CN111862027A - Textile flaw detection method based on low-rank sparse matrix decomposition - Google Patents
Textile flaw detection method based on low-rank sparse matrix decomposition Download PDFInfo
- Publication number
- CN111862027A CN111862027A CN202010678742.5A CN202010678742A CN111862027A CN 111862027 A CN111862027 A CN 111862027A CN 202010678742 A CN202010678742 A CN 202010678742A CN 111862027 A CN111862027 A CN 111862027A
- Authority
- CN
- China
- Prior art keywords
- low
- rank
- textile
- matrix
- flaw
- 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.)
- Granted
Links
- 239000011159 matrix material Substances 0.000 title claims abstract description 98
- 239000004753 textile Substances 0.000 title claims abstract description 83
- 238000001514 detection method Methods 0.000 title claims abstract description 71
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 53
- 230000007547 defect Effects 0.000 claims abstract description 54
- 238000000034 method Methods 0.000 claims abstract description 19
- 230000011218 segmentation Effects 0.000 claims abstract description 9
- 238000000638 solvent extraction Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 17
- 238000009826 distribution Methods 0.000 claims description 9
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 3
- 230000000737 periodic effect Effects 0.000 abstract 1
- 239000004744 fabric Substances 0.000 description 9
- 238000012360 testing method Methods 0.000 description 3
- 230000003190 augmentative effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003416 augmentation Effects 0.000 description 1
- 238000005311 autocorrelation function Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
-
- 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
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- 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
- G06T2207/20024—Filtering details
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30124—Fabrics; Textile; Paper
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Treatment Of Fiber Materials (AREA)
Abstract
The invention relates to the technical field of textile detection, in particular to a textile flaw detection method based on low-rank sparse matrix decomposition, which comprises the following steps of firstly partitioning a periodic textile according to period information to obtain flaw prior for guiding a low-rank decomposition model, and then adding Laplace regularization to increase the distance between a background and a defect area; finally, the generated sparse matrix is segmented by adopting an optimal threshold segmentation algorithm to complete defect detection; the invention provides a textile detection method based on low-rank sparse matrix decomposition, which is characterized in that flaw priors are determined through a block segmentation method and added into a low-rank decomposition model to construct a weighted low-rank decomposition model, so that the detection precision of large flaw blocks is increased, the distance between a flaw area and a background is increased through a Laplace regularization term, and the detection precision and robustness of flaws are further improved.
Description
Technical Field
The invention relates to the technical field of textile detection, in particular to a textile flaw detection method based on low-rank sparse matrix decomposition.
Background
P textiles always develop various defects in their production process, and textile defects are one of the main factors affecting the quality of textiles. Therefore, flaw detection is an indispensable step in textile production. However, due to the complex and varied texture and the various defect types of the textile image, certain challenges are brought to the research of the defect detection algorithm.
Currently, textiles can be divided into two main categories: the first is a plain weave fabric (such as plain and twill) with a simple structure and no complex patterns; another type is a textile image having a periodically varying pattern, where one basic pattern is called a block or a period, and different types of fabrics (e.g., box and star patterns) have different period sizes and shapes.
For the first category of textiles, there are many mature algorithms that can be applied in practice, and they can be roughly classified into the following categories: 1) statistical methods including autocorrelation functions, mathematical morphology, morphological filters, etc.; 2) spectral methods including Gabor filtering, wavelet transform, fourier transform; 3) the training method comprises the following steps: including neural networks and the like; 4) model methods, including autoregressive models, Markov random field models, and the like. Among them, the statistical method and the spectrum method have a large false detection when detecting a large or small flaw.
The second category of textiles still presents certain challenges to the current detection due to the complexity of the texture elements. Recent detection algorithms include wavelet pre-processed golden image subtraction (WGIS), Brinell Band (BB), Ruled Band (RB), ER algorithm, TC algorithm, LSG algorithm, and the like. Although WGIS can detect large defects, it is ineffective for some small defects. For BB and RB, the upper and lower bands of the Bollinger band are sensitive to any slight variations in the input data, and thus are difficult to apply to practical industrial production, for example, 70 types of defects in the fabric. ER is a sports spirit based defect detection method, i.e. a fair competition between different image patches. And the TC divides the fabric into blocks according to the cycle size of the fabric, corrects the fabric image by using a block template, and finally extracts block texture characteristics for defect detection. LSG utilizes morphological analysis (MCA) to automatically segment the mesh and then converts the defect detection problem into a mesh similarity problem, although LSG achieves a good effect, its adaptability to different types of defects needs to be further improved.
A low rank decomposition (LR) model may decompose an original image into a low rank portion representing the image background and a sparse portion corresponding to the defect region. Patterned fabric images with complex texture elements have a high visual redundancy and defect areas appear prominent in the fabric background. In consideration of the characteristics, the low-rank decomposition model is more suitable for detecting the textile flaws.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: in order to solve the problems that some large uniform flaw blocks cannot be detected in the prior art, and when the background of the textile and the flaw area are relatively consistent or the texture of the textile image is complex, the conventional method is difficult to separate the large uniform flaw blocks from the flaw area, a textile detection method based on low-rank sparse matrix decomposition is provided.
The technical scheme adopted by the invention for solving the technical problems is as follows: a textile flaw detection method based on low-rank sparse matrix decomposition comprises the following steps:
s1: inputting a flawless textile image containing a periodically changing pattern;
s2: determining the size of a pattern period template in the textile image, and partitioning the pure textile image according to the size of the pattern period template to obtain a plurality of training feature blocks; specifically, the size of each feature block is the same as the size of the pattern period template;
S3: and extracting Gabor characteristics of each training characteristic block, calculating the Chebyshev distance between the training characteristic blocks, and constructing a characteristic distance matrix.
S4: calculating the average Chebyshev distance d between the training characteristic blocks1;
S5: inputting a textile image to be detected, and partitioning the textile image to be detected according to the size of the pattern period template to obtain a plurality of detection characteristic blocks; extracting Gabor characteristics of each detection characteristic block, calculating the Chebyshev distance between the detection characteristic blocks, and constructing a characteristic distance matrix; calculating the average Chebyshev distance d between the detection feature blocks2(ii) a Wherein when d2>d1Marking the textile image to be detected as a flaw block, or marking the textile image to be detected as a pure block, and obtaining a flaw priori as a result; namely, the flaw prior is divided into a training stage and a testing stage, steps S1-S4 are the training stage to obtain the average chebyshev distance d1 between the training feature blocks, S5 is the testing stage to obtain the average chebyshev distance d2 between the testing feature blocks, and the distance between the feature blocks is used for representing the similarity between the feature blocks;
s6: extracting Gabor characteristics of each detection characteristic block in the textile image to be detected to form a characteristic matrix F;
S7: the flaw priori is subjected to dimension conversion and then is placed into a weighted low-rank decomposition and Laplace regularization model for low-rank decomposition;
s8: and performing optimal threshold segmentation on the sparse matrix obtained by low-rank decomposition to obtain a final detection result.
Further, in step 6, the method for constructing the feature matrix F is to preprocess the textile image to be detected by using a Gabor filter to generate Gabor features, and then segment the textile image to be detected into N detection feature blocks by using a Simple Linear Iterative Clustering (SLIC) algorithm, where P is { P ═ P1,P2,…,PN}, each detection feature block PiBy feature vectors fiIndicating that the textile image to be detected can be represented as a feature matrix F ═ F1,f2,…,fNTherein ofM is the dimension of the detected feature block,is a real number domain.
Further, in step 7, the method for constructing the weighted low rank decomposition and laplacian regularization term model is to put the obtained flaw priors into an initial low rank decomposition model through dimension conversion for guiding the decomposition of the weighted low rank decomposition model, increase the detection rate of large homogeneous flaw blocks, improve the detection accuracy of the model, construct the weighted low rank decomposition model as,wherein W represents flaw prior, F represents a characteristic matrix of the textile, B represents a low-rank matrix and represents a background area of the textile image to be detected, and S represents a sparse matrix and represents a flaw area.
When the difference between the background and the defects is not obvious or the texture of the textile images is complex, the correlation between the background and the defects is strong, so when the situation is faced, the detection accuracy of the method based on the weighted low-rank decomposition model is not ideal, and further, in step 7, after the weighted low-rank decomposition model is constructed, laplacian regularization is introduced to enlarge the distance between the background and the defects so as to distinguish the defect regions in the background, and the weighted low-rank decomposition and laplacian regularization term model is defined as:beta is the Lagrangian multiplier and theta is the Laplacian regulare.
In order to facilitate the solution, further, in step 7, after defining the low rank decomposition of the weight and the laplacian regularization term model, an auxiliary variable H is introduced to separate the objective function, and the low rank of the weightThe decomposition and Laplace regularization term model is described asWherein W represents flaw prior, F represents a characteristic matrix of the textile, B represents a low-rank matrix representing a background area of the textile image to be detected, S represents a sparse matrix representing a flaw area, and M represents a defect areaFFor a laplacian matrix, Tr () represents the trace operation of the matrix.
Further, in step S8, after the feature matrix F is decomposed into a low rank matrix B and a sparse matrix S, l of each column in the sparse matrix is obtained 1The norm represents the significance of each detected feature block, and the greater the significance is, the greater the probability of containing a flaw is, thereby generating a flaw distribution map S; and denoising the flaw distribution map, converting the flaw distribution map into a gray image G, and finally segmenting the gray image G by using an optimal threshold segmentation algorithm to obtain a final binarization detection result.
The invention has the beneficial effects that: according to the textile detection method based on low-rank sparse matrix decomposition, the defect prior is determined through a block segmentation method, the defect prior is added into a low-rank decomposition model, a weighted low-rank decomposition model is constructed, the detection precision of a large defect block is improved, the distance between a defect area and a background is increased through a Laplace regularization term, the detection precision and robustness of the defect are improved, and therefore the detection of the large defect block and the detection of textiles with the background and the defect area relatively consistent or the image texture is complex are achieved.
Drawings
The invention is further illustrated with reference to the following figures and examples.
FIG. 1 is a block flow diagram of the detection method of the present invention;
FIG. 2 is a defect prior map of the present invention;
FIG. 3 is a low rank decomposition diagram of the present invention;
FIG. 4 is a graph showing the results of the detection according to the present invention;
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Unless defined otherwise, technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention belongs. Meanwhile, in order to make the present specification clearer and more concise, a detailed definitional description of functions and constructions well known in the art will be omitted.
Example 1
As shown in fig. 1, a textile defect detection method based on low-rank sparse matrix decomposition includes the following steps:
s1: inputting a flawless textile image containing a periodically changing pattern;
s2: determining the size of a pattern period template in the textile image, and partitioning the pure textile image according to the size of the pattern period template to obtain a plurality of training feature blocks; specifically, the size of each feature block is the same as the size of the pattern period template;
s3: extracting Gabor characteristics of each training characteristic block, calculating the Chebyshev distance between the training characteristic blocks, and constructing a characteristic distance matrix;
in this step, a Gabor filter is used to extract the Gabor feature of each training feature block, and a Gabor filter bank defining a two-dimensional Gabor transform function is as follows:
Wherein the parameter σxAnd σyA shape factor representing a Gaussian surface; θ represents a direction; g0Represents the center frequency; the parameters θ are defined here as 0, π/4, π/2 and 3 π/4; x and y are coordinates.
S4: calculating the average Chebyshev distance d between the training characteristic blocks1;
S5: inputting a textile image to be detected, and partitioning the textile image to be detected according to the size of the pattern period template to obtain a plurality of detection characteristic blocks; extracting Gabor characteristics of each detection characteristic block, calculating the Chebyshev distance between the detection characteristic blocks, and constructing a characteristic distance matrix;
in this step, a Gabor filter is used to extract Gabor features of each detection feature block, and a Gabor filter bank defining a two-dimensional Gabor transform function is as follows:
calculating the average Chebyshev distance d between the detection feature blocks2;
Wherein when d2>d1When the textile image to be detected is marked as a defective block, otherwise, the textile image to be detected is marked as a blank block, the obtained result is a defect prior, and the defect prior is shown in figure 2;
s6: extracting Gabor characteristics of each detection characteristic block in the textile image to be detected to form a characteristic matrix F;
preprocessing a textile image to be detected by using a Gabor filter to generate Gabor characteristics, and then segmenting the textile image to be detected into N detection characteristic blocks by using a Simple Linear Iterative Clustering (SLIC) algorithm, wherein the N detection characteristic blocks are expressed as P ═ P { (P }) 1,P2,…,PN}, each detection feature block PiBy feature vectors fiIndicating that the textile image to be detected can be represented as a feature matrix F ═ F1,f2,…,fNTherein ofM is the dimension of the detected feature block,is a real number domain;
s7: the flaw prior is subjected to dimension conversion and then is put into a weighted low-rank decomposition and Laplace regularization model for low-rank decomposition, and the decomposition process is shown in figure 3;
the method comprises the following specific steps:
the initial low rank decomposition model is:
f represents a characteristic matrix of the textile, B represents a low-rank matrix and represents a background region of the textile image to be detected, S represents a sparse matrix and represents a flaw region, and lambda is a Lagrange multiplier;
in order to increase the detection precision of large defect blocks and improve the detection precision of the model, defect prior is added, and a weighted low-rank decomposition model is constructed as follows:
wherein W represents flaw prior, F represents a characteristic matrix of the textile, B represents a low-rank matrix and represents a background area of the textile image to be detected, S represents a sparse matrix and represents a flaw area, and lambda is a Lagrange multiplier;
the lagrange function of the augmentation is:
mu is a Lagrange multiplier, W represents flaw prior, F represents a characteristic matrix of the textile, B represents a low-rank matrix and represents a background area of the textile image to be detected, S represents a sparse matrix and represents a flaw area, lambda is the Lagrange multiplier, Tr represents a trace of the matrix, | | | | | | represents a norm, | | | | | Y-pass 1Represents 1 norm, | | | purpleFRepresents the F norm;
the correlation between the background and the defects is strong when the difference between the background and the defects is not significant or the texture of the fabric image is complex. Therefore, in the face of this situation, the detection accuracy of the method based on the weighted low rank decomposition model is not ideal. To solve this problem, laplacian regularization is introduced to enlarge the distance between the background and the defect in order to distinguish the flaw area in the background, which is defined as follows:
wherein N represents the number of region blocks, beta is a Lagrangian multiplier, theta is a Laplacian regular operator, S represents a sparse matrix and represents a defect region, and S represents a defect regioniRepresents the ith block sparse matrix, ai,jElements representing affinity matrix A, MFIs Laplace matrix, Tr represents the trace of the matrix;
affinity matrix a is defined as follows:
wherein, ai,jElements representing an affinity matrix A, IiDenotes the ith block image region, fiRepresenting the average gray scale of the ith block image area, sigma representing the variance of the whole image, and V representing the set of adjacent block pairs on the image;
laplace matrix MFThe (i, j) th item of (a) is defined as:
on the basis of a weighted low-rank decomposition model, Laplace regularization is added to construct a weighted low-rank decomposition and Laplace regularization term model, and the model is defined as follows:
The method comprises the following steps that beta is a Lagrange multiplier, theta is a Laplace regularization operator, W represents a flaw priori, F represents a characteristic matrix of a textile, B represents a low-rank matrix and represents a background area of a textile image to be detected, and S represents a sparse matrix and represents a flaw area;
to facilitate the solution, an auxiliary variable H is introduced to separate the objective function, and the low rank decomposition of the weights and the laplacian regularization term can be described as:
s.t.F=B+S,S=H
beta is a Lagrange multiplier, B represents a low-rank matrix and represents a background area of a to-be-detected textile image, W represents defect prior, F represents a characteristic matrix of the textile, S represents a sparse matrix and represents a defect area, and M represents the defect areaFIs Laplace matrix, Tr () represents the trace operation of solving the matrix;
currently, there are several mainstream algorithms for solving low rank decomposition, such as the accelerated near-end gradient (APG), the Augmented Lagrange Multiplier (ALM), and the Alternating Direction Method (ADM). In consideration of the efficiency and precision of the algorithm, the ADM method is adopted to solve the convex optimization model low-rank sparse matrix decomposition model, and the minimum augmented Lagrange function is as follows:
wherein, Y1And Y2The method comprises the steps of taking a Lagrange multiplier matrix as the matrix, taking lambda, mu and beta as Lagrange multipliers, representing a low-rank matrix by B, representing a background area of a to-be-detected textile image, representing defect prior by W, representing a characteristic matrix of the textile, representing a sparse matrix by S, representing a defect area, and M FThe matrix is a Laplace matrix, Tr represents a trace of the matrix, and H is an auxiliary variable matrix;
since the above problem is unconstrained, other variables (e.g., S, H) can be fixed, the optimal values of B, S, H can be iteratively searched respectively, and then the lagrange multiplier is updated;
s8: performing optimal threshold segmentation on the sparse matrix obtained by low-rank decomposition to obtain a final detection result;
decomposing the characteristic matrix F into a low-rank matrix B and a sparse matrix S, and solving each sparse matrix in the sparse matrixColumn of1Norm to represent the saliency of each detected feature block:
Sal(Ii)=||Si||1
wherein S isiRepresenting the ith sparse matrix, IiA saliency map corresponding to the ith sparse matrix is represented;
if | | | Si||1The larger the defect distribution map is, the ith sparse matrix most possibly contains defects, and the defect distribution map generated by the sparse matrix S is subjected to noise reduction to obtain a new defect distribution
G is a convolution template, represents convolution operation, and o represents dot product operation;
and finally, segmenting the G by using an optimal threshold segmentation algorithm to obtain a final binarization detection result, wherein the detection result is shown in figure 4.
In light of the foregoing description of the preferred embodiment of the present invention, it is to be understood that numerous changes and modifications may be made without departing from the spirit and scope of the invention as defined by the appended claims. The technical scope of the present invention is not limited to the content of the specification, and must be determined according to the scope of the claims.
Claims (6)
1. A textile flaw detection method based on low-rank sparse matrix decomposition is characterized by comprising the following steps: the method comprises the following steps:
s1: inputting a flawless textile image containing a periodically changing pattern;
s2: determining the size of a pattern period template in the textile image, and partitioning the pure textile image according to the size of the pattern period template to obtain a plurality of training feature blocks;
s3: extracting Gabor characteristics of each training characteristic block, calculating the Chebyshev distance between the training characteristic blocks, and constructing a characteristic distance matrix;
s4: calculating the average Chebyshev distance d between the training characteristic blocks1;
S5: inputting a textile image to be detected, and partitioning the textile image to be detected according to the size of the pattern period template to obtain a plurality of detection characteristic blocks; extracting Gabor characteristics of each detection characteristic block, calculating the Chebyshev distance between the detection characteristic blocks, and constructing a characteristic distance matrix; calculating the average Chebyshev distance d between the detection feature blocks2(ii) a Wherein when d2>d1Marking the textile image to be detected as a flaw block, or marking the textile image to be detected as a pure block, and obtaining a flaw priori as a result;
s6: extracting Gabor characteristics of each detection characteristic block in the textile image to be detected to form a characteristic matrix F;
S7: the flaw priori is subjected to dimension conversion and then is placed into a weighted low-rank decomposition and Laplace regularization model for low-rank decomposition;
s8: and performing optimal threshold segmentation on the sparse matrix obtained by low-rank decomposition to obtain a final detection result.
2. The textile defect detection method based on low-rank sparse matrix factorization of claim 1, characterized in that: in step 6, the method for constructing the feature matrix F is to preprocess the textile image to be detected by using a Gabor filter to generate Gabor features, and then divide the textile image to be detected into N detection feature blocks by using a Simple Linear Iterative Clustering (SLIC) algorithm, wherein the detection feature blocks are expressed as P ═ { P ═ P { (P)1,P2,…,PNEach detection bitSign block PiBy feature vectors fiIndicating that the textile image to be detected can be represented as a feature matrix F ═ F1,f2,…,fNTherein ofM is the dimension of the detected feature block,is a real number domain.
3. The textile defect detection method based on low-rank sparse matrix factorization of claim 2, characterized by: in step 7, the method for constructing the low-rank decomposition of the weight and Laplace regularization term model is to put the acquired flaw prior into an initial low-rank decomposition model through dimension conversion, construct the low-rank decomposition model of the weight as, B represents a low-rank matrix and represents a background area of the textile image to be detected, W represents a flaw prior, S represents a sparse matrix and represents a flaw area, F represents a characteristic matrix of the textile image to be detected, and lambda is a Lagrange multiplier.
4. The textile defect detection method based on low-rank sparse matrix factorization of claim 3, wherein: in step 7, after constructing the weighted low-rank decomposition model, introducing laplacian regularization to enlarge the distance between the background and the defect, and defining the weighted low-rank decomposition and laplacian regularization term model as follows:beta is the Lagrangian multiplier and theta is the Laplacian regulare.
5. The textile defect detection method based on low-rank sparse matrix factorization of claim 3, characterized in thatIn the following steps: in step 7, after defining the weighted low-rank decomposition and Laplace regularization term model, introducing an auxiliary variable H to separate a target function, wherein the weighted low-rank decomposition and Laplace regularization term model are described asWherein M isFFor a laplacian matrix, Tr () represents the trace operation of the matrix.
6. The textile defect detection method based on low rank sparse matrix factorization of any of claims 3-5, wherein: in step S8, after the feature matrix F is decomposed into a low-rank matrix B and a sparse matrix S, l of each column in the sparse matrix is obtained 1The norm represents the significance of each detected feature block, and the greater the significance is, the greater the probability of containing a flaw is, thereby generating a flaw distribution map S; and denoising the flaw distribution map, converting the flaw distribution map into a gray image G, and finally segmenting the gray image G by using an optimal threshold segmentation algorithm to obtain a final binarization detection result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010678742.5A CN111862027B (en) | 2020-07-15 | 2020-07-15 | Textile flaw detection method based on low-rank sparse matrix decomposition |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010678742.5A CN111862027B (en) | 2020-07-15 | 2020-07-15 | Textile flaw detection method based on low-rank sparse matrix decomposition |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111862027A true CN111862027A (en) | 2020-10-30 |
CN111862027B CN111862027B (en) | 2024-02-20 |
Family
ID=72983458
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010678742.5A Active CN111862027B (en) | 2020-07-15 | 2020-07-15 | Textile flaw detection method based on low-rank sparse matrix decomposition |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111862027B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112308882A (en) * | 2020-11-10 | 2021-02-02 | 苏州兴钊防务研究院有限公司 | Video moving object detection method and device based on structured decomposition |
CN113033318A (en) * | 2021-03-01 | 2021-06-25 | 深圳大学 | Human body action detection method and device and computer readable storage medium |
CN116109640A (en) * | 2023-04-13 | 2023-05-12 | 常州微亿智造科技有限公司 | Workpiece surface small defect detection method in industrial detection |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101556603B1 (en) * | 2014-12-30 | 2015-10-01 | 연세대학교 산학협력단 | Apparatus and Method for Image Seperation using Rank Prior |
CN107358245A (en) * | 2017-07-19 | 2017-11-17 | 安徽大学 | Method for detecting image collaborative salient region |
CN107392921A (en) * | 2017-07-14 | 2017-11-24 | 西安邮电大学 | A kind of semi-supervised multi-object clustering image partition method based on Chebyshev's distance |
CN107705306A (en) * | 2017-10-26 | 2018-02-16 | 中原工学院 | A kind of fabric defect detection method based on multiple features matrix low rank decomposition |
CN109712113A (en) * | 2018-11-28 | 2019-05-03 | 中原工学院 | A kind of fabric defect detection method based on cascade low-rank decomposition |
US20190209035A1 (en) * | 2018-01-11 | 2019-07-11 | Zhejiang University | Method for noninvasive imaging of cardiac electrophysiological based on low rank and sparse constraints |
CN110188770A (en) * | 2019-05-17 | 2019-08-30 | 重庆邮电大学 | A kind of non-convex low-rank well-marked target detection method decomposed based on structure matrix |
CN111046868A (en) * | 2019-11-26 | 2020-04-21 | 广东石油化工学院 | Target significance detection method based on matrix low-rank sparse decomposition |
-
2020
- 2020-07-15 CN CN202010678742.5A patent/CN111862027B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101556603B1 (en) * | 2014-12-30 | 2015-10-01 | 연세대학교 산학협력단 | Apparatus and Method for Image Seperation using Rank Prior |
CN107392921A (en) * | 2017-07-14 | 2017-11-24 | 西安邮电大学 | A kind of semi-supervised multi-object clustering image partition method based on Chebyshev's distance |
CN107358245A (en) * | 2017-07-19 | 2017-11-17 | 安徽大学 | Method for detecting image collaborative salient region |
CN107705306A (en) * | 2017-10-26 | 2018-02-16 | 中原工学院 | A kind of fabric defect detection method based on multiple features matrix low rank decomposition |
US20190209035A1 (en) * | 2018-01-11 | 2019-07-11 | Zhejiang University | Method for noninvasive imaging of cardiac electrophysiological based on low rank and sparse constraints |
CN109712113A (en) * | 2018-11-28 | 2019-05-03 | 中原工学院 | A kind of fabric defect detection method based on cascade low-rank decomposition |
CN110188770A (en) * | 2019-05-17 | 2019-08-30 | 重庆邮电大学 | A kind of non-convex low-rank well-marked target detection method decomposed based on structure matrix |
CN111046868A (en) * | 2019-11-26 | 2020-04-21 | 广东石油化工学院 | Target significance detection method based on matrix low-rank sparse decomposition |
Non-Patent Citations (6)
Title |
---|
JUNJIE CAO: "Fabric defect inspection using prior knowledge guided least squares regression", MULTIMED TOOLS APPL, pages 4141 - 4155 * |
刘威;常兴治;梁久祯;贾靓;顾程熙;: "基于局部最优分析的纺织品瑕疵检测方法", 模式识别与人工智能, no. 02, 15 February 2018 (2018-02-15) * |
杨恩君: "基于低秩分解的织物疵点检测", 纺织学报, vol. 41, no. 5 * |
纪旋;梁久祯;侯振杰;常兴治;刘威;: "基于模板校正与低秩分解的纺织品瑕疵检测方法", 模式识别与人工智能, no. 03, 15 March 2019 (2019-03-15) * |
马晓迪: "基于迹表示和正则化的显著目标检测算法", 计算机辅助设计与图形学学报, vol. 30, no. 11, pages 2018 - 2025 * |
高广帅: "基于低秩稀疏矩阵分解的织物疵点检测算法研", 中国优秀硕士学位论文全文数据库工程科技Ⅰ辑, no. 7, pages 59 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112308882A (en) * | 2020-11-10 | 2021-02-02 | 苏州兴钊防务研究院有限公司 | Video moving object detection method and device based on structured decomposition |
CN113033318A (en) * | 2021-03-01 | 2021-06-25 | 深圳大学 | Human body action detection method and device and computer readable storage medium |
CN113033318B (en) * | 2021-03-01 | 2023-09-26 | 深圳大学 | Human body motion detection method, device and computer readable storage medium |
CN116109640A (en) * | 2023-04-13 | 2023-05-12 | 常州微亿智造科技有限公司 | Workpiece surface small defect detection method in industrial detection |
CN116109640B (en) * | 2023-04-13 | 2023-08-18 | 常州微亿智造科技有限公司 | Workpiece surface small defect detection method in industrial detection |
Also Published As
Publication number | Publication date |
---|---|
CN111862027B (en) | 2024-02-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | Deformable patterned fabric defect detection with fisher criterion-based deep learning | |
CN107705306B (en) | Fabric defect detection method based on multi-feature matrix low-rank decomposition | |
CN113240626B (en) | Glass cover plate concave-convex type flaw detection and classification method based on neural network | |
CN112070727B (en) | Metal surface defect detection method based on machine learning | |
CN108932518B (en) | Shoe print image feature extraction and retrieval method based on visual bag-of-words model | |
CN105844278A (en) | Multi-feature fused fabric scanning pattern recognition method | |
CN111862027B (en) | Textile flaw detection method based on low-rank sparse matrix decomposition | |
Valliammal et al. | A novel approach for plant leaf image segmentation using fuzzy clustering | |
CN111091134A (en) | Method for identifying tissue structure of colored woven fabric based on multi-feature fusion | |
CN105809173A (en) | Bionic vision transformation-based image RSTN (rotation, scaling, translation and noise) invariant attributive feature extraction and recognition method | |
Liang et al. | Automatic defect detection of texture surface with an efficient texture removal network | |
Liu et al. | A novel color-texture descriptor based on local histograms for image segmentation | |
CN111563577B (en) | Unet-based intrinsic image decomposition method for skip layer frequency division and multi-scale identification | |
Sulistianingsih et al. | Classification of batik image using grey level co-occurrence matrix feature extraction and correlation based feature selection | |
CN115147341A (en) | Fabric surface defect classification depth network method based on biological vision inspiration | |
Dong et al. | Fusing multilevel deep features for fabric defect detection based NTV-RPCA | |
CN107895371A (en) | Textile flaw detection method based on peak value coverage values and Gabor characteristic | |
CN113838040A (en) | Detection method for defect area of color texture fabric | |
CN116842210B (en) | Textile printing texture intelligent retrieval method based on texture features | |
CN101630369B (en) | Pedestrian detection method based on wavelet fractal characteristic | |
CN107945164A (en) | Textile flaw detection method based on peak threshold, rotational alignment and composite character | |
Yu et al. | Research on CNN Algorithm for Monochromatic Fabric Defect Detection | |
Kang et al. | Automatic classification of woven fabric structure based on computer vision techniques | |
Chavan et al. | Synthetic image generation for visual particle size distribution estimation based on U-NET convolutional networks | |
Zhang et al. | Deep Neural Network with Strip Pooling for Image Classification of Yarn-Dyed Plaid Fabrics. |
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 |