 Deformation measurement method based on image feature enhanced digital volume image correlation method
CN114700941A (en) * 2022-03-28 2022-07-05 中科合肥智慧农业协同创新研究院 Strawberry picking method based on binocular vision and robot system
CN114700941B (en) * 2022-03-28 2024-02-27 中科合肥智慧农业协同创新研究院 Strawberry picking method based on binocular vision and robot system
CN114419045A (en) * 2022-03-30 2022-04-29 武汉中导光电设备有限公司 Method, device and equipment for detecting defects of photoetching mask plate and readable storage medium
CN114757913A (en) * 2022-04-15 2022-07-15 电子科技大学 Display screen defect detection method
CN114926671A (en) * 2022-06-16 2022-08-19 福州大学 NLED/QLED/OLED pixel defect positioning method and system based on template matching
CN115205223B (en) * 2022-06-22 2023-03-14 锋睿领创(珠海)科技有限公司 Visual inspection method and device for transparent object, computer equipment and medium
CN115205223A (en) * 2022-06-22 2022-10-18 锋睿领创(珠海)科技有限公司 Visual detection method and device for transparent object, computer equipment and medium
CN115184380A (en) * 2022-08-01 2022-10-14 国网上海市电力公司 Printed circuit board welding spot abnormity detection method based on machine vision
CN115184380B (en) * 2022-08-01 2024-04-26 国网上海市电力公司 Method for detecting abnormity of welding spots of printed circuit board based on machine vision
CN115049713A (en) * 2022-08-11 2022-09-13 武汉中导光电设备有限公司 Image registration method, device, equipment and readable storage medium
CN115049713B (en) * 2022-08-11 2022-11-25 武汉中导光电设备有限公司 Image registration method, device, equipment and readable storage medium
CN116486126B (en) * 2023-06-25 2023-10-27 合肥联宝信息技术有限公司 Template determination method, device, equipment and storage medium
CN116486126A (en) * 2023-06-25 2023-07-25 合肥联宝信息技术有限公司 Template determination method, device, equipment and storage medium
CN116539638A (en) * 2023-07-04 2023-08-04 季华实验室 Doping concentration measuring method, device, electronic equipment and storage medium
CN117250208A (en) * 2023-11-20 2023-12-19 青岛天仁微纳科技有限责任公司 Machine vision-based nano-imprint wafer defect accurate detection system and method
CN117250208B (en) * 2023-11-20 2024-02-06 青岛天仁微纳科技有限责任公司 Machine vision-based nano-imprint wafer defect accurate detection system and method
CN117392518A (en) * 2023-12-13 2024-01-12 南京耀宇视芯科技有限公司 Low-power-consumption visual positioning and mapping chip and method thereof
CN117392518B (en) * 2023-12-13 2024-04-09 南京耀宇视芯科技有限公司 Low-power-consumption visual positioning and mapping chip and method thereof

Similar Documents

Publication Publication Date Title
CN106600600A (en) Wafer defect detection method based on characteristic matching
Li et al. Automatic pavement crack detection by multi-scale image fusion
CN103077512B (en) Based on the feature extracting and matching method of the digital picture that major component is analysed
Guo et al. Efficient center voting for object detection and 6D pose estimation in 3D point cloud
CN107230203B (en) Casting defect identification method based on human eye visual attention mechanism
CN109993094A (en) Fault in material intelligent checking system and method based on machine vision
CN104867137B (en) A kind of method for registering images based on improvement RANSAC algorithms
CN105184225B (en) A kind of multinational banknote image recognition methods and device
CN106557740B (en) The recognition methods of oil depot target in a kind of remote sensing images
CN110866430A (en) License plate recognition method and device
CN110473201A (en) A kind of automatic testing method and device of disc surface defect
CN108509950B (en) Railway contact net support number plate detection and identification method based on probability feature weighted fusion
Xie et al. Fabric defect detection method combing image pyramid and direction template
CN113240623B (en) Pavement disease detection method and device
CN107016394A (en) A kind of decussating fibers characteristic point matching method
CN108182705A (en) A kind of three-dimensional coordinate localization method based on machine vision
CN110443261A (en) A kind of more figure matching process restored based on low-rank tensor
CN114549446A (en) Cylinder sleeve defect mark detection method based on deep learning
Li et al. A two-stage crack detection method for concrete bridges using Convolutional Neural Networks
CN114648445B (en) Multi-view high-resolution point cloud splicing method based on feature point extraction and fine registration optimization
CN115937518A (en) Pavement disease identification method and system based on multi-source image fusion
CN116052110B (en) Intelligent positioning method and system for pavement marking defects
CN109829511B (en) Texture classification-based method for detecting cloud layer area in downward-looking infrared image
CN111882573A (en) Cultivated land plot extraction method and system based on high-resolution image data
CN110889418A (en) Gas contour identification method

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20170426