CN113643275B - Ultrasonic defect detection method based on unsupervised manifold segmentation - Google Patents
Ultrasonic defect detection method based on unsupervised manifold segmentation Download PDFInfo
- Publication number
- CN113643275B CN113643275B CN202110999452.5A CN202110999452A CN113643275B CN 113643275 B CN113643275 B CN 113643275B CN 202110999452 A CN202110999452 A CN 202110999452A CN 113643275 B CN113643275 B CN 113643275B
- Authority
- CN
- China
- Prior art keywords
- segmentation
- dimensional
- ultrasonic
- matrix
- defect
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000007547 defect Effects 0.000 title claims abstract description 64
- 230000011218 segmentation Effects 0.000 title claims abstract description 43
- 238000001514 detection method Methods 0.000 title claims abstract description 22
- 238000000034 method Methods 0.000 claims abstract description 22
- 229920000642 polymer Polymers 0.000 claims abstract description 12
- 230000009467 reduction Effects 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 47
- 238000002592 echocardiography Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 238000002604 ultrasonography Methods 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 8
- 238000009826 distribution Methods 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 239000003086 colorant Substances 0.000 claims description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 238000012285 ultrasound imaging Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims description 3
- 238000012800 visualization Methods 0.000 claims description 3
- 238000012935 Averaging Methods 0.000 claims description 2
- 230000000007 visual effect Effects 0.000 claims description 2
- 238000003709 image segmentation Methods 0.000 abstract description 6
- 229920000049 Carbon (fiber) Polymers 0.000 abstract description 5
- 239000004917 carbon fiber Substances 0.000 abstract description 5
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 abstract description 5
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 238000000605 extraction Methods 0.000 abstract description 3
- 238000003384 imaging method Methods 0.000 abstract description 3
- 238000009659 non-destructive testing Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000011208 reinforced composite material Substances 0.000 description 2
- 229920005989 resin Polymers 0.000 description 2
- 239000011347 resin Substances 0.000 description 2
- 239000004809 Teflon Substances 0.000 description 1
- 229920006362 Teflon® Polymers 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000003822 epoxy resin Substances 0.000 description 1
- 238000002347 injection Methods 0.000 description 1
- 239000007924 injection Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000007769 metal material Substances 0.000 description 1
- 229920000647 polyepoxide Polymers 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 239000000243 solution Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000001721 transfer moulding Methods 0.000 description 1
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G06T3/06—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10132—Ultrasound image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
Abstract
An ultrasonic defect detection method based on unsupervised manifold segmentation belongs to the technical field of nondestructive detection of internal defects of carbon fiber polymers. It comprises the following steps: 1) Acquiring a polymer internal defect image dataset; 2) Pre-processing defect ultrasonic data; 3) Unified manifold approximation and projection dimension reduction; 4) Unsupervised image segmentation; 5) And (5) visualizing the defects. The invention uses unified manifold approximation and projection, saves local information and global data structure, carries out high-quality dimension reduction on data, and carries out pixel-level segmentation and clustering by combining an unsupervised image segmentation method, thereby completing defect region extraction, improving the identifiability of ultrasonic imaging of defects in the polymer and being beneficial to improving the accuracy of ultrasonic nondestructive testing.
Description
Technical Field
The invention belongs to the technical field of nondestructive detection of internal defects of carbon fiber polymers, and particularly relates to an ultrasonic defect detection method based on unsupervised manifold segmentation.
Background
The carbon fiber reinforced composite material (CFRP) has the high-quality characteristics of low density, high strength, high chemical stability and the like, is always a competitive substitute for the traditional metal material, and has important application in the industrial fields of aviation, automobiles and the like. However, the CFRP structure is easily defective and damaged during the manufacturing and use process due to factors such as the manufacturing process and external impact. To ensure the integrity and reliability of CFRP, it is important to conduct non-destructive testing during the manufacture and use of CFRP materials.
A number of ultrasound imaging data analysis methods have been developed, however, the training of classifiers for these algorithms requires a large amount of historical data, and it is currently difficult to obtain large sample data sets, and some image processing algorithms are not substantially straightforward for composite ultrasound defect detection. Therefore, it is particularly important to realize automatic detection of unsupervised ultrasound.
Disclosure of Invention
Aiming at the problems in the prior art, the invention aims to provide an ultrasonic defect detection method based on unsupervised manifold segmentation, which fully utilizes the advantages of manifold learning reduction and unsupervised image segmentation and realizes efficient detection of a model.
The invention provides the following technical scheme: an ultrasonic defect detection method based on an unsupervised manifold segmentation, wherein the unsupervised manifold segmentation Ums (Unsupervised manifold segmentation) comprises unified manifold approximation, projection dimension reduction and unsupervised semantic segmentation, and comprises the following steps:
(1) Acquiring a polymer internal defect image dataset
Transmitting continuous ultrasonic pulses at fixed intervals to the surface of a workpiece by an ultrasonic phased detector, collecting ultrasonic images from the inside of a polymer, transmitting sound waves in the workpiece, reflecting the sound waves on the upper surface and the lower surface of a test medium and an interface formed by the medium and the defect, capturing ultrasonic echoes by a receiver, and finally storing the obtained ultrasonic defect images into a three-dimensional matrix to be recorded;
(2) Defect ultrasound data preprocessing
Converting an ultrasonic defect image recorded in a three-dimensional rectangular conversion form into a two-dimensional matrix, and preprocessing the two-dimensional matrix to balance echoes of defects with different depths and weaken the influence of surface and bottom echoes;
(3) Unified manifold approximation and projection dimension reduction
Taking the preprocessed two-dimensional matrix in the step (2) as input, and constructing a fuzzy topological representation by a unified manifold approximation and projection algorithm through calculating the membership of the nearest neighbor of each point, optimizing the low-dimensional representation in a low-dimensional space, and measuring the low-dimensional representation by adopting cross entropy;
(4) Unsupervised semantic segmentation
Averaging the dimension-reduced two-dimensional matrix along the row direction, converting the dimension-reduced two-dimensional matrix into a two-dimensional matrix, carrying out self-adaptive clustering according to the colors and textures of pixels, distributing consistent semantic labels, carrying out super-pixel segmentation, pre-classifying images, calculating a feature map by using a deep convolution network until the semantic segmentation result is consistent with the pre-classified result, and finally carrying out combined block visual output;
(5) Defect visualization
The results of the unsupervised manifold segmentation Ums of the ultrasound defect image are output and the defect is visualized.
The ultrasonic defect detection method based on the unsupervised manifold segmentation is characterized in that the process of the step (2) is as follows:
step 2.1: three-dimensional ultrasound matrix X to be recorded 0 ∈R H×X×Y Conversion to a two-dimensional matrix X 1 Wherein H, X, Y represent the sampling time point, the horizontal scanning direction, and the vertical scanning direction positions, respectively;
step 2.2: robust normalization method for two-dimensional matrix X 1 Preprocessing to balance echoes of defects with different depths and weaken the influence of surface and bottom echoes, wherein the calculation formula is as follows:
where x represents a certain value of the sample, mean is the median of the sample, and IQR is the quartile range of the sample.
The ultrasonic defect detection method based on the unsupervised manifold segmentation is characterized in that the specific process of the step (3) is as follows:
step 3.1: and 2, performing neighbor point calculation on each pixel value in the ultrasonic data matrix subjected to pretreatment in the step 2.2, wherein the calculation formula is as follows:
in which x is i And x j For sample points in the expanded ultrasound imaging data matrix, where subscripts i, j are sample point numbers, d (x i ,x j ) For point x i And point x j Arbitrary distance between them, p i Representing the distance, sigma, from the ith data point to its first nearest neighbor i Representing a normalization factor;
step 3.2: the initial low-dimensional coordinates are distributed by using the graph Laplace, then the graph Laplace matrix is formed by constructing the Laplace matrix through matrix algebra, and finally the Laplace matrix is decomposed to obtain each x i K nearest neighbor set of (2)Wherein->For each x i Determining ρ i Sum sigma i :
To bind together points with a measure of local variation, a calculation of the distribution is performed using a symmetrization of the high-dimensional probability:
p ij ＝p i|j +p j|i -p i|j p j|i ；
step 3.3: by usingCalculating low-dimensional distance probability of curve cluster of (2)Initializing low-dimensional representation by embedding line spectrum without normalization, q ij Representing y i Is y j The distribution probability of the neighborhood is expressed as follows:
q ij ＝(1+a(y i -y j ) 2b ) -1
wherein the super parameters a and b are empirical parameters, y i ，y j Is x i ，x j Mapping in a low dimensional space;
step 3.4: the low-dimensional representation is optimized and measured using cross entropy, as follows:
the ultrasonic defect detection method based on the unsupervised manifold segmentation is characterized in that the specific process of the step (4) is as follows:
step 4.1: the dimension-reduced two-dimensional matrix is represented as an average value according to the row directionConverted into a size n x ×n y Performing self-adaptive clustering according to the colors and textures of pixels, distributing consistent semantic labels, performing super-pixel segmentation by using a Philippine tile algorithm, pre-classifying the images, distributing consistent semantic labels if semantic information of small areas after super-pixel segmentation is not different, initializing a convolutional neural network, and keeping variance and mean value of each layer;
step 4.2: using an initialized convolutional neural network, taking the label with the maximum value as the corresponding pixel according to the feature map, counting the category with the maximum occurrence number in each cluster according to the semantic segmentation clustering result, marking the pixels with the same semantic information in all pixels in the cluster as small blocks, and finally merging all the small blocks into a large block;
step 4.3: and calculating a loss function, adopting random gradient descent update parameters, and iterating for T times to converge until the semantic segmentation result is consistent with the pre-classification result.
By adopting the technology, compared with the prior art, the invention has the following beneficial effects:
the invention uses unified manifold approximation and projection to save as many local and more global data structures as possible, carries out high-quality dimension reduction on the data, and carries out pixel-level segmentation and clustering by combining an unsupervised image segmentation method, thereby completing defect region extraction, improving the identifiability of ultrasonic imaging of defects in the polymer and being beneficial to improving the accuracy of ultrasonic nondestructive detection.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a schematic diagram showing the defects of CFRP of a carbon fiber reinforced composite material according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of the three-dimensional ultrasonic matrix converted into a two-dimensional matrix according to the present invention;
fig. 4 is a graph of the analysis results of the unsupervised manifold segmentation Ums method.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings and examples of the present invention. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
On the contrary, the invention is intended to cover any alternatives, modifications, equivalents, and variations as may be included within the spirit and scope of the invention as defined by the appended claims. Further, in the following detailed description of the present invention, certain specific details are set forth in order to provide a better understanding of the present invention. The present invention will be fully understood by those skilled in the art without the details described herein.
Referring to fig. 1 to 4, an ultrasonic defect detection method based on unsupervised manifold segmentation, the method comprising the steps of:
(1) The process of acquiring the polymer internal defect image data set is as follows:
the CFRP specimens were prepared using a tree vacuum assisted resin transfer molding process and six internal defects were prefabricated manually. The CFRP sample was made of 78 layers of carbon fiber sheet and epoxy resin and had a thickness of 20cm. Before resin injection, 6 square teflon tapes were inserted with defect sizes of 20.0mm×20.0mm×0.5mm.
An ultrasonic phased detector is used to transmit successive, fixed-interval ultrasonic pulses to the surface of the workpiece to acquire ultrasonic images from the interior of the polymer. The acoustic wave propagates inside the workpiece and is reflected at the upper and lower surfaces of the test medium and at the interface between the medium and the defect. Ultrasound echoes are captured using a receiver. And finally, storing the obtained ultrasonic image as a three-dimensional matrix for recording.
(2) The defect ultrasonic data preprocessing comprises the following steps:
step 2.1: three-dimensional ultrasound matrix X to be recorded 0 ∈R H×X×Y Conversion to a two-dimensional matrix X 1 Where H, X, Y denote the sampling time point, the horizontal scanning direction, and the vertical scanning direction positions, respectively.
Step 2.2: in ultrasonic detection, ultrasonic energy is gradually attenuated with the increase of the propagation distance, deep defect signals are likely to be shielded by shallow defects, and the return values of the defect signals are greatly influenced by surface echoes and bottom echoes, so that defect detection becomes difficult. Thus, a robust normalization method is employed to balance the echo of different depth defects and attenuate the effects of surface and bottom echo. The calculation formula is as follows:
where x is a value of the sample, mean is the median of the sample, and IQR is the quartile range of the sample.
(3) The unified manifold approximation and projection dimension reduction process comprises the following steps:
step 3.1: and 2, performing neighbor point calculation on each pixel value in the ultrasonic data matrix subjected to pretreatment in the step 2.2, wherein the calculation formula is as follows:
in which x is i And x j For sample points in the expanded ultrasound imaging data matrix, where subscripts i, j are sample point numbers, d (x i ,x j ) For point x i And point x j Arbitrary distance between them, p i Representing the distance, sigma, from the ith data point to its first nearest neighbor i Representing the normalization factor.
Step 3.2: the initial low-dimensional coordinates are distributed by using the graph Laplace, then the graph Laplace matrix is formed by constructing the Laplace matrix through matrix algebra, and finally the Laplace matrix is decomposed to obtain each x i K nearest neighbor set of (2)Wherein->For each x i Determining ρ i Sum sigma i :
To bind together points with a measure of local variation, a calculation of the distribution is performed using a symmetrization of the high-dimensional probability:
p ij ＝p i|j +p j|i -p i|j p j|i
step 3.3: by usingThe low-dimensional distance probability is calculated by the curve cluster of the (2), the low-dimensional representation is initialized by spectrum embedding,does not need to be normalized, q ij Representing y i Is y j The distribution probability of the neighborhood is expressed as follows:
q ij ＝(1+a(yi-yj) 2b ) -1
wherein the default super-parameters a=1.93, b=0.79, xi, x j Mapping to low-dimensional space corresponds to yi, y j 。
Step 3.4: optimizing the low-dimensional representation, enabling the fuzzy topological representation of the low-dimensional space to be similar as much as possible, and measuring the fuzzy topological representation by adopting cross entropy:
(4) Unsupervised image segmentation, the process is as follows:
the matrix after dimension reduction is averaged in the row direction to be expressed asConverted into a size n x ×n y Is a 2-dimensional matrix full-size image of (c). If the colors and textures of the pixels are similar and the positions are close, consistent semantic tags are allocated, super-pixel segmentation is carried out by using a Philippine tile algorithm, the images are pre-classified, and if the semantic information of the small areas is not different, the consistent semantic tags are allocated. And initializing a neural network, and maintaining the variance and the mean value of each layer.
Step 4.2: and combining the self-encoder, using a four-layer convolution network feature diagram, and taking the label with the maximum value as the label of the corresponding pixel according to the feature diagram. For the traditional semantic segmentation clustering result, counting the class with the largest occurrence number in each cluster, marking all pixels in the cluster as small blocks by using pixels containing the same semantic information, and finally merging all small blocks into a large block.
Step 4.3: and calculating a loss function, adopting random gradient descent update parameters, and iterating for 64 times to converge until the semantic segmentation result is consistent with the pre-classification result as much as possible.
(5) Defect visualization
Outputting the result of the unsupervised manifold segmentation Ums dimension reduction segmentation of the ultrasonic defect and visualizing the defect.
The method has good defect evaluation performance, can well process the original noise, has good performance in the aspects of boundary and background separation and defect shape, and can clearly detect 6 defects. The defect information is complete, the outline of the edge of the defect is clear, and the accuracy of defect evaluation is facilitated. This suggests that the unsupervised manifold segmentation Ums is a viable and excellent method for analyzing ultrasound test data.
The method of the invention uses unified manifold approximation and projection to store as many local and more global data structures as possible. The unsupervised image segmentation method is added to carry out pixel-level segmentation and clustering, so that defect region extraction is completed, the identifiability of ultrasonic detection imaging of the defects in the polymer is improved, and the accuracy of ultrasonic nondestructive detection is improved.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
Claims (2)
1. An ultrasonic defect detection method based on unsupervised manifold segmentation, the unsupervised manifold segmentation Ums comprises unified manifold approximation, projection dimension reduction and unsupervised semantic segmentation, and is characterized in that: the method comprises the following steps:
(1) Acquiring a polymer internal defect image dataset
Transmitting continuous ultrasonic pulses at fixed intervals to the surface of a workpiece by an ultrasonic phased detector, collecting ultrasonic images from the inside of a polymer, transmitting sound waves in the workpiece, reflecting the sound waves on the upper surface and the lower surface of a test medium and an interface formed by the medium and the defect, capturing ultrasonic echoes by a receiver, and finally storing the obtained ultrasonic defect images into a three-dimensional matrix to be recorded;
(2) Defect ultrasound data preprocessing
Converting an ultrasonic defect image recorded in a three-dimensional rectangular conversion form into a two-dimensional matrix, and preprocessing the two-dimensional matrix to balance echoes of defects with different depths and weaken the influence of surface and bottom echoes;
the process of (2) is as follows:
step 2.1: three-dimensional ultrasound matrix X to be recorded 0 ∈R H×X×Y Conversion to a two-dimensional matrix X 1 Wherein H, X, Y represent the sampling time point, the horizontal scanning direction, and the vertical scanning direction positions, respectively;
step 2.2: robust normalization method for two-dimensional matrix X 1 Preprocessing to balance echoes of defects with different depths and weaken the influence of surface and bottom echoes, wherein the calculation formula is as follows:
where x represents a certain value of the sample, mean is the median of the sample, and IQR is the quartile range of the sample;
(3) Unified manifold approximation and projection dimension reduction
Taking the preprocessed two-dimensional matrix in the step (2) as input, and constructing a fuzzy topological representation by a unified manifold approximation and projection algorithm through calculating the membership of the nearest neighbor of each point, optimizing the low-dimensional representation in a low-dimensional space, and measuring the low-dimensional representation by adopting cross entropy;
the specific process of the (3) is as follows:
step 3.1: and 2, performing neighbor point calculation on each pixel value in the ultrasonic data matrix subjected to pretreatment in the step 2.2, wherein the calculation formula is as follows:
in which x is i And x j For sample points in the expanded ultrasound imaging data matrix, where subscripts i, j are sample point numbers, d (x i ,x j ) For point x i And point x j Arbitrary distance between them, p i Representing the distance, sigma, from the ith data point to its first nearest neighbor i Representing a normalization factor;
step 3.2: the initial low-dimensional coordinates are distributed by using the graph Laplace, then the graph Laplace matrix is formed by constructing the Laplace matrix through matrix algebra, and finally the Laplace matrix is decomposed to obtain each x i K nearest neighbor set of (2)Wherein->For each x i Determining ρ i Sum sigma i :
To bind together points with a measure of local variation, a calculation of the distribution is performed using a symmetrization of the high-dimensional probability:
p ij ＝p i|j +p j|i -p i|j p j|i ；
step 3.3: by usingThe low-dimensional distance probability is calculated by the curve cluster of (a), the low-dimensional representation is initialized by spectrum embedding, normalization is not needed, and q is not needed ij Representing y i Is y j The distribution probability of the neighborhood is expressed as follows:
q ij ＝(1+a(y i -y j ) 2b ) -1
wherein the super parameters a and b are empirical parameters, y i ，y j Is x i ，x j Mapping in a low dimensional space;
step 3.4: the low-dimensional representation is optimized and measured using cross entropy, as follows:
(4) Unsupervised semantic segmentation
Averaging the dimension-reduced two-dimensional matrix along the row direction, converting the dimension-reduced two-dimensional matrix into a two-dimensional matrix, carrying out self-adaptive clustering according to the colors and textures of pixels, distributing consistent semantic labels, carrying out super-pixel segmentation, pre-classifying images, calculating a feature map by using a deep convolution network until the semantic segmentation result is consistent with the pre-classified result, and finally carrying out combined block visual output;
(5) Defect visualization
The results of the unsupervised manifold segmentation Ums of the ultrasound defect image are output and the defect is visualized.
2. The ultrasonic defect detection method based on the unsupervised manifold segmentation as set forth in claim 1, wherein the specific process of (4) is as follows:
step 4.1: the dimension-reduced two-dimensional matrix is represented as an average value according to the row directionConverted into a size n x ×n y Performing self-adaptive clustering according to the colors and textures of pixels, distributing consistent semantic labels, performing super-pixel segmentation by using a Philippine tile algorithm, pre-classifying the images, distributing consistent semantic labels if semantic information of small areas after super-pixel segmentation is not different, initializing a convolutional neural network, and keeping variance and mean value of each layer;
step 4.2: using an initialized convolutional neural network, taking the label with the maximum value as the corresponding pixel according to the feature map, counting the category with the maximum occurrence number in each cluster according to the semantic segmentation clustering result, marking the pixels with the same semantic information in all pixels in the cluster as small blocks, and finally merging all the small blocks into a large block;
step 4.3: and calculating a loss function, adopting random gradient descent update parameters, and iterating for T times to converge until the semantic segmentation result is consistent with the pre-classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110999452.5A CN113643275B (en) | 2021-08-29 | 2021-08-29 | Ultrasonic defect detection method based on unsupervised manifold segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110999452.5A CN113643275B (en) | 2021-08-29 | 2021-08-29 | Ultrasonic defect detection method based on unsupervised manifold segmentation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113643275A CN113643275A (en) | 2021-11-12 |
| CN113643275B true CN113643275B (en) | 2024-03-01 |
Family
ID=78424307
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110999452.5A Active CN113643275B (en) | 2021-08-29 | 2021-08-29 | Ultrasonic defect detection method based on unsupervised manifold segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113643275B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114266298B (en) * | 2021-12-16 | 2022-07-08 | 盐城工学院 | Image segmentation method and system based on consistent manifold approximation and projection clustering integration |
| CN114494175B (en) * | 2022-01-21 | 2024-05-03 | 厦门大学 | Interactive space segmentation method for mass spectrum imaging data |
| CN114581969A (en) * | 2022-01-21 | 2022-06-03 | 厦门大学 | Face position information-based standing and sitting motion detection method |
| CN114429771A (en) * | 2022-04-02 | 2022-05-03 | 武汉地震工程研究院有限公司 | Intelligent detection method and system for bonding defects of steel beam and CFRP (carbon fiber reinforced plastics) plate |
| CN117589889B (en) * | 2024-01-19 | 2024-04-02 | 江苏省特种设备安全监督检验研究院 | Ultrasonic phased array microcrack detection method and system based on finite element model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019104767A1 (en) * | 2017-11-28 | 2019-06-06 | 河海大学常州校区 | Fabric defect detection method based on deep convolutional neural network and visual saliency |
| CN112057079A (en) * | 2020-08-07 | 2020-12-11 | 中国科学院深圳先进技术研究院 | Behavior quantification method and terminal based on state and map |
| CN112489016A (en) * | 2020-11-30 | 2021-03-12 | 浙江工业大学 | Local sensitivity discriminant analysis method for composite material defect thermal image |
| CN113160195A (en) * | 2021-04-28 | 2021-07-23 | 浙江工业大学 | Composite material defect nondestructive testing method based on thermal imaging |
-
2021
- 2021-08-29 CN CN202110999452.5A patent/CN113643275B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019104767A1 (en) * | 2017-11-28 | 2019-06-06 | 河海大学常州校区 | Fabric defect detection method based on deep convolutional neural network and visual saliency |
| CN112057079A (en) * | 2020-08-07 | 2020-12-11 | 中国科学院深圳先进技术研究院 | Behavior quantification method and terminal based on state and map |
| CN112489016A (en) * | 2020-11-30 | 2021-03-12 | 浙江工业大学 | Local sensitivity discriminant analysis method for composite material defect thermal image |
| CN113160195A (en) * | 2021-04-28 | 2021-07-23 | 浙江工业大学 | Composite material defect nondestructive testing method based on thermal imaging |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113643275A (en) | 2021-11-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113643275B (en) | Ultrasonic defect detection method based on unsupervised manifold segmentation | |
| CN112700429B (en) | Airport pavement underground structure disease automatic detection method based on deep learning | |
| CN109685768B (en) | Pulmonary nodule automatic detection method and system based on pulmonary CT sequence | |
| CN109871902B (en) | SAR small sample identification method based on super-resolution countermeasure generation cascade network | |
| CN113436169B (en) | Industrial equipment surface crack detection method and system based on semi-supervised semantic segmentation | |
| CN112001218B (en) | Three-dimensional particle class detection method and system based on convolutional neural network | |
| CN112861919A (en) | Underwater sonar image target detection method based on improved YOLOv3-tiny | |
| CN113504306B (en) | Steel rail defect detection method based on ultrasonic phased array low-rank matrix recovery | |
| CN111738332A (en) | Underwater multi-source acoustic image substrate classification method and system based on feature level fusion | |
| CN112699717A (en) | SAR image generation method and generation device based on GAN network | |
| CN111709487B (en) | Underwater multi-source acoustic image substrate classification method and system based on decision-level fusion | |
| CN113538331A (en) | Metal surface damage target detection and identification method, device, equipment and storage medium | |
| CN109410139A (en) | Inside a kind of historical relic and surface disease digital assay appraisal procedure | |
| Sun et al. | Probabilistic neural network based seabed sediment recognition method for side-scan sonar imagery | |
| Hua et al. | Defect identification method of carbon fiber sucker rod based on GoogLeNet-based deep learning model and transfer learning | |
| CN111738278A (en) | Underwater multi-source acoustic image feature extraction method and system | |
| CN115047068A (en) | Composite material internal defect detection method based on local linear embedding and cluster segmentation | |
| Sun et al. | Automated quality characterization for composites using hybrid ultrasonic imaging techniques | |
| CN113534146B (en) | Automatic detection method and system for radar video image target | |
| CN115471727A (en) | Composite material defect detection method based on transfer learning | |
| CN115223033A (en) | Synthetic aperture sonar image target classification method and system | |
| Wang et al. | Sonar objective detection based on dilated separable densely connected CNNs and quantum-behaved PSO algorithm | |
| CN111738277B (en) | Underwater shallow stratum profile image feature extraction method and system | |
| CN114324599B (en) | Periodic reconstruction-based ultrasonic C-scan quantification method for debonded area of honeycomb sandwich plate | |
| CN117523548B (en) | Three-dimensional model object extraction and recognition method based on neural network |
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 |