Titanium-steel clad plate interface defect detection method based on x-ray imaging

Abstract

The application belongs to the technical field of image processing, and particularly relates to a titanium steel composite plate interface defect detection method based on X-ray imaging, which comprises the following steps: collecting an X-ray image of a composite plate interface, and performing image direction regularization through Fourier transform and affine transformation; using empirical mode decomposition on each row of pixel sequences of the regularized image to obtain a rough background image; processing the rough background image through an anisotropic Gaussian kernel to obtain a filtered background image; obtaining a residual image according to the filtered background image and the regularized image; obtaining a defect index of the composite plate through the residual image, and detecting the interface defects of the composite plate according to the defect index. The application constructs a background model through empirical mode decomposition and performs filtering in combination with an anisotropic Gaussian kernel, effectively separates the ripple background and defect signals of the composite plate interface, alleviates the confusion problem of the ripple and the defects, and improves the accuracy of the defect detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a method for detecting interface defects of titanium-steel composite plates based on X-ray imaging. BACKGROUND

[0002] Titanium-steel composite plates are widely used in chemical industry, marine engineering and other fields due to their excellent comprehensive performance. Due to the particularity of the use scene of the composite plate, the composite plate is required to have good quality, and therefore the quality detection of the composite plate is very important. The composite plate is often made by explosion welding process, and the core of the process is to realize the combination of the interfaces of different metal materials, and therefore the quality of the interface combination of the composite plate is directly related to the safety and service life in the use process.

[0003] X-ray detection is a common non-destructive testing method for evaluating the interface quality of the composite plate. By observing the two-dimensional X-ray transmission image of the composite plate, the detector can intuitively analyze the image to identify and locate the micro defects such as incomplete welding and inclusions on the interface.

[0004] However, there are great challenges in realizing the automatic detection of the defects of the composite plate. Due to the inherent characteristics of the explosion welding process, the background of the X-ray image of the composite plate is not uniform and flat, but presents a clear corrugated texture. This corrugated texture is caused by the wavy plastic deformation of the metal interface under the explosion impact, and it is not a defect itself, but it appears as a strong background interference with complex morphology and dramatic gray scale change in the image. The real defect signal is usually weak and also has high frequency characteristics, and it is easily covered or confused by this strong and non-stationary corrugated background. In addition, the randomness of the workpiece placement angle during imaging leads to the uncertainty of the corrugated direction, making it difficult for conventional image processing algorithms to effectively distinguish the background and the defects, and easily misidentifying the corrugation as a defect or missing the real defect, thereby resulting in the problems of low accuracy and poor reliability of the existing automatic detection technology of the defects of the composite plate. SUMMARY

[0005] To solve the technical problems of non-stationary wave background texture and defect signal confusion in the composite plate X-ray image, and low defect detection accuracy and easy to miss detection and misjudgment caused by inconsistent wave direction, the application provides a titanium steel composite plate interface defect detection method based on X-ray imaging, comprising: acquiring a composite plate interface X-ray image, performing image direction regularization on the composite plate interface X-ray image through Fourier transform and affine transformation to obtain a corrected image; taking any row in the corrected image as a target row, using empirical mode decomposition on a target sequence composed of pixel gray values in the target row to decompose the target sequence into several IMF components and a residual term, removing a preset number of IMF components, and superimposing the remaining IMF components and the residual term to obtain a background model sequence of the target row; combining the background model sequences of each row in the corrected image into a rough background image in order; processing the rough background image through an anisotropic Gaussian kernel to obtain a filtered background image, wherein the standard deviation of the anisotropic Gaussian kernel in the wave texture direction is less than the standard deviation perpendicular to the wave texture direction; obtaining a residual index of each pixel point according to the difference between the pixel value of each pixel point in the filtered background image and the pixel value of the corresponding pixel point in the corrected image, and combining the residual indexes of all pixel points into a residual image; performing threshold segmentation on the residual image to obtain a defect mask; determining a defect area by analyzing the connected components of the defect mask; obtaining a defect index of the composite plate according to the area of the defect area and the residual index of the pixel points in the defect area; and performing composite plate interface defect detection according to the defect index.

[0006] The application first performs Fourier transform and affine transformation on the X-ray image to realize the direction regularization of the composite plate wave texture, overcoming the problem of inconsistent wave direction caused by different acquisition angles; further, the application processes each row of the regularized image using empirical mode decomposition, removes the initial IMF component representing high-frequency defect signals, reconstructs the remaining components and residual terms, and constructs a rough background image that does not contain defect information; at the same time, to solve the problem of row discontinuity artifacts caused by row-by-row decomposition, the application uses an anisotropic Gaussian kernel for filtering, which uses a larger standard deviation for smoothing perpendicular to the wave direction to eliminate artifacts, and uses a smaller standard deviation to retain wave details in the wave direction to obtain an accurate filtered background image. Finally, by subtracting the texture corrected image from the filtered background image, the defect signal is effectively separated from the strong noise and non-stationary wave background, solving the technical problem of wave and defect confusion, and combining the defect area and severity to construct a defect index to realize high-precision defect detection.

[0007] Preferably, the image direction normalization of the composite board interface X-ray image through Fourier transform and affine transformation to obtain a corrected map comprises: performing Fourier transform on the composite board interface X-ray image to obtain a frequency spectrum; determining a rotation angle used by the affine transformation according to the frequency spectrum, and rotating the composite board interface X-ray image to obtain the corrected map.

[0008] The present application converts the directionality of the ripple texture features in the image into the frequency domain by performing Fourier transform on the composite board interface X-ray image. In the frequency domain, the periodic wave structure is converted into a high-energy bright line with energy concentration, and the direction feature is clearer and easier to identify on the frequency spectrum than in the original image space domain. Therefore, the main direction of the ripple can be more accurately captured by analyzing the frequency spectrum, which provides an accurate angle basis for the subsequent affine transformation rotation, thereby ensuring the accuracy of the image direction normalization and avoiding the interference of inconsistent ripple directions on the subsequent row-by-row background modeling.

[0009] Preferably, the determination of the rotation angle used by the affine transformation according to the frequency spectrum comprises: using Radon transform on the frequency spectrum to obtain an angle at which the Radon transform result obtains a maximum peak value, and determining the angle as the bright line direction angle in the frequency spectrum; and using 90 degrees minus the bright line direction angle to obtain the rotation angle used by the affine transformation.

[0010] Preferably, the background model sequence satisfies the relationship: ; in the formula, is the background model sequence of the target row, is the first IMF component of the target sequence, is a residual term of the target sequence, is the number of IMF components removed from the target sequence, is the number of IMF components obtained by performing empirical mode decomposition on the target sequence.

[0011] The present application realizes the separation of the high-frequency defect signal and the low-frequency waveform background trend by removing the first m IMF components representing the high-frequency defect signal and only superimposing the subsequent IMF components and the residual term representing the overall trend, which can accurately reconstruct the background model containing only the waveform trend without defect information, and provides a pure background reference for subsequent extraction of the defect signal.

[0012] Preferably, the removal of the preset number of IMF components comprises: removing the first preset number of IMF components.

[0013] Preferably, the residual index satisfies the relationship: ; in the formula, The defect index of the pixel point with the coordinate of (x, y) in the modified image is calculated according to the following formula: The gray value of the pixel point with the coordinate of (x, y) in the modified image is calculated according to the following formula: The gray value of the pixel point with the coordinate of (x, y) in the modified image is calculated according to the following formula: The gray value of the pixel point with the coordinate of (x, y) in the modified image is calculated according to the following formula: The gray value of the pixel point with the coordinate of (x, y) in the modified image is calculated according to the following formula: The gray value of the pixel point with the coordinate of (x, y) in the modified image is calculated according to the following formula: The gray value of the pixel point with the coordinate of (x, y) in the modified image is calculated according to the following formula:

[0014] The present application highlights the defect signal by modifying the absolute difference between the gray value of the image and the gray value of the filtered background image as a measure of the defect signal, and adjusting the absolute difference by the gray value of the filtered background image, so that the position of the defect on the composite plate can also obtain a higher response value in the final residual image, making it more prominent, thereby improving the detectability of the defect.

[0015] Preferably, the threshold segmentation of the residual image comprises: calculating the segmentation threshold by the Otsu algorithm, and segmenting the residual image.

[0016] Preferably, the defect region is a region with an area greater than a preset area threshold in the result of connected component analysis.

[0017] Preferably, the defect index satisfies the following relationship: ; in the formula, is the defect index of the composite plate, is the sum of the areas of all defect regions, is the area of the modified image, is the sum of the residual indexes of the pixel points corresponding to all defect regions, is an adjustment factor, is an exponential function with a natural constant as the base.

[0018] The present application not only considers the area ratio of the defect region to the total image area, but also evaluates the severity of the defect by the average defect residual index of the defect region, thereby realizing comprehensive evaluation of the severity of the defect of the composite plate. The present application further introduces an exponential function to process the average defect severity, so that the defect index increases exponentially with the increase of the severity, and the final defect index can more sensitively reflect the existence of severe defects, thereby providing a more reliable basis for defect judgment.

[0019] Preferably, the composite plate interface defect detection according to the defect index comprises: determining that the composite plate has an interface defect in response to the defect index being greater than a preset defect threshold.

[0020] The beneficial effects of the present application are that: the present application solves the interference problem of inconsistent ripple texture direction in the X-ray image by global direction regularization through Fourier transform and affine transformation, and provides a standardized image basis for subsequent row-by-row signal analysis; the present application introduces an empirical mode decomposition technique to adaptively decompose each row of pixel sequences. By removing the initial IMF component representing the high-frequency defect signal and reconstructing the remaining component, an individual waveform background model can be accurately established for each row, and effective separation of the defect signal and the background trend is achieved; in addition, the present application also introduces an anisotropic Gaussian kernel to smooth the direction perpendicular to the ripple while maintaining the sharpness of the ripple direction, thereby obtaining a smooth and realistic global filtered background image without losing the details of the waveform, and solving the row-to-row jump artifact problem caused by independent row-by-row decomposition; the present application not only locates defects through Otsu and connected domain analysis, but also constructs a defect index that comprehensively considers the defect area and the defect severity, and uses an exponential function to amplify the weight of severe defects, so that the evaluation result is closer to the phenomenon that severe defects are more harmful to the composite plate, and the composite plate interface defect detection can be effectively performed. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flowchart schematically showing a titanium steel composite plate interface defect detection method based on X-ray imaging in the present application;

[0022] Figure 2 is a modified image schematically shown in the present application;

[0023] Figure 3 is a residual image schematically shown in the present application;

[0024] Figure 4 is a defect mask schematically shown in the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0026] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0027] The embodiments of the present application disclose a titanium steel composite plate interface defect detection method based on X-ray imaging, with reference to Figure 1 , comprising steps S1 to S5:

[0028] S1, collect the composite board interface X-ray image, and perform image direction normalization through Fourier transform and affine transformation.

[0029] It should be noted that the ripple texture in the composite board interface X-ray image has obvious directionality, however, the ripple texture of the composite board in the image may not extend in the horizontal direction, which will affect the detection of the composite board interface defects in the subsequent steps, therefore, the image direction normalization is performed through Fourier transform and affine transformation.

[0030] Specifically, through an X-ray DR (Digital Radiography) device, a two-dimensional transmission image of the composite board is collected to obtain the composite board interface X-ray image. The composite board interface X-ray image is subjected to two-dimensional fast Fourier transform to convert the composite board interface X-ray image into a frequency spectrum image, the Radon transform is used on the frequency spectrum image to obtain an angle at which the Radon transform result obtains a maximum peak value, and the angle is taken as the bright line direction angle in the frequency spectrum image; the horizontal right is taken as 0 degree, and the angle is sequentially increased along the counterclockwise direction; the ripple texture direction angle of the composite board interface X-ray image is obtained by using 90 degrees minus the bright line direction angle, and the composite board interface X-ray image is rotated to obtain a corrected image through affine transformation, wherein the rotation angle is the ripple texture direction angle.

[0031] Exemplarily, if the ripple texture in the two-dimensional transmission image of the composite board is horizontal, the bright line direction angle is 90 degrees, and the ripple texture direction angle is 0 degree, and the rotation angle is 0 degree at this time; if the ripple texture in the two-dimensional transmission image of the composite board is vertical, the bright line direction angle is 0 degree, and the ripple texture direction angle is 90 degrees, and the two-dimensional transmission image of the composite board will be counterclockwise rotated by 90 degrees to rotate the ripple texture in the image to the horizontal state; if the ripple texture in the two-dimensional transmission image of the composite board is inclined by 10 degrees, the bright line direction angle is 100 degrees, and the ripple texture direction angle is -10 degrees, and the rotation angle is -10 degrees at this time, and the two-dimensional transmission image of the composite board will be clockwise rotated by 10 degrees to rotate the ripple texture to the horizontal state.

[0032] It should be noted that in the frequency spectrum, the wave structure periodically and directionally extended in the clad interface X-ray image will be converted into a series of concentrated high-energy frequency components, which are usually arranged in a high-energy bright line passing through the center of the frequency spectrum. The direction of the high-energy bright line in the frequency domain is perpendicular to the main extension direction of the wave structure in the clad interface X-ray image. Therefore, the direction of the high-energy bright line in the frequency spectrum is determined by Radon transform, which is calculated by integrating the frequency spectrum along different angles (from 0° to 179°). Since the wave energy is concentrated on the high-energy bright line, when the projection angle is consistent with the angle of the high-energy bright line, the integral value will reach a maximum peak. Therefore, by finding the angle that makes the Radon transform result reach the maximum value, the direction of the high-energy bright line in the frequency spectrum can be determined, and the direction of the ripple texture in the clad interface X-ray image can be corrected, so that the direction of the ripple texture is horizontal.

[0033] Exemplarily, Figure 2 The corrected image in the present application can be seen from the figure that the ripple texture has approached to the horizontal state, thereby providing a data basis for subsequent empirical mode decomposition.

[0034] S2, using empirical mode decomposition on the sequence of gray scale values of each row of pixels in the corrected image, a rough background image is obtained according to the decomposition result.

[0035] It should be noted that after the image direction is regularized, each row of the image can be regarded as a composite signal superimposed by low-frequency wave trend, high-frequency defect signal and noise. Due to the characteristics of explosive welding process, this wave trend is usually not smooth; therefore, the present application uses empirical mode decomposition on the sequence of gray scale values of each row of pixels in the corrected image, and a rough background image is obtained according to the decomposition result.

[0036] Specifically, taking any row in the corrected image as a target row, the sequence of gray scale values of the pixels in the target row is a target sequence, and the target sequence is decomposed into a plurality of IMF components and a residual term by using empirical mode decomposition. The first three IMF components are removed, the other IMF components are superimposed, and then the residual term is superimposed to obtain the background model sequence of the target row; the background model sequences of each row in the corrected image are combined in the order of their appearance in the image to obtain the rough background image.

[0037] Exemplarily, the target sequence is decomposed into three IMF components by using empirical mode decomposition, the first three IMF components are removed, the other IMF components are superimposed, and then the residual term is superimposed to obtain the background model sequence of the target sequence.

[0038] ​In one embodiment, the background model sequence satisfies the following relation:

[0039] ;

[0040] In the formula, The background model sequence for the target row. For the target sequence One IMF component, For the residual term of the target sequence, This represents the number of IMF components removed from all IMF components of the target sequence. This represents the number of IMF components obtained from Empirical Mode Decomposition (EMD) of the target sequence.

[0041] Among them, the first to be decomposed Each IMF component contains high-frequency defect signals. The combination of the remaining IMF components after removing these IMF components and the residual terms represents the waveform trend of the target line signal.

[0042] It should be added that, , and Both are sequences. Adding two sequences together results in another sequence. The operation of adding two sequences together involves adding the values ​​in the sequences one by one.

[0043] S3. The coarse background image is processed by anisotropic Gaussian kernel to obtain a filtered background image.

[0044] It should be noted that during the acquisition of the coarse background image, the empirical mode decomposition (EMD) of the correction image is performed independently row by row, which leads to a lack of correlation between the decomposition results of the target sequence and the adjacent rows of the target row. This independence may cause the output coarse background image to exhibit physically unrealistic stripe artifacts or slight grayscale jumps in the direction perpendicular to the ripples. Therefore, this invention processes the coarse background image using an anisotropic Gaussian kernel.

[0045] Specifically, an anisotropic Gaussian kernel is set, with its standard deviation in the horizontal direction being smaller than that in the vertical direction. The anisotropic Gaussian kernel is then used to process the rough background image to obtain a filtered background image.

[0046] For example, the horizontal standard deviation of the anisotropic Gaussian kernel is 0.5 and the vertical standard deviation is 5.

[0047] The standard deviation in the horizontal direction is relatively small to maintain the sharpness of the waveform structure along the ripple direction; the standard deviation in the vertical direction is relatively large to perform cross-line smoothing and eliminate inter-line jump artifacts in the rough background image.

[0048] S4. Obtain the residual image based on the filtered background image and the correction image.

[0049] It should be noted that the filtered background image represents the normal waveform background, while the corrected image contains all the feature information of the composite board interface. By subtracting the corrected image from the filtered background image, the defects on the composite board interface can be extracted, thereby realizing the detection of defects on the composite board interface. Therefore, this invention obtains the residual image based on the filtered background image and the corrected image.

[0050] Specifically, the residual index of each pixel is obtained based on the difference between the pixel value of each pixel in the filtered background image and the pixel value of the corresponding pixel in the correction image, and the residual indices of all pixels are combined to form a residual image.

[0051] Specifically, the residual index satisfies the following relationship:

[0052] ;

[0053] In the formula, To correct the coordinates in the graph The residual index of the pixel, To correct the coordinates in the graph The grayscale value of the pixel. The coordinates in the filtered background image are The grayscale value of the pixel. To prevent division by zero errors in parameters, this embodiment... The value is 0.001, and the implementers can adjust it according to the actual situation. .

[0054] in, This represents the absolute grayscale difference between the corrected image and the filtered background image. The larger the value, the more significant the difference between the corrected image and the filtered background image in terms of coordinates. The greater the difference at the interface, the better the coordinates of the composite board. The more likely the corresponding location is to have interface defects, the larger the residual index will be; The smaller the value, the more accurate the comparison between the corrected image and the filtered background image on the coordinate system. The smaller the difference at the interface, the better the coordinates of the composite board. The more likely the corresponding location is to be free of defects, the smaller the residual index; to further highlight the defect area in the correction plot, by... This method makes the pixels corresponding to the defective locations on the composite board more prominent in the residual map.

[0055] For example, Figure 3 The residual image in this invention shows that the strong periodic ripple background in the original image is effectively suppressed, while the areas in the image that differ significantly from the background model are highlighted, thus effectively separating the defect signal from the complex ripple interference.

[0056] S5, obtaining a defect index of the composite board through the residual image, and performing interface defect detection of the composite board according to the defect index.

[0057] It should be noted that the normal and high-contrast ripple background in the residual image has been stripped, and the defect signal is highlighted, and the confusion problem of the ripple and the defect in the correction image has been well alleviated, and the defects existing in the interface of the composite board can be well detected through the residual image, so that the defect mask image is obtained through the residual image.

[0058] Specifically, the Otsu algorithm is used for the residual image, and the best segmentation threshold is automatically calculated, the pixel points greater than the best segmentation threshold in the residual image are marked as 1, and the pixel points less than or equal to the best segmentation threshold are marked as 0, and a defect mask is obtained; the connected component analysis is performed on the defect mask, and each independent area is obtained. The area of each area is obtained, and the area greater than the area threshold is taken as a defect area, and the defect index is obtained according to the area of each defect area and the residual index of the pixel points corresponding to the defect area.

[0059] Further, when the defect index is greater than a defect threshold, the composite board has an interface defect.

[0060] Exemplarily, the area threshold is 5 pixels, and the defect threshold is 0.6.

[0061] Specifically, the defect index satisfies the relationship:

[0062] ;

[0063] In the formula, is the defect index of the composite board, is the sum of the areas of all defect areas, is the area of the correction image, is the sum of the residual indexes of the pixel points corresponding to all defect areas, is an adjustment factor, is an exponential function with a natural constant as the base, and in the embodiment is 2, and the value of can be adjusted according to actual conditions.

[0064] Among them, represents the area ratio of the defect area in the residual image, and the greater the value, the greater the defect area of the composite board, and the greater the defect index of the composite board; the smaller the value, the smaller the defect area of the composite board, and the smaller the defect index of the composite board. The average defect severity of the defect area, that is, the average residual index of each pixel point in the defect area, the greater the value, the more likely there is a serious welding defect in the manufacturing process of the clad plate, and the greater the defect index of the clad plate; the smaller the value, the smaller the defect index of the clad plate.

[0065] It should be noted that the serious welding defect in the clad plate is more harmful to the use of the clad plate, The output value of the function increases with the increase of the input value, so by Further adjustment is made to the The clad plate with serious welding defects has a greater defect index.

[0066] Exemplarily, Figure 4 For the defect mask in the application, it can be seen that only a few discrete white areas are reserved in the image, which accurately correspond to the positions where the clad plate is not tightly fused in the correction map, and then the area of the white areas can be used for clad plate interface defect detection.

Claims

1. A method for detecting interface defects of a titanium-steel clad plate based on X-ray imaging, characterized by, The method comprises the following steps: An interface X-ray image of the composite board is acquired, and the interface X-ray image of the composite board is subjected to image direction normalization through Fourier transform and affine transformation to obtain a corrected image; Taking any row in the corrected image as a target row, an empirical mode decomposition is used on a target sequence composed of pixel gray values in the target row to decompose the target sequence into a plurality of IMF components and a residual term, after removing a preset number of IMF components, the remaining IMF components and the residual term are superimposed to obtain a background model sequence of the target row; background model sequences of each row in the corrected image are sequentially combined into a rough background image; the rough background image is processed through an anisotropic Gaussian kernel to obtain a filtered background image, wherein a standard deviation of the anisotropic Gaussian kernel in the ripple texture direction is less than a standard deviation perpendicular to the ripple texture direction; According to the difference between the pixel value of each pixel point in the filtered background image and the pixel value of the corresponding pixel point in the correction image, the residual index of each pixel point is obtained, satisfying the relationship: ; in the formula, is the residual index of the pixel point with coordinates in the correction image, is the gray value of the pixel point with coordinates in the correction image, is the gray value of the pixel point with coordinates in the filtered background image, is a parameter for preventing division by zero error; Residual indicators of all pixel points are combined into a residual image; the residual image is subjected to threshold segmentation to obtain a defect mask; The defect mask is subjected to connected component analysis to determine a defect region; According to the area of the defect region and the residual index of the pixel points in the defect region, a defect index of the composite board is obtained, satisfying a relationship: ; in the formula, is the defect index of the composite board, is the sum of the areas of all the defect regions, is the area of the corrected graph, is the sum of the residual indices of the pixel points corresponding to all the defect regions, is an adjustment factor, is an exponential function with a natural constant as the base number; Interface defects of the composite board are detected according to the defect index.

2. The X-ray imaging-based titanium steel clad plate interface defect detection method of claim 1, wherein, The method of normalizing the image direction of the interface X-ray image of the composite board through Fourier transform and affine transformation to obtain the corrected image comprises the following steps: Fourier transform is performed on the interface X-ray image of the composite board to obtain a frequency spectrum; a rotation angle used for affine transformation is determined according to the frequency spectrum, and the interface X-ray image of the composite board is rotated to obtain the corrected image. 3.The X-ray imaging-based titanium steel clad plate interface defect detection method according to claim 2, characterized in that, The method of determining the rotation angle used for affine transformation according to the frequency spectrum comprises the following steps: a Radon transform is used on the frequency spectrum to obtain an angle at which a maximum peak value of the Radon transform result is obtained, and the angle is determined as a bright line direction angle in the frequency spectrum; 90 degrees is subtracted from the bright line direction angle to obtain the rotation angle used for affine transformation. 4.The X-ray imaging-based titanium steel clad plate interface defect detection method according to claim 1, wherein, The background model sequence satisfies the following relationship: ； wherein is a background model sequence of the target row, is the th IMF component of the target sequence, is a residual term of the target sequence, is the number of IMF components removed from all IMF components of the target sequence, is the number of IMF components removed from all IMF components of the target sequence, is the number of IMF components obtained by empirical mode decomposition of the target sequence. 5.The X-ray imaging-based titanium steel clad plate interface defect detection method according to claim 1, wherein, The method of removing the preset number of IMF components comprises the following steps: a preset number of IMF components that are decomposed first are removed. 6.The X-ray imaging-based titanium steel clad plate interface defect detection method according to claim 1, wherein, The method of performing threshold segmentation on the residual image comprises the following steps: an Otsu algorithm is used to calculate a segmentation threshold, and the residual image is segmented. 7.The X-ray imaging-based titanium steel clad plate interface defect detection method according to claim 1, wherein, The defect region is a region with an area greater than a preset area threshold in the result of the connected component analysis. 8.The X-ray imaging-based titanium steel clad plate interface defect detection method of claim 1, wherein, The method of detecting the interface defects of the composite board according to the defect index comprises the following steps: in response to the defect index being greater than a preset defect threshold, it is determined that the composite board has interface defects.

Images