A visual detection method for micro-cracks on the surface of an aviation composite material

Abstract

The present application relates to the technical field of image analysis detection, and particularly relates to a kind of visual detection method of aviation composite material surface microcrack.This method generates candidate direction aggregation window according to the nominal center spacing of the same group of fiber bundle and filters according to image size, and the actual direction aggregation window is determined by the direction consistency evaluation value after gradient direction doubling angle processing;The gradient direction doubling angle in the actual direction aggregation window is weighted and aggregated, and the fiber weaving texture reference direction field is constructed, the reference axial set includes the reference gradient direction and its orthogonal direction on the surface where there are two groups of orthogonal fiber bundle;Crack response value is determined by dark value loss and minimum direction deviation angle, candidate pixels are determined according to response value relative ordering, and after direction guided aggregation, the aggregated candidate connected region that meets the condition is determined as crack region.The method suppresses the edge interference consistent with the fiber texture direction, and combines the broken high response segment.

Description

Technical Field

[0001] This invention relates to the field of image analysis and detection technology, specifically to a visual detection method for microcracks on the surface of aerospace composite materials. Background Technology

[0002] During the manufacturing, assembly, and service of aerospace composite skins, small-width, irregularly extending microcracks are prone to develop on their surfaces. These microcracks intertwine with surface features such as light and dark stripes, weaving intersections, stains, and scratches in carbon fiber fabrics, resulting in low contrast and variable orientation. Existing visual inspection methods primarily rely on grayscale thresholding, template comparison, or manual interpretation. Grayscale thresholding methods struggle to distinguish microcracks from normal fiber dark stripes, manual interpretation is inefficient and inconsistent, and methods based on fixed windows or global texture statistics struggle to adaptively determine the appropriate directional statistical range for local fiber arrangement, leading to missed or false detections.

[0003] In response to the above situation, the technical problems to be solved in this application include: First, on a surface with periodic fiber weave texture, how to establish a reference gradient direction benchmark for normal fiber texture in each local area to suppress edges consistent with the texture direction and retain abnormal edges that diagonally cross the texture; Second, how to adaptively determine the physical range of the direction statistical window so that it covers multiple fiber bundle intervals in the same group to resist local noise, but is not too large to mix in different weave directions; Third, how to deterministically merge high-response segments of the same microcrack that are broken due to being obscured by the light and dark stripes of the fiber bundle, to avoid the same crack being divided into multiple short regions. The technical problems of this application are for the detection of slender dark defects such as cracks and the suppression of interference from normal fiber texture; when microcracks and scratches are both slender dark defects, further differentiation between the two requires additional information such as three-dimensional morphology, phase deflection, or polarization, which is not the technical problem solved by this application. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to provide a visual inspection method for microcracks on the surface of aerospace composite materials. The specific technical solution adopted is as follows:

[0005] A visual inspection method for microcracks on the surface of aerospace composite materials, the method comprising:

[0006] Acquire grayscale images of the surface of aerospace composite material skin, as well as image width, image height, spatial resolution, and nominal center-to-center spacing of fiber bundles in the same group;

[0007] The pixel side length of the candidate direction aggregation window is determined based on the nominal center spacing of the fiber bundles in the same group; the candidate direction aggregation windows are filtered according to the image width and image height; for the filtered candidate direction aggregation windows, the gradient magnitude and gradient direction of the pixels within the window are calculated; the gradient direction of each pixel is doubled by an angle; the direction consistency evaluation value of each candidate direction aggregation window is determined using the gradient magnitude as the weight; the candidate direction aggregation window with the largest direction consistency evaluation value is determined as the actual direction aggregation window.

[0008] The gradient direction of pixels within the actual direction aggregation window is doubled in angle; weighted aggregation is performed using the gradient magnitude as the weight to obtain the fiber weave texture reference gradient direction of each center pixel; the fiber weave texture reference direction field is formed by the set of all center pixel positions and their corresponding reference axes; when there are two sets of orthogonal fiber bundles on the surface to be inspected, the set of reference axes includes the reference gradient direction and its orthogonal direction;

[0009] For each central pixel within the fiber weave texture reference direction field, calculate the dark value loss and the minimum directional deviation angle between the central pixel gradient direction axis and each axis in the corresponding reference axis set; determine the crack response value based on the dark value loss and the directional deviation angle; construct a crack response field from all central pixels and their corresponding crack response values; determine candidate pixels based on the relative order of crack response values; establish an initial candidate connected region; perform directional guided aggregation on the initial candidate connected region according to the directional consistency condition, extension relationship condition, and spacing condition to obtain an aggregated candidate connected region; when the aggregated candidate connected region satisfies the joint crack determination condition, the aggregated candidate connected region is determined as a crack region.

[0010] Furthermore, grayscale images are acquired by a spatially calibrated area array industrial camera under uniform diffuse illumination. The grayscale images are saved in uncompressed grayscale TIFF format. A single trigger generates a complete surface image. Image width and height are obtained by reading the width and height fields from the image file header. Spatial resolution is obtained by reading the pixel equivalent field from the camera's spatial calibration record. The calibration period covers the entire acquisition timeframe.

[0011] Tracing information is bound to the grayscale image. This tracing information is not involved in orientation calculations, response calculations, or crack category determination. Tracing information includes a unique image identifier, original file location information, acquisition time, and acquisition device number. The tracing information is transmitted read-only along with the data objects in each step.

[0012] Determining the nominal center-to-center spacing of fiber bundles within the same group involves: retrieving the center-to-center spacing of fiber bundles from controlled documents corresponding to the model, material type, and material batch of the component under test. Controlled documents include material technical specifications, fiber fabric supply technical documents, or component manufacturing process documents. Record the controlled document identifier, document version, and applicable material batch. When the controlled document only records the repeat length of a complete braided unit, the center-to-center spacing of a single fiber bundle within the same group is calculated based on the number of intervals between fiber bundles within the same group contained in a complete braided unit. When the controlled document does not record the nominal center-to-center spacing of fiber bundles within the same group, measurements are taken on a reference specimen with the same material type and braided structure as the component under test. Measure the actual distance between the center lines of multiple adjacent fiber bundles along a direction perpendicular to the extension direction of the fiber bundles within the same group. Take the median of the measured distances as the nominal center-to-center spacing of the fiber bundles within the same group. Measurements are performed using spatially calibrated microscopic imaging equipment or equivalent dimensional measuring equipment. Record the measurement results, measuring equipment, spatial calibration information, material batch, and applicable scope in the actual measurement calibration record. When there are two or more groups of fiber bundles with different nominal center-to-center distances on the surface to be inspected, the nominal center-to-center distance of each group of fiber bundles is obtained separately. The nominal center-to-center distance with the largest value is taken as the nominal center-to-center distance of the same group of fiber bundles.

[0013] Furthermore, with Indicates the candidate spacing multiple; Starting from 2, increase sequentially; side length of the candidate direction aggregation window. Not less than The smallest odd-numbered pixel side length corresponding to the number of pixels corresponding to the nominal center-to-center spacing of the same fiber bundle, i.e.:

[0014]

[0015] In the formula, the parameters Integer pixel side length of the candidate direction aggregation window; parameter Indicates parameters Belongs to the set of positive integers; parameters Indicates the nominal center-to-center spacing of fiber bundles in the same group; parameter Indicates spatial resolution; parameters Represents the AND relation; parameters This represents the modulo operation;

[0016] Candidate Spacing Multiple Set satisfy:

[0017]

[0018] In the formula, the parameters Indicates image width; parameter Indicates image height; parameter This represents the smaller of the image width and image height.

[0019] When the candidate spacing multiple set is empty, output the processing status "Image size does not meet the direction aggregation condition".

[0020] Furthermore, the gradient magnitude and gradient direction are calculated using the central difference method; the gradient direction is referenced to the positive horizontal axis of the image as 0 radians; the range of values ​​for the gradient direction is... ;

[0021] For the center pixel Centered on, with side length as The candidate direction aggregation window has a total of pixels. Pixels within the window are numbered in row-major order; the first The image coordinates of pixels are Weighted sum of sinusoidal direction components Cosine direction component weighted sum With gradient modulus total for:

[0022]

[0023]

[0024]

[0025] In the formula, the parameter Indicates the first Gradient magnitude of pixels; parameters Indicates the first Gradient direction of each pixel; parameters Represents the sine function; parameters Represents the cosine function;

[0026] Directional Concentration for:

[0027]

[0028] In the formula, the parameters It is a dimensionless quantity, and its range of values ​​is 1. .

[0029] Furthermore, the half-side length of the candidate direction aggregation window Effective center pixel set for:

[0030]

[0031] In the formula, the parameter , These represent the x and y coordinates of the center pixel, respectively.

[0032] Directional Consistency Evaluation Value Set of effective center pixels Concentration in all directions The median; when the effective center pixel set When the number of inward concentrations is even, the arithmetic mean of the two middle values ​​after sorting is taken as the median.

[0033] When multiple candidate spacing multiples correspond to the same maximum directional consistency evaluation value, the candidate spacing multiple with the smallest value is taken as the adaptive directional aggregation spacing multiple. When the directional consistency evaluation values ​​of all candidate spacing multiples are lower than the lower limit of directional consistency. If the braided structure recorded in the controlled document provides a priori fiber axis, then the smallest spacing multiple in the candidate spacing multiple set is taken as the adaptive direction aggregation spacing multiple Np, and the priori fiber axis is recorded as the priori reference axis to continue the subsequent steps; otherwise, the processing status "fiber braided texture reference direction cannot be determined" is output.

[0034] Actual direction aggregation window side length In the formula, the parameters Indicates the adaptive directional aggregation spacing multiple; parameter This indicates the side length of the aggregation window in the actual direction.

[0035] Furthermore, for the central pixel The actual orientation aggregation window centered on the center, the total number of pixels in the window is Pixels within the window are numbered in row-major order; the first The image coordinates of pixels are The weighted sum of the sinusoidal direction components corresponding to the actual direction aggregation window. Weighted sum of cosine direction components for:

[0036]

[0037]

[0038] In the formula, the parameters Indicates the first Gradient magnitude of pixels; parameters Indicates the first Gradient direction of each pixel;

[0039] Effective direction pixel set for:

[0040]

[0041] In the formula, the parameters This represents the set of effective center pixels where the actual direction aggregation window can be fully expanded and all pixels within the window can undergo center difference; ;

[0042] When the prior reference axis and the effective direction pixel set do not exist If empty, output the processing status "Fiber weave texture reference direction cannot be determined" and terminate;

[0043] For each valid center pixel, the fiber weave texture references the gradient direction. A robust iterative weighted solution is adopted, namely:

[0044]

[0045]

[0046]

[0047] In the formula, Indicates the number of iterations; This is a robustness parameter, with a value of, for example, 3; This represents the arctangent function whose resultant direction is determined by the sign of the weighted sum of the sine and cosine direction components; after iteration, the value is taken as... .

[0048] When the prior reference axis exists, the robust iterative weighted solution is not performed; instead, the solution is directly derived from the prior reference axis. The reference axis set is defined as follows: When the surface under inspection contains only a single fiber bundle or is a unidirectional fabric, the reference axis set is... When there are two sets of orthogonal fiber bundles on the surface to be inspected, the reference axial set is: When the prior reference axis does not exist, the robust iterative weighted solution is performed to obtain the reference gradient direction. The reference axial set is (Single fiber bundle or unidirectional fabric), or (Two sets of orthogonal fiber bundles).

[0049] Furthermore, dark value loss for:

[0050]

[0051]

[0052] In the formula, Represents the center pixel Pixel intensity; Indicated by The arithmetic mean of the intensity of all pixels within the actual direction aggregation window centered on the center; Indicated by Centered on, with side length as The arithmetic mean of the intensity of all pixels within a square neighborhood. It is a natural number, with values ​​such as 1 (i.e., a 3×3 neighborhood), and 2k+1 is not greater than half the side length of the actual direction aggregation window; pixels in the neighborhood that exceed the image boundary are padded by mirroring the nearest boundary. This indicates taking the maximum value.

[0053] Direction deviation angle for:

[0054]

[0055] In the formula, Represents the center pixel The gradient direction; Represents the center pixel The corresponding set of reference axes; This indicates a modulo operation with an output range of 1 / 2. ; This indicates taking the absolute value; when the reference axis set contains only one axis, The range of values ​​is When the set of reference axes contains two orthogonal axes, The range of values ​​is .

[0056] Crack response value for:

[0057]

[0058] In the formula, the parameters This represents the square of the sine of the directional deviation angle; when the gradient magnitude of the center pixel is zero, its gradient direction is undefined, so the crack response value of the center pixel is set to 0.

[0059] Furthermore, candidate pixels are determined based on crack response values, including:

[0060] Set of effective direction pixels All crack response values ​​are arranged in ascending order; the sorted sequence is denoted as... In the formula, the parameters This represents the total number of center pixels within the set of valid directional pixels;

[0061] The natural interval between adjacent sorted response values ​​is , ;

[0062] The location of the maximum natural interval is determined as the adaptive splitting point of the response distribution; splitting point index. Indicates that the splitting interval is located at and Between; when the maximum natural interval appears in multiple positions in parallel, the position with the smallest index is taken as the split position index; when all crack response values ​​are equal, the processing status "no crack candidate detected" is output and the process is terminated;

[0063] Sort index The corresponding response value is used as the low-value body; the sorting index is used. The corresponding response value is taken as the high-value tail; the response value and Equal pixels are all included in the high-value tail;

[0064] Pixels corresponding to high tail response values ​​are identified as candidate pixels; candidate pixel set Composed of all candidate pixels; Recorded as ; As the adaptive response threshold.

[0065] Furthermore, the initial candidate connected regions are numbered in row priority order according to the smallest pixel coordinates within the region;

[0066] For each initial candidate connected region, determine the minimum bounding rectangle based on the coordinates of all candidate pixels within the region; the dominant direction of the region is the axial direction corresponding to the longer side of the minimum bounding rectangle; the direction difference is... The two long sides are taken as the same axis;

[0067] For any two regions to be compared whose dominant direction can be determined, the allowable deviation for direction merging is... for:

[0068]

[0069] In the formula, and Let represent the lengths of the longer side of the minimum bounding rectangle of the two regions, respectively. Represents the arctangent function. This indicates taking the minimum value in the set;

[0070] When the minimum axial deviation between the dominant directions of the two regions to be compared is not greater than When the condition of directional consistency is met, the minimum axial deviation between the direction connecting the adjacent boundaries of two regions to be compared and the dominant direction of the region is not greater than 1. When the extension relationship condition is met, the spacing condition is met; when the minimum pixel spacing between adjacent boundaries of regions does not exceed half the side length of the actual direction aggregation window, the spacing condition is met; when the direction consistency condition, extension relationship condition and spacing condition are all met, the pixel paths between the adjacent boundaries of the two regions to be compared are connected along the dominant direction of the region; the pixel paths are recorded as a sequence of connected pixels.

[0071] Calculate the Euclidean distance between the pair of pixels with the smallest Euclidean distance among the boundary pixels of two regions to obtain the minimum pixel spacing between adjacent boundaries of regions; directly compare half the side length of the actual direction aggregation window with the minimum pixel spacing in the real number range; take the direction of the line connecting the two pixels corresponding to the minimum pixel spacing as the direction connecting the adjacent boundaries of the two regions to be compared; when the two pixels corresponding to the minimum pixel spacing are not unique, take the line direction with the smaller axial deviation from the dominant direction of the two regions; if it still cannot be uniquely determined, do not perform merging;

[0072] The pixel path is generated using digital linear interpolation; the pixel path uses the smallest spacing pixel pair as its endpoints.

[0073] After each region merge, the minimum bounding rectangle is redefined based on the candidate pixels and connected pixels within the merged region; the dominant direction, long side length, and short side length of the region are updated; and the allowable deviation for directional merging used in subsequent comparisons is recalculated.

[0074] Region search and merging are performed in ascending order of the initial candidate connected region numbers;

[0075] Single-pixel regions and degenerate regions that cannot form a unique long side direction are not included in the directional guidance aggregation; single-pixel regions and degenerate regions are retained as independent candidate regions.

[0076] Furthermore, the first [unit / item] after aggregation will be [completed / adjusted]. The set of original candidate pixel coordinates obtained by crack response filtering in the aggregated candidate connected regions is denoted as . The set of connected pixel coordinates formed during the directional guidance aggregation process is denoted as... Complete set of pixel coordinates satisfy:

[0077]

[0078] In the formula, the parameter Represents a union; when no fragment merging occurs. It is an empty set;

[0079] Based on the complete set of pixel coordinates Determine the first The minimum bounding rectangle of the aggregated candidate connected regions; the length of the longer side is denoted as . The length of the shorter side is denoted as The axial direction corresponding to the longer side is denoted as During the calculation, each pixel is treated as a pixel unit with a side length of one pixel.

[0080] When all of the following conditions are met, the first The following candidate connected regions were identified as crack regions: (1) (2) Slenderness condition In the formula (3) Minimum length condition. In the formula , The minimum physical length of the crack. For spatial resolution; (4) Mean dark value deficit condition (5) Gradient verticality condition In the formula , representing the minimum axial deviation between the candidate pixel gradient axis and the dominant direction of the region; (6) when no fragment merging occurs, the dominant direction of the source initial candidate connected region can be determined; (7) when fragment merging occurs, the two regions to be compared involved in each merging satisfy the direction consistency condition, extension relationship condition, and spacing condition; (8) connected pixels are not used as original candidate pixels in crack response statistics. Among them, The average dark value loss threshold is measured in gray levels and has a value of, for example, 10. The gradient verticality threshold is dimensionless and takes a value of, for example, 0.5. When the candidate pixel distribution of the aggregated candidate connected region exhibits a significant curved shape along the dominant direction of the region, the region is divided into several continuous sub-segments along its dominant direction, the regional axis of each sub-segment is determined, and the gradient verticality condition is applied to each sub-segment.

[0081] When no region is identified as a crack region, output the processing status "No crack region detected";

[0082] The candidate connected regions identified as crack regions are encapsulated into crack region records. The crack region records include the original candidate pixel coordinate sequence, the connected pixel coordinate sequence, the source initial candidate connected region number, the dominant direction of the aggregated region, the length of the long side and the length of the short side of the minimum circumscribed rotation rectangle, the crack response value sequence corresponding to the original candidate pixels, and the original image traceability index.

[0083] The present invention has the following beneficial effects:

[0084] This application determines the side length of the candidate orientation aggregation window based on the nominal center-to-center spacing of fiber bundles within the same group. The window covers at least multiple fiber bundle intervals, and the edges of multiple fiber bundles in the same direction participate in orientation statistics together, thereby reducing the impact of cracks, blemishes, and noise on orientation estimation. The median of the orientation concentration of each candidate window is used as the orientation consistency evaluation value, and the window with the highest evaluation value is selected as the actual orientation aggregation window. Therefore, the orientation statistics range varies with the image content, covering a sufficient number of fiber bundle intervals without mixing regions with different weaving directions into the same window. After doubling the angle weighting and aggregation of the gradient directions within the actual window, the difference between the two sides of the same bright and dark stripe is approximately... The gradient directions are aligned to the same axis and no longer cancel each other out. The resulting reference direction field marks the normal fiber texture gradient axis of each central pixel at that location. The crack response is determined by the dark value loss and the direction deviation angle. Only pixels with both dark value loss and direction deviation have a high response. Pixels with only a single feature, such as normal fiber bundle dark stripes and reflective points, have a response of zero or close to zero. Microcracks that diagonally cross the texture are retained. Candidate pixels are screened out according to the relative order of their response values. Candidate connected regions are then aggregated by direction guidance. High-response segments of the same crack that are truncated by the bright and dark stripes of the fiber bundle are connected into continuous candidate regions, thus reducing the number of short regions that the same crack is split into. When the aggregated candidate connected region meets the joint criteria of elongation, minimum length, average dark value loss, and gradient perpendicularity, the aggregated candidate connected region is determined as the crack region. At this point, each step of the calculation and the parallel cases are processed according to the determined rules, and the detection results obtained from the same image are consistent. It should be noted that this method detects slender, dark-colored defects such as cracks in a single grayscale image; when microcracks and scratches are both slender, dark-colored defects, they are isomorphic in terms of grayscale and directional features, and this method does not distinguish between them separately. Attached Figure Description

[0085] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0086] Figure 1 This is a flowchart of a visual inspection method for microcracks on the surface of aerospace composite materials, provided as an embodiment of the present invention. Detailed Implementation

[0087] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a visual inspection method for microcracks on the surface of aerospace composite materials proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0088] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0089] The following description, in conjunction with the accompanying drawings, details the specific scheme of the visual inspection method for microcracks on the surface of aerospace composite materials provided by the present invention.

[0090] Please see Figure 1 This illustrates a visual inspection method for microcracks on the surface of aerospace composite materials according to an embodiment of the present invention. The method includes:

[0091] Step S1: Obtain a grayscale image of the surface of the aerospace composite material skin, as well as the image width, image height, spatial resolution, and nominal center-to-center spacing of the fiber bundles in the same group.

[0092] In the implementation scenario of surface inspection of aerospace composite materials, the side length of the subsequent candidate direction polymerization window needs to be converted into pixel side length according to the actual physical scale of fiber bundle arrangement. The correspondence between pixel distance and actual size, and the arrangement scale of fiber bundles in the same group should be determined before the recognition begins. The source of collected information and parameters also needs to be recorded so as to verify and review the recognition results.

[0093] In one embodiment of the present invention, a spatially calibrated area array industrial camera acquires grayscale images of the surface of an aerospace composite material skin under uniform diffuse illumination. The images are saved in uncompressed grayscale TIFF format, and a complete surface image is formed in a single trigger. Pixel fields and file header fields are read through the image decoding interface to obtain pixel intensity, image width, and image height. The spatial resolution is obtained by reading the pixel equivalent field from the camera spatial calibration record, and the calibration validity period covers the current acquisition period. Tracing information such as the image's unique identifier, original file location information, acquisition time, and acquisition device number are bound to the current image. This tracing information is only used for object location identification, parameter source verification, and final crack record review; it does not participate in direction calculation, response calculation, or crack category determination, and is read-only transmitted with each step of the data object. The nominal center-to-center spacing of the same group of fiber bundles is read from the controlled documents, including material technical specifications, fiber fabric supply technical documents, or component manufacturing process documents. When the controlled documents only record the repeat length of a complete braided unit, the center-to-center spacing of a single group of fiber bundles is calculated based on the number of intervals of the same group of fiber bundles contained in a complete braided unit. When the controlled documents do not record this parameter, the actual distance between the center lines of multiple groups of adjacent fiber bundles is measured along the direction perpendicular to the extension of the same group of fiber bundles on the reference sample, and the median is taken as the parameter. The measurement results, along with the measuring equipment, spatial calibration information, material batch, and applicable scope, are written into the actual measurement calibration record. When there are multiple groups of fiber bundles with different nominal center-to-center spacings on the surface to be inspected, the nominal center-to-center spacing with the largest value is taken as the nominal center-to-center spacing of the same group of fiber bundles.

[0094] After this step, the conversion relationship between pixels and actual size, as well as the arrangement scale of fiber bundles, have been determined. Subsequent candidate orientation aggregation windows can be generated according to the actual physical scale, and the recognition results can be traced back to the acquisition device and parameter source. Based on these fundamental parameters, the step of determining the orientation aggregation window is then performed.

[0095] Step S2: Determine the pixel side length of the candidate direction aggregation window based on the nominal center spacing of the fiber bundles in the same group; filter the candidate direction aggregation windows according to the image width and image height; calculate the gradient magnitude and gradient direction of the pixels in the selected candidate direction aggregation windows; double the angle of the gradient direction of each pixel; determine the direction consistency evaluation value of each candidate direction aggregation window using the gradient magnitude as the weight; determine the candidate direction aggregation window with the largest direction consistency evaluation value as the actual direction aggregation window.

[0096] When observing only the edge of a single fiber bundle, the calculated direction is easily affected by cracks, blemishes, and image noise. The direction statistics window needs to cover multiple intervals within the same fiber bundle group, allowing the edges of multiple fiber bundles in the same direction to participate in the statistics. However, an excessively large window may simultaneously include areas with different weaving directions or regions where the local direction changes. Furthermore, the gradient directions on both sides of the same light and dark stripe differ by approximately [missing information]. Directly summing the gradient directions would cause the components on both sides to cancel each other out. Therefore, this step generates candidate direction aggregation windows sequentially according to multiples of the nominal center spacing of the same group of fiber bundles, compares the concentration of gradient directions within each candidate window, and determines the actual direction aggregation window from these comparisons.

[0097] Preferably, in one embodiment of the present invention, with Indicates the candidate spacing multiple. Starting from 2, the physical side length of the candidate direction aggregation window increases sequentially. The nominal center-to-center spacing of the same group of fiber bundles is used as the lower limit. Candidate values The corresponding candidate direction aggregation window side length is used This indicates that the unit is pixels. Take the smallest odd number that is not less than the result of the physical scale conversion, that is:

[0098]

[0099] In the formula, the parameters Integer pixel side length of the candidate direction aggregation window; parameter Indicates parameters Belongs to the set of positive integers; parameters Indicates the nominal center-to-center spacing of fiber bundles in the same group; parameter Indicates spatial resolution; parameters Represents the AND relation; parameters This indicates the modulo operation.

[0100] In the formula, express The physical length corresponding to the nominal center-to-center spacing of fiber bundles in the same group, in micrometers. Divide by ,Bundle The physical length of the spacing between fiber bundles in the same group, converted into the number of pixels, serves as the lower limit of the window side length, ensuring that the window covers a sufficient number of fiber bundle spacings on a physical scale; condition By restricting the side length to an odd number, the window is centered on a single pixel, with the same number of pixels covering both sides of the central pixel. Take the smallest integer that satisfies both of the above constraints, so that the window under each candidate multiple is as small as possible and the local statistical range is as concentrated as possible.

[0101] Candidate Spacing Multiple Set satisfy:

[0102]

[0103] In the formula, the parameters Indicates image width; parameter Indicates image height; parameter This represents the smaller of the image width and image height.

[0104] In the formula, the condition The physical side length of the candidate window must be no less than the nominal center-to-center distance between two fiber bundles in the same group, so that the edges of repeatedly occurring fiber bundles in the same direction participate in the direction statistics; condition The image's shorter side must accommodate the widths of both candidate windows and the boundary margin required for the center difference, preventing the candidate windows from occupying most of the image's shorter side and losing their local directional statistical significance. These two conditions are based on... Established at the same time Only then is it included in the set. When the candidate spacing multiple set is empty, the processing status "Image size does not meet the orientation aggregation condition" is output.

[0105] For each candidate value Calculate pixel gradient magnitude in the current image and pixel gradient direction The gradient magnitude and gradient direction are calculated using the central difference method, a current technique in image processing. The gradient direction is defined with the positive horizontal axis of the image as the 0-radian reference, and its value range is... The subsequent reference gradient direction and the gradient direction use the same angle reference.

[0106] For the center pixel Centered on, with side length as The candidate direction aggregation window has a total of pixels. Pixels within the window are numbered in row-major order; the first The image coordinates of pixels are Weighted sum of sinusoidal direction components Cosine direction component weighted sum With gradient modulus total for:

[0107]

[0108]

[0109]

[0110] In the formula, the parameters Indicates the first Gradient magnitude of pixels; parameters Indicates the first Gradient direction of each pixel; parameters Represents the sine function; parameters This represents the cosine function.

[0111] In the above formulas for the weighted sum of sinusoidal direction components, the weighted sum of cosine direction components, and the total gradient magnitude, the gradient direction of each window pixel is first doubled. The result is further decomposed into sine and cosine components, multiplied by the gradient magnitude, and then summed over all pixels within the window. The gradient magnitude serves as a weight, ensuring that the edges of fiber bundles with significant grayscale changes contribute more to the statistical results, while low-gradient, flat regions contribute less. Doubling the angle makes the phase difference approximately... The gradient directions are mapped to the same axis, and the edge components on both sides of the same bright and dark stripe accumulate in the same direction and do not cancel each other out. The sine and cosine are summed separately to preserve the position of the resultant direction in the two-dimensional direction domain. The sum of gradient magnitudes within the window is used as the normalized denominator for the directional concentration.

[0112] Candidate window at center pixel The directional concentration at a location is defined as ,and:

[0113]

[0114] In the formula, the parameter It is a dimensionless quantity, and its range of values ​​is 1. .

[0115] In the formula, the numerator is and The composite magnitude represents the net intensity accumulated along a certain axis in the gradient direction within the window; the denominator is the sum of gradient magnitudes, representing the total number of gradients participating in the statistics within the window. The ratio of the two measures the concentration of gradient directions within the window. When all gradient directions within the window are in the same direction, the composite magnitude equals the total number of gradients, and the ratio is 1; when the gradient directions are uniformly distributed, the components in each direction cancel each other out, the composite magnitude approaches 0, and the ratio approaches 0. There is no gradient within the time window, and the directional concentration is defined as 0.

[0116] Candidate direction aggregation window half-side length ; with the center pixel The range of pixel coordinates covered by the candidate window centered on it is and ,in and Let x and y be the x and y coordinates of any pixel within the window. Since the central difference method is used to calculate the gradient magnitude and gradient direction of each pixel within the window, the pixels within the window must satisfy... and This ensures that the left, right, top, and bottom adjacent pixels can be read. The set of center pixels is defined as the set of center pixels that simultaneously satisfy the conditions that the candidate window is completely located inside the image and all pixels within the window can be subjected to center differencing. The effective set of center pixels. for:

[0117]

[0118] In the formula, the parameters , These represent the x and y coordinates of the center pixel, respectively.

[0119] In the formula, and The left and right boundaries of the common constraint window are located inside the image, and boundary margins required for the center difference are reserved. and The same constraints are applied to the ordinate. For center pixels satisfying the above coordinate range, their candidate windows are completely located within the image, and each pixel within the window can read its left, right, and top / bottom adjacent pixels, allowing for center difference analysis. An empty set indicates that the candidate window side length is too large or the current image size is insufficient to support the direction statistics for that candidate value; therefore, a direction consistency evaluation value is not calculated for that candidate value. The necessary size condition for a single candidate window to form a non-empty valid set of center pixels is... ,gather In the definition These are conservative engineering screening criteria set to preserve additional space for statistical margins, rather than minimum size criteria directly derived from the full expansion of a single candidate window and the calculation of the central difference.

[0120] right Candidate values ​​are obtained by taking the median of the concentration in all directions. Corresponding directional consistency evaluation value Normal fiber weave textures typically occupy the main area of ​​the image; taking the median can reduce the impact of local noise such as small cracks and blemishes on the candidate window evaluation results. The largest candidate value is determined as the adaptive orientation aggregation spacing multiple of the current image. .

[0121] Preferably, when When the number of inward concentrations is even, the median is the arithmetic mean of the two middle values ​​after sorting; when multiple candidate values ​​correspond to the same maximum... When the candidate value with the smallest value is selected, a unique result is obtained for the same input, and a small local statistical range is retained under the condition of the same directional consistency; when the directional consistency evaluation values ​​of all candidate spacing multiples are lower than the lower limit of directional consistency. If the current image fails to form an effective gradient for directional statistics based solely on local directional aggregation, and the weave structure recorded in the controlled file provides a priori fiber axis, then the smallest spacing multiple in the candidate spacing multiple set is taken as the adaptive directional aggregation spacing multiple Np, and the priori fiber axis is recorded as the priori reference axis to continue subsequent steps; otherwise, the system outputs a processing status of "fiber weave texture reference direction is uncertain" and does not generate a crack region record. The lower limit of directional consistency is a preset threshold of directional concentration, dimensionless, and its value range is [insert range here]. .

[0122] get Afterwards, The corresponding candidate orientation aggregation window is determined as the orientation aggregation window actually used in the current image, and the actual orientation aggregation window side length is... The actual total number of pixels within the window The actual direction aggregation window can be fully expanded and perform center difference on the effective set of center pixels. .

[0123] Step S3: Double the angle of the gradient direction of the pixels in the actual direction aggregation window; perform weighted aggregation with gradient magnitude as weight to obtain the fiber weave texture reference gradient direction of each center pixel; the fiber weave texture reference direction field is formed by the set of all center pixel positions and the corresponding reference axis; when there are two sets of orthogonal fiber bundles on the surface to be inspected, the set of reference axis includes the reference gradient direction and its orthogonal direction.

[0124] Step S2 determines the range of local direction statistics through the actual direction aggregation window. This step uniformly describes the gradients of multiple pixels within the window as a local reference gradient direction. The gradient directions on both sides of the same light and dark stripe within the window differ by approximately Direct averaging would cause the components on both sides to cancel each other out. Therefore, the gradient direction is doubled before aggregation, and the sine and cosine components are summed separately to preserve the position of the doubled angle in the two-dimensional direction domain. For windows that cannot form a unique synthesis axis, no reference gradient direction is generated.

[0125] In one embodiment of the present invention, for any center pixel ,by Construct a structure with a side length of [missing information] at the center. The actual direction aggregation window numbers all pixels within the window in row priority order. , will the The image coordinates of each pixel are defined as follows: The actual direction aggregation window corresponds to the doubled angle sine weighted sum. Weighted sum of cosines of doubled angles for:

[0126]

[0127]

[0128] In the formula, the parameters Indicates the first Gradient magnitude of pixels; parameters Indicates the first The gradient direction of each pixel.

[0129] In the formula, the gradient magnitude is used as the weight. The gradient direction of each pixel within the window is doubled, and the sum of the sine and cosine components is calculated separately. Gradient magnitude weighting ensures that the edges of fiber bundles with significant grayscale changes contribute more to the reference direction, while low-gradient, flat regions contribute less. Doubling the angle makes the difference between the two sides of the same bright and dark stripe approximately... The gradient direction is mapped to the same axis, and the components on both sides accumulate in the same direction without canceling each other out. The sine and cosine are summed separately to preserve the position of the resultant direction in the two-dimensional direction domain, avoiding the problem of crossing angle boundaries when directly averaging the angles.

[0130] Effective direction pixel set for:

[0131]

[0132] In the formula, the parameters This represents the set of effective center pixels where the actual direction aggregation window can be fully expanded and all pixels within the window can undergo center difference; .

[0133] In the formula, Let be the squared magnitude of the synthesized direction component within the window. A value greater than 0 indicates the presence of a non-zero synthesized direction component within the window, capable of forming a unique synthesized axis; if and If both values ​​are 0, there is no valid gradient within the window, or gradient components along different axes completely cancel each other out, making it impossible to obtain a meaningful reference gradient direction; therefore, the corresponding center pixel is not included. When the prior reference axis and the set of valid direction pixels do not exist. If empty, output the processing status "Fiber weave texture reference direction is undetermined" and terminate the subsequent recognition process of the current image.

[0134] For each valid center pixel, the fiber weave texture references the gradient direction. A robust iterative weighted solution is adopted, namely:

[0135]

[0136]

[0137]

[0138] In the formula, Indicates the number of iterations; This is a robustness parameter, with a value of, for example, 3; This represents the arctangent function whose resultant direction is determined by the sign of the weighted sum of the sine and cosine direction components; after iteration, the value is taken as... .

[0139] When the prior reference axis exists, the robust iterative weighted solution is not performed; instead, the solution is directly derived from the prior reference axis. The reference axis set is defined as follows: When the surface under inspection contains only a single fiber bundle or is a unidirectional fabric, the reference axis set is... When there are two sets of orthogonal fiber bundles on the surface to be inspected, the reference axial set is: When the prior reference axis does not exist, the robust iterative weighted solution is performed to obtain the reference gradient direction. The reference axial set is (Single fiber bundle or unidirectional fabric), or (Two sets of orthogonal fiber bundles).

[0140] The fiber weave texture reference orientation field establishes a gradient axis reference for the normal fiber texture that should be presented in the local region for each valid center pixel in the current image to be detected. Representing coordinates The reference gradient direction is formed by the convergence of the edges of multiple fiber bundles in the same direction. This directional field itself does not directly determine the crack, but rather provides a pixel-by-pixel comparison benchmark that varies with the local texture direction to distinguish between normal periodic fiber textures and abnormal microcracks. At this point, the crack response calculation step begins.

[0141] Step S4: For each center pixel in the fiber weave texture reference direction field, calculate the dark value loss and the minimum directional deviation angle between the center pixel gradient direction axis and each axis in the corresponding reference axis set; determine the crack response value by the dark value loss and the directional deviation angle; construct the crack response field by all center pixels and their corresponding crack response values; determine candidate pixels according to the relative sorting of crack response values; establish an initial candidate connected region; perform directional guided aggregation on the initial candidate connected region according to the directional consistency condition, extension relationship condition, and spacing condition to obtain an aggregated candidate connected region; when the aggregated candidate connected region satisfies the joint crack determination condition, determine the aggregated candidate connected region as the crack region.

[0142] Based on the reference orientation field obtained in step S3, the actual gradient direction of the edge of a normal fiber bundle is usually close to the reference gradient direction, while the edge of a microcrack obliquely piercing the fiber weave texture tends to deviate from the reference gradient direction. Under uniform diffuse illumination, the interior of a microcrack or near its edge usually exhibits a dark value deficiency relative to the local surface. Relying solely on pixel brightness reduction cannot distinguish between microcracks and normal fiber dark stripes, and relying solely on orientation deviation cannot eliminate blemishes, low-gradient regions, and local brightness fluctuations. Therefore, this step combines the local dark value deficiency with the degree of orientation deviation to calculate the crack response. A true microcrack may be obscured by the bright and dark stripes of the fiber bundle, appearing as multiple separate high-response segments in the response field. If connected regions are directly output based solely on adjacency, the same true crack will be incorrectly segmented into multiple short regions. Fiber bundle intersections, local blemishes, scratches, illumination variations, and imaging noise may also form local high-value pixels in the response field. Therefore, this step first filters candidate pixels according to the relative order of response values, then establishes connected regions for the candidate pixels and performs orientation-guided aggregation, and finally determines the crack region.

[0143] In one embodiment of the present invention, the effective direction pixel set Each center pixel within Local dark value loss Defined as:

[0144]

[0145]

[0146] In the formula, the parameters Represents the center pixel Pixel intensity; parameters Indicated by The arithmetic mean of the intensities of all pixels within the actual direction aggregation window centered on the center; parameters Indicated by Centered on, with side length as The arithmetic mean of the intensity of all pixels within a square neighborhood. The value is a natural number, for example, 1 (i.e., a 3×3 neighborhood), and 2k+1 is not greater than half the side length of the actual direction aggregation window; pixels in the neighborhood that exceed the image boundary are padded by mirroring the nearest boundary; parameters This indicates taking the maximum value.

[0147] In the formula, Indicated by Centered The brightness difference between the neighborhood mean and the window mean; when the brightness difference is positive, the neighborhood is darker than the local mean. A one-sided truncation is applied to this difference, where the dark value deficit is set to 0 when the neighborhood is equal to or brighter than the local mean. The one-sided truncation corresponds to the imaging characteristic of microcracks exhibiting a dark line structure under uniform diffuse illumination. Locally bright pixels do not form a dark value deficit, and reflective points and locally bright fiber regions do not acquire a crack response due to large brightness differences.

[0148] Direction deviation angle for:

[0149]

[0150] In the formula, the parameters Represents the center pixel gradient direction; parameters Represents the center pixel Corresponding reference axis set; parameters This indicates a modulo operation with an output range of 1 / 2. ;parameter Indicates taking the absolute value; parameter The range of values ​​is (Single reference axis) or (Biorthogonal reference axis).

[0151] In the formula, both the gradient direction and the reference gradient direction are axial, differing by a certain distance. The directions correspond to the same axis. Translate the direction difference. Later Take the mold, then subtract And take the absolute value, folding the direction difference to... ,difference The direction difference is reduced to 0, and the phase difference is... Directional difference attributed to Eliminate numerically close azimuths when the azimuth angle is at either end of the azimuth range. The problem is that the actual axes are close to each other.

[0152] Crack response value for:

[0153]

[0154] In the formula, the parameters This represents the square of the sine of the directional deviation angle; when the gradient magnitude of the center pixel is zero, its gradient direction is undefined, so the crack response value of the center pixel is set to 0.

[0155] In the formula, the crack response is the product of the dark value deficit and the square of the sine of the directional deviation angle. Approaching 0 When the value approaches zero, the edge response aligned with the reference direction is suppressed. near hour When the value is close to 1, edge responses deviating from the reference direction are preserved. have It is periodic and insensitive to the sign of the orientation difference, corresponding to the half-period relationship of the gradient axis. The product form requires that the pixel simultaneously has dark value deficit and orientation deviation; the response is 0 when either factor is 0. All center pixels are considered. and the corresponding crack response value The crack response field of the current image is formed by combining these elements.

[0156] Candidate pixels were determined based on the crack response value, including:

[0157] Set of effective direction pixels All crack response values ​​are arranged in ascending order; the sorted sequence is denoted as... In the formula, the parameters This represents the total number of center pixels within the set of valid directional pixels; when When no crack candidate is detected, the processing status is output and the process terminates.

[0158] The natural interval between adjacent sorted response values ​​is , ;

[0159] The location of the maximum natural interval is determined as the adaptive splitting point of the response distribution; splitting point index. Indicates that the splitting interval is located at and Between; when the maximum natural interval appears in multiple positions in parallel, the position with the smallest index is taken as the split position index; when all crack response values ​​are equal, the processing status "no crack candidate detected" is output and the process is terminated;

[0160] Sort index The corresponding response value is used as the low-value body; the sorting index is used. The corresponding response value is taken as the high-value tail; the response value and Equal pixels are all included in the high-value tail;

[0161] Pixels corresponding to high tail response values ​​are identified as candidate pixels; candidate pixel set Composed of all candidate pixels; Recorded as ; As the adaptive response threshold.

[0162] In obtaining the candidate pixel set Next, initial candidate connected regions are established according to the eight-adjacency relationship. For any two candidate pixels, if the two candidate pixels are adjacent in the horizontal, vertical, or diagonal direction, the two candidate pixels are determined to have a direct connectivity relationship. All candidate pixels connected to each other through direct connectivity or continuously passed connectivity relationships are grouped into the same initial candidate connected region. The initial candidate connected regions are numbered according to the row priority of the smallest pixel coordinates within the region, so that the same set of candidate pixels always obtains the same region number and processing order.

[0163] For each initial candidate connected region The minimum bounding rectangle is determined based on the coordinates of all candidate pixels within the region, and the dominant direction of the region is used. It is expressed in radians, taking the axial direction corresponding to the longer side of the smallest circumscribed rotating rectangle, with the direction difference being... The two long sides represent the same axis. The dominant directions of a region are compared according to their axial relationship, without distinguishing between the forward and reverse directions of the extension. If the initial candidate connected region contains only one pixel, or the spatial distribution of pixels within the region cannot form a unique long side direction, the corresponding region is marked as a region with an undetermined direction. Regions with undetermined directions do not participate in direction-guided aggregation, but are still retained as independent candidate regions.

[0164] Perform direction-guided aggregation on any two regions to be compared whose dominant regional direction is determined. Allowable deviation for direction merging. In each comparison of the initial candidate connected regions With the initial candidate connected region The time is adaptively determined based on the geometric dimensions of the region in the current image, that is:

[0165]

[0166] In the formula, and Let represent the lengths of the longer side of the minimum bounding rectangle of the two regions, respectively. Represents the arctangent function. This indicates taking the minimum value in the set.

[0167] In the formula, Indicates that the length of the long side is The directional resolution error caused by a one-pixel lateral offset of the boundary in the region. Similarly, the sum of the two is the allowable deviation for the direction of this comparison, and then... The upper bound is used for truncation. The shorter the longer side, the greater the directional error caused by pixel discretization, and the larger the allowable deviation. After the regions are merged, the allowable deviation for directional merging used in subsequent comparisons is recalculated based on the length of the longer side of the updated minimum bounding rotation rectangle.

[0168] When the minimum axial deviation between the dominant directions of the two regions to be compared is not greater than When the condition of directional consistency is met, the minimum axial deviation between the direction connecting the adjacent boundaries of two regions to be compared and the dominant direction of the region is not greater than 1. When the extension relationship condition is met, the spacing condition is met. When the minimum pixel spacing between adjacent boundaries of a region does not exceed half the side length of the actual direction aggregation window, the spacing condition is met. When the direction consistency condition, extension relationship condition, and spacing condition are all met, the pixel paths between adjacent boundaries of two regions to be compared are connected along the dominant direction of the region, and the pixel paths are recorded as the connected pixel sequence generated by region aggregation. The connected pixels are only used to form a continuous candidate region coordinate description and do not change the crack response field that has already been formed.

[0169] Preferably, the Euclidean distance between the pair of pixels with the smallest Euclidean distance among the boundary pixels of two regions is calculated to obtain the minimum pixel spacing between adjacent boundaries of regions; half of the side length of the actual direction aggregation window is directly compared with the minimum pixel spacing in the real number range without rounding; the direction of the line connecting the two pixels corresponding to the minimum pixel spacing is used as the direction connecting the adjacent boundaries of the two regions to be compared; when the two pixels corresponding to the minimum pixel spacing are not unique, the direction of the line with the smaller axial deviation from the dominant direction of the two regions is taken; when it still cannot be uniquely determined, merging is not performed; the pixel path is generated by digital straight line interpolation, with the minimum spacing pixel pair as the endpoint.

[0170] After each region merge, the minimum bounding rectangle is redefined based on the candidate pixels and connected pixels within the merged region. The dominant direction, length of the long side, and length of the short side of the region are updated. The search continues for other initial candidate connected regions that have not yet been merged into the current region and satisfy the conditions of directional consistency, extension relationship, and spacing, until no further mergeable regions remain. Region search and merging are performed in ascending order of the initial candidate connected region numbers, ensuring that the same input data yields a unique aggregation result. After all regions are aggregated, the resulting set of candidate connected regions is encapsulated. For each aggregated candidate connected region, the encapsulation content records the original candidate pixel coordinate sequence, the connected pixel coordinate sequence formed during the directional guidance aggregation process, the source initial candidate connected region number, the dominant direction of the aggregated region, the lengths of the long and short sides of the minimum bounding rectangle, the crack response value sequence corresponding to the original candidate pixels, and the original image tracing index.

[0171] The first after aggregation The set of original candidate pixel coordinates obtained by crack response filtering in the aggregated candidate connected regions is denoted as . The set of connected pixel coordinates formed during the directional guidance aggregation process is denoted as... Complete set of pixel coordinates satisfy:

[0172]

[0173] In the formula, the parameters Represents a union; when no fragment merging occurs. It is an empty set.

[0174] In the formula, the complete pixel set is the union of the original candidate pixel set and the connected pixel set. The connected pixels are only used to connect the separated high-response segments, without changing the attribution of the original candidate pixels or supplementing the crack response value for the connected pixels. All pixels in the image are derived from the candidate pixel set. The pixels in the image are only used to describe the connection locations formed by directional guidance aggregation between different high-response fragments. When no fragment merging occurs in the current region... It is an empty set. Based on the complete set of pixel coordinates. Determine the first The minimum bounding rectangle of the aggregated candidate connected regions is denoted by its long side length and short side length as follows: and All units are pixels, and the axial direction corresponding to the longer side is denoted as... During calculation, each pixel is treated as a pixel unit with a side length of one pixel to avoid the short side length of a single pixel width region being zero.

[0175] When all of the following conditions are met, the first The aggregated candidate connected regions were identified as crack regions:

[0176] (1) (2) Slenderness condition In the formula For example, the slenderness ratio threshold. (3) Minimum length condition In the formula , For example, the minimum physical length of the crack (4) Average dark value deficit condition (5) Gradient verticality condition In the formula , representing the minimum axial deviation between the gradient axis of the candidate pixel and the dominant direction of the region; (6) when no fragment merging occurs, the dominant direction of the source initial candidate connected region can be determined; (7) when fragment merging occurs, the two regions to be compared involved in each merging satisfy the direction consistency condition, extension relationship condition and spacing condition; (8) connected pixels are not used as original candidate pixels to participate in crack response statistics. Among them, conditions (2) and (3) ensure that the candidate region has sufficient elongation and reaches the minimum physical length, condition (4) ensures that the region as a whole has a stable dark value loss, and condition (5) ensures that the gradient direction of the candidate pixel is approximately perpendicular to the long axis of the region, that is, it satisfies the direction characteristics of the crack edge gradient. For single pixel regions, clumped regions with the same long and short sides, and degenerate regions that cannot form a unique dominant direction, they are not identified as crack regions. This judgment process only acts on the final output and does not modify the crack response field, candidate pixel set, initial candidate connected region or direction guidance aggregation result in reverse. Wherein, Dmin is the average dark value loss threshold, with the dimension of gray level and a value of, for example, 10; Smin is the gradient verticality threshold, dimensionless and a value of, for example, 0.5; when the candidate pixel distribution of the aggregated candidate connected region has a significant curved shape along the dominant direction of the region, the region is divided into several continuous sub-segments along its dominant direction, the regional axis of each sub-segment is determined, and the gradient verticality condition is applied to each sub-segment.

[0177] If no region is identified as a crack area, output the processing status "No crack area detected".

[0178] At this point, pixels that simultaneously possess both dark value deficit and directional deviation features receive higher crack responses, while pixels with only a single feature receive lower responses. High-response segments of the same crack, which are truncated by the light and dark stripes of the fiber bundle, are aggregated and connected into continuous candidate regions through directional guidance. Aggregated candidate connected regions that meet the judgment criteria are recorded as crack regions for output.

[0179] This completes the visual inspection of microcracks on the surface of aerospace composite materials.

[0180] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0181] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A visual inspection method for microcracks on the surface of aerospace composite materials, characterized in that, The method includes: Acquire grayscale images of the surface of aerospace composite material skin, as well as image width, image height, spatial resolution, and nominal center-to-center spacing of fiber bundles in the same group; The pixel side length of the candidate direction aggregation window is determined based on the nominal center spacing of the fiber bundles in the same group; the candidate direction aggregation windows are filtered according to the image width and image height; for the filtered candidate direction aggregation windows, the gradient magnitude and gradient direction of the pixels within the window are calculated; the gradient direction of each pixel is doubled by an angle; the direction consistency evaluation value of each candidate direction aggregation window is determined using the gradient magnitude as the weight; the candidate direction aggregation window with the largest direction consistency evaluation value is determined as the actual direction aggregation window. The gradient direction of the pixels within the actual direction aggregation window is doubled in angle; weighted aggregation is performed using the gradient magnitude as the weight to obtain the fiber weave texture reference gradient direction of each center pixel; the fiber weave texture reference direction field is formed by the set of all center pixel positions and their corresponding reference axes; when there are two sets of orthogonal fiber bundles on the surface to be inspected, the set of reference axes includes the reference gradient direction and its orthogonal direction; For each central pixel within the fiber weave texture reference direction field, calculate the dark value loss and the minimum directional deviation angle between the central pixel gradient direction axis and each axis in the corresponding reference axis set; determine the crack response value based on the dark value loss and the directional deviation angle; construct a crack response field from all central pixels and their corresponding crack response values; determine candidate pixels according to the relative order of the crack response values; establish an initial candidate connected region; perform directional guidance aggregation on the initial candidate connected region according to the directional consistency condition, extension relationship condition, and spacing condition to obtain an aggregated candidate connected region; when the aggregated candidate connected region satisfies the joint crack determination condition, determine the aggregated candidate connected region as a crack region.

2. The visual inspection method for microcracks on the surface of aerospace composite materials according to claim 1, characterized in that, The grayscale image was acquired by a spatially calibrated area array industrial camera under uniform diffuse illumination. The grayscale image is saved in uncompressed grayscale TIFF format. A single trigger generates a complete surface image. The image width and height are obtained by reading the width and height fields from the image file header. The spatial resolution is obtained by reading the pixel equivalent field from the camera spatial calibration record. The calibration period covers the current acquisition time. The traceability information is bound to the grayscale image. This traceability information is not involved in direction calculation, response calculation, or crack category determination. The traceability information includes a unique image identifier, original file location information, acquisition time, and acquisition device number. This traceability information is transmitted read-only along with the data objects in each step. The determination of the nominal center spacing of the same group of fiber bundles includes: reading the center spacing of the same group of fiber bundles from a controlled document corresponding to the model, material type, and material batch of the component under test. The controlled document includes material technical specifications, fiber fabric supply technical documents, or component manufacturing process documents. The controlled document identifier, document version, and applicable material batch are recorded. When the controlled document only records the repeat length of a complete braided unit, the center spacing of a single group of fiber bundles is calculated based on the number of intervals between the same group of fiber bundles contained in a complete braided unit. When the controlled document does not record the nominal center spacing of the same group of fiber bundles, a reference sample with the same material type and braided structure as the component under test is measured. The actual distance between the center lines of multiple groups of adjacent fiber bundles is measured along a direction perpendicular to the extension of the same group of fiber bundles. The median of the measured distances is taken as the nominal center spacing of the same group of fiber bundles. The measurement is performed using a spatially calibrated microscopic imaging device or an equivalent size measuring device. The measurement results, measuring device, spatial calibration information, material batch, and applicable scope are written into the actual measurement calibration record. When there are two or more groups of fiber bundles with different nominal center-to-center distances on the surface to be inspected, the nominal center-to-center distance of each group of fiber bundles is obtained separately. The nominal center-to-center distance with the largest value is taken as the nominal center-to-center distance of the same group of fiber bundles.

3. The visual inspection method for microcracks on the surface of aerospace composite materials according to claim 1, characterized in that, by Indicates the candidate spacing multiple; Starting from 2, increase sequentially; side length of the candidate direction aggregation window. Not less than The smallest odd-numbered pixel side length corresponding to the number of pixels corresponding to the nominal center-to-center spacing of the same fiber bundle, i.e.: In the formula, the parameters Integer pixel side length of the candidate direction aggregation window; parameter Indicates parameters Belongs to the set of positive integers; parameters Indicates the nominal center-to-center spacing of fiber bundles in the same group; parameter Indicates spatial resolution; parameters Represents the AND relation; parameters This represents the modulo operation; Candidate Spacing Multiple Set satisfy: In the formula, the parameters Indicates image width; parameter Indicates image height; parameter This represents the smaller of the image width and image height. When the set of candidate spacing multiples is empty, output the processing status "Image size does not meet the direction aggregation condition".

4. The visual inspection method for microcracks on the surface of aerospace composite materials according to claim 3, characterized in that, The gradient magnitude and gradient direction are calculated using the central difference method; the gradient direction is referenced to the positive horizontal axis of the image as 0 radians; the value range of the gradient direction is... ; For the center pixel Centered on, with side length as The candidate direction aggregation window has a total of pixels. Pixels within the window are numbered in row-major order; the first The image coordinates of pixels are ; Sine direction component weighted sum Cosine direction component weighted sum With gradient modulus total for: In the formula, the parameters Indicates the first Gradient magnitude of pixels; parameters Indicates the first Gradient direction of each pixel; parameters Represents the sine function; parameters Represents the cosine function; Directional Concentration for: In the formula, the parameters It is a dimensionless quantity, and its range of values ​​is 1. .

5. The visual inspection method for microcracks on the surface of aerospace composite materials according to claim 4, characterized in that, Candidate direction aggregation window half-side length Effective center pixel set for: In the formula, the parameters , These represent the x and y coordinates of the center pixel, respectively. Directional Consistency Evaluation Value Set of effective center pixels Concentration in all directions The median; when the effective center pixel set When the number of inward concentrations is even, the arithmetic mean of the two middle values ​​after sorting is taken as the median. When multiple candidate spacing multiples correspond to the same maximum directional consistency evaluation value, the candidate spacing multiple with the smallest value is taken as the adaptive directional aggregation spacing multiple. When the directional consistency evaluation values ​​of all candidate spacing multiples are lower than the lower limit of directional consistency. If the braided structure recorded in the controlled document provides a priori fiber axis, then the smallest spacing multiple in the candidate spacing multiple set is taken as the adaptive direction aggregation spacing multiple Np, and the priori fiber axis is recorded as the priori reference axis to continue the subsequent steps; otherwise, the processing status "fiber braided texture reference direction cannot be determined" is output. Actual direction aggregation window side length In the formula, the parameters Indicates the adaptive directional aggregation spacing multiple; parameter This indicates the side length of the aggregation window in the actual direction.

6. The visual inspection method for microcracks on the surface of aerospace composite materials according to claim 5, characterized in that, For the center pixel The actual orientation aggregation window centered on the center, the total number of pixels in the window is Pixels within the window are numbered in row-major order; the first The image coordinates of pixels are ; Weighted sum of sinusoidal direction components corresponding to the actual direction aggregation window Weighted sum of cosine direction components for: In the formula, the parameters Indicates the first Gradient magnitude of pixels; parameters Indicates the first Gradient direction of each pixel; Effective direction pixel set for: In the formula, the parameters This represents the set of effective center pixels where the actual direction aggregation window can be fully expanded and all pixels within the window can undergo center difference; ; When the prior reference axis and the effective direction pixel set do not exist If empty, output the processing status "Fiber weave texture reference direction cannot be determined" and terminate; For each valid center pixel, the fiber weave texture references the gradient direction. A robust iterative weighted solution is adopted, namely: In the formula, Indicates the number of iterations; This is a robustness parameter, with a value of, for example, 3; This represents the arctangent function whose resultant direction is determined by the sign of the weighted sum of the sine and cosine direction components; after iteration, the value is taken as... . When the prior reference axis exists, the robust iterative weighted solution is not performed; instead, the solution is directly derived from the prior reference axis. The reference axis set is defined as follows: When the surface under inspection contains only a single fiber bundle or is a unidirectional fabric, the reference axis set is... When there are two sets of orthogonal fiber bundles on the surface to be inspected, the reference axial set is: When the prior reference axis does not exist, the robust iterative weighted solution is performed to obtain the reference gradient direction. The reference axial set is (Single fiber bundle or unidirectional fabric), or (Two sets of orthogonal fiber bundles).

7. The visual inspection method for microcracks on the surface of aerospace composite materials according to claim 6, characterized in that, Dark value loss for: In the formula, Represents the center pixel Pixel intensity; Indicated by The arithmetic mean of the intensity of all pixels within the actual direction aggregation window centered on the center; Indicated by Centered on, with side length as The arithmetic mean of the intensity of all pixels within a square neighborhood. It is a natural number, with values ​​such as 1 (i.e., a 3×3 neighborhood), and 2k+1 is not greater than half the side length of the actual direction aggregation window; pixels in the neighborhood that exceed the image boundary are padded by mirroring the nearest boundary. This indicates taking the maximum value. Direction deviation angle for: In the formula, Represents the center pixel The gradient direction; Represents the center pixel The corresponding set of reference axes; This indicates a modulo operation with an output range of 1 / 2. ; This indicates taking the absolute value; when the reference axis set contains only one axis, The range of values ​​is When the set of reference axes contains two orthogonal axes, The range of values ​​is . Crack response value for: In the formula, the parameters This represents the square of the sine of the directional deviation angle; when the gradient magnitude of the center pixel is zero, its gradient direction is undefined, so the crack response value of the center pixel is set to 0.

8. The visual inspection method for microcracks on the surface of aerospace composite materials according to claim 7, characterized in that, Candidate pixels were determined based on the crack response value, including: Set of effective direction pixels All crack response values ​​are arranged in ascending order; the sorted sequence is denoted as... In the formula, the parameters This represents the total number of center pixels within the set of valid directional pixels; when When no crack candidate is detected, the processing status is output and the process terminates; when When no crack candidate is detected, the processing status is output and the process terminates. The natural interval between adjacent sorted response values ​​is , ; The location of the maximum natural interval is determined as the adaptive splitting point of the response distribution; splitting point index. Indicates that the splitting interval is located at and Between; when the maximum natural interval appears in multiple positions in parallel, the position with the smallest index is taken as the split position index; when all crack response values ​​are equal, the processing status "no crack candidate detected" is output and the process is terminated; Sort index The corresponding response value is used as the low-value body; the sorting index is used. The corresponding response value is taken as the high-value tail; the response value and Equal pixels are all included in the high-value tail; Pixels corresponding to high tail response values ​​are identified as candidate pixels; candidate pixel set Composed of all candidate pixels; Recorded as ; As the adaptive response threshold.

9. The visual inspection method for microcracks on the surface of aerospace composite materials according to claim 1, characterized in that, The initial candidate connected regions are numbered in row priority order according to the smallest pixel coordinates within the region; For each initial candidate connected region, determine the minimum bounding rectangle based on the coordinates of all candidate pixels within the region; the dominant direction of the region is the axial direction corresponding to the longer side of the minimum bounding rectangle; the direction difference is... The two long sides are taken as the same axis; For any two regions to be compared whose dominant direction can be determined, the allowable deviation for direction merging is... for: In the formula, and Let represent the lengths of the longer side of the minimum bounding rectangle of the two regions, respectively. Represents the arctangent function. This indicates taking the minimum value in the set; When the minimum axial deviation between the dominant directions of the two regions to be compared is not greater than When the condition of consistent direction is met, the condition for determining direction consistency is met. When the minimum axial deviation between the direction connecting the adjacent boundaries of two regions to be compared and the dominant direction of the region is no greater than When the extension relationship condition is met, the spacing condition is met; when the minimum pixel spacing between adjacent boundaries of a region does not exceed half the side length of the actual direction aggregation window, the spacing condition is met. When the direction consistency condition, extension relationship condition and spacing condition are all met, connect the pixel paths between the adjacent boundaries of the two regions to be compared along the dominant direction of the region. Record the pixel path as a sequence of connected pixels; Calculate the Euclidean distance between the pair of pixels with the smallest Euclidean distance among the boundary pixels of two regions to obtain the minimum pixel spacing between adjacent boundaries of regions; actual Half the side length of the direction aggregation window is directly compared with the minimum pixel pitch within the real number range; the direction of the line connecting the two pixels corresponding to the minimum pixel pitch is used as the direction connecting the adjacent boundaries of the two regions to be compared; when the two pixels corresponding to the minimum pixel pitch are not unique, the direction of the line connecting the two regions with the smaller axial deviation from the dominant direction of the region is taken; when it still cannot be uniquely determined, merging is not performed. The pixel path is generated using digital linear interpolation; The pixel path uses the smallest spacing pixel pair as its endpoint; After each region merge, the minimum bounding rectangle is redefined based on the candidate pixels and connected pixels within the merged region. Update the dominant direction, long side length, and short side length of the region; recalculate the allowable deviation for directional merging used in subsequent comparisons; Region search and merging are performed in ascending order of the initial candidate connected region numbers; Single-pixel regions and degenerate regions that cannot form a unique long side direction do not participate in directional guidance aggregation; the single-pixel regions and the degenerate regions are retained as independent candidate regions.

10. The visual inspection method for microcracks on the surface of aerospace composite materials according to claim 9, characterized in that, The first after aggregation The set of original candidate pixel coordinates obtained by crack response filtering in the aggregated candidate connected regions is denoted as . ; The set of connected pixel coordinates formed during the directional guidance aggregation process is denoted as... Complete set of pixel coordinates satisfy: In the formula, the parameters Represents a union; when no fragment merging occurs. It is an empty set; Based on the complete set of pixel coordinates Determine the first The minimum bounding rectangle of the aggregated candidate connected regions; the length of the longer side is denoted as . The length of the shorter side is denoted as The axial direction corresponding to the longer side is denoted as During the calculation, each pixel is treated as a pixel unit with a side length of one pixel. When all of the following conditions are met, the first The following candidate connected regions were identified as crack regions: (1) (2) Slenderness condition In the formula (3) Minimum length condition. In the formula , The minimum physical length of the crack. For spatial resolution; (4) Mean dark value deficit condition (5) Gradient verticality condition In the formula , representing the minimum axial deviation between the candidate pixel gradient axis and the dominant direction of the region; (6) when no fragment merging occurs, the dominant direction of the source initial candidate connected region can be determined; (7) when fragment merging occurs, the two regions to be compared involved in each merging satisfy the direction consistency condition, extension relationship condition, and spacing condition; (8) connected pixels are not used as original candidate pixels in crack response statistics. Among them, The average dark value loss threshold is measured in gray levels and has a value of, for example, 10. The gradient verticality threshold is dimensionless and takes a value of, for example, 0.

5. When the candidate pixel distribution of the aggregated candidate connected region exhibits a significant curved shape along the dominant direction of the region, the region is divided into several continuous sub-segments along its dominant direction, the regional axis of each sub-segment is determined, and the gradient verticality condition is applied to each sub-segment. If no region is identified as a crack region, output the processing status "No crack region detected"; The aggregated candidate connected regions identified as crack regions are encapsulated as crack region records; the crack region records include the original candidate pixel coordinate sequence, the connected pixel coordinate sequence, the source initial candidate connected region number, the dominant direction of the aggregated region, the length of the long side and the length of the short side of the minimum circumscribed rotation rectangle, the crack response value sequence corresponding to the original candidate pixels, and the original image tracing index.

Images