CN113702439A - Infrared nondestructive testing method based on iterative generation of sparse principal component model - Google Patents
Infrared nondestructive testing method based on iterative generation of sparse principal component model Download PDFInfo
- Publication number
- CN113702439A CN113702439A CN202110963492.4A CN202110963492A CN113702439A CN 113702439 A CN113702439 A CN 113702439A CN 202110963492 A CN202110963492 A CN 202110963492A CN 113702439 A CN113702439 A CN 113702439A
- Authority
- CN
- China
- Prior art keywords
- infrared thermal
- thermal image
- image data
- images
- data set
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000009659 non-destructive testing Methods 0.000 title claims abstract description 22
- 238000001931 thermography Methods 0.000 claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims abstract description 10
- 238000000605 extraction Methods 0.000 claims abstract description 9
- 230000009467 reduction Effects 0.000 claims abstract description 7
- 230000007547 defect Effects 0.000 claims description 63
- 239000011159 matrix material Substances 0.000 claims description 31
- 239000013598 vector Substances 0.000 claims description 17
- 238000007781 pre-processing Methods 0.000 claims description 5
- 239000000126 substance Substances 0.000 claims description 4
- 238000005520 cutting process Methods 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims description 3
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 claims description 2
- 238000002591 computed tomography Methods 0.000 abstract 1
- 238000001228 spectrum Methods 0.000 abstract 1
- 239000002131 composite material Substances 0.000 description 15
- 230000003042 antagnostic effect Effects 0.000 description 8
- 230000006870 function Effects 0.000 description 8
- 238000005457 optimization Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 6
- 238000001514 detection method Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 4
- 229920000049 Carbon (fiber) Polymers 0.000 description 3
- 239000004917 carbon fiber Substances 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 3
- 230000003190 augmentative effect Effects 0.000 description 2
- 239000011162 core material Substances 0.000 description 2
- 230000002950 deficient Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 239000011208 reinforced composite material Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- 238000007792 addition Methods 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000032798 delamination Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 230000003647 oxidation Effects 0.000 description 1
- 238000007254 oxidation reaction Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N25/00—Investigating or analyzing materials by the use of thermal means
- G01N25/72—Investigating presence of flaws
Abstract
The invention discloses an infrared thermal imaging nondestructive testing method based on an iteration generation sparse principal component model, which utilizes an SPCT (spin-spectrum computed tomography) method to perform denoising, dimensionality reduction and feature extraction on a preprocessed infrared thermal image data set to obtain a visualized component loading image; fusing the component loading image with the preprocessed infrared thermal image; training the generative countermeasure neural network by using the fused infrared thermal images, selecting a plurality of generated false thermal images and the initial infrared thermal image to splice into an amplified infrared thermal image data set, and performing iterative cycle by using the amplified infrared thermal image data set as the initial infrared thermal image data set. The invention solves the problem of insufficient data set of the original thermal image, and simultaneously utilizes the SPCT method to further improve the signal-to-noise ratio of the thermal image, thereby having good practical application prospect.
Description
Technical Field
The invention belongs to the technical field of defect detection, and particularly relates to an infrared nondestructive detection method based on an iteration generation sparse principal component model.
Background
In recent years, research on composite materials has become mature, and various composite materials with excellent performance are developed, including Carbon Fiber Reinforced Composite (CFRP), which has the advantages of high strength, light weight, corrosion resistance, oxidation resistance, high temperature resistance, and the like, and thus has a wide application prospect in the fields of aerospace, automobiles, and the like. However, during the preparation and use of the composite material, various defects or damages may be generated on the surface and inside of the composite material, so that a safety hazard is generated and even serious consequences which are difficult to predict are caused. Therefore, the method has very important significance for quickly, accurately and efficiently detecting whether the composite material has defects and the positions, the shapes and the sizes of the defects.
Defects in composite materials mainly include: debonding, erosion, cracks, bubbles, inclusions, laminate delamination, core material deformation and the like. In addition, damage such as surface scratches, surface cracks, through-holes, and crush of the core material may occur during use. Some of these defects and damage can be observed with the naked eye on the surface of the composite material; and some defects exist in the composite material and cannot be directly detected by naked eyes, and at the moment, the internal defects can be identified by using a nondestructive detection technology.
At present, there are many nondestructive testing methods for composite materials, such as infrared thermal imaging testing, ultrasonic testing, X-ray testing, eddy current testing, and the like. The infrared thermal imaging nondestructive testing technology has the advantages of simple arrangement, low cost, high speed, no contact, no pollution, large testing area, intuitive result and the like, thereby being widely applied to the field of nondestructive testing of composite materials. However, due to the limitations of the sensitivity of the sensor, the resolution of the infrared thermal image, the background noise, the measurement noise, and the like, the original infrared thermal image generally has the problems of low signal-to-noise ratio, blurred defect edge, low detection accuracy, and the like, and it is difficult to directly identify the defect with naked eyes, so it is very necessary to enhance the defect feature information, reduce the background noise, and improve the defect feature extraction capability.
Researchers have proposed many effective thermal image data processing methods to improve the visibility of defects, such as pulse phase thermal imaging, penalty least squares, thermal imaging sequence reconstruction, principal component thermal imaging (PCT) and its extension, independent principal component thermal imaging, and so on. In recent years, an emerging deep learning algorithm is being widely applied to various fields. However, the application of the deep learning algorithm in the field of nondestructive testing is not mature at present. Moreover, many current research works only use a single thermal image data analysis method, and in a complex engineering environment, it is difficult to meet the actual requirements only using a single method.
Therefore, it is necessary to provide a non-destructive testing technique that is suitable for deep learning in infrared thermal imaging in combination with other models.
Disclosure of Invention
The invention aims to meet the actual requirement of nondestructive testing of internal defects of a composite material, and provides a composite material nondestructive testing method based on an iteration generation sparse principal component model aiming at the limitation of the existing nondestructive testing technology.
An infrared thermal imaging nondestructive testing method based on an iterative generation sparse principal component model comprises the following steps:
(1) collecting a plurality of infrared thermal images of an object to be detected to form an initial infrared thermal image data set;
(2) preprocessing the initial infrared thermal image data set to obtain a preprocessed infrared thermal image data set;
(3) carrying out denoising, dimension reduction and feature extraction on the preprocessed infrared thermal image data set by utilizing a PCT method (sparse principal component thermal imaging (SPCT) method) for applying sparse constraint improvement on loading to obtain a visual defect feature loading image;
(4) extracting R, G, B component loading images of the defect feature loading images, and calculating the signal-to-noise ratio of the component loading images:
when the signal-to-noise ratio meets the requirement, outputting a defect feature loading image or/and R, G, B component loading images;
when the signal-to-noise ratio does not meet the requirement, selecting a corresponding component loading image according to the light-dark relation between the defect in the original thermal image and the background, and fusing the component loading image and the preprocessed infrared thermal image to obtain a fused infrared thermal image;
(5) training the generative countermeasure neural network by using the fused infrared thermal images, selecting a plurality of generated false thermal images and the initial infrared thermal image to splice into an amplified infrared thermal image data set, and returning to the step (2) as the initial infrared thermal image data set.
In the step (1), a plurality of original infrared thermal images of the object to be detected can be acquired by using an existing acquisition system of the infrared thermal image data set, then uniform size cutting processing is carried out, a part of unnecessary background area is removed, only the initial infrared thermal images of the interested area containing all defects are reserved, and the initial infrared thermal image data set is formed. By this procedure, each infrared thermal image was made to be H × W in size. Wherein H represents an image length; w is the image width.
In the step (2), during pretreatment:
first, an initial infrared thermal image data set is converted into a three-dimensional thermal image data matrix corresponding to the size and number of initial infrared thermal images: i.e., an hxwxn three-dimensional data matrix, and converting the three-dimensional thermal image data matrix into an n-row two-dimensional thermal image data matrix, where n is the number of initial infrared thermal images. Each row in the dimensional thermal image data matrix is a row vector of length H x W. In this new two-dimensional matrix, each row represents one piece of raw thermal image data and each element represents a pixel value of the thermal image.
And then, carrying out z-score standardization treatment on the obtained two-dimensional thermal image data matrix to obtain a preprocessed infrared thermal image data set.
In the step (3), when a PCT method (i.e., SPCT method) for applying sparse constraint improvement to load is used to perform denoising, dimension reduction, and feature extraction on the infrared thermographic image data set, it can be represented as:
subject to PTP＝I
wherein the content of the first and second substances,
when the objective function takes the minimum valueThe values of the variables p, Q; i | · | purple windFRepresents F norm, | ·| non-conducting phosphor2Represents L2Norm, | · | luminance1Represents L1Norm, δ and λ are tuning parameters. P ═ p1p2......pk]Represents a load matrix comprising k load vectors, pkDenotes the kth load vector, Q is a sparse approximation of p, Q ═ Q1q2......qk]Containing k load vectors, qjThe j-th loading vector forming Q, and X is the preprocessed infrared thermal image data set obtained in the step (2); and I is an identity matrix.
The invention is realized by introducing an L2The norm further regularization solves the problem of lack of interpretability that exists for PCT that lacks load-imposed sparse constraint improvement.
The SPCT method is utilized to process the preprocessed two-dimensional thermal image data matrix to generate a sparse loading matrix Q, so that pixels with almost same surface temperature of the thermal image are combined together. And each column of the loading matrix Q can be visualized into loading images, so that most defect characteristic information is ensured to be displayed in the loading images.
One or more loaded images containing all or most of the defect features may be obtained by step (3). One or more of the loaded images with the most prominent defective features can be selected for subsequent fusion. In general, a visual feature defect loading image containing all defect features can be directly obtained.
Due to the fact that the original infrared thermal image data after preprocessing are subjected to normalization processing, the brightness range of the obtained loading image is far smaller than 0-255. And decomposing a three-dimensional matrix representing the three-channel loaded image with the size of H multiplied by W multiplied by 3 obtained after SPCT into three groups of different two-dimensional matrixes H multiplied by W according to R, G, B channels. And then visualizing the three groups of two-dimensional matrixes respectively to obtain three single-channel images of the loaded image. The brightness range can be expanded to 0-255 by extracting R, G, B components of the loaded image, so that the contrast between the defect area and the image background is obviously improved, the signal to noise ratio is further improved, and the recognition efficiency is accelerated.
Preferably, in the step (4), the component loading image of the defect feature loading image with the most obvious defect feature is selected to be fused with the preprocessed infrared thermal image. Preferably, we select the feature defect loading image containing the most feature defect information as the loading image of this step.
When a single defect feature loading image containing all defect features does not exist, two or more defect feature loading images can be selected to be fused with the preprocessed infrared thermal image.
In step (4), the selected number of false thermal images is equivalent to the number of thermal images in the initial data set, and the images are loaded. Preferably, the ratio of the number of selected false thermal images to the number of thermal images in the initial data set is 1: 0.8-1.2.
Preferably, the method of the present invention can be used for defect detection of a plurality of materials, such as various composite materials (e.g., carbon fiber reinforced composite materials), single component materials, and the like.
The invention provides a composite material infrared thermal imaging nondestructive testing method based on an iteration generated sparse principal component model. The problem of insufficient original thermal image data set can be solved by generating a series of false thermal images to expand the data set by utilizing a generative confrontation neural network which is one of deep learning algorithms which are gradually raised in recent years. There are many other problems in the original thermal image, such as low contrast between defect and background, high noise in background and measurement, low definition, low signal-to-noise ratio, etc. if the original thermal image is directly used to input into the generative antagonistic neural network, the generated false thermal image still has many problems as the original thermal image. In order to solve the problems, the invention utilizes a sparse principal component thermal imaging (SPCT) method to perform dimensionality reduction, denoising, feature extraction and reconstruction visualization on an original thermal image data set, and can reduce the dimensionality of dozens of original thermal images into dozens of thermal images containing defect feature information, wherein all defect features are integrated in one thermal image, most background noise is removed, and the signal-to-noise ratio is improved; then R, G, B components of the thermal image integrated with all the defect characteristic information are respectively extracted, and the contrast between the defect and the background is further improved; the signal-to-noise ratio of the thermal image is further improved at the moment; then selecting one of the thermal images of the R component or the B component to perform image fusion with the original thermal image, wherein the fused thermal image contains both the information of the original thermal image and the enhanced defect characteristic information; and then the fused thermal image data set is used as the input of a generating type countermeasure neural network for training, a series of false thermal images and original thermal images are generated in the training process and spliced into an extended data set, and then SPCT is carried out to form a closed loop iteration process, so that the signal-to-noise ratio of the thermal images is obviously improved finally, and the method has a good practical application prospect.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a diagram of a generative antagonistic neural network;
FIG. 3 is a plurality of raw infrared thermal images collected in an example;
FIG. 4 is a diagram illustrating a plurality of principal component images obtained in step three of the present embodiment;
FIG. 5 is a defect feature loading image and its corresponding component loading image taken from FIG. 4;
FIG. 6 is a plurality of principal component images obtained using an augmented dataset according to an embodiment;
FIG. 7 is a defect feature loading image and its corresponding component loading image taken from FIG. 6.
Detailed Description
The technical solution of the present invention will be further explained with reference to the accompanying drawings in the following embodiments. It is to be understood that the specific embodiments described are merely illustrative of some, and not restrictive, of the embodiments of the invention. Any other embodiments obtained by persons skilled in the art without inventive work based on the embodiments of the present invention shall fall within the scope of protection of the present invention.
Example 1
Taking the infrared thermal imaging nondestructive testing process for the internal defects of the carbon fiber reinforced composite material as an example, the invention is further explained as follows:
the invention provides an infrared thermal imaging nondestructive testing method based on an iteration generation sparse principal component model, as shown in figure 1, the method comprises the following steps:
the method comprises the following steps: a data set is collected. And setting an acquisition system of the infrared thermal image data set, and acquiring infrared thermal images of a plurality of objects to be detected. And performing uniform size cutting processing on the infrared thermal image data set, removing a part of unnecessary background area, and only reserving the interested area containing all defects, so that the size of each infrared thermal image is H multiplied by W. Wherein H represents an image length; w is the image width. After the step is finished, obtaining an initial infrared thermal image to form an initial infrared thermal image data set. FIG. 3 is a partial initial infrared thermographic image acquired in this example.
Step two: and (4) preprocessing data. The initial infrared thermal image dataset is converted into a three-dimensional data matrix of size H x W x n. Wherein n represents the number of infrared thermal images. The three-dimensional thermal image data matrix is converted to a two-dimensional thermal image data matrix of n rows, where each row is a row vector of length H W. In this new two-dimensional matrix of thermal image data, each row represents one piece of raw thermal image data and each element represents a pixel value of the thermal image. The two-dimensional thermographic data matrix is then z-score normalized such that the mean of each variable is 0 and the variance is 1.
Step three: and (5) feature extraction. And denoising, dimension reduction, feature extraction and visualization are carried out on the preprocessed infrared thermal image by using an SPCT algorithm, and the signal-to-noise ratio is calculated.
First, the problem of extracting a Principal Component (PC) using principal component thermal imaging (PCT) can be expressed as follows:
wherein the content of the first and second substances,
representing the value of P when the objective function is maximum; subject to P non-woven 21 or less represents a constraint condition, P is a load vector with dimension H multiplied by W multiplied by 1, X is the thermal image data matrix after expansion, | | |2Represents L2And (4) norm. The first PC (t denotes the first PC) can be calculated by t-XP, which is a linear combination of X. Other PCs can be computed in a similar manner, with their relationship to each other being orthogonal. Generally, the first few PCs are sufficient to extract most of the feature information in the large volume of thermal image data.
Then, by improving the PCT method by applying a sparsity constraint to each loading, the final result can be made more explanatory and the background noise and measurement noise can be further eliminated on a PCT basis. In general, when computing the load vector of X by loading the PCT method that imposes sparse constraint refinement, the optimization problem to obtain the first PC can be generally expressed as:
subject to ||P||2≤1 (2)
where γ is an optimization parameter that can adjust the sparsity of P. I | · | purple wind0Is L0A norm, which represents the number of non-zero elements in the vector, is used to represent sparsity.
However L0The norm generally does not account for population effects, which would likely result in the optimization problem in equation (2) to randomly select sparse loading of variables. This condition often lacks interpretability by introducing an L2Further regularization of the norm solves this problem. And further obtaining an expression of a final loading sparse constraint application improved PCT method (SPCT):
subject to PTP＝I
wherein the content of the first and second substances,
represents the value of the variable p, Q when the objective function takes the minimum value; i | · | purple windFIndicating that F-norms δ and λ are tuning parameters. P ═ p1p2......pk]Represents a load matrix comprising k load vectors, pkDenotes the kth load vector, Q is a sparse approximation of p, Q ═ Q1q2......qk]Containing k load vectors, qjIs the jth load vector that makes up Q.
The two-dimensional thermal image data matrix is processed by the SPCT method to generate a sparse loading matrix Q (i.e., one or more PCs ahead) so that pixels with almost the same temperature on the thermal image surface are combined together. And visualizing each column of the sparse loading matrix Q into a loading image to ensure that most defect characteristic information is displayed in the loading images, and thus obtaining a final defect characteristic loading image.
Through the third step, one or more defect feature loading images containing most or all of the defect feature information can be obtained.
Step four: r, G, B components are extracted. And selecting one or more defect feature loading images with the most obvious defect features, respectively extracting R, G, B component loading images of the defect feature loading images, visualizing the images, and calculating the signal-to-noise ratio.
To objectively evaluate the performance of the model, a wide range of signal-to-noise ratio (SNR) metrics are employed in the field of thermal image data processing. The absolute value of SNR is calculated as follows
Wherein M isdefMean pixel value, M, representing defective areanRepresents the average pixel value of the background region, and σnRepresenting the standard deviation of the pixel values of the background area. The SNR is a dimensionless index reflecting the contrast between the defect area and the background area. In general, a greater SNR indicates greater signal strengthThe stronger the ability to identify defects. In this embodiment, we choose to calculate the signal-to-noise ratio of the component-loaded image as the output condition. And when the signal-to-noise ratio of one component loading image meets the requirement, outputting a defect feature loading image and a corresponding component loading image.
Because the original infrared thermal image data (namely the two-dimensional thermal image data matrix) processed by the SPCT is subjected to normalization processing, the brightness range of the obtained loaded image is far less than 0-255. And decomposing a three-dimensional matrix representing the three-channel loaded image with the size of H multiplied by W multiplied by 3 obtained after SPCT into three groups of different two-dimensional matrixes H multiplied by W according to R, G, B channels. And then visualizing the three groups of two-dimensional matrixes respectively to obtain three single-channel images of the loaded image, namely R, G, B component loaded images. The brightness range can be expanded to 0-255 by extracting R, G, B components of the loaded image, so that the contrast of the defect area and the image background is obviously improved.
Step five: and (5) image fusion. Selecting a component loading image of an R channel or a B channel, and carrying out image fusion with the preprocessed infrared thermal image according to different weights, wherein the weight loading image can be expressed as follows:
F(x)＝αf0(x)+(1-α)f1(x),α∈(0,1)
wherein x represents a pixel point in the image, α represents a weight, f0(x)、f1(x) Representing a pre-processed infrared thermal image or a component-loaded image, respectively.
Step six: and (5) data amplification. And training the fused infrared thermal images as the input of a generating type antagonistic neural network, finally generating a plurality of false thermal images, and finally selecting a plurality of false thermal images (the number of the false thermal images is equal to that of the initial infrared thermal images, for example, is 0.8-1.2 times of the number of the initial infrared thermal images) to be spliced with the cut initial thermal images to form an augmented data set for training the generating type antagonistic neural network.
Specifically, the general structure of the generative antagonistic neural network is shown in fig. 2 (LOSS in the figure represents a LOSS function), and the generative antagonistic neural network comprises two parts, namely a generative model G and a discriminator model D, which game and learn with each other, and finally generate a very good result.
The input of the generator model is a line of random noise, the output of the generator model is a false infrared thermal image, and the purpose of continuous training of the generator model is to generate the infrared thermal image similar to a real data set as much as possible, so that the discriminator model cannot judge the authenticity; the discriminator model is a binary model, the input of the discriminator model is an infrared thermal image in a real data set and a false thermal image generated by the generator model, the output of the discriminator model is used for judging whether the input infrared thermal image is real or not, and the aim of continuous training of the discriminator model is to judge the authenticity of the input image as far as possible.
In the process that the generator model and the discriminator model mutually game, the whole network is continuously optimized. When the discrimination result of the discriminator on the input is 0.5, the discriminator can not judge whether the input image is true or false, and Nash balance is achieved, and the generation type antagonistic neural network can generate the infrared thermal image which is false or false.
The loss function/objective function of the resulting antagonistic neural network is expressed as follows:
this is a maximum and minimum optimization problem, optimizing D first and then G, and is essentially two optimization problems, which can be broken down into the following two formulas:
a first stage: optimization of D
When the real thermal image label is 1 and the generated thermal image label is 0, it is expected that the real thermal image is closer to 1, i.e. D (x) tends to 1, and the generated thermal image is closer to 0, i.e. D (g (z)) tends to 0, so that the objective function is increased.
And a second stage: optimization of G
Since the real thermal image tag is 1 and the generated thermal image tag is 0, it is expected that D (g (z)) tends toward 1, so that the objective function is reduced.
During training, the generator needs to fix the discriminator, so that the two modules are trained respectively and alternately. Gaming occurs because the discriminator wants to discriminate the generated false thermal images, and the generator continually optimizes the network so that the generated thermal images are more and more realistic until nash balance is achieved.
Step seven: and (6) circularly iterating. And taking the amplified infrared thermal image data set as an initial infrared thermal image data set, substituting the initial infrared thermal image data set into the second step to the sixth step for continuous processing to form a closed loop iteration structure. And when the signal-to-noise ratio calculated in the fourth step reaches the expectation, the loop can be exited, and the result is output.
The foregoing is merely illustrative of the overall structure of the invention, and many other embodiments of the invention are possible, and those skilled in the art will appreciate that various modifications, additions and substitutions are possible, without departing from the scope and spirit of the invention as disclosed in the accompanying claims.
Claims (7)
1. An infrared thermal imaging nondestructive testing method based on an iteration generation sparse principal component model is characterized by comprising the following steps:
(1) collecting a plurality of infrared thermal images of an object to be detected to form an initial infrared thermal image data set;
(2) preprocessing the initial infrared thermal image data set to obtain a preprocessed infrared thermal image data set;
(3) denoising, dimensionality reduction and feature extraction are carried out on the preprocessed infrared thermal image data set by using a sparse principal component thermal imaging method, and a visual defect feature loading image is obtained;
(4) extracting R, G, B component loading images of the defect feature loading images, and calculating the signal-to-noise ratio of the component loading images:
when the signal-to-noise ratio meets the requirement, outputting a defect feature loading image or/and R, G, B component loading images;
when the signal-to-noise ratio does not meet the requirement, selecting a corresponding component loading image according to the light-dark relation between the defect in the original thermal image and the background, and fusing the component loading image and the preprocessed infrared thermal image to obtain a fused infrared thermal image;
(5) training the generative countermeasure neural network by using the fused infrared thermal images, selecting a plurality of generated false thermal images and the initial infrared thermal image to splice into an amplified infrared thermal image data set, and returning to the step (2) as the initial infrared thermal image data set.
2. The infrared thermal imaging nondestructive testing method based on iterative generation of sparse principal component model according to claim 1, wherein in step (1), a plurality of original infrared thermal images of the object to be tested are firstly collected, then a uniform size cutting process is performed to remove a part of unnecessary background area, only the initial infrared thermal image of the interested area containing all defects is reserved, and the initial infrared thermal image data set is composed.
3. The infrared thermal imaging nondestructive testing method based on the iterative generation of the sparse principal component model as claimed in claim 1, wherein in the step (2), during the preprocessing:
firstly, converting an initial infrared thermal image data set into a three-dimensional thermal image data matrix corresponding to the size and the number of the initial infrared thermal images, and converting the three-dimensional thermal image data matrix into a two-dimensional thermal image data matrix with n rows, wherein n is the number of the initial infrared thermal images;
and then, carrying out z-score standardization treatment on the obtained two-dimensional thermal image data matrix to obtain a preprocessed infrared thermal image data set.
4. The infrared thermal imaging nondestructive testing method based on iterative generation of sparse principal component model as claimed in claim 1, wherein in step (3), denoising, dimension reduction and feature extraction are performed on the infrared thermal image data set by using the sparse principal component thermal imaging method, which can be expressed as:
subject to PTP＝I
wherein the content of the first and second substances,
represents the value of the variable p, Q when the objective function takes the minimum value; i | · | purple windFRepresents F norm, | ·| non-conducting phosphor2Represents L2Norm, | · | luminance1Represents L1Norm, δ and λ are tuning parameters. P ═ p1p2......pk]Represents a load matrix comprising k load vectors, pkDenotes the kth load vector, Q is a sparse approximation of p, Q ═ Q1q2......qk]Containing k load vectors, qjThe j-th loading vector forming Q, and X is the preprocessed infrared thermal image data set obtained in the step (2); and I is an identity matrix.
5. The infrared thermal imaging nondestructive testing method based on iterative generation of sparse principal component model according to claim 1, wherein in step (4), the component loading image of the defect feature loading image having the most obvious defect feature is selected to be fused with the preprocessed infrared thermal image.
6. The infrared thermal imaging nondestructive testing method based on iterative generation of a sparse principal component model according to claim 5, wherein when there is no single defect feature loading image containing all defect features, two or more defect feature loading images can be selected to be fused with the preprocessed infrared thermal image.
7. The infrared thermographic nondestructive testing method based on iterative generation of a sparse principal component model of claim 1, wherein in step (4), the selected number of false thermal images is equivalent to the number of thermal images in the initial dataset, and the images are loaded.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110963492.4A CN113702439A (en) | 2021-08-20 | 2021-08-20 | Infrared nondestructive testing method based on iterative generation of sparse principal component model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110963492.4A CN113702439A (en) | 2021-08-20 | 2021-08-20 | Infrared nondestructive testing method based on iterative generation of sparse principal component model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113702439A true CN113702439A (en) | 2021-11-26 |
Family
ID=78653771
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110963492.4A Pending CN113702439A (en) | 2021-08-20 | 2021-08-20 | Infrared nondestructive testing method based on iterative generation of sparse principal component model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113702439A (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150097944A1 (en) * | 2011-03-31 | 2015-04-09 | Sage Electrochromics, Inc. | System and method for detecting and repairing defects in an electrochromic device using thermal imaging |
| CN106875345A (en) * | 2016-12-26 | 2017-06-20 | 浙江工业大学 | Non local TV model images denoising method based on singular value weighting function |
| CN107122725A (en) * | 2017-04-18 | 2017-09-01 | 深圳大学 | A kind of face identification method and its system based on joint sparse discriminant analysis |
| CN110414362A (en) * | 2019-07-02 | 2019-11-05 | 安徽继远软件有限公司 | Electric power image data augmentation method based on production confrontation network |
| CN111242873A (en) * | 2020-01-21 | 2020-06-05 | 北京工业大学 | Image denoising method based on sparse representation |
| CN111325748A (en) * | 2020-03-20 | 2020-06-23 | 哈尔滨工业大学 | Infrared thermal image nondestructive testing method based on convolutional neural network |
| CN112396110A (en) * | 2020-11-20 | 2021-02-23 | 南京大学 | Method for generating anti-cascade network augmented image |
| CN112461892A (en) * | 2020-11-02 | 2021-03-09 | 浙江工业大学 | Infrared thermal image analysis method for nondestructive detection of composite material defects |
| CN112507966A (en) * | 2020-12-23 | 2021-03-16 | 杭州电子科技大学 | Color face recognition method based on quaternion non-convex penalty sparse principal component analysis |
| CN112798648A (en) * | 2020-12-09 | 2021-05-14 | 浙江工业大学 | Composite material defect detection method based on generated kernel principal component thermal image analysis |
| 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
- 2021-08-20 CN CN202110963492.4A patent/CN113702439A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150097944A1 (en) * | 2011-03-31 | 2015-04-09 | Sage Electrochromics, Inc. | System and method for detecting and repairing defects in an electrochromic device using thermal imaging |
| CN106875345A (en) * | 2016-12-26 | 2017-06-20 | 浙江工业大学 | Non local TV model images denoising method based on singular value weighting function |
| CN107122725A (en) * | 2017-04-18 | 2017-09-01 | 深圳大学 | A kind of face identification method and its system based on joint sparse discriminant analysis |
| CN110414362A (en) * | 2019-07-02 | 2019-11-05 | 安徽继远软件有限公司 | Electric power image data augmentation method based on production confrontation network |
| CN111242873A (en) * | 2020-01-21 | 2020-06-05 | 北京工业大学 | Image denoising method based on sparse representation |
| CN111325748A (en) * | 2020-03-20 | 2020-06-23 | 哈尔滨工业大学 | Infrared thermal image nondestructive testing method based on convolutional neural network |
| CN112461892A (en) * | 2020-11-02 | 2021-03-09 | 浙江工业大学 | Infrared thermal image analysis method for nondestructive detection of composite material defects |
| CN112396110A (en) * | 2020-11-20 | 2021-02-23 | 南京大学 | Method for generating anti-cascade network augmented image |
| CN112798648A (en) * | 2020-12-09 | 2021-05-14 | 浙江工业大学 | Composite material defect detection method based on generated kernel principal component thermal image analysis |
| CN112507966A (en) * | 2020-12-23 | 2021-03-16 | 杭州电子科技大学 | Color face recognition method based on quaternion non-convex penalty sparse principal component analysis |
| CN113160195A (en) * | 2021-04-28 | 2021-07-23 | 浙江工业大学 | Composite material defect nondestructive testing method based on thermal imaging |
Non-Patent Citations (1)
| Title |
|---|
| JING JIE，SHIQING DAI，BEIPING HOU等: "Defect Detection in Composite Products Based on Sparse Moving Window Principal Component Thermography", ADVANCES IN POLYMER TECHNOLOGY, vol. 2020, 7 February 2020 (2020-02-07), pages 1 - 12 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shipway et al. | Automated defect detection for fluorescent penetrant inspection using random forest | |
| CN110108754B (en) | Structured sparse decomposition-based light-excitation infrared thermal imaging defect detection method | |
| CN103234976B (en) | Based on the online visible detection method of tricot machine Fabric Defect of Gabor transformation | |
| CN103955922B (en) | Method for detecting flaws of printed fabric based on Gabor filter | |
| CN112070727B (en) | Metal surface defect detection method based on machine learning | |
| CN107229918A (en) | A kind of SAR image object detection method based on full convolutional neural networks | |
| CN110930387A (en) | Fabric defect detection method based on depth separable convolutional neural network | |
| CN112798648B (en) | Composite material defect detection method based on generated kernel principal component thermal image analysis | |
| CN106530288A (en) | Fabric defect detection method based on deep learning algorithm | |
| CN111161224A (en) | Casting internal defect grading evaluation system and method based on deep learning | |
| Xiao et al. | Development of a CNN edge detection model of noised X-ray images for enhanced performance of non-destructive testing | |
| CN112488211A (en) | Fabric image flaw classification method | |
| CN110895814A (en) | Intelligent segmentation method for aero-engine hole detection image damage based on context coding network | |
| CN115170529A (en) | Multi-scale tiny flaw detection method based on attention mechanism | |
| Hu et al. | Automatic defect detection from X-ray scans for aluminum conductor composite core wire based on classification neutral network | |
| CN113506239A (en) | Strip steel surface defect detection method based on cross-stage local network | |
| Fan et al. | Application of YOLOv5 neural network based on improved attention mechanism in recognition of Thangka image defects | |
| CN113838040A (en) | Detection method for defect area of color texture fabric | |
| Chen et al. | A mixed samples-driven methodology based on denoising diffusion probabilistic model for identifying damage in carbon fiber composite structures | |
| CN117274899A (en) | Storage hidden danger detection method based on visible light and infrared light image feature fusion | |
| Sahasrabudhe et al. | Structured spatial domain image and data comparison metrics | |
| CN113702439A (en) | Infrared nondestructive testing method based on iterative generation of sparse principal component model | |
| Ghamisi et al. | Anomaly detection in automated fibre placement: learning with data limitations | |
| Yang et al. | Defect detection scheme of pins for aviation connectors based on image segmentation and improved RESNET-50 | |
| CN116524356A (en) | Ore image small sample target detection method and system |
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 | ||
| CB02 | Change of applicant information |
Address after: 310023 No. 318 stay Road, Xihu District, Zhejiang, Hangzhou Applicant after: ZHEJIANG University OF SCIENCE AND TECHNOLOGY Address before: 310023 No. 318, Ho Ho Road, Hangzhou, Zhejiang Applicant before: ZHEJIANG University OF SCIENCE AND TECHNOLOGY |
|
| CB02 | Change of applicant information |