Stepped eddy current thermal imaging defect reconstruction method based on SM-IGABEA model

Abstract

The invention discloses a stepped eddy current thermal imaging defect reconstruction method based on an SM-IGABEA model, and the method comprises the steps: firstly carrying out the SVD reconstruction and noise reduction of a collected infrared thermal image sequence, then calculating the third-order statistic skewness of all pixel points, and compressing and reconstructing a skewness image; in order to reduce the influence of non-uniform background of an image on defect measurement precision, a genetic algorithm is adopted to search for a maximum scale factor capable of enabling the background to become uniform, a series of non-uniform backgrounds are obtained by calculating convolution of a skewness image and each scale factor, and the backgrounds are subtracted from the skewness image to obtain a series of background-removed images; then, compression reconstruction of the background-removed image is realized through a third-order nonlinear weight; according to the reconstructed image, defect form segmentation can be realized through first-order differential, and defect depth measurement can be realized through a skewness mean value of a segmentation domain.

Description

Technical Field

[0001] The present invention belongs to the technical field of defect detection, and in particular relates to a step-type eddy current thermal imaging defect reconstruction method based on an SM-IGABEA model. Background Art

[0002] Oil storage tanks are key equipment for crude oil precipitation, storage and transportation. During long-term use, they are corroded by the internal environment medium. This causes local thinning of the tank body or water permeation and perforation of the tank body, which seriously affects the safe production of the enterprise. The oil storage tank is large in size, and it is difficult and expensive to clean and overhaul it at one time. There has been a lack of efficient and reliable non-destructive testing methods for the inspection of oil storage tanks. Stepped eddy current thermal imaging technology has broad application prospects in non-destructive testing. It realizes quantitative analysis through the relationship between defect information in the sample and the surface temperature field distribution. It has the advantages of no pollution, non-contact, and can be inspected in some places with explosion-proof requirements.

[0003] However, due to the weak excitation conditions of the step eddy current thermal imaging technology, its temperature rise is low and the heating time is long, and the blurring effect makes it difficult to quantify the defects it detects. The size of the defect will affect the tensile strength of the tank, and in severe cases, it will cause leakage in the equipment. Therefore, the quantitative evaluation of the defect size is of great significance for assessing the degree of damage. In the past, the depth of the defect was measured by the time peak of the temperature response on the surface of the tank, but this method has extremely high requirements for the frame rate of the infrared thermal imager, and this type of infrared thermal imager is often very expensive, which brings difficulties to the application of SECT technology in engineering. On the other hand, in actual industrial inspections, the surface of the inspected object is often accompanied by problems such as inconsistent emissivity, blurring effects caused by spatial thermal radiation and long-term heating, which makes the background domain of the compressed image obtained by the mainstream reconstruction method uneven, the defect edge is blurred, and it is difficult to achieve effective segmentation of the defect. Therefore, reducing the frame rate requirements of the infrared thermal imager and eliminating the influence of measurement noise and background unevenness are of great research value for the reconstruction of defects inside the tank. Summary of the invention

[0004] The purpose of the present invention is to address the problems in the above-mentioned prior art and provide a stepped eddy current thermal imaging defect reconstruction method based on the SM-IGABEA model. The SM-IGABEA model is used to reconstruct the image to reduce the frame rate requirements of the infrared thermal imager, eliminate the influence of measurement noise and background unevenness, achieve defect segmentation through the first-order differential of the reconstructed image, and achieve defect depth measurement through the segmentation domain mean and the geometric characteristics of the defect.

[0005] The technical solution adopted by the present invention is as follows:

[0006] The step-type eddy current thermal imaging defect reconstruction method based on the SM-IGABEA model is characterized by comprising the following steps:

[0007] S1, obtaining infrared thermal image sequence;

[0008] The specimen with known defects at room temperature is heated by using the stepped eddy current thermal imaging nondestructive testing platform, and P frames of infrared thermal images are collected during the entire heating stage to form an infrared image time series S, where the size of each frame of infrared thermal image is M×N, where M and N are the length and height of the infrared thermal image, respectively.

[0009] S2, singular value decomposition denoising and reconstruction;

[0010] S2.1. Discretize into multiple thermal images X by time i , a single image X of shape M×N i Decomposed into the product of three matrices through the SVD method, namely;

[0011] X new i =URV T

[0012] S2.2, then select the first k largest singular values ​​to reconstruct and obtain the de-noised and reconstructed discrete heat map X new i , so that the noise information on a single heat map is eliminated to a great extent, that is;

[0013]

[0014] S2.3, decomposition and denoising reconstructed discrete heat map X new i In chronological order, reassemble into a denoised infrared image time series S new ;

[0015] S3, the compression of thermal image sequence is realized through the third-order statistic skewness (i.e., SM model);

[0016] S3.1. Transform the infrared image time series S new Decompose into one-dimensional time series data P(x,y) according to the pixel position, and calculate the skewness value Ske(P(x,y)) corresponding to each pixel P(x,y) point through the third-order statistics, that is;

[0017]

[0018] in, is the mean value of the one-dimensional time series temperature data of pixel P(x,y), and N is the number of frames;

[0019] S3.2. The skewness values ​​Ske(P(x,y)) of all pixels are spliced ​​and reorganized according to their positions (x,y) to achieve compression of the thermal image sequence. The compressed matrix is ​​the skewness matrix Ske.

[0020] S4, using the improved Gaussian background elimination algorithm (IGABEA) method to achieve the removal of uneven background;

[0021] S4.1. Calculate the convolution of the skewness matrix Ske and the Gaussian function with scale factor σ ranging from 1 to n, that is;

[0022]

[0023] Among them, σ i is the i-th size factor, L i (x,y,σ i ) is the background image estimated using the i-th size factor;

[0024] S4.2. Estimation of the background matrix L through the skewness matrix Ske and different size factors i (x,y,σ i ) to obtain the skewness image sequence D after removing the uneven background i (x,y,σ i ),Right now;

[0025] D i (x,y,σ i )=|Ske(P(x,y))-L i (x,y,σ i )|

[0026] S4.3, using the third-order nonlinear convergence factor to achieve the skewness image sequence D after removing the uneven background i (x,y,σ i ) compression and reconstruction, that is;

[0027]

[0028] Among them, D new i (x, y) is the value of the pixel position (x, y) after compression and reconstruction with the background removed, which is reconstructed into a matrix form D according to the position new ;

[0029] S5, genetic algorithm to achieve the construction of the best compression reconstruction map;

[0030] S5.1. Calculate S4.3 in the horizontal and vertical directions to reconstruct the skewness matrix D new The first-order differentials of are dx and dy respectively. Find the pixel point where dx and dy are both 0 and the left and right or upper and lower differential values ​​of the extracted position have opposite signs. The location of this pixel point is the defect center location, and D at this location is extracted. new The value is Ske d matrix;

[0031] S5.2. Extract the pixel points where both dx and dy are 0 and the left and right or top and bottom differential values ​​of the extracted position have the same sign. Extract the D new The value is Ske s The matrix is ​​used as the signal intensity of the background domain, and the standard deviation σ of these pixels is calculated ref ;

[0032] S5.3. Calculate the signal-to-noise ratio (SNR) of each defect in the compressed reconstruction result image under the current scale factor σ, that is;

[0033]

[0034] S5.4, setting the optimization variable of the genetic algorithm to the scale factor σ, and the optimization target to the signal-to-noise ratio SNR of the defect domain;

[0035] S5.5, repeat S4.1-S5.3 until the genetic algorithm converges, thereby obtaining the defect-enhanced matrix D new ;

[0036] S6, defect segmentation based on first-order differential mean;

[0037] S6.1. Calculate the enhanced matrix D new The first-order differentials d in the longitudinal and transverse directions through the defect center R x and d R y;

[0038] S6.2. Extract the first-order differential d R x and d R The four positions of the maximum and minimum values ​​in y, extract the corresponding D of these four positions new The reconstruction value of

[0039] S6.3. Calculate the mean value V of the four extracted reconstruction values mean , in the matrix D new Higher than V mean The defect area is considered to be assigned a value of 1. new Below V mean The area where the background is located is assigned a value of 0, thus completing the defect segmentation task;

[0040] S7, measurement of defect cutting plane;

[0041] S7.1. Calculate the first-order differential d in S6.2. R The distance between the maximum and minimum values ​​of x is scaled according to the ratio of the pixel point to the actual distance to obtain the width W of the defect d ;

[0042] S7.2. Calculate the first-order differential d in S6.2.R The distance between the maximum and minimum values ​​of y is scaled according to the ratio of the pixel point to the actual distance to obtain the height H of the defect d ;

[0043] S8, defect depth reconstruction based on ternary linear regression;

[0044] S8.1. Extract the defect area position in S (6.3) and calculate the skewness mean Ske of the defect area in the skewness matrix Ske in S (3.2) mean d , extract the width W of S7.1-S7.2 d With height H d ;

[0045] S8.2. Will Ske mean d , W d With H d Input the fitting model to achieve defect depth D d of reconstruction, i.e.;

[0046]

[0047] Where T s is the thickness of the steel plate.

[0048] Furthermore, in S1, the frequency of the alternating current is 50-200 Hz, the current is less than or equal to 20 A, and the sampling frequency of the thermal imager is less than or equal to 10 Hz.

[0049] Further, in S5.1, if the first-order differential values ​​in the horizontal and vertical directions are both 0, and the first-order differential values ​​at the left and right or upper and lower pixel positions have different signs, it indicates that the defect center is at that position.

[0050] Furthermore, in S6.3, the average value V greater than the four reconstructed values mean , in image D new Higher than V mean The area where the defect is located is assigned a value of 1. new Below V mean The area considered to be the background is assigned a value of 0.

[0051] Furthermore, in S8.2, the defect domain skewness mean Ske mean d The depth of the defect is measured by combining the length and width of the defect.

[0052] The beneficial effects of the present invention are:

[0053] The present invention proposes a step-type eddy current thermal imaging defect reconstruction method based on the SM-IGABEA model. The SM-IGABEA model is used to reconstruct the image, effectively eliminating the influence of measurement noise and background unevenness, and realizing the reconstruction and quantification of defects. It also reduces the dependence of the measurement method on the high frame rate of the infrared thermal imager, effectively reducing the application cost of the technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flowchart of a step-type eddy current thermal imaging defect reconstruction method based on the SM-IGABEA model provided in Example 1 of the present invention;

[0055] Figure 2 The step-type eddy current thermal imaging test system used in Example 1 of the present invention;

[0056] Figure 3 A schematic diagram of the dimensions of an exciter in a stepped eddy current thermal imaging test system used in Example 1 of the present invention;

[0057] Figure 4 Schematic diagram of defects of the material under test in Example 1 of the present invention;

[0058] Figure 5 The thermal imaging image with the maximum thermal contrast of the material under test in Example 1 of the present invention;

[0059] Figure 6 A background inhomogeneity map reconstructed from the skewness of the material under test in Embodiment 1 of the present invention;

[0060] Figure 7 This is a defect reconstructed image of the material being tested in Example 1 of the present invention.

[0061] Figure 8 A three-dimensional stereogram reconstructed from defects of the material being tested in Example 1 of the present invention.

[0062] Fig. 9 This is a defect segmentation image of the material being tested in Example 1 of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0064] Embodiment 1:

[0065] This embodiment provides a step-by-step eddy current thermal imaging defect reconstruction method based on the SM-IGABEA model. The process is as follows: Figure 1 As shown, the following steps are included:

[0066] S1. Based on Figure 2 The eddy current thermography test system shown in the figure has coil dimensions such as Figure 3 As shown, a 1.8 mm solid copper coil is wound around a ferrite yoke with an outer diameter of 50 mm × 25 mm. The number of turns of the coil is 360. The heating source passes an alternating current with a frequency of 50 Hz and an amplitude of 10 A into the coil. The size of the yoke is as follows: Figure 3 As shown, it consists of two identical detection cross arms and two identical coil cross arms. The detection cross arm is 450mm long, 25mm wide and 25mm high. The coil arm is 250mm long, 50mm wide and 250mm high. The two poles are 200mm apart. The size of the sample to be tested is as follows: Figure 4 As shown, the material under test is pressure vessel steel Q235, with a size of 600mm×400mm; in this example, the area of ​​the material under test (i.e., the ROI area) located between the two inner ends of the coil and the two poles of the magnetic yoke is excited, wherein the size of the ROI area is 200mm×300mm, the corrosion defect is composed of four sides with the same length of 15mm, and the remaining thickness of the steel plate is 3.5mm, 4.0mm, 4.5mm and 5.0mm respectively; the original thermal image sequence S of the material under test in the ROI area changing with time is collected by an infrared thermal imager, the frame rate is 10fps, and the size of each frame image is M×N; wherein, M=400, N=700; the image sequence recording time consists of an excitation time of 30S and a cooling time of 15S;

[0067] S2, singular value decomposition denoising reconstruction, the collected infrared thermal image time data sequence S, the maximum thermal contrast Figure 5 As shown, there is a lot of noise, which seriously affects the reconstruction of defects;

[0068] S2.1. Discretize into multiple thermal images X by time i , a single image X of shape M×N i Decomposed into the product of three matrices through the SVD method, namely;

[0069] X i =U i R i V i T

[0070] Among them, U i is the ith discrete heatmap X i The M×M orthogonal matrix, R i is the ith discrete heatmap X i In the M×N diagonal matrix (diagonal elements are singular values, and the rest of the elements are 0), V i T The i-th discrete heat map X i The transpose of V in V i is an N×N orthogonal matrix;

[0071] S2.2, then select the first k largest singular values ​​to reconstruct and obtain the de-noised and reconstructed discrete heat map X new i , so that the noise information on a single heat map is eliminated to a great extent, that is;

[0072]

[0073] S2.3, decomposition and denoising reconstructed discrete heat map X new i In chronological order, reassemble into a denoised infrared image time series S new ;

[0074] S new =[X new 1,X new 2,X new 3...X new 450 ]

[0075] S3, the compression of thermal image sequence is realized through the third-order statistic skewness (i.e., SM model);

[0076] S3.1. The denoised infrared image time series S new Decompose the one-dimensional time series data P(x,y) according to the pixel position, and calculate the skewness value Ske(P(x,y)) corresponding to the point P(x,y) through the third-order statistic skewness, that is;

[0077]

[0078] in, is the mean temperature of P(x,y), N is the number of frames;

[0079] S3.2. The skewness values ​​Ske(P(x,y)) of all pixels are spliced ​​and reorganized according to their positions (x,y) to achieve compression of the thermal image sequence. The compressed matrix is ​​the skewness matrix Ske, see Skewness Compression Reconstruction Figure 6 As shown in the figure, although the noise is reduced, the background near the defect area is extremely uneven, making it difficult to achieve accurate defect segmentation using threshold segmentation and other methods;

[0080]

[0081] S4, using the improved Gaussian background elimination algorithm (IGABEA) method to achieve the removal of uneven background;

[0082] S4.1. Calculate the convolution of the skewness matrix Ske and the Gaussian function with scale factor σ ranging from 1 to n, that is;

[0083]

[0084] Among them, σ i is the i-th size factor, L i (x,y,σ i ) is the i-th size factor σ i estimated background matrix;

[0085] S4.2. Estimation of the background matrix L by the skewness matrix Ske and different size factors σ i (x,y,σ i ) to obtain the skewness image sequence D after removing the uneven background i (x,y,σ i ),Right now;

[0086] D i (x,y,σ i )=|Ske(P(x,y))-L i (x,y,σ i )|

[0087] S4.3, using the third-order nonlinear convergence factor to achieve the skewness image sequence D after removing the uneven background i (x,y,σ i ) compression and reconstruction, that is;

[0088]

[0089] Among them, D new (x, y) is the value of the pixel position (x, y) after compression and reconstruction with the background removed, which is reconstructed into a matrix form D according to the position new ,Right now;

[0090]

[0091] S5, genetic algorithm to achieve the construction of the optimal compression reconstruction matrix;

[0092] S5.1. Calculate S4.3 in the horizontal and vertical directions to reconstruct the skewness matrix D new The first-order differential of is:

[0093]

[0094] Where ΔP is the difference between actual pixel units. dx is the first-order differential in the horizontal direction;

[0095]

[0096] Among them, dy is the first-order differential in the longitudinal direction;

[0097] Calculate that dx and dy are both 0, and extract the pixel point with opposite signs of left and right differential values ​​and up and down differential values. The location of this pixel point is the defect center location, and extract D at this location. new The value is Ske d matrix;

[0098] Ske d =[D new (x1,y1),D new (x2,y2),...D new (x dnum ,y dnum )]

[0099] Among them, d num For Ske d The number of defect center pixels contained in ;

[0100] S5.2. Extract the pixels where dx and dy are both 0 and the left and right differential values ​​have the same sign as the top and bottom differential values ​​of the extracted position, and extract the D new The value is Ske s The matrix is ​​used as the signal intensity of the background domain;

[0101] Ske s =[D new (x1,y1),D new (x2,y2),...D new (x Pnum ,y Pnum )]

[0102] Among them, P num For Ske s The number of pixels contained in

[0103] Calculate Ske s The standard deviation σ ref ;

[0104]

[0105] in, For Ske s The mean of Ske i s For Ske s The i-th value in is D new (x i ,y i ).

[0106] S5.3. Calculate the sum of the signal-to-noise ratios (SNR) of each defect in the compressed reconstruction result image under the current scale factor σ, that is;

[0107]

[0108] In this example, the number of defects is 4, so num is 4;

[0109] S5.4, setting the optimization variable of the genetic algorithm to the scale factor σ, and the optimization target to the sum of the signal-to-noise ratios (SNR) of the defect domain;

[0110] S5.5, repeat S4.1-S5.3 until the SNR calculated by the genetic algorithm is the largest, extract the corresponding maximum scale factor σ, and obtain the matrix D after compression and reconstruction by the SM-IGABEA model new , matrix D new See Background Removal Figure 7 and Figure 8 As shown, Figure 8 Compared to Figure 6 The uniformity in the background domain is significantly improved, and the defect domain is more prominent;

[0111] S6, defect segmentation based on first-order differential mean;

[0112] S6.1. Calculate the matrix D new The first-order differential d in the longitudinal and transverse directions through the center of each defect R x and d R y;

[0113] S6.2. Extract the first-order differential d R x and d R The four positions of the maximum and minimum values ​​in y are extracted in the matrix D new The reconstruction value of

[0114] S6.3. Calculate the mean V of the four reconstruction values mean , in the matrix D new Higher than V mean The defect area is considered to be assigned a value of 1. new Below V mean The area where the background is located is assigned a value of 0, thereby completing the defect segmentation task, see defect segmentation Figure 8 ;

[0115] S7, measurement of defect cutting plane;

[0116] S7.1. Calculate the first-order differential d in S6.2. R The distance between the maximum and minimum values ​​of x is scaled according to the ratio of the pixel point to the actual distance to obtain the width W of the defect d ;

[0117] S7.2. Calculate the first-order differential d in S6.2. RThe distance between the maximum and minimum values ​​of y is scaled according to the ratio of the pixel point to the actual distance to obtain the height H of the defect d ;

[0118] S8, defect depth reconstruction based on ternary linear regression;

[0119] S8.1. Extract the defect domain position in S (6.3), and calculate the skewness mean Ske of the defect area in the skewness matrix Ske in S (3.2) according to the defect domain position mean d , extract the width W of S7.1-S7.2 d With height H d ;

[0120] S8.2. Will Ske mean d , W d With H d Input the fitting model to achieve defect depth D d of reconstruction, i.e.;

[0121]

[0122] Among them, T s is the thickness of the steel plate;

[0123] From the maximum thermal contrast Figure 5 and Figure 7 It can be seen from the comparison that the present invention can effectively reduce noise and improve the signal-to-noise ratio of the defect position. Figure 6 and Figure 8 It can be seen from the comparison that the present invention can effectively reduce the problem of uneven defect domain background caused by the SM method, thereby achieving accurate quantification of defects. Fig. 9 It can be seen that there is still some noise on the edge of the segmented defect, but it does not affect the recognition of the true shape of the defect. The depth and geometric features of the defect can also be quantified through S7 and S8. In addition, since the temperature time series data is only related to the distribution of the time series data, the requirement for the frame rate of the infrared thermal imager is relatively low, which greatly reduces the cost of the application of this method and engineering practice.

[0124] Although the above describes the illustrative specific embodiments of the present invention to facilitate those skilled in the art to understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A step-type eddy current thermal imaging defect reconstruction method based on the SM-IGABEA model, characterized in that: The steps include: S1, obtaining infrared thermal image sequence; The specimen with known defects at room temperature is heated by using the stepped eddy current thermal imaging nondestructive testing platform, and P frames of infrared thermal images are collected during the entire heating stage to form an infrared image time series S, where the size of each frame of infrared thermal image is M×N, where M and N are the length and height of the infrared thermal image, respectively. S2, singular value decomposition denoising and reconstruction; S2.

1. Discretize into multiple thermal images X by time i , a single image X of shape M×N i Decomposed into the product of three matrices through the SVD method, namely; X new i =URV T S2.2, then select the first k largest singular values ​​to reconstruct and obtain the de-noised and reconstructed discrete heat map X new i , so that the noise information on a single heat map is eliminated to a great extent, that is; X new i =U m×m i R m×k i V k×n i T S2.3, decomposition and denoising reconstructed discrete heat map X new i In chronological order, reassemble into a denoised infrared image time series S new ; S3, the compression of thermal image sequence is realized through the third-order statistic skewness (i.e., SM model); S3.

1. Convert the infrared image time series S new Decompose into one-dimensional time series data P(x,y) according to the pixel position, and calculate the skewness value Ske(P(x,y)) corresponding to each pixel P(x,y) point through the third-order statistics, that is; in, is the mean value of the one-dimensional time series temperature data of pixel P(x,y), and N is the number of frames; S3.

2. The skewness values ​​Ske(P(x,y)) of all pixels are spliced ​​and reorganized according to their positions (x,y) to achieve compression of the thermal image sequence. The compressed matrix is ​​the skewness matrix Ske. S4, using the improved Gaussian background elimination algorithm (IGABEA) method to achieve the removal of uneven background; S4.

1. Calculate the convolution of the skewness matrix Ske and the Gaussian function with scale factor σ ranging from 1 to n, that is; Among them, σ i is the i-th size factor, L i (x,y,σ i ) is the background image estimated using the i-th size factor; S4.

2. Estimation of the background matrix L through the skewness matrix Ske and different size factors i (x,y,σ i ) to obtain the skewness image sequence D after removing the uneven background i (x,y,σ i ),Right now; D i (x,y,σ i )=|Ske(P(x,y))-L i (x,y,σ i )| S4.3, using the third-order nonlinear convergence factor to achieve the skewness image sequence D after removing the uneven background i (x,y,σ i ) compression and reconstruction, that is; Among them, D new i (x, y) is the value of the pixel position (x, y) after compression and reconstruction with the background removed, which is reconstructed into a matrix form D according to the position new ; S5, genetic algorithm to achieve the construction of the best compression reconstruction map; S5.

1. Calculate S4.3 in the horizontal and vertical directions to reconstruct the skewness matrix D new The first-order differentials of are dx and dy respectively. Find the pixel point where dx and dy are both 0 and the left and right or upper and lower differential values ​​of the extracted position have opposite signs. The location of this pixel point is the defect center location, and D at this location is extracted. new The value is Ske d matrix; S5.

2. Extract the pixel points where both dx and dy are 0 and the left and right or top and bottom differential values ​​of the extracted position have the same sign. Extract the D new The value is Ske s The matrix is ​​used as the signal intensity of the background domain, and the standard deviation σ of these pixels is calculated ref ; S5.

3. Calculate the signal-to-noise ratio (SNR) of each defect in the compressed reconstruction result image under the current scale factor σ, that is; S5.4, setting the optimization variable of the genetic algorithm to the scale factor σ, and the optimization target to the signal-to-noise ratio SNR of the defect domain; S5.5, repeat S4.1-S5.3 until the genetic algorithm converges, thereby obtaining the defect-enhanced matrix D new ; S6, defect segmentation based on first-order differential mean; S6.

1. Calculate the enhanced matrix D new The first-order differentials d in the longitudinal and transverse directions through the defect center R x and d R y; S6.

2. Extract the first-order differential d R x and d R The four positions of the maximum and minimum values ​​in y, extract the corresponding D of these four positions new The reconstruction value of S6.

3. Calculate the mean value V of the four extracted reconstruction values mean , in the matrix D new Higher than V mean The defect area is considered to be assigned a value of 1. new Below V mean The area where the background is located is assigned a value of 0, thus completing the defect segmentation task; S7, measurement of defect cutting plane; S7.

1. Calculate the first-order differential d in S6.

2. R The distance between the maximum and minimum values ​​of x is scaled according to the ratio of the pixel point to the actual distance to obtain the width W of the defect d ; S7.

2. Calculate the first-order differential d in S6.

2. R The distance between the maximum and minimum values ​​of y is scaled according to the ratio of the pixel point to the actual distance to obtain the height H of the defect d ; S8, defect depth reconstruction based on ternary linear regression; S8.

1. Extract the defect area position in S (6.3) and calculate the skewness mean Ske of the defect area in the skewness matrix Ske in S (3.2) mean d , extract the width W of S7.1-S7.2 d With height H d ; S8.

2. Will Ske mean d , W d With H d Input the fitting model to achieve defect depth D d of reconstruction, i.e.; Where T s is the thickness of the steel plate.

2. According to the SM-IGABEA model-based step-type eddy current thermal imaging defect reconstruction method of claim 1, it is characterized in that: In S1, the alternating current frequency is 50-200 Hz, the current is less than or equal to 20 A, and the thermal imager sampling frequency is less than or equal to 10 Hz.

3. The step-type eddy current thermal imaging defect reconstruction method based on the SM-IGABEA model according to claim 1 is characterized in that: In S5.1, if the first-order differential values ​​in the horizontal and vertical directions are both 0, and the first-order differential values ​​at the left and right or upper and lower pixel positions have different signs, it indicates that the defect center is at that position.

4. The step-type eddy current thermal imaging defect reconstruction method based on the SM-IGABEA model according to claim 1 is characterized in that: In S6.3, the average value V greater than the four reconstructed values mean , in image D new Higher than V mean The area where the defect is located is assigned a value of 1. new Below V mean The area considered to be the background is assigned a value of 0.

5. The step-type eddy current thermal imaging defect reconstruction method based on the SM-IGABEA model according to claim 1 is characterized in that: In S8.2, the defect domain skewness mean Ske mean d The depth of the defect is measured by combining the length and width of the defect.

Images