CN107341499A - It is a kind of based on non-formaldehyde finishing and ELM fabric defect detection and sorting technique - Google Patents
It is a kind of based on non-formaldehyde finishing and ELM fabric defect detection and sorting technique Download PDFInfo
- Publication number
- CN107341499A CN107341499A CN201710381898.5A CN201710381898A CN107341499A CN 107341499 A CN107341499 A CN 107341499A CN 201710381898 A CN201710381898 A CN 201710381898A CN 107341499 A CN107341499 A CN 107341499A
- Authority
- CN
- China
- Prior art keywords
- fabric defect
- image
- fabric
- defect image
- elm
- 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
- 230000007547 defect Effects 0.000 title claims abstract description 188
- 239000004744 fabric Substances 0.000 title claims abstract description 168
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000001514 detection method Methods 0.000 title claims abstract description 22
- WSFSSNUMVMOOMR-UHFFFAOYSA-N Formaldehyde Chemical compound O=C WSFSSNUMVMOOMR-UHFFFAOYSA-N 0.000 title abstract 6
- 238000012549 training Methods 0.000 claims abstract description 40
- 230000011218 segmentation Effects 0.000 claims abstract description 38
- 238000000605 extraction Methods 0.000 claims abstract description 16
- 239000013598 vector Substances 0.000 claims description 15
- 230000004927 fusion Effects 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 11
- 230000002159 abnormal effect Effects 0.000 claims description 7
- 238000005516 engineering process Methods 0.000 claims description 6
- 238000003709 image segmentation Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 4
- 230000035945 sensitivity Effects 0.000 claims description 4
- 230000000877 morphologic effect Effects 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 238000003909 pattern recognition Methods 0.000 abstract description 2
- 230000009395 genetic defect Effects 0.000 abstract 1
- 238000000354 decomposition reaction Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000002950 deficient Effects 0.000 description 2
- 238000010191 image analysis Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 239000004753 textile Substances 0.000 description 2
- 230000037303 wrinkles Effects 0.000 description 2
- 241000218220 Ulmaceae Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000003703 image analysis method Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 210000004205 output neuron Anatomy 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- BJORNXNYWNIWEY-UHFFFAOYSA-N tetrahydrozoline hydrochloride Chemical compound Cl.N1CCN=C1C1C2=CC=CC=C2CCC1 BJORNXNYWNIWEY-UHFFFAOYSA-N 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000011179 visual inspection Methods 0.000 description 1
- 238000009941 weaving Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of based on the detection of non-formaldehyde finishing and ELM fabric defect and sorting technique, belong to computer vision, pattern-recognition and image application field.Fabric defect image is inputted first, realizes that fabric defect image is split, the fabric defect image after being split；Secondly, the shape of fabric defect image and genetic defects image, textural characteristics after extraction segmentation, realize the feature extraction of fabric defect image；Then using fabric defect set of image characteristics to be sorted and label as training set, ELM graders is trained, obtain ELM classifier parameters；Finally, by the grader after training, merged according to Bayesian probability, realize fabric defect image classification, export fabric defect image classification result.The detection and sorting technique that the present invention uses have higher accuracy rate.
Description
Technical Field
The invention relates to a fabric defect detection and classification method based on unsupervised segmentation and ELM (Extreme Learning Machines, ELM for short), and belongs to the field of computer vision, pattern recognition and image application.
Background
In modern textile production, fabrics suffer from a number of defects, such as hole defects, oil spots, lack of warp, lack of weft, missed stitches, creases, scratches, etc. The defects of the fabric are caused by the faults of a weaving machine or unqualified yarn quality, and therefore, the detection and classification of the defects of the fabric are a key link for controlling the quality of the textile. The traditional fabric defect detection and classification are mainly finished through manual visual inspection, and the detection and classification result depends on subjective judgment of a detector. On the one hand, much work is very tedious; on the other hand, the long observation causes fatigue of the detecting personnel, thereby causing the fabric defect detection and classification errors. Therefore, an intelligent fabric defect detection and classification method becomes an effective method for improving the quality of the fabric.
The precondition requirement of fabric defect classification is fabric defect detection, and the existing fabric defect detection method is mainly realized by combining image analysis with a threshold segmentation method. For example(<Development of a machine visionsystem:real-time fabric defect detection and classification with neuralnetworks>,2014, 105(6): 575-585) proposes a frequency domain method combined with a dual threshold value method and a morphology method to realize the fabric defect detection. Lucia Bissi (<Automated defect detection in uniform and structuredfabrics using Gabor filters and PCA>,2013, 24(7): 838 and 845) establishing a non-defective fabric texture model through a wavelet domain hidden markov random field, obtaining model parameters through training a non-defective image, determining whether the fabric is defective or not through a maximum expectation thought, and detecting the defective fabric image through a threshold value method in a segmentation way. Jing J (<Supervised defectdetection on textile fabrics via optimal Gabor filter>,2013, 44(1): 40-57) proposed a supervised fabric defect detection method using Gabor filters of genetic algorithms to adjust matching to defect free fabric texture information. And then detecting the fabric defect image based on the adjusted optimal Gabor filter, and finally segmenting the fabric defect image by a threshold value method. At present, the detection methods are supervised, and a large number of images of the non-defective fabric are required for auxiliary detection, so that the overall efficiency of the system is not improved. The fabric defect detection method is based on unsupervised fabric defect segmentation of local patch approximation. Compared with the known method, the method avoids the problem of needing a large number of fabric images and has better segmentation effect.
In the aspects of defect classification and feature extraction, the known feature extraction method is mainly used for extracting texture attributes of fabric defects based on a frequency domain method, and the feature attributes are single and are not beneficial to improving the classification accuracy. The feature extraction method does not need image preprocessing operation, extracts texture features by a wavelet packet decomposition technology meeting the requirement of multi-resolution, extracts shape features after image segmentation by adopting a Hu invariant moment method, and improves the accuracy of classification of subsequent fabric defects. In classifier selection, the known method is based on a BP neural network and an SVM classifier. Kuo C F J (< Image analysis of finished fabrics using the fabric packets >, (2015)) utilizes an NN classifier to achieve fabric defect classification. Li W (< Yarn-driven Yarn defect classification and classification using combined defect and defect vector machine >, 2014, 105 (2): 163-. The classification results of the methods show that the classification accuracy is not ideal, and the training process of the classifier is complicated. Therefore, a method for improving the classification accuracy of fabric defects in both the aspect of feature extraction and classification is needed.
Disclosure of Invention
The invention provides a fabric defect detection and classification method based on unsupervised segmentation and ELM, which is used for effectively detecting and classifying fabric defect images so as to meet the requirement of fabric quality control in the current industrial production.
The technical scheme of the invention is as follows: firstly, inputting a fabric defect image, realizing fabric defect image segmentation, and obtaining a segmented fabric defect image; secondly, extracting the shape and texture characteristics of the segmented fabric defect image and the original defect image to realize the characteristic extraction of the fabric defect image; then, taking the defect image feature set and the labels of the fabric to be classified as training sets, and training an ELM classifier to obtain parameters of the ELM classifier; and finally, classifying the fabric defect images through the trained classifier according to Bayesian probability fusion, and outputting the classification result of the fabric defect images.
The method comprises the following specific steps:
step1, inputting fabric defect image G'By extracting the image G'Patch { xiSet up data matrix X ═ X1,x2,…,xi,…xn]，xi∈RwWherein w represents the image G'Patch xiDimension, n denotes image G'Patch xiTotal number; based on the Euclidean distance E (i) | | xi-x||2(x is { x)iMean value of) eliminating outlier of data matrix X to obtain training data Xc＝[x1,x2,…,xi,…,xc]I is more than or equal to 1 and less than c and less than n, wherein c represents training data XcMiddle image G'Patch xiTotal number; finding a dictionary D, representing the training data X with the least square errorcEach point of (1) is represented by(aiIs each x of dimension giCoefficient vector of) is passed throughThe second iteration is solved, and a dictionary D ═ D after learning is generated1,d2,…,di,…,dg]，di∈Rw(i ≦ g < n), where g represents the total number of elements in the post-learning dictionary D; then, the original patch x is computedψAnd learning dictionary based D approximation patchPixel difference of(gamma is a user-defined parameter used to control segmentation sensitivity,weight terms used to reduce the influence of those pixels that are far from the center pixel ψ) to construct an anomaly map; finally, using the 2D maximum entropy andmorphologic operation is carried out on the defect area in the segmentation abnormal image to obtain a segmented fabric defect image G″；
Step2, extracting the fabric defect image G after segmentation by adopting Hu invariant moment″A shape feature; fabric defect image G decomposed by wavelet packet based on optimal wavelet packet technology'Obtaining fabric defect image G'Calculating Shannon entropy to extract texture features of the fabric defect image G';
step3, collecting G (G) of fabric defect images to be classified1,G2,...GzProcessing by a Step1 and a Step2 method to obtain a shape and texture feature set and a label of the fabric defect, and inputting the shape and texture feature set and the label into two ELM-OAA classifiers for classification training;
step4, fabric defect image G based on extraction'Texture feature of (3) and fabric defect image G after segmentation″The trained ELM-OAA classifier and Bayesian probability fusion are adopted to realize the fabric defect image G'Classifying, outputting fabric defect image G'Classifying results;
for marking a defect image G of a non-label fabric'From fabric defect images G'Selecting a single feature from shape and texture features Kt (Kt 2)Respectively input into each trained ELM-OAA classifier, and marking fabric defect categories by Bayesian probability fusion to finally realize a fabric defect image G'And (6) classifying.
The invention has the beneficial effects that:
1. the known image analysis method for detecting the fabric defects is mainly supervised, needs a large number of non-defective fabric images for auxiliary detection, is not beneficial to improving the overall efficiency of the system, and has limitation. The invention is unsupervised fabric defect segmentation based on local patch approximation, avoids the problem that a large number of non-defective fabric images are needed, and has better segmentation effect.
2. Most of the known feature extraction methods are based on texture attributes of fabric defects, but the feature attributes are relatively single and are not beneficial to the accuracy of classification. The feature extraction method does not need image preprocessing operation, extracts texture features by a wavelet packet decomposition technology meeting the requirement of multi-resolution, extracts shape features after image segmentation by adopting a Hu invariant moment method, and improves the accuracy of classification of subsequent fabric defects.
3. Most of the known fabric defect classification methods are based on BP neural networks and SVM classifiers, only one characteristic attribute is trained, and the training process is complicated. In the method, an ELM classifier which is simple and time-saving in training is selected, and two different types of features are trained by adopting an OAA decomposition method: and the texture features and the shape features are finally combined with two feature attribute classification results by adopting Bayesian probability, so that the accuracy is high.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is an exemplary diagram of an unsupervised segmentation process of a fabric defect image according to the present invention;
FIG. 3 is a Gaussian symmetric function surface plot of a defect image of a fabric according to the present invention;
FIG. 4 is a graph of the fabric defect image segmentation result of the present invention;
FIG. 5 is a PR graph showing the segmentation result of a fabric defect image test sample according to the present invention;
FIG. 6 is an exemplary diagram of a defect image classification process of a fabric according to the present invention;
FIG. 7 is an exploded view of a secondary wavelet packet of a portion of a fabric defect image in accordance with the present invention;
FIG. 8 is a schematic diagram of ELM-OAA probability fusion in accordance with the present invention;
FIG. 9 is a comparison chart of classification results of different types of fabric defects by different methods.
Detailed Description
The invention is further described with reference to the following figures and specific examples.
Example 1: as shown in fig. 1-8, a fabric defect detection and classification method based on unsupervised segmentation and ELM. Firstly, inputting a fabric defect image, realizing fabric defect image segmentation, and obtaining a segmented fabric defect image; secondly, extracting the shape and texture characteristics of the segmented fabric defect image and the original defect image to realize the characteristic extraction of the fabric defect image; then, taking the defect image feature set and the labels of the fabric to be classified as training sets, and training an ELM classifier to obtain parameters of the ELM classifier; and finally, classifying the fabric defect images through the trained classifier according to Bayesian probability fusion, and outputting the classification result of the fabric defect images. The detection and classification method adopted by the invention has higher accuracy.
The method comprises the following specific steps:
step1, inputting fabric defect image G'By extracting the image G'Patch { xiSet up data matrix X ═ X1,x2,...,xi,...,xn]，xi∈RwWherein w represents the image G'Patch xiDimension, n denotes image G' patch xiTotal number; based on the Euclidean distance E (i) | | xi-x||2(x is { x)iMean value of) eliminating outlier of data matrix X to obtain training data Xc＝[x1,x2,...,xi,...,xc]I is more than or equal to 1 and less than c and less than n, wherein c represents training data XcMiddle image G'Patch xiTotal number; finding a dictionary D, substituting the least square errorTable training data XcEach point of (1) is represented by(aiIs each x of dimension giCoefficient vector of) is passed throughThe second iteration is solved, and a dictionary D ═ D after learning is generated1,d2,...,di,...,dg]，di∈Rw(i ≦ g < n), where g represents the total number of elements in the post-learning dictionary D; then, the original patch x is computedψAnd learning dictionary based D approximation patchPixel difference of(gamma is a user-defined parameter used to control segmentation sensitivity,weight terms used to reduce the influence of those pixels that are far from the center pixel ψ) to construct an anomaly map; finally, segmenting the defect area in the abnormal graph by using 2D maximum entropy and morphological operation to obtain a segmented fabric defect image G″；
Step2, extracting the fabric defect image G after segmentation by adopting Hu invariant moment″A shape feature; decomposing the fabric defect image G ' by using a wavelet packet based on an optimal wavelet packet technology to obtain an optimal decomposed wavelet packet coefficient of the fabric defect image G ', and calculating Shannon entropy to extract texture features of the fabric defect image G ';
step3, collecting G (G) of fabric defect images to be classified1,G2,...GzProcessing by a Step1 and a Step2 method to obtain a shape and texture feature set and a label of the fabric defect, and inputting the shape and texture feature set and the label into two ELM-OAA classifiers for classification training;
step4, fabric defect image G based on extraction'Texture feature of (3) and fabric defect image G after segmentation″The trained ELM-OAA classifier and Bayesian probability fusion are adopted to realize the fabric defect image G'Classifying, outputting fabric defect image G'Classifying results;
for marking a defect image G of a non-label fabric'From fabric defect images G'Selecting a single feature from shape and texture features Kt (Kt 2)Respectively input into each trained ELM-OAA classifier, and marking fabric defect categories by Bayesian probability fusion to finally realize a fabric defect image G'And (6) classifying.
Example 2: as shown in fig. 1, the method comprises the following specific steps: step1, inputting fabric defect image G'By extracting the image G'Patch { xiSet up data matrix X ═ X1,x2,…,xi,…,xn]，xi∈RwW represents an image G'Patch xiN denotes the image G'Patch xiX contains n column vectors of dimension w; based on the Euclidean distance E (i) | | xi-x||2Where x is { xiMean of the data matrix X, eliminating outliers of the data matrix X to obtain training data, Xc＝[x1,x2,…,xi,…,xc]，1≤i≤c＜n，xi∈RwW represents an image G'Patch xiWherein c represents training data XcMiddle image G'Patch xiTotal number, i.e. XcC column vectors containing dimension w; finding a dictionary D, representing the training data X with the least square errorcEach point of (1) is formed by a non-convex function formula(aiIs each x of dimension kiCoefficient vector of (D) by assuming that the variable D or a is known, the non-convex function becomes a convex function, according to an iterative optimization formulaA post-learning dictionary D ═ D is generated1,d2,...,di,...,dk]，di∈Rw(i ≦ k < n), where k represents the total number of elements in the post-learning dictionary D,denotes the number of iterative solutions, A ═ a1,a2,...,ai,...,an](ii) a Then, the original patch x is computedψAnd learning dictionary based D approximation patchPixel difference ofTo construct an anomaly map, gamma is a user-defined parameter used to control segmentation sensitivity,is a weight term to reduce the influence of those pixels that are far away from the central pixel psi,is calculated by a two-dimensional symmetrical Gaussian function,wherein the standard deviation σ 'determines the shape of the curve, and σ' is 2 in this embodiment to ensure that more defect regions are found in the abnormal graph for controlling the patch pixel ψ'A weighted distribution of (a); finally, segmenting the defect area in the abnormal graph by using 2D maximum entropy and morphological operation to obtain a segmented fabric defect image G″。
In the foregoing step, since the outliers have been eliminated, the learned dictionary D captures only the structural features of the defect-free fabric regions, and in the constructed abnormal map,because the patch size p is 28, the mask size 28 × 28 is 28, so as to ensure that more defect regions are found in the anomaly map, and the obtained two-dimensional gaussian symmetric function surface map is shown in fig. 3, where fig. 3 shows the weightsSimilar to a low pass filter, it has a maximum center weight and then the weight decreases as the radius increases. After Step1, Matlab is used to perform simulation experiments, and the flow chart of the unsupervised segmentation method for the defect images of the hollow defect fabrics is shown in FIG. 2.
Calculating the segmentation result and 3 measurement standards of the real defect area, wherein the accuracy is as follows: precision, recall: recall, comprehensive index: f-measure, the calculation formula is as follows:
wherein N isrIs an active pixel of the segmentation result; n is a radical ofgtThe method uses two parameters, namely the number of k elements of a learning dictionary and the patch size p × p, to verify the influence of the two parameters by calculating Precision, Recall and F-measure of different k and pThe fabric defect images are all used for manually marking the real areas of the defects and used as the basis for calculating the accuracy rate and the precision rate. Figure 5 shows the average PR profile of all fabric defect samples. In fig. 5, when k and p are too small, i.e., k is 1 and p < 15, curve PR shows that the overall performance of the unsupervised segmentation method is poor and relatively stable, see F-measure in fig. 5 (c); when k and p are not selected too small, i.e., k < 15 and p > 15, the segmentation performance is not greatly affected; when k is too large, i.e., k is 15, the effect on the segmentation accuracy is not good, see Precision in fig. 5 (a). The reason why the above phenomenon occurs is: on the one hand, k is too small, which increases the likelihood of fitting potential defect regions when fitting fabric texture, reducing defect significance in the relief pattern. On the other hand, p is too small, and when the defect size is larger than the patch size, the fabric texture cannot be sufficiently captured, and discrimination is poor. By combining the analysis of fig. 5, the method of the present invention has a certain robustness under the premise of not extreme parameter selection.
In this embodiment, the unsupervised segmentation result is statistically analyzed by using the defect image to be classified (hole defect, linear scratch, dot scratch, oil stain, dent, broken weft, and wrinkle) as input. Fig. 4 shows some visualized segmentation results, where the parameter k is 7 and the patch size is 28 × 28.
Step2, extracting the fabric defect image G after segmentation by adopting Hu invariant moment″A shape feature; fabric defect image G decomposed by wavelet packet based on optimal wavelet packet technology'Obtaining fabric defect image G'Calculating Shannon entropy to extract fabric defect image G'The texture features of (1);
based on the fabric defect image G after segmentation″Shape classification, we mainly use Hu invariant moment, the geometric moment m of order o + qoq(o and q are each any non-negative integer):
where F (x ', y') is the image gray value function and x 'and y' are the image coordinates, respectively. Because of the geometrical moment moqThe central moment of order (o + q) is:
whereinAndrespectively, are the coordinates of the center of mass of the image,the normalized central moment is:
ηoq＝uoq/u00 ee＝(o+q)/2 o+q＝2,3,...
geometric moments and central moments can describe regions of shape, but cannot describe rotational invariance. Therefore, 7 invariant moments φ of Hu are usedi(o + q. ltoreq.3) can satisfy the invariance of translation and rotation.
φ1＝η20+η02
φ2＝(η20-η02)2+4η11 2
φ3＝(η30-3η12)2+(3η21-η03)2
φ4＝(η30+η12)2+(η21+η03)2
φ5＝(η30-3η12)(η30+η12)[(η30+η12)2-3(η21+η03)2]
+(3η21-η03)(η21-η03)[3(η30+η12)2-(η21+η03)2]
φ6＝(η20-η02)[(η30+η12)2-3(η21+η03)2]
+4η11(η30+η12)(η21+η03)
φ7＝(3η21-η03)(η30+η12)[(η30+η12)2-3(η21+η03)2]
-(η30-3η12)(η21+η03)[3(η30+η12)2-(η21+η03)2]
Because the range of 7 moments is large, data needs to be compressed. To avoid potential negative effects, the actual invariant moment IjCalculated using the following formula:
Ij＝log10|φj|(j＝1,2,...,7)
based on the fabric defect image G after segmentation″The shape characteristics of (a) were obtained by a simulation experiment using Matlab, and the results are shown in table 1.
TABLE 1 shape characteristics of different fabric defect images after segmentation
For the original defect image, in the invention, the extraction of the texture features adopts multi-scale wavelet packet transformation. Wherein the wavelet packet decomposition image is based on the optimal wavelet packet basis tree. Each image quiltIs decomposed into 4(B-1, 2, … B) sub-image, at resolution B level, the sub-image pixels beingThe original image and the subimages are considered as parent and child nodes of a tree, the shannon entropy of a Q × Q image is defined as follows:
where ω (u, v) represents the wavelet packet coefficient at the image (u, v), continuing to decompose each child node if the sum of the entropies of the 4 child nodes is less than the entropy of the parent node; otherwise, the decomposition is stopped.
After Step2, inputting the characteristics of the extracted fabric defect image to be classified into a classifier which is trained subsequently for classification. An exemplary diagram of a specific classification flow in the method of the present invention is shown in fig. 6 (taking a hole defect as an example). In the example, a simulation experiment is carried out by using Matlab with the input defect image (hole defect) of the fabric to be classified as input. We present an optimal wavelet packet decomposition diagram for feature extraction, as shown in fig. 7.
Step3, collecting G (G) of fabric defect images to be classified1,G2,...GzProcessing the fabric defects by a Step1 and a Step2 to obtain a shape and texture feature set and a label of the fabric defects, and inputting the shape and texture feature set and the label into two ELM-OAA classifiers for classification training, wherein G'Belonging to a fabric defect image set G to be classified;
to train the ELM-OAA classifier, we first normalize these features, given a visual feature vector f with η dimensionsi″＝(fi″1,fi″2,...,fi″η),1, 2., z, the sample mean vector μ and the standard deviation vector σ are calculated as follows:
whereinIs the number of features. Normalized feature vector f'i″The calculation formula is as follows:
on the ELM-OAA classifier, a classification problem comprising a training set of seven classes of defect types, namely hole defects, linear scratches, dot scratches, oil stains, pits, broken picks, wrinkles, uses seven two-class ELM classifiers, each of which is independently trained according to an ELM classifier training algorithm. Each of the two classes of ELMs has the same input data si″But target data ti″Is different. For example, for the x-th'Two class ELM classifier, target output data ti″Is divided into two subsets: for a group belonging to the x-th'The sample label for a class is 1 and the samples belonging to other classes are labeled-1. The x' th class of two-class ELM classifier has an output value of yxIn a Single Hidden Layer feed-forward network (SLFNs), the weight β of the output neuron after the X' second-class ELM classifier shares L Hidden nodesxWe define an output weight vector βx' isβx' the calculation formula is as follows:
βx'＝(Hx')+tx'
wherein t isxIs defined asNx'is the number of training sets participating in the x' th two-class classifier. Hx'The definition is as follows:
because SLFNs represent a collective output of seven classes, similar to the multiple outputs of each two-class ELM classifier, we need to integrate them into the final class. The classification method in the invention adopts classification and output yi'(s), i' 1, …,7 are identical, meaning that the training sample(s) isi″,ti″) The total loss of (c) is minimal for all class labels. Training sample(s)i″,ti″) The total loss of (d) is calculated as follows:
where M is the 7 × 7 matrix, diagonal element +1, other elements-1. yx'(s) is the x-th'The output value, Θ, of the two-class classifier is an exponential loss function. Training samples(s) based on the above formulai″,ti″) The final output of (c) is calculated as:
finally, the output weight vector β for the seven-class two-class ELM classifier7 x' training is complete. After Step3, the ELM-OAA diagram is shown in FIG. 8.
Step4, fabric defect image G based on extraction'The trained ELM-OAA classifier and Bayes probability fusion are adopted to realize the fabric defect image G'Classifying, outputting fabric defect image G'Classifying results;
in the invention, two fabric defect characteristics are involved in the classifier training, so that two ELM-OAA classifiers are finally trained. In order to improve the classification accuracy, the Bayesian probability fusion is selected to integrate the prediction results of each feature participating in the ELM-OAA classifier training, so that the mode that the fabric defect image category is directly determined by a single feature attribute is avoided. For a fabric defect image to be classified, firstly, a single feature is selected from shape features and texture features Kt (Kt ═ 2) of the fabric defect imageAnd then input to each trained ELM-OAA classifier. It provides a [0,1 ]]Prediction of range as a posterior probabilityIn the invention, different characteristic conditions are assumed to be independent, wherein the posterior probability calculation corresponding to different characteristic values of Kt is defined as follows:
according to Bayes theoryIn the present invention, probability fusion of multiple features is defined as follows:
wherein,prior probability P (t)i″) Assumed to be uniformly distributed, the final fusion result is calculated as follows:
after Step4, a classification result of the fabric defects can be obtained, and a specific classification method flow chart is shown in fig. 6 (taking hole defects as an example). In the example, a simulation experiment is carried out by using Matlab with the input defect image (hole defect) of the fabric to be classified as input. We show a graph of the classification results of different classes of fabric defects using shape, texture and the method of the present invention, respectively, as shown in fig. 9. In order to illustrate whether the fabric defect image to be classified is correctly classified, a triangle is used for representing correct classification in fig. 9, and a circle is used for representing wrong classification. When the texture in the original fabric defect image is similar or the shape in the segmented fabric defect image is similar, the classification error can be generated by adopting the texture or shape single feature classification, and the classification accuracy of the fabric defect can be improved by adopting the classification method provided by the invention. The comparison results of the classification accuracy of different fabric defect images by using single characteristics and training and testing of the method are shown in table 2.
TABLE 2 test Classification accuracy for different training methods
It can be seen from table 2 that the accuracy of the classification method of the present invention is higher than that of the classification method of a single shape or texture, i.e., the classification accuracy of the fabric defect is improved.
In order to better illustrate the good efficiency of using the ELM classification and OAA decomposition method, the method adopts the training CPU time to evaluate the classification efficiency of the fabric defect image, the method is compared with other methods, the comparison result is shown in Table 3, and the classification time of the fabric defect image adopting the method is less than that of other methods, namely the classification efficiency is relatively higher as shown in Table 3.
TABLE 3 efficiency of different methods for classifying fabric defect images
| Type of classifier | Training time (T) |
| SVM-OAO | 13.96s |
| SVM-OAA | 120.16s |
| ELM-OAO | 5.86s |
| The method of the invention | 3.73s |
While the present invention has been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art.
Claims (3)
1. A fabric defect detection and classification method based on unsupervised segmentation and ELM is characterized in that: firstly, inputting a fabric defect image, realizing fabric defect image segmentation, and obtaining a segmented fabric defect image; secondly, extracting the shape characteristics of the segmented fabric defect image and the texture characteristics of the fabric defect image to realize the characteristic extraction of the fabric defect image; secondly, training an ELM classifier by taking the feature set and the label of the fabric defect image to be classified as a training set to obtain ELM classifier parameters; and finally, classifying the fabric defect images by the trained classifier according to Bayesian probability fusion, and outputting the classification result of the fabric defect images.
2. The unsupervised segmentation and ELM-based fabric defect detection and classification method of claim 1, characterized in that: the method comprises the following specific steps:
step1, input fabric defect image G', patch { x ] by extracting image GiSet up data matrix X ═ X1,x2,…,xi,…,xn]，xi∈RwW denotes the image G' patch xiN denotes the image G' patch xiX contains n column vectors of dimension w; based on the Euclidean distance E (i) | | xi-x||2Where x is { xiMean of the data matrix X, eliminating outliers of the data matrix X to obtain training data, Xc＝[x1,x2,…,xi,…,xc]，1≤i≤c＜n，xi∈RwW denotes the image G' patch xiWherein c represents training data XcMedium image G' Patch xiTotal number, i.e. XcC column vectors containing dimension w; finding a dictionary D, representing the training data X with the least square errorcEach point of (1) is formed by a non-convex function formula(aiIs each x of dimension kiCoefficient vector of (D) by assuming that the variable D or a is known, the non-convex function becomes a convex function, according to an iterative optimization formulaA post-learning dictionary D ═ D is generated1,d2,...,di,...,dk]，di∈Rw(i ≦ k < n), where k represents the total number of elements in the post-learning dictionary D,representing iterationsNumber of solving, a ═ a1,a2,...,ai,...,an](ii) a Then, the original patch x is computedψAnd learning dictionary based D approximation patchPixel difference ofTo construct an anomaly map, gamma is a user-defined parameter used to control segmentation sensitivity,is a weight term to reduce the influence of those pixels that are far away from the central pixel psi,is calculated by a two-dimensional symmetrical Gaussian function,wherein the standard deviation σ ' determines the curve shape, and σ ' is 2 in this embodiment, so as to ensure that more defect regions are found in the abnormal graph, and the defect regions are used for controlling the weighted distribution on the patch pixel ψ '; finally, segmenting the defect area in the abnormal graph by using 2D maximum entropy and morphological operation to obtain a segmented fabric defect image G”；
Step2, extracting the fabric defect image G after segmentation by adopting Hu invariant moment”A shape feature; decomposing the fabric defect image G ' by using a wavelet packet based on an optimal wavelet packet technology to obtain an optimal decomposed wavelet packet coefficient of the fabric defect image G ', and calculating Shannon entropy to extract texture features of the fabric defect image G ';
step3, collecting G (G) of fabric defect images to be classified1,G2,...GzProcessing by a Step1 and a Step2 method to obtain a shape and texture feature set and a label of the fabric defect, and inputting the shape and texture feature set and the label into two ELM-OAA classifiers for classification training;
step4, texture based on extracted fabric defect image GPhysical characteristics and segmented fabric defect image G”The trained ELM-OAA classifier and Bayesian probability fusion are adopted to realize the classification of the fabric defect image G 'and output the classification result of the fabric defect image G'.
3. The unsupervised segmentation and ELM-based fabric defect detection and classification method of claim 2, characterized in that: marking a defect image G of the unlabeled fabric at the steps 3 and 4'From fabric defect images G'Texture feature of (3) and fabric defect image G after segmentation”Selecting a single feature from the shape features Kt, Kt ═ 2Respectively input into each trained ELM-OAA classifier, and marking fabric defect categories by Bayesian probability fusion to finally realize a fabric defect image G'And (6) classifying.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710381898.5A CN107341499B (en) | 2017-05-26 | 2017-05-26 | Fabric defect detection and classification method based on unsupervised segmentation and ELM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710381898.5A CN107341499B (en) | 2017-05-26 | 2017-05-26 | Fabric defect detection and classification method based on unsupervised segmentation and ELM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107341499A true CN107341499A (en) | 2017-11-10 |
| CN107341499B CN107341499B (en) | 2021-01-05 |
Family
ID=60219905
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710381898.5A Active CN107341499B (en) | 2017-05-26 | 2017-05-26 | Fabric defect detection and classification method based on unsupervised segmentation and ELM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107341499B (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108038516A (en) * | 2017-12-27 | 2018-05-15 | 中山大学 | White embryo cloth flatness stage division based on low-dimensional image coding with integrated study |
| CN108154499A (en) * | 2017-12-08 | 2018-06-12 | 东华大学 | A kind of woven fabric texture flaw detection method based on K-SVD study dictionaries |
| CN108920900A (en) * | 2018-06-21 | 2018-11-30 | 福州大学 | The unsupervised extreme learning machine Feature Extraction System and method of gene expression profile data |
| CN109598287A (en) * | 2018-10-30 | 2019-04-09 | 中国科学院自动化研究所 | The apparent flaws detection method that confrontation network sample generates is generated based on depth convolution |
| CN109635866A (en) * | 2018-12-10 | 2019-04-16 | 杭州帝视科技有限公司 | The method for handling intestines image |
| CN109919939A (en) * | 2019-03-27 | 2019-06-21 | 王合山 | A kind of yarn faults detection method and device based on genetic algorithm |
| CN110245633A (en) * | 2019-06-20 | 2019-09-17 | 广东工业大学 | A kind of remote sensing images method for detecting abnormality and device with rotational invariance |
| CN110263780A (en) * | 2018-10-30 | 2019-09-20 | 腾讯科技（深圳）有限公司 | Realize the method, apparatus and equipment of isomery figure, spatial configuration of molecules property identification |
| CN110362063A (en) * | 2019-07-15 | 2019-10-22 | 山东建筑大学 | Based on the global fault detection method and system for keeping unsupervised core extreme learning machine |
| CN110490858A (en) * | 2019-08-21 | 2019-11-22 | 西安工程大学 | A kind of fabric defect Pixel-level classification method based on deep learning |
| CN111161246A (en) * | 2019-12-30 | 2020-05-15 | 歌尔股份有限公司 | Product defect detection method, device and system |
| CN111260614A (en) * | 2020-01-13 | 2020-06-09 | 华南理工大学 | Convolutional neural network cloth flaw detection method based on extreme learning machine |
| WO2020113773A1 (en) * | 2018-12-04 | 2020-06-11 | 深圳码隆科技有限公司 | Image recognition technology-based spinning machine fault monitoring system and method |
| CN111612788A (en) * | 2020-06-22 | 2020-09-01 | 创新奇智(上海)科技有限公司 | Defect identification method and device and electronic equipment |
| CN112001909A (en) * | 2020-08-26 | 2020-11-27 | 北京科技大学 | Powder bed defect visual detection method based on image feature fusion |
| CN112424827A (en) * | 2018-07-20 | 2021-02-26 | 株式会社N－Tech | Method for creating tagged image data, method for inspecting tagged image data, program for creating tagged image data, device for inspecting tagged image data, and storage medium |
| CN112541889A (en) * | 2020-12-02 | 2021-03-23 | 杭州安脉盛智能技术有限公司 | Method for detecting surface defects of complex texture object of multi-level model |
| CN113638104A (en) * | 2021-10-14 | 2021-11-12 | 海门市恒昌织带有限公司 | Intelligent yarn cleaning control method and system for bobbin winder |
| CN114155186A (en) * | 2021-10-08 | 2022-03-08 | 四川大学 | Unsupervised learning-based defect detection system and method |
| CN115496892A (en) * | 2022-11-07 | 2022-12-20 | 合肥中科类脑智能技术有限公司 | Industrial defect detection method and device, electronic equipment and storage medium |
| EP4160524A1 (en) * | 2021-10-01 | 2023-04-05 | AQC Industry | Method for detecting at least one defect on a support, device and computer program associated |
| CN117893467A (en) * | 2023-12-06 | 2024-04-16 | 江苏新丝路纺织科技有限公司 | Textile defect type identification method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103593670A (en) * | 2013-10-14 | 2014-02-19 | 浙江工业大学 | Copper sheet and strip surface defect detection method based on-line sequential extreme learning machine |
| CN104361352A (en) * | 2014-11-13 | 2015-02-18 | 东北林业大学 | Solid wood panel defect separation method based on compressed sensing |
| CN105044121A (en) * | 2015-08-17 | 2015-11-11 | 东华大学 | Online flaw detection device and online flaw detection method of circular knitting machine |
| CN105335763A (en) * | 2015-12-07 | 2016-02-17 | 东华大学 | Fabric defect classification method based on improved extreme learning machine |
| CN106650823A (en) * | 2016-12-30 | 2017-05-10 | 湖南文理学院 | Probability extreme learning machine integration-based foam nickel surface defect classification method |
-
2017
- 2017-05-26 CN CN201710381898.5A patent/CN107341499B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103593670A (en) * | 2013-10-14 | 2014-02-19 | 浙江工业大学 | Copper sheet and strip surface defect detection method based on-line sequential extreme learning machine |
| CN104361352A (en) * | 2014-11-13 | 2015-02-18 | 东北林业大学 | Solid wood panel defect separation method based on compressed sensing |
| CN105044121A (en) * | 2015-08-17 | 2015-11-11 | 东华大学 | Online flaw detection device and online flaw detection method of circular knitting machine |
| CN105335763A (en) * | 2015-12-07 | 2016-02-17 | 东华大学 | Fabric defect classification method based on improved extreme learning machine |
| CN106650823A (en) * | 2016-12-30 | 2017-05-10 | 湖南文理学院 | Probability extreme learning machine integration-based foam nickel surface defect classification method |
Non-Patent Citations (1)
| Title |
|---|
| MD. MOZAHARUL MOTTALIB,ET AL.: "Fabric Defect Classification with Geometric Features Using Bayesian Classifier", 《PROCEEDINGS OF 2015 3RD INTERNATIONAL CONFERENCE ON ADVANCES IN ELECTRICAL ENGINEERING》 * |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108154499A (en) * | 2017-12-08 | 2018-06-12 | 东华大学 | A kind of woven fabric texture flaw detection method based on K-SVD study dictionaries |
| CN108038516A (en) * | 2017-12-27 | 2018-05-15 | 中山大学 | White embryo cloth flatness stage division based on low-dimensional image coding with integrated study |
| CN108920900A (en) * | 2018-06-21 | 2018-11-30 | 福州大学 | The unsupervised extreme learning machine Feature Extraction System and method of gene expression profile data |
| CN112424827A (en) * | 2018-07-20 | 2021-02-26 | 株式会社N－Tech | Method for creating tagged image data, method for inspecting tagged image data, program for creating tagged image data, device for inspecting tagged image data, and storage medium |
| CN112424827B (en) * | 2018-07-20 | 2024-08-27 | 株式会社N－Tech | Method for producing image data with label, method for checking, memory storing program, device for producing image data with label, and checking device |
| CN110263780A (en) * | 2018-10-30 | 2019-09-20 | 腾讯科技（深圳）有限公司 | Realize the method, apparatus and equipment of isomery figure, spatial configuration of molecules property identification |
| CN110263780B (en) * | 2018-10-30 | 2022-09-02 | 腾讯科技（深圳）有限公司 | Method, device and equipment for realizing identification of properties of special composition picture and molecular space structure |
| US11854671B2 (en) | 2018-10-30 | 2023-12-26 | Tencent Technology (Shenzhen) Company Limited | Method and apparatus for identifying heterogeneous graph and property of molecular space structure and computer device |
| CN109598287B (en) * | 2018-10-30 | 2021-06-08 | 中国科学院自动化研究所 | Appearance flaw detection method for resisting network sample generation based on deep convolution generation |
| CN109598287A (en) * | 2018-10-30 | 2019-04-09 | 中国科学院自动化研究所 | The apparent flaws detection method that confrontation network sample generates is generated based on depth convolution |
| WO2020088439A1 (en) * | 2018-10-30 | 2020-05-07 | 腾讯科技（深圳）有限公司 | Method for identifying isomerism graph and molecular spatial structural property, device, and computer apparatus |
| WO2020113773A1 (en) * | 2018-12-04 | 2020-06-11 | 深圳码隆科技有限公司 | Image recognition technology-based spinning machine fault monitoring system and method |
| CN109635866B (en) * | 2018-12-10 | 2021-07-23 | 杭州帝视科技有限公司 | Method of processing an intestinal image |
| CN109635866A (en) * | 2018-12-10 | 2019-04-16 | 杭州帝视科技有限公司 | The method for handling intestines image |
| CN109919939B (en) * | 2019-03-27 | 2023-09-29 | 王合山 | Yarn defect detection method and device based on genetic algorithm |
| CN109919939A (en) * | 2019-03-27 | 2019-06-21 | 王合山 | A kind of yarn faults detection method and device based on genetic algorithm |
| CN110245633A (en) * | 2019-06-20 | 2019-09-17 | 广东工业大学 | A kind of remote sensing images method for detecting abnormality and device with rotational invariance |
| CN110362063B (en) * | 2019-07-15 | 2020-07-03 | 山东建筑大学 | Fault detection method and system based on global maintenance unsupervised kernel extreme learning machine |
| CN110362063A (en) * | 2019-07-15 | 2019-10-22 | 山东建筑大学 | Based on the global fault detection method and system for keeping unsupervised core extreme learning machine |
| CN110490858A (en) * | 2019-08-21 | 2019-11-22 | 西安工程大学 | A kind of fabric defect Pixel-level classification method based on deep learning |
| CN110490858B (en) * | 2019-08-21 | 2022-12-13 | 西安工程大学 | Fabric defective pixel level classification method based on deep learning |
| CN111161246A (en) * | 2019-12-30 | 2020-05-15 | 歌尔股份有限公司 | Product defect detection method, device and system |
| CN111161246B (en) * | 2019-12-30 | 2024-05-14 | 歌尔股份有限公司 | Product defect detection method, device and system |
| US11741593B2 (en) | 2019-12-30 | 2023-08-29 | Goertek Inc. | Product defect detection method, device and system |
| CN111260614A (en) * | 2020-01-13 | 2020-06-09 | 华南理工大学 | Convolutional neural network cloth flaw detection method based on extreme learning machine |
| CN111612788A (en) * | 2020-06-22 | 2020-09-01 | 创新奇智(上海)科技有限公司 | Defect identification method and device and electronic equipment |
| CN111612788B (en) * | 2020-06-22 | 2023-07-04 | 创新奇智(上海)科技有限公司 | Defect identification method and device and electronic equipment |
| CN112001909A (en) * | 2020-08-26 | 2020-11-27 | 北京科技大学 | Powder bed defect visual detection method based on image feature fusion |
| CN112001909B (en) * | 2020-08-26 | 2023-11-24 | 北京科技大学 | Powder bed defect visual detection method based on image feature fusion |
| CN112541889A (en) * | 2020-12-02 | 2021-03-23 | 杭州安脉盛智能技术有限公司 | Method for detecting surface defects of complex texture object of multi-level model |
| WO2023052537A1 (en) * | 2021-10-01 | 2023-04-06 | Aqc Industry | Method for detecting at least one defect on a support, device and computer program associated |
| EP4160524A1 (en) * | 2021-10-01 | 2023-04-05 | AQC Industry | Method for detecting at least one defect on a support, device and computer program associated |
| CN114155186B (en) * | 2021-10-08 | 2023-05-12 | 四川大学 | Defect detection system and method based on unsupervised learning |
| CN114155186A (en) * | 2021-10-08 | 2022-03-08 | 四川大学 | Unsupervised learning-based defect detection system and method |
| CN113638104B (en) * | 2021-10-14 | 2022-02-08 | 海门市恒昌织带有限公司 | Intelligent yarn cleaning control method and system for bobbin winder |
| CN113638104A (en) * | 2021-10-14 | 2021-11-12 | 海门市恒昌织带有限公司 | Intelligent yarn cleaning control method and system for bobbin winder |
| CN115496892A (en) * | 2022-11-07 | 2022-12-20 | 合肥中科类脑智能技术有限公司 | Industrial defect detection method and device, electronic equipment and storage medium |
| CN117893467A (en) * | 2023-12-06 | 2024-04-16 | 江苏新丝路纺织科技有限公司 | Textile defect type identification method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107341499B (en) | 2021-01-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107341499B (en) | Fabric defect detection and classification method based on unsupervised segmentation and ELM | |
| Li et al. | Automatic fabric defect detection with a wide-and-compact network | |
| CN108562589B (en) | Method for detecting surface defects of magnetic circuit material | |
| Xu et al. | A weakly supervised surface defect detection based on convolutional neural network | |
| CN110264448B (en) | Insulator fault detection method based on machine vision | |
| CN110827260B (en) | Cloth defect classification method based on LBP characteristics and convolutional neural network | |
| Weimer et al. | Learning defect classifiers for textured surfaces using neural networks and statistical feature representations | |
| CN105678788B (en) | A kind of fabric defect detection method based on HOG and low-rank decomposition | |
| CN113160192A (en) | Visual sense-based snow pressing vehicle appearance defect detection method and device under complex background | |
| CN112446429B (en) | CGAN (Carrier grade Access network) -based routing inspection image data small sample expansion method | |
| Kuo et al. | Application of computer vision in the automatic identification and classification of woven fabric weave patterns | |
| WO2021087014A1 (en) | Bbp assisted defect detection flow for sem images | |
| Liu et al. | Unsupervised segmentation and elm for fabric defect image classification | |
| Liu et al. | A classification method of glass defect based on multiresolution and information fusion | |
| CN110414616B (en) | Remote sensing image dictionary learning and classifying method utilizing spatial relationship | |
| Zeng et al. | Steel sheet defect detection based on deep learning method | |
| CN114565607B (en) | Fabric defect image segmentation method based on neural network | |
| Lien et al. | Product surface defect detection based on deep learning | |
| Saleh et al. | Tire defect detection model using machine learning | |
| Luo et al. | Alphanumeric character recognition based on BP neural network classification and combined features | |
| Yang et al. | Fabric defect classification using wavelet frames and minimum classification error training | |
| CN117067112A (en) | Water cutting machine and control method thereof | |
| CN109829511B (en) | Texture classification-based method for detecting cloud layer area in downward-looking infrared image | |
| Zhuang et al. | Multi-component attention-based convolution network for color difference recognition with wavelet entropy strategy | |
| CN109858397A (en) | A kind of faint IR target recognition method based on adaptive modeling |
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 |