A composite material ultrasonic scan image defect feature extraction method and system

Abstract

The application belongs to the technical field of image processing, and particularly relates to a composite material ultrasonic scanning image defect feature extraction method and system, which comprises the following steps: obtaining an ultrasonic scanning image of a composite material and performing filtering processing; constructing a gradient outer product matrix of neighborhood pixel points of a pixel point and accumulating to obtain a local structure matrix; performing feature decomposition on the local structure matrix to obtain an eigenvalue, and calculating a structure coherence factor according to the eigenvalue; combining a gray value and the structure coherence factor to construct a coupling potential, highlighting a low-gray and disordered defect signal through a nonlinear gain; and using a high coupling potential point as an anchor point to perform region segmentation and extract a defect block. The application effectively solves the problem of missing detection of low-contrast defects in a strong texture background by using the essential difference between background texture and defects in physical topology, and significantly improves the precision of defect feature extraction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a composite material ultrasonic scan image defect feature extraction method and system. BACKGROUND

[0002] Carbon fiber reinforced polymer (CFRP) is widely used in aerospace and other fields due to its excellent performance. In ultrasonic C-scan nondestructive testing of CFRP, the image background usually presents high-frequency and high-contrast texture interference due to the periodic weaving structure of the fiber bundle, while the real engineering defects, such as delamination and impact damage, appear as edge blurred and irregularly shaped dark spots. In order to realize automatic recognition and quantification of defects, the industry often introduces superpixel segmentation technology to pre-segment the image into a number of perceptually meaningful sub-regions to replace the inefficient pixel-level processing.

[0003] However, the traditional superpixel segmentation algorithm mainly relies on the gray similarity and spatial proximity between pixels to generate superpixel blocks by local clustering, and filters out potential defect regions according to the average gray of the superpixel blocks.

[0004] However, in the composite material ultrasonic scan image, the background texture itself is composed of alternating bright peaks and dark valleys, and the gray value of the valley area is very low. For those hidden defects with weak contrast, their gray features are often very close to the dark valleys of the background texture. This gray confusion leads to the inability of the traditional algorithm to distinguish the dark of the background from the dark of the defect in essence, and it is easy to misclassify the pixels in the defect area as part of the background texture, which directly leads to the complete neglect of a large number of small or low-contrast defects, causing serious missed detection, and thus significantly reducing the accuracy of defect feature extraction. SUMMARY

[0005] To solve the technical problem of low defect feature extraction accuracy caused by the existing superpixel segmentation algorithm being disturbed by strong texture and the gray confusion between defects and background valleys, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for extracting defect features from ultrasonic scanning images of composite materials, comprising: acquiring an ultrasonic scanning image of a composite material and performing filtering processing to obtain a processed image; for any pixel in the processed image: calculating the weight value of each neighboring pixel using a two-dimensional Gaussian formula; weighting the gradient outer product matrix of the neighboring pixel according to the weight value of each neighboring pixel to determine the structure matrix of the neighboring pixel; accumulating the structure matrices of all neighboring pixels to obtain a local structure matrix; solving for the eigenvalues ​​of the local structure matrix based on the principle of linear algebra; determining the structural coherence factor according to the ratio of the difference to the sum of the two eigenvalues; determining the coupling potential energy according to the structural coherence factor and the gray value of the current pixel; constructing a coupling potential energy map using the coupling potential energy of all pixels, and extracting pixels with high coupling potential energy at a predetermined proportion in the coupling potential energy map and marking them as anchor points; performing a superpixel segmentation algorithm on the coupling potential energy map based on all anchor points to obtain multiple segmented blocks; and marking the segmented block as a defect block when the mean of the coupling potential energy of all pixels in the segmented block is greater than a set threshold.

[0007] This invention evaluates the flow order of local textures by constructing a structure matrix containing gradient outer products and solving for eigenvalues; it constructs a coupled potential energy model using structural coherence factors and grayscale values, and uses a nonlinear gain mechanism to highlight low-grayscale and structurally disordered defect signals while suppressing low-grayscale but structurally ordered background textures; and then uses high-potential-energy anchor points to guide superpixel segmentation, ensuring the topological integrity of the defect region, thus fundamentally solving the technical problem of missed and false detections of weak defects under strong texture interference.

[0008] Preferably, the step of acquiring the ultrasonic scan image of the composite material and performing filtering processing to obtain the processed image includes: converting the ultrasonic scan image to grayscale and performing convolution processing on the grayscale image using a Gaussian filter with a 3×3 filter kernel to obtain the processed image.

[0009] Preferably, the step of calculating the weight value of each neighboring pixel of the current pixel using the two-dimensional Gaussian formula includes: constructing a system centered on each pixel. A neighborhood window is defined as all pixels within the neighborhood window that are considered as the neighboring pixels of the given pixel; wherein... It is an odd number greater than 9; calculate the cutoff radius from its center to its edge based on the size of the neighborhood window. According to the Gaussian distribution The scale parameters are obtained by back-reasoning the criteria. ;based on The weight value of each neighboring pixel is calculated using the two-dimensional Gaussian formula.

[0010] Preferably, determining the structure matrix of the neighboring pixels includes: based on the neighboring pixels in... and The gradient value in the direction is used to construct a gradient vector, the outer product of which is equal to the product of the gradient vector and its transpose; the structure matrix of the neighboring pixels is then calculated. , In the formula, The current pixel's th The weight values ​​of each neighboring pixel; , The current pixel's th The neighboring pixels are in direction and The gradient value in the direction.

[0011] This invention constructs a gradient outer product matrix and uses squaring operations to eliminate the influence of gradient sign, quantifying edge energy and texture tilt. Combined with distance weighting, it effectively suppresses noise and preserves macroscopic texture flow characteristics, providing an accurate geometric and statistical feature basis for distinguishing background texture from defects.

[0012] Preferably, the gradient value is calculated using the Sobel operator.

[0013] Preferably, the eigenvalues ​​of the local structure matrix satisfy the expression: ; In the formula, Principal eigenvalues; Secondary eigenvalues; The values ​​are taken from the diagonal elements in the local structure matrix; The values ​​are taken from the off-diagonal elements in the local structure matrix.

[0014] This invention achieves texture energy decoupling through feature decomposition, and uses primary and secondary feature values ​​to characterize the energy intensity perpendicular to and along the texture direction, respectively, accurately reflecting the flow intensity, and providing data basis for identifying backgrounds with a single flow direction and defects with disordered distribution.

[0015] Preferably, the structural coherence factor satisfies the expression: In the formula, The structural coherence factor of the current pixel; , These are the principal and secondary eigenvalues ​​of the local structure matrix of the current pixel. These are the preset hyperparameters.

[0016] This invention eliminates the influence of overall grayscale amplitude fluctuations by calculating the feature value ratio, achieving feature amplitude independence. This factor reflects the local structural orderliness, effectively distinguishes between ordered backgrounds and loose defects, avoids misjudging dark textures as defects, and improves the accuracy of feature extraction.

[0017] Preferably, the coupling potential energy satisfies the expression: In the formula, This represents the coupling potential energy of the current pixel. This represents the grayscale value of the current pixel. The structural coherence factor of the current pixel; This is the gain coefficient; It is a natural exponential function.

[0018] This invention integrates grayscale and structural disorder, and utilizes a nonlinear gain mechanism to generate a high response only when pixels are simultaneously dark and disordered. This physically filters out ordered background dark texture interference, significantly improves the signal-to-noise ratio, and ensures the effective capture of weak defects with low contrast.

[0019] Preferably, the preset ratio is 5%; the method for obtaining the set threshold is: to collect a standard, defect-free scanned image and calculate the average value of the coupling potential energy of all its pixels as the set threshold.

[0020] Secondly, the present invention provides a composite material ultrasonic scanning image defect feature extraction system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned composite material ultrasonic scanning image defect feature extraction method is implemented.

[0021] By adopting the above technical solution, a computer program is generated from the above-mentioned method for extracting defect features from ultrasonic scanning images of composite materials, and stored in a memory for loading and execution by a processor. This allows for the creation of a terminal device based on the memory and processor, making it convenient to use.

[0022] The beneficial effects of this invention are as follows:

[0023] (1) This invention uses high potential energy anchor points to guide regional aggregation, uses the defect core to lock the surrounding weak signals, and uses a segmentation strategy based on topological continuity to overcome the regional fragmentation problem caused by traditional methods, ensuring that the extracted defect contours are complete and natural, and truly restore the physical boundaries.

[0024] (2) This invention addresses the problem of grayscale confusion in strong texture backgrounds by using structural order to construct an aggregate barrier, effectively excluding low-grayscale but structurally ordered background valleys from defects, achieving high-sensitivity capture of low-contrast hidden defects and reducing the false detection rate. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a method for extracting defect features from ultrasonic scanning images of composite materials according to the present invention;

[0026] Figure 2This is a schematic illustration of an ultrasonic scanning grayscale image of a composite material;

[0027] Figure 3 This is an illustration of the results of extracting defect features using traditional methods;

[0028] Figure 4 This is a schematic illustration of the defect feature extraction results of the method of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] This invention discloses a method for extracting defect features from ultrasonic scanning images of composite materials, referring to... Figure 1 This includes steps S1-S5:

[0032] S1. Acquire ultrasonic scanning images of the composite material and perform filtering to obtain the processed image.

[0033] It should be noted that the essence of an ultrasonic C-scan image is a mapping of sound wave amplitude. Due to the reflection of the microstructure inside the material, the original image is often accompanied by high-frequency speckle noise, which manifests as drastic random jumps between pixels. If the gradient is calculated directly on the original image, the speckle noise will be treated as strong edges, which will seriously interfere with the subsequent judgment of the flow direction of macroscopic texture. Therefore, microscale smoothing must be performed to preserve macroscopic texture and suppress microscopic speckle.

[0034] Specifically, the composite material component to be tested is placed in a water immersion ultrasonic scanning system, and a 5MHz focusing probe is used for scanning. The maximum amplitude of the echo within the time gate is extracted and linearly mapped to a grayscale value of 0-255 to obtain an ultrasonic scanning image. The ultrasonic scanning image is then converted to grayscale to obtain an ultrasonic scanning grayscale image.

[0035] Furthermore, a Gaussian filter is used to convolve the ultrasonic scanning grayscale image to obtain the processed image, ensuring that the gradient calculated subsequently reflects the fiber orientation rather than electronic noise; in this embodiment of the invention, the Gaussian filter kernel is 3×3.

[0036] At this point, the processed image has been obtained.

[0037] S2. For any pixel in the processed image: calculate the weight value of each neighboring pixel; based on the weight value of each neighboring pixel, weight the gradient outer product matrix of that neighboring pixel to determine the structure matrix of that neighboring pixel; accumulate the structure matrices of all neighboring pixels to obtain the local structure matrix; based on the principle of linear algebra, solve for the eigenvalues ​​of the local structure matrix.

[0038] It should be noted that, within the background region, due to the directional arrangement of the fiber bundles, the gray-level gradient direction in its local neighborhood exhibits a high degree of consistency, and the gradient direction is mainly perpendicular to the texture extension direction; while within the defect region, affected by the ultrasonic scattering effect, the gray-level gradient direction in its local neighborhood exhibits a chaotic and random distribution; if the gray-level gradient information of a single pixel is relied upon alone, it will lead to an inability to accurately distinguish between background texture features and defect features. Therefore, a statistical integration of the gradient in the local neighborhood is introduced to evaluate the local flow direction statistical characteristics of the pixel.

[0039] Specifically, the Sobel operator is used to calculate the value of each pixel in the processed ultrasound scan image. gradient values ​​in the direction and The gradient value in the direction.

[0040] To obtain the neighboring pixels of each pixel, the following method is used: construct a network centered on each pixel. The neighborhood window takes all pixels within the neighborhood window as the neighboring pixels of the given pixel.

[0041] in, The local analysis range of a pixel is defined to capture gradient changes within that local area. The background texture of CFRP material is periodic; the width of one brightness cycle of a single fiber bundle or texture is approximately 5-10 pixels. Therefore, at least 1-1.5 texture cycles must be covered within the neighborhood window to ensure the capture of gradient change features. For odd numbers greater than 9, this invention will The value is set to 19, which can be adjusted by the implementer as needed. It should be added that for pixels in the edge region where a complete neighborhood window cannot be obtained, the gradient information of the pixel itself is used for completion processing.

[0042] To obtain the weight value of each neighboring pixel of the current pixel, the method is to calculate the cutoff radius from the center to the edge based on the size of the neighborhood window. According to the Gaussian distribution The scale parameters are obtained by back-reasoning the criteria. ;based on The weight value of each neighboring pixel is calculated using the two-dimensional Gaussian formula.

[0043] To obtain the outer product of the gradients of each neighboring pixel of the current pixel, the method is as follows: based on the gradients of that neighboring pixel... and The gradient value in the direction is used to construct the gradient vector, whose outer product matrix is ​​equal to the product of the gradient vector and the transpose of the gradient vector.

[0044] Based on the weight value of each neighboring pixel within the current pixel's neighborhood, the outer product matrix of the gradients of these neighboring pixels is weighted to determine the structure matrix of each neighboring pixel; the structure matrix satisfies the expression:

[0045]

[0046] In the formula, The current pixel's th The structure matrix of the neighboring pixels; The current pixel's th The weight values ​​of each neighboring pixel; The current pixel's th The neighboring pixels are in Gradient value in the direction; The current pixel's th The neighboring pixels are in The gradient value in the direction.

[0047] in, Reflecting the current pixel's first Instantaneous gradient geometric features at neighboring pixels, diagonal elements and The squaring operation eliminates the influence of the gradient sign, quantifying the edge energy intensity of the point in the orthogonal direction; off-diagonal elements. It records the covariance of the horizontal and vertical gradients, quantifying the directional tilt of the edge; this matrix is ​​geometrically equivalent to the outer product of the gradient vectors, describing the micro-texture orientation at this single point. Reflects the current pixel's... The neighboring pixels are based on their physical distance to the first... The contribution of local texture flow statistics at each pixel is determined based on the principle of image spatial correlation, and the distance from the i-th pixel is considered. The closer the neighboring pixels are, the higher the consistency of their texture structure with the center point. By utilizing the radial decay characteristic of the Gaussian function, nearby pixels are given higher weights. This ensures that when performing local flow direction statistics, we can both use large neighborhood information to suppress random noise and prevent interference from distant heterogeneous textures on local features through distance weighting, thereby achieving accurate estimation of local texture flow direction.

[0048] Furthermore, the structure matrices of all neighboring pixels of the current pixel are summed to obtain the local structure matrix of the current pixel. To decouple the texture flow intensity of a pixel, the eigenvalues ​​of the local structure matrix of that pixel are solved based on the principles of linear algebra. The eigenvalues ​​of the local structure matrix satisfy the expression: , ,in, The main feature value reflects the direction of the most drastic gray-level change in the local area, that is, the energy intensity perpendicular to the direction of the texture stripes; As a secondary eigenvalue, it reflects the energy intensity perpendicular to the long axis direction, that is, along the direction of the texture stripes; The values ​​are taken from the diagonal elements of the local structure matrix of that pixel. The values ​​are taken from the off-diagonal elements of the local structure matrix of the pixel.

[0049] At this point, we have obtained two eigenvalues ​​of the local structure matrix for each pixel.

[0050] S3. Determine the structural coherence factor based on the ratio of the difference between two eigenvalues ​​to the sum of the eigenvalues.

[0051] It should be noted that traditional segmentation algorithms often fall into the misconception that low grayscale is a defect when faced with a strong textured background, and are prone to oversegmentation at the black valleys of the background. However, from the perspective of physical structure, there is an essential topological difference between the background texture and the defect area. Therefore, a structural coherence factor is introduced to evaluate the structural order of the local area, thereby providing boundary constraints for the subsequent adaptive region aggregation.

[0052] Specifically, the structural coherence factor of the current pixel is calculated based on the difference ratio between the two eigenvalues ​​of the local structure matrix of the current pixel; the structural coherence factor satisfies the expression:

[0053]

[0054] In the formula, The structural coherence factor of the current pixel; , These are the principal and secondary eigenvalues ​​of the local structure matrix of the current pixel. These are the preset hyperparameters.

[0055] in, The difference reflects the dominant intensity of the local gradient flow at the current pixel. The larger the difference, the more significantly the energy in the dominant direction is higher than that in the orthogonal direction, meaning the more definite the directionality of the local texture. This reflects the total energy intensity of the local gradient of the current pixel. Dividing by this term eliminates the influence of overall grayscale amplitude fluctuations caused by uneven illumination or gain changes, achieving amplitude independence of the feature. Adding a preset hyperparameter is to avoid the denominator being 0, making... Take 0.01; In summary, As it approaches 1, at this time... Much larger This indicates that there is a single and significant dominant flow direction in the local area of ​​the current pixel, which physically corresponds to a well-structured background texture. Its highly ordered structure should be considered a no-go zone for the aggregation process; conversely, it indicates that its structure is loose and disordered, which is the core feature of a potential defect area.

[0056] At this point, the structural coherence factor of each pixel has been obtained.

[0057] S4. Determine the coupling potential energy based on the structural coherence factor and the gray value of the current pixel.

[0058] It should be noted that, considering the presence of random Gaussian noise points or electronic speckle during ultrasonic scanning, these features also exhibit disorder, resulting in a low structural coherence factor and a tendency to generate false alarms. In contrast, actual engineering defects possess two significant physical characteristics: firstly, defects block or absorb ultrasonic waves, typically resulting in lower grayscale values ​​in the defect area; secondly, scattering effects disrupt the periodicity of fiber textures at the defect location, leading to structural disorder. Therefore, a multidimensional feature space was constructed. Only when a pixel simultaneously satisfies both strong attenuation and strong disorder will its potential energy be nonlinearly activated, thereby automatically filtering out background textures and random noise through a physical mechanism.

[0059] Specifically, the coupling potential energy of the current pixel is determined based on its grayscale value and structural coherence factor; the coupling potential energy satisfies the expression:

[0060]

[0061] In the formula, This represents the coupling potential energy of the current pixel. This represents the grayscale value of the current pixel. The structural coherence factor of the current pixel; This is the gain coefficient; It is a natural exponential function.

[0062] in, This value reflects the degree of energy attenuation of the ultrasonic signal at the current pixel. The larger the value, the lower the gray value of the current pixel, which means that the current pixel is more likely to be blocked or absorbed by defects during the propagation of the sound wave. The structural coherence factor reflects the current pixel's structure. The smaller this value, the more disordered the local texture arrangement of the current pixel is, and the greater the possibility of sound wave scattering effect. In summary, if the current pixel simultaneously satisfies the conditions of dark grayscale and disordered structure, the exponential term will generate a drastic nonlinear gain, making the potential energy value at the defect much higher than the potential energy value of the dark background texture, thereby achieving high signal-to-noise ratio feature enhancement. The gain coefficient is used to adjust the amplification factor of structural disorder on potential energy. In this embodiment, an empirical value of 4.0 is given, which can be adjusted by the implementer according to the actual situation.

[0063] At this point, the coupling potential energy of each pixel has been obtained.

[0064] S5. Construct a coupling potential energy map using the coupling potential energy of all pixels, extract pixels with high coupling potential energy at a predetermined ratio from the coupling potential energy map and mark them as anchor points; perform segmentation processing on the coupling potential energy map based on all anchor points to obtain defect blocks.

[0065] It should be noted that after the above processing, the real defect core is characterized by an extremely high potential energy peak, while the background texture and noise are significantly suppressed. Therefore, the point with the highest potential energy is regarded as the anchor point, and the anchor point is used to lock the local area where similar potential energy is located, thereby ensuring that the extracted defect is topologically continuous and avoiding the region fragmentation caused by traditional segmentation methods.

[0066] Specifically, a coupling potential energy map is constructed using the coupling potential energy of all pixels, and the coupling potential energy of all pixels is sorted in descending order. Pixels with coupling potential energy greater than the preset top 5% are extracted and marked as anchor points. At the same time, a superpixel segmentation algorithm is performed on the coupling potential energy map based on all anchor points to obtain all segmented blocks.

[0067] Calculate the mean value of the coupling potential energy of all pixels in each segmentation block, and mark the segmentation block as a defective block when the mean value is greater than a set threshold.

[0068] The threshold is obtained as follows: a standard, defect-free scanned image is acquired, and the average value of the coupling potential energy of all its pixels is calculated as the threshold; and the preset proportional quantile can be adjusted according to the actual signal-to-noise ratio.

[0069] For example, Figure 2 The image shows an ultrasonic scanning grayscale image of the composite material. The background exhibits a significant high-frequency periodic texture and contains several irregularly shaped, blurred dark spots and other engineering defects. Figure 3 The image shows the results of defect feature extraction using traditional methods. The white patches are the extracted defect features. The low-contrast hidden defects in the image cannot be distinguished from the background by the algorithm. These faint defect areas are incorrectly clustered and merged into huge background superpixel blocks, resulting in most potential defects failing to be extracted.Figure 4 The image shows the result of defect feature extraction by the method of the present invention. The white patches are the extracted defect features. By utilizing the high orderliness of the background texture, an aggregation barrier was successfully constructed. Although the dark texture in the image has a low grayscale, it is effectively excluded from the aggregation area due to its ordered structure. The extracted defect area, namely the white patches, has a complete outline and natural shape, which fits the physical boundary of the real defect. Even weak defects can be effectively captured.

[0070] This invention also discloses a composite material ultrasonic scanning image defect feature extraction system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a composite material ultrasonic scanning image defect feature extraction method according to the present invention.

[0071] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for extracting defect features from ultrasonic scanning images of composite materials, characterized in that, include: The ultrasonic scanning images of the composite material are acquired and filtered to obtain the processed image; For any pixel in the processed image: calculate the weight value of each neighboring pixel using the two-dimensional Gaussian formula; Based on the weight value of each neighboring pixel, the gradient outer product matrix of that neighboring pixel is weighted to determine the structure matrix of that neighboring pixel; The structure matrices of all neighboring pixels are summed to obtain the local structure matrix; the eigenvalues ​​of the local structure matrix are solved based on the principles of linear algebra. The structural coherence factor is determined by the ratio of the difference between two eigenvalues ​​to their sum. The coupling potential is determined based on the structural coherence factor and the gray value of the current pixel. Construct a coupling potential energy map using the coupling potential energy of all pixels, and extract pixels with high coupling potential energy at a predetermined ratio from the coupling potential energy map and mark them as anchor points; A superpixel segmentation algorithm is performed on the coupled potential energy map based on all anchor points to obtain multiple segmentation blocks; when the mean of the coupled potential energy of all pixels in the segmentation block is greater than a set threshold, the segmentation block is marked as a defect block; The coupling potential energy satisfies the following expression: In the formula, This represents the coupling potential energy of the current pixel. This represents the grayscale value of the current pixel. The structural coherence factor of the current pixel; This is the gain coefficient; It is a natural exponential function.

2. The method for extracting defect features from ultrasonic scanning images of composite materials according to claim 1, characterized in that, The process of acquiring and filtering the ultrasonic scan image of the composite material to obtain the processed image includes: The ultrasound scan image was converted to grayscale, and then a Gaussian filter with a 3×3 kernel was used to convolve the grayscale image to obtain the processed image.

3. The method for extracting defect features from ultrasonic scanning images of composite materials according to claim 1, characterized in that, The calculation of the weight value of each neighboring pixel of the current pixel using the two-dimensional Gaussian formula includes: Build with each pixel as the center A neighborhood window is defined as all pixels within the neighborhood window that are considered as the neighboring pixels of the given pixel; wherein... It is an odd number greater than 9; calculate the cutoff radius from its center to its edge based on the size of the neighborhood window. According to the Gaussian distribution The scale parameters are obtained by back-reasoning the criteria. ;based on The weight value of each neighboring pixel is calculated using the two-dimensional Gaussian formula.

4. The method for extracting defect features from ultrasonic scanning images of composite materials according to claim 1, characterized in that, The determination of the structure matrix of the neighboring pixels includes: Based on the neighboring pixels and The gradient value in the direction is used to construct a gradient vector, the outer product of which is equal to the product of the gradient vector and its transpose; the structure matrix of the neighboring pixels is then calculated. , In the formula, The current pixel's th The weight values ​​of each neighboring pixel; , The current pixel's th The neighboring pixels are in direction and The gradient value in the direction.

5. The method for extracting defect features from ultrasonic scanning images of composite materials according to claim 4, characterized in that, The gradient value is calculated using the Sobel operator.

6. The method for extracting defect features from ultrasonic scanning images of composite materials according to claim 1, characterized in that, The eigenvalues ​​of the local structure matrix satisfy the expression: ； ； In the formula, Principal eigenvalues; Secondary eigenvalues; The values ​​are taken from the diagonal elements in the local structure matrix; The values ​​are taken from the off-diagonal elements in the local structure matrix.

7. The method for extracting defect features from ultrasonic scanning images of composite materials according to claim 1, characterized in that, The structural coherence factor satisfies the expression: ； In the formula, The structural coherence factor of the current pixel; , These are the principal and secondary eigenvalues ​​of the local structure matrix of the current pixel. These are the preset hyperparameters.

8. The method for extracting defect features from ultrasonic scanning images of composite materials according to claim 1, characterized in that, The preset ratio is 5%; the method for obtaining the set threshold is: to collect a standard, defect-free scanned image and calculate the average value of the coupling potential energy of all its pixels as the set threshold.

9. A system for extracting defect features from ultrasonic scanning images of composite materials, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a method for extracting defect features from ultrasonic scanning images of composite materials according to any one of claims 1-8.

Images