A method and system for generating image primitives for composite hole machining damage detection

Abstract

The application discloses a kind of composite hole processing damage detection image-based element generation method and system, wherein the method comprises: obtaining composite hole processing export area image, to obtain effective material area image;From effective material area image, extract several features, fuse feature to obtain fusion feature;The fusion feature is divided into several local sub-regions, and initial seed reference point is determined in each local sub-region;With initial seed reference point as center, execute near neighbor search, and screen seed expansion node;Unattributed pixel node, initial seed reference point and seed expansion node are constructed as sparse region propagation graph;Initial seed reference point and seed expansion node are regarded as known constraint node, and the region membership degree of unattributed pixel node is propagated and solved;According to region membership degree, determine the region label of unattributed pixel node, output the region-based element generation result of hole processing export area image.

Description

Technical Field

[0001] This application relates to the field of composite material processing quality inspection and digital image processing technology, and in particular to an image primitive generation method and system for detecting hole processing damage in composite materials. Background Technology

[0002] Composite materials possess characteristics such as high temperature resistance, thermal shock resistance, ablation resistance, and high specific strength, and are commonly used in critical load-bearing components and thermal protection components in aerospace. These components typically require hole machining for bolting, riveting, or assembly positioning. The quality of hole machining directly affects the contact state of the hole edges, the local load transfer path, and the structural reliability during service.

[0003] Because composite materials consist of a multiphase structure including a matrix phase, fiber phase, interface phase, or surface coating, different phase regions exhibit significant differences in hardness, brittleness, toughness, removal methods, and interfacial bonding strength. During hole machining, the hole exit area is prone to various types of damage, such as matrix breakage, fiber pull-out, interface peeling, and coating collapse. These damaged areas often exhibit characteristics such as grayscale overlap, texture interlacing, gradual boundary changes, and localized weak contrast in images.

[0004] Traditional thresholding, edge detection, and conventional region growing methods typically rely on intensity differences or sharp edges. When the grayscale distribution of the coating collapse area overlaps with that of the original surface, or when the textures of fibrous phase damage and matrix phase damage are similar, single grayscale or gradient criteria are prone to region confusion and boundary shift. On the other hand, when using pixels directly as processing units for subsequent classification and recognition, the number of nodes is large, noise has a significant impact, and the stability of category determination in local weak boundary regions is insufficient.

[0005] Image region pre-segmentation methods can aggregate a large number of pixels into locally consistent regions, reducing the input size of subsequent models while preserving local boundary and structural information. However, conventional graph propagation methods are quite sensitive to the initial seed position. If the seed falls on a strong edge, a sudden change in texture, or the boundary of a phase region, subsequent region attribution propagation is easily interfered with. Furthermore, conventional methods typically perform edge connections and probability propagation across the entire image pixel grid, which can easily retain a large number of connections that are weakly related to region attribution, leading to computational redundancy and high solution overhead.

[0006] Therefore, a method for generating region primitives that adapts to the complex textures, weak boundaries, and coexistence of multiphase regions in composite material hole processing damage images is needed. This method should be able to adaptively adjust feature contributions based on local image content, guide seed reference points to avoid unstable boundary regions, and incorporate high-confidence regions into known constraint nodes in advance through a seed expansion mechanism. This improves boundary fitting capability while reducing the solution scale of the region propagation graph. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this application provides an image primitive generation method and system for detecting hole processing damage in composite materials. This method utilizes the local differences in color, gradient, and texture information in hole processing damage images to construct a fused feature field, and uses initial seed reference points and seed expansion nodes to jointly constrain the region propagation process, thereby generating region primitives suitable for subsequent identification and quantitative evaluation without the need for manual labeling of damage boundaries.

[0008] This invention is particularly applicable to visual inspection of hole processing quality in fiber-reinforced ceramic matrix composites, fiber-reinforced resin matrix composites, and other composite materials with multiphase heterogeneous structures. It can be used as a front-end processing method for subsequent damage classification and identification, damage area calculation, damage boundary extraction, and processing quality evaluation.

[0009] On the one hand, a method for generating image primitives for detecting hole processing damage in composite materials is provided, including: The image of the hole processing exit area of ​​the composite material is obtained, and the image of the hole processing exit area is preprocessed to obtain the effective material area image; Several features are extracted from the image of the effective material region, and all extracted features are fused to obtain fused features; The fusion features are divided into several local sub-regions, and an initial seed reference point is determined in each local sub-region. Perform nearest neighbor search centered on the initial seed reference point, and filter the seed expansion nodes; The unassigned pixel nodes, the initial seed reference point, and the seed expansion nodes are constructed into a sparse region propagation graph, and the edge connection relationships in the propagation graph are determined. Using the initial seed reference point and seed expansion node as known constraint nodes, the region membership degree of unassigned pixel nodes is solved by propagation. Based on the region membership degree, the region label of the unassigned pixel node is determined, and the region primitive generation result of the composite material hole processing exit region image is output.

[0010] On the other hand, an image primitive generation system for detecting hole processing damage in composite materials is provided, comprising: The image acquisition and preprocessing module is configured to: acquire an image of the hole processing exit area of ​​the composite material, and preprocess the image of the hole processing exit area to obtain an effective material area image; The multi-source feature fusion module is configured to: extract several features from the effective material region image, and fuse all the extracted features to obtain fused features; The seed reference point determination module is configured to divide the fused features into several local sub-regions and determine an initial seed reference point in each local sub-region. The seed expansion module is configured to perform nearest neighbor search centered on the initial seed reference point and filter seed expansion nodes; The sparse region propagation module is configured to: construct a sparse region propagation graph from unclaimed pixel nodes, initial seed reference points, and seed expansion nodes, and determine the edge connectivity relationships in the propagation graph; The region primitive output module is configured to: use the initial seed reference point and seed expansion node as known constraint nodes, solve the region membership degree propagation problem for unassigned pixel nodes; determine the region label of the unassigned pixel node based on the region membership degree, and output the region primitive generation result of the composite material hole processing exit region image.

[0011] Furthermore, an electronic device is also provided, including: Memory, used for non-transitory storage of computer-readable instructions; and Processor, for executing the computer-readable instructions, When the computer-readable instructions are executed by the processor, they perform the method described in the first aspect above.

[0012] In another aspect, a storage medium is also provided for non-transitory storage of computer-readable instructions, wherein when the non-transitory computer-readable instructions are executed by a computer, the method described in the first aspect is performed.

[0013] In another aspect, a computer program product is also provided, including a computer program that, when run on one or more processors, is used to implement the method described in the first aspect above.

[0014] The above technical solution has the following advantages or beneficial effects: This invention constructs a fusion feature field based on local statistical differences, which can adapt to the situation of gray-level overlap and texture similarity in hole processing damage images of multiphase regions of composite materials, and improve the ability to distinguish weak boundary regions. This invention determines the initial seed reference point through local stability constraints and spatial compactness constraints, so that the seed is located as close as possible to the local consistent region, reducing the propagation error caused by the seed falling on strong edges or category transition regions. This invention incorporates high-confidence pixels into known constraint nodes in advance through a seed expansion mechanism, which can reduce the scale of unknown nodes and enhance the stability of region propagation. The sparse region propagation map of this invention only retains the local connectivity related to the region attribution judgment, which can improve the solution efficiency while maintaining the boundary fitting ability. The region primitive results output by this invention include region members, boundaries, centers and adjacency relationships, which can be directly used for subsequent damage classification and recognition, damage boundary extraction, damage area calculation and hole processing quality evaluation. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0016] Figure 1 This is a schematic diagram of multi-source feature fusion and local weight adaptation in this invention; Figure 2 A schematic diagram of the effective area and seed distribution of the hole processing exit damage image provided by the present invention; Figure 3 This is a schematic diagram of seed expansion in this invention; Figure 4 This is a schematic diagram of the sparse region propagation map construction in this invention; Figure 5 This is a schematic diagram of the region primitive generation results and adjacency relationships in this invention; Figure 6 This is a schematic diagram of the system module composition provided by the present invention. Detailed Implementation

[0017] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, 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 application pertains.

[0018] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of this application. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] As described in the background section, the exit area of ​​composite material hole processing often suffers from problems such as hole background, weak boundaries, and complex fiber textures. To address this issue, this invention proposes an image primitive generation method for composite material hole processing damage detection. This method involves multi-source feature fusion, stable seed point distribution determination, seed point expansion, and sparse region propagation within the effective material area, thereby obtaining a region-level representation suitable for damage identification and quantification.

[0020] Example 1 This embodiment provides a method for generating image primitives for detecting hole processing damage in composite materials; A method for generating image primitives for detecting hole processing damage in composite materials, comprising: S101: Obtain an image of the hole processing exit area of ​​the composite material, and preprocess the hole processing exit area image to obtain an effective material area image; S102: Extract several features from the effective material region image, and fuse all extracted features to obtain fused features; S103: Divide the fused features into several local sub-regions, and determine the initial seed reference point in each local sub-region; S104: Perform nearest neighbor search centered on the initial seed reference point and select seed expansion nodes; S105: Construct a sparse region propagation graph from unassigned pixel nodes, initial seed reference point, and seed expansion nodes, and determine the edge connectivity in the propagation graph; S106: Using the initial seed reference point and seed expansion node as known constraint nodes, solve the region membership degree propagation for unassigned pixel nodes; determine the region label of the unassigned pixel nodes based on the region membership degree, and output the region primitive generation result of the composite material hole processing exit region image.

[0021] Further, in step S101: acquiring an image of the hole machining exit area of ​​the composite material, wherein the hole machining exit area refers to the hole opening and surrounding area on the side of the composite material workpiece through which the drill bit exits along the feed direction. During image acquisition, the camera optical axis is oriented towards the exit side surface, and the hole opening and the damaged area around the hole are within the imaging field of view.

[0022] Further, in step S101: the image of the hole processing exit area is preprocessed to obtain an image of the effective material area. The preprocessing includes: hole area screening, noise suppression, and brightness uniformity processing.

[0023] Furthermore, the screening of the porous area specifically includes: A hole mask is generated based on the prior of the circular hole, grayscale distribution, connected component morphology, or a combination thereof, and blank areas inside the hole, background areas with brightness below a set threshold, or low-information areas are excluded from the graph node set.

[0024] Specifically, the image of the hole processing exit area is converted into a grayscale image, a grayscale histogram is plotted, and a grayscale threshold is determined using the valley between the two peaks of the bimodal histogram or the maximum inter-class variance method. Candidate dark areas are then generated based on the grayscale threshold. :

[0025] in, Indicates pixel position; Represents pixels grayscale value; Indicates the grayscale threshold; Represents pixels It belongs to the candidate dark region. Represents pixels It does not belong to the candidate dark region.

[0026] For each candidate dark region, sequentially perform closing operations, region filling, and connected component labeling, and calculate the area, perimeter, geometric center, equivalent radius, and roundness of each connected component. Among these, the... Roundness of connected components The calculation formula is:

[0027] in, Indicates the first The area of ​​each connected region; Indicates the first The perimeter of a connected region; Indicates the first The roundness of a connected domain. The closer it is to 1, the better. The closer a connected component is to a circle.

[0028] Select the connected region with the largest area, a roundness not less than the set roundness threshold, and a geometric center and equivalent radius that conform to the prior range of the circular hole to be processed as the hole region, and set the mask value in this region to 1, and set the mask value in other positions to 0.

[0029] The effective material region is determined based on the hole mask. Subsequent feature extraction, seed initialization, graph node and edge establishment, and region membership propagation are all performed only within the effective material region with a mask value of 0. Pixels with a mask value of 1 do not establish graph nodes and related edges, thereby eliminating the interference of blank spaces, dark backgrounds, and low-information regions inside the hole on the generation of region primitives and reducing the size of the propagation graph.

[0030] The low-information region refers to a region where the local gray-level mean is no higher than a set mean threshold, the local gray-level variance is no higher than a set variance threshold, and the local average gradient is no higher than a set gradient threshold. To avoid accidentally deleting normal smooth regions of the material surface, only the aforementioned low-information regions connected to the hole region or located within the prior range of the circular hole are incorporated into the hole mask. This ensures that blank areas, dark background areas, or other low-information regions within the hole do not participate in subsequent node construction, resulting in an effective material region image. This step avoids interference from the hole background on the region primitive generation results.

[0031] Furthermore, the noise suppression specifically includes: The image of the hole machining exit area is converted to the YCbCr color space, and a 3×3 median filter is applied to the luminance component. For any pixel, its luminance value is replaced with the median of the luminance values ​​in its 3×3 neighborhood; for pixels with incomplete neighborhoods at image boundaries, the neighborhood is filled in using a mirroring method. The chrominance component remains unchanged, and the filtered luminance component is recombined with the original chrominance component to obtain the noise-suppressed image. Noise suppression is used to remove isolated impulse noise while preserving grayscale abrupt changes at hole edges and damage boundaries.

[0032] Furthermore, the brightness uniformity processing specifically includes: Within the effective material region, the 1st and 99th percentile values ​​of the brightness component of the image after noise suppression are statistically analyzed and used as the lower limit and upper limit of brightness stretching, respectively. Pixel values ​​below the lower limit of brightness stretching are truncated to the lower limit of a preset brightness range, and pixel values ​​above the upper limit of brightness stretching are truncated to the upper limit of a preset brightness range. Pixel values ​​between the two brightness stretching limits are then linearly mapped. Subsequently, the processed luminance component is recombined with the original chrominance component to obtain an effective material region image with luminance uniformity.

[0033] Using percentile values ​​instead of minimum and maximum brightness values ​​to determine the stretching range can reduce the impact of a few abnormal bright or dark spots on the brightness uniformity result.

[0034] Furthermore, the extraction of several features from the effective material region image includes: extracting color features, gradient features, and texture features for each pixel in the image; wherein, the color features include: luminance components and chromaticity components; the gradient features are used to characterize the intensity changes of hole edge chipping, local breakage, or coating edge; and the texture features are used to characterize the local structural differences of the fiber phase, matrix phase, and interface region.

[0035] Furthermore, such as Figure 1 As shown, S102 fuses all the extracted features to obtain fused features. The fusion adopts weighted fusion, and the weight of the weighted fusion is determined based on the statistical differences in the local sub-regions.

[0036] The effect of the above technical solution is that by combining color features, gradient features, and texture features, a basic feature description for each effective pixel is formed. Considering that the dominant features of different local regions are not consistent, this embodiment determines the fusion weight based on the degree of difference of various features within the local sub-region. Features with significant differences in the current local sub-region and which contribute significantly to region differentiation are assigned higher weights; features with weak differences or that are easily affected by noise are given lower contributions. The resulting fused feature field can simultaneously express grayscale, boundary, and texture information, and avoids the insufficient adaptability caused by using globally fixed weights. Determining the fusion weight based on local statistical differences ensures that features with high discriminative power in the current local sub-region receive higher weights, while reducing the influence of features that contribute weakly to region attribution.

[0037] Furthermore, the weights for the weighted fusion are determined based on statistical differences within local sub-regions, specifically including: Let the first Each local subregion contains 1 effective pixel, pixel The basic features consist of luminance, two chromaticity components, gradient, and texture features; for the third 3D feature, calculate the feature in the 6th dimension. Cumulative standardized differences within each local subregion :

[0038] in, Indicates the first Within the local sub-region, the first The standardized cumulative difference of a dimensional feature; Indicates the first The first effective pixel 3D eigenvalues; The first effective pixel at or near the geometric center of a local sub-region. 3D eigenvalues; Indicates the first The variance of the dimensional feature within this local subregion; To prevent positive numbers with a denominator of zero.

[0039] Adaptive weights are determined based on the standardized cumulative differences of each feature dimension. :

[0040] in, Indicates the first Within the local sub-region, the first Weights of dimensional features; A set of indices representing all feature dimensions; This represents the summation index. The sum of the feature weights for each dimension is 1.

[0041] The greater the standardized difference of a certain feature dimension, the more significant its distinguishing effect on the current local region, and therefore the higher its weight. Finally, each feature dimension is multiplied by its corresponding weight to obtain the fused feature of the pixel.

[0042] Further, S103: dividing the fused features into several local sub-regions, specifically including: Based on the total number of effective pixels and the number of predefined region primitives Determine the scale of the reference area. : , Starting from the top left corner of the bounding rectangle of the effective material region, along the row and column directions of the image according to the reference region scale. Establish a non-overlapping regular mesh to divide the effective material region into... A local sub-region of equal area.

[0043] Pixels within the hole mask are not included in the candidate pixel set; portions of the image edge that are not a complete grid are incorporated into the nearest neighboring local sub-region. Each valid pixel belongs to only one local sub-region, and the average coordinates of the valid pixels within each local sub-region are used as the geometric center of that local sub-region.

[0044] Further, S103: Determining the initial seed reference point in each local sub-region is to determine the initial seed reference point in each local sub-region based on local stability constraints and spatial compactness constraints.

[0045] The local stability constraint is used to evaluate the degree of change in fusion features between candidate pixels and their neighboring pixels, and the spatial compactness constraint is used to limit the degree to which the initial seed reference point deviates from the center of the local sub-region.

[0046] For the Candidate pixels within a local sub-region Local stability constraints Represented as:

[0047] in, This represents the local stability constraint of the candidate pixel; This represents the 3×3 effective neighborhood of a candidate pixel; and The fusion features of neighboring pixels and candidate pixels respectively represent the 1st, 2nd, and 3rd digits of the fusion features. One component; This represents the set of indices for all fused feature dimensions. The smaller the value, the more stable the fusion features within the neighborhood of the candidate pixel.

[0048] Space compactness constraints Represented as:

[0049] in, Indicates spatial compactness constraints; , () represents the row and column coordinates of the candidate pixel; , () represents the row and column coordinates of the geometric center of a local subregion; Indicates the scale of the reference area. The smaller the value, the closer the candidate pixel is to the center of the local sub-region.

[0050] Within each local sub-region, the local stability and spatial compactness of candidate pixels are considered comprehensively. Local stability measures the degree of change between candidate pixels and neighboring pixels in the fused feature space, while spatial compactness ensures that the seed reference point is located as close as possible within the local sub-region.

[0051] Candidate pixels whose local feature changes are less than a set threshold and whose spatial positions meet the compactness requirements are selected as initial seed reference points.

[0052] Furthermore, the determination of the initial seed reference point within each local sub-region based on local stability constraints and spatial compactness constraints specifically includes: Traverse all candidate pixels that do not belong to the hole mask within each local sub-region, calculate local stability constraints and spatial compactness constraints respectively, and determine the initial seed reference point. :

[0053] in, Indicates the first Initial seed reference point for each local sub-region; Indicates the first A local sub-region; Indicates the effective material area; This represents the regularization factor used to balance local stability constraints and spatial compactness constraints.

[0054] The candidate pixel with the smallest objective function value is selected as the initial seed reference point; when multiple candidate pixels have the same objective function value, the candidate pixel closest to the center of the local sub-region is selected first.

[0055] Further, S104: performing nearest neighbor search centered on the initial seed reference point, specifically includes: By appending the fused features of each pixel with row and column coordinates normalized to the scale of the reference region, an enhanced feature that simultaneously contains fused features and spatial location is obtained.

[0056] For each local sub-region, a KD-Tree index is built using the enhanced features of all valid pixels within it; using the enhanced features of the initial seed reference point as the query point, a radius nearest neighbor query is performed to obtain candidate extended pixels located within the preset search radius.

[0057] The search scope is limited to the local sub-region to which the initial seed reference point belongs and the effective material region, in order to avoid expanding across local sub-regions or falling into the hole region.

[0058] Nearest neighbor retrieval is achieved using KD-Tree, approximate nearest neighbor index, or local window retrieval, and candidate extended pixels are obtained within a preset search radius.

[0059] Further, S104: Filtering seed expansion nodes involves filtering seed expansion points based on fusion feature similarity, spatial proximity, and local attribution constraints, specifically including: For a retrieved candidate expanded pixel, if the difference in fused features between it and the initial seed reference point is less than a set threshold, and its spatial distance is within a preset search radius, then the candidate expanded pixel is determined as a seed expanded node. The seed expanded node and the initial seed reference point together serve as known constraint nodes for subsequent region propagation.

[0060] When the same candidate extended pixel simultaneously meets the extension conditions of multiple initial seed reference points, it is preferentially assigned to the initial seed reference point with the smallest difference in fused features; if the difference is less than the preset tolerance, the candidate extended pixel is retained as an unassigned pixel node, and its label is determined by the subsequent region membership propagation stage.

[0061] Furthermore, the step of selecting seed expansion points based on fusion feature similarity, spatial proximity, and local affiliation constraints specifically includes: For candidate pixels obtained by KD-Tree retrieval, calculate the comprehensive nearest neighbor distance between the candidate pixel and the initial seed reference point. :

[0062] in, Indicates candidate pixels Compared with the initial seed reference point The combined proximity distance between them; and The first and second parts of the fused features of the candidate pixels and the initial seed reference point are respectively represented by the first and second parts of the fused features. One component; and These represent the spatial coordinates of the two respectively; Indicates the spatial constraint factor; This indicates the baseline region scale. The combined proximity distance reflects both differences in fusion characteristics and spatial distance.

[0063] If a candidate pixel is located within the local sub-region and effective material region to which the initial seed reference point belongs, has not yet been assigned to another seed, and its comprehensive proximity distance is not greater than the preset search radius, then it is determined as a seed expansion node and assigned the same region label as the initial seed reference point.

[0064] When the same candidate pixel simultaneously meets the expansion conditions of multiple initial seed reference points, it is assigned to the initial seed reference point with the smallest comprehensive neighbor distance; if the difference between the minimum and the second smallest value is not greater than the set tolerance, the pixel is retained as an unassigned pixel node.

[0065] Further, S105: constructing a sparse region propagation map from unassigned pixel nodes, initial seed reference points, and seed expansion nodes, specifically includes: The pixel nodes within the effective material region are divided into an initial seed reference point set, a seed expansion node set, and an unassigned pixel node set, with the three sets not overlapping.

[0066] Retrieve valid pixels in the four or eight neighborhoods of an unassigned pixel node as the center: when adjacent nodes are both unassigned pixel nodes, establish local adjacency edges between unassigned nodes; when adjacent nodes are initial seed reference points or seed expansion nodes, establish constraint connection edges between unassigned nodes and known constraint nodes.

[0067] No connections are established between initial seed reference points, between seed expansion nodes, between initial seed reference points and seed expansion nodes, or between long-distance connections beyond the local neighborhood. At the same time, nodes inside the hole mask and their associated edges are deleted, thus obtaining a sparse region propagation graph that retains only the connection relationships required for region affiliation propagation.

[0068] The pixel nodes are divided into an initial seed reference point set, a seed expansion node set, and an unassigned pixel node set. A sparse region propagation graph is constructed based on the local adjacency relationships between unassigned pixel nodes and the connection relationships between unassigned pixel nodes and their neighboring seed constraint nodes. The sparse region propagation graph does not calculate redundant connections between distant pixels, between different seed expansion nodes, or that are irrelevant to region assignment, thereby reducing the graph construction size.

[0069] Furthermore, S105: Determining the edge connection relationships in the propagation graph is based on the differences in fusion features and spatial constraints.

[0070] The edge connections of the propagation graph include: local adjacency edges between unassigned pixel nodes, and constraint connection edges between unassigned pixel nodes and the initial seed reference point or seed expansion node.

[0071] For edge connections in a graph, edge weights can be determined based on the differences in fusion features between nodes, spatial distance, and local adjacency. Nodes with smaller feature differences and closer spatial distances have higher connection weights; nodes with larger feature differences or greater spatial distances have lower connection weights or are not connected.

[0072] Furthermore, the edge connectivity relationships in the propagation graph are determined based on the differences in fusion features and spatial constraints, specifically including: For any two nodes that satisfy the local adjacency condition and Calculate the edge weights between the two:

[0073] in, Represents a node and nodes Edge weights between them; and This represents the fused feature vector of the two nodes; and Represents the spatial coordinates of the two nodes; and These represent the scale parameters for feature differences and spatial differences, respectively. Indicates the scale of the reference area; This represents a positive constant to prevent edge weights from being zero. This indicates a space constraint switch.

[0074] For two adjacent unassigned pixel nodes, let For the constrained connection edges between unassigned pixel nodes and the initial seed reference point or seed expansion node, let ,in This represents the spatial constraint factor.

[0075] The smaller the difference in fusion characteristics between nodes and the closer the spatial distance, the greater the edge weight; nodes that do not meet the local adjacency condition are not connected.

[0076] Further, S106: using the initial seed reference point and seed expansion node as known constraint nodes, the region membership degree of the unassigned pixel nodes is solved by propagation, specifically including: The initial seed reference point and its seed extension nodes are combined into a set of known constraint nodes, and each sub-extension node inherits the region label of the corresponding initial seed reference point.

[0077] The graph Laplacian matrix is ​​calculated based on the edge weight matrix of the propagation graph, and then the graph Laplacian matrix is ​​divided into blocks according to the order of known constrained nodes and unassigned pixel nodes.

[0078] The expression for solving the region membership degree of unassigned pixel nodes is:

[0079] in, This represents the Laplacian submatrix corresponding to the unassigned pixel node; This represents the coupling submatrix between unassigned pixel nodes and known constrained nodes; A label indicator matrix representing known constraint nodes; This represents the membership matrix of the unassigned pixel node region to be solved.

[0080] After obtaining the region membership, the membership of each unassigned pixel node with respect to all region labels is normalized so that the sum of its membership is 1. Seed expansion nodes participate in propagation as known boundary conditions and are not considered as unknown variables in the solution.

[0081] The region membership degree propagation solution is implemented using random walk, graph Laplacian regularization, energy minimization, or equivalent graph propagation methods. The initial seed reference point and seed expansion nodes are treated as known constraint nodes, and the region membership degree is propagated to solve for unassigned pixel nodes. This solution can be implemented using random walk, graph Laplacian regularization, or equivalent energy minimization methods. Since the seed expansion nodes are already included in the propagation as known constraint nodes and are not considered unknown variables in the solution, the size of unknown nodes can be reduced and the solution efficiency improved.

[0082] Further, S106: Based on the region membership degree, determine the region label of the unassigned pixel node, and output the region primitive generation result of the composite material hole processing exit region image, specifically including: For each unassigned pixel node, its region label is determined according to the principle of maximum membership. :

[0083] in, Indicates unclaimed pixel nodes The area label, Represents a node For the first Normalized membership degree of each region label This indicates the total number of region labels.

[0084] All node labels are remapped back to the original image coordinates, and a connectivity check is performed on each label region.

[0085] If there is an isolated connected component for the same label that does not contain the corresponding initial seed reference point, then the sum of the edge weights between the component and each of its neighboring regions is calculated and merged into the neighboring region with the largest sum of edge weights.

[0086] Finally, a set of pixels with the same label and spatial connectivity is defined as a region primitive, and the region primitive label map, the set of region member pixels, the region boundary, the region center, the region adjacency relationship and the pixel membership matrix are output.

[0087] For each unassigned pixel node, its region label is determined according to the maximum membership principle, and the final region primitive is obtained by combining spatial connectivity constraints. The output results include a region primitive label map, region boundary, region center, set of region member pixels, region adjacency relationship, and pixel membership prior. The above results can be used for subsequent damage classification and identification, damage boundary extraction, damage area calculation, and hole processing quality evaluation.

[0088] The following specific embodiments illustrate and verify the solution of this application.

[0089] This embodiment uses the image of the hole exit area of ​​carbon fiber reinforced resin matrix composite (CFRP) as an example to illustrate the seed-expanded region primitive generation method described in this invention. After drilling, the hole exit area of ​​the CFRP sample is prone to delamination damage. Due to the obvious fiber texture and anisotropic grayscale variation on the CFRP surface, the damaged area at the hole edge and the normal material area often exhibit weak boundaries, texture interlacing, and unstable local contrast. Relying solely on grayscale thresholding or edge detection is insufficient to obtain stable region segmentation results.

[0090] First, an image of the CFRP hole processing exit area is acquired, and the image undergoes brightness normalization, noise suppression, and hole region removal processing to obtain an effective material region image to be processed. Subsequently, color channel responses, gradient responses, and texture responses are extracted from the effective material region image, and the contribution weights of different features in the fusion process are determined based on local statistical differences, forming an adaptive fusion feature field. For example... Figure 1 As shown, the color channel, gradient response, and texture response reflect the brightness differences, hole edge structure changes, and fiber texture differences in the CFRP image, respectively. After being constrained by local statistical differences, they together constitute a fused feature field. This fused feature field can enhance the distinguishability between the hole edge damage area and the surrounding material area, and reduce the instability caused by changes in illumination and fiber texture interference to the single grayscale feature.

[0091] After obtaining the fused feature field, candidate seed positions are determined within the effective material region according to a preset region scale. A hole mask is then used to exclude dark background regions within the holes, ensuring that subsequent seed reference point determination and region propagation processes are concentrated within the hole edge damage and its adjacent material region. For example... Figure 2As shown, the initial candidate seed points are distributed in the effective material region around the hole. The hole region does not participate in the subsequent graph node construction, thereby avoiding interference from the low-information region inside the hole to the region primitive generation result.

[0092] Then, an initial seed reference point is determined in the fused feature field, and seed expansion is performed centered on the initial seed reference point. For example... Figure 3 As shown, the system selects pixels within each local region that are similar in features to the initial seed reference point, spatially adjacent, and have high confidence levels, and uses these pixels as seed expansion nodes. This process allows some locally consistent regions to form a stable set of known nodes in advance, avoiding the influence of noise or weak boundaries on region propagation due to insufficient seed points.

[0093] Furthermore, the initial seed reference point, seed expansion nodes, and unassigned pixel nodes are collectively constructed into a sparse region propagation map. For example... Figure 4 As shown, the initial seed reference point and seed expansion nodes are considered known nodes, while unassigned pixels are considered unknown nodes whose affiliation needs to be determined. Only local adjacency relationships and candidate connectivity relationships relevant to region affiliation are retained in the figure, thus reducing redundant connections between distant, unrelated pixels. The region affiliation result for each pixel can be obtained by solving the membership degree of unassigned pixel nodes to different region labels.

[0094] Finally, the region primitive generation results of the CFRP hole machining exit damage image are output according to the maximum membership principle. For example... Figure 5 As shown, the generated region primitives can achieve good adhesion along hole edge damage, fiber texture abrupt change areas, and local damage boundaries, while maintaining consistency within the region. The region primitive results can be further used for boundary extraction of CFRP hole machining damage, damage area calculation, damage category identification, and hole machining quality evaluation.

[0095] This embodiment demonstrates that, in response to the problems of obvious fiber texture, discontinuous damage boundaries, and complex gray-scale changes at the hole edges in CFRP hole processing exit images, this invention can generate stable regional primitive results by combining multi-source feature fusion, seed point distribution within the effective region, seed expansion, and sparse region propagation, providing reliable regional-level input for subsequent damage identification and quantitative analysis.

[0096] This invention acquires images of the outlet region of composite material hole processing and filters out the cavity inside the hole, dark background, and low-information areas to obtain an effective material region image. It extracts the color, gradient, and texture features of the effective pixels and generates a fused feature field based on local statistical differences. Within each local sub-region, initial seed reference points are determined based on local stability and spatial compactness. Nearest neighbor retrieval is performed using the initial seed reference points as the center, and pixels that satisfy the consistency conditions with the seed reference points in both features and space are determined as seed expansion nodes. A sparse region propagation map is constructed based on unassigned pixels, the initial seed reference points, and the seed expansion nodes, and the region membership degree is solved. The region primitive label map, region boundaries, and region adjacency relationships are output according to the maximum membership degree and connectivity constraints. This method can improve the fit and generation stability of region primitive boundaries and reduce the computational cost of full-image propagation in composite material hole processing outlet damage images with complex textures, weak boundaries, and gray-level overlap. It can serve as a front-end processing method for damage classification and recognition, damage boundary extraction, and damage quantification evaluation.

[0097] Example 2 This embodiment provides an image primitive generation system for detecting hole processing damage in composite materials; An image primitive generation system for detecting hole processing damage in composite materials includes: The image acquisition and preprocessing module is configured to: acquire an image of the hole processing exit area of ​​the composite material, and preprocess the image of the hole processing exit area to obtain an effective material area image; The multi-source feature fusion module is configured to: extract several features from the effective material region image, and fuse all the extracted features to obtain fused features; The seed reference point determination module is configured to divide the fused features into several local sub-regions and determine an initial seed reference point in each local sub-region. The seed expansion module is configured to perform nearest neighbor search centered on the initial seed reference point and filter seed expansion nodes; The sparse region propagation module is configured to: construct a sparse region propagation graph from unclaimed pixel nodes, initial seed reference points, and seed expansion nodes, and determine the edge connectivity relationships in the propagation graph; The region primitive output module is configured to: use the initial seed reference point and seed expansion node as known constraint nodes, solve the region membership degree propagation problem for unassigned pixel nodes; determine the region label of the unassigned pixel node based on the region membership degree, and output the region primitive generation result of the composite material hole processing exit region image.

[0098] It should be noted that the image acquisition and preprocessing module, multi-source feature fusion module, seed reference point determination module, seed expansion module, sparse region propagation module, and region primitive output module described above correspond to steps S101 to S106 in Embodiment 1. The examples and application scenarios implemented by these modules and their corresponding steps are the same, but they are not limited to the content disclosed in Embodiment 1. It should also be noted that these modules, as part of the system, can be executed in a computer system, such as a set of computer-executable instructions.

[0099] The system can be deployed in industrial vision inspection equipment, edge computing devices, or general-purpose computer workstations. The system hardware may include an image acquisition unit, a computing and processing unit, a storage unit, a display and interaction unit, and a communication interface. The image acquisition unit may consist of an industrial camera, a lens, and a light source, used to acquire images of the outlet area of ​​the composite material hole processing. The computing and processing unit may employ a CPU, GPU, or edge computing device to perform image preprocessing, multi-source feature fusion, seed reference point determination, seed expansion, and sparse region propagation calculations. The storage unit stores the original image, preprocessed image, fused feature data, seed node data, and region primitive generation results. The display and interaction unit displays the region primitive generation results and allows setting the region scale, seed expansion range, and output parameters. The communication interface is used for data interaction with a host computer, industrial camera, or subsequent damage identification system.

[0100] When the system is working, it first completes parameter settings and module initialization, and then the image acquisition unit acquires the image of the CFRP hole processing exit area; subsequently, the image acquisition and preprocessing module outputs the effective material area image, and the multi-source feature fusion module generates a fused feature field; based on this, the seed reference point determination module generates an initial set of seed reference points, and the seed expansion module further generates a set of seed expansion nodes; then, the sparse region propagation module constructs a sparse region propagation map based on the initial seed reference points, seed expansion nodes, and unassigned pixel nodes, and obtains the region membership degree of each unassigned pixel node; finally, the region primitive output module generates the region primitive result based on the membership degree, and outputs the region boundary, region center, and region adjacency relationship, while the data management and human-computer interaction module completes the result display and data saving.

[0101] like Figure 6 As shown, the system carries various software functional modules through a hardware platform, and completes the generation of regional primitives for hole processing damage images in the order of image acquisition, image preprocessing, multi-source feature fusion, seed reference point determination, seed expansion, sparse region propagation, and regional primitive output.

[0102] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0103] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0104] Example 3 This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein, the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory to cause the electronic device to perform the method described in Embodiment 1.

[0105] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0106] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0107] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.

[0108] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0109] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0110] Example 4 This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the method described in Embodiment 1.

[0111] Example 5 This embodiment also provides a computer program product, including a computer program that, when executed by a processor, implements the method in Embodiment 1.

[0112] This application also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0113] The computer program code used to implement the methods of this application may be written in one or more programming languages. This computer program code may be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0114] In the context of this application, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0115] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0116] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for generating image primitives for detecting hole processing damage in composite materials, characterized in that, include: The image of the hole processing exit area of ​​the composite material is obtained, and the image of the hole processing exit area is preprocessed to obtain the effective material area image; Several features are extracted from the image of the effective material region, and all extracted features are fused to obtain fused features; The fusion features are divided into several local sub-regions, and an initial seed reference point is determined in each local sub-region. Perform nearest neighbor search centered on the initial seed reference point, and filter the seed expansion nodes; The unassigned pixel nodes, the initial seed reference point, and the seed expansion nodes are constructed into a sparse region propagation graph, and the edge connection relationships in the propagation graph are determined. Using the initial seed reference point and seed expansion node as known constraint nodes, the region membership degree of unassigned pixel nodes is solved by propagation. Based on the region membership degree, determine the region label of the unassigned pixel node, and output the region primitive generation result of the composite material hole processing exit region image.

2. The image primitive generation method for detecting hole processing damage in composite materials as described in claim 1, characterized in that, All extracted features are fused to obtain fused features. The fusion adopts weighted fusion, and the weights of the weighted fusion are determined based on the statistical differences in local sub-regions. The weights for the weighted fusion are determined based on statistical differences within local sub-regions, specifically including: Let the first Each local subregion contains 1 effective pixel, pixel The basic features consist of luminance, two chromaticity components, gradient, and texture features; for the third 3D feature, calculate the feature in the 6th dimension. Cumulative standardized differences within each local subregion : in, Indicates the first Within the local sub-region, the first The standardized cumulative difference of a dimensional feature; Indicates the first The first effective pixel 3D eigenvalues; The first effective pixel at or near the geometric center of a local sub-region. 3D eigenvalues; Indicates the first The variance of the dimensional feature within this local subregion; To prevent positive numbers with a denominator of zero; Adaptive weights are determined based on the standardized cumulative differences of each feature dimension. : in, Indicates the first Within the local sub-region, the first Weights of dimensional features; A set of indices representing all feature dimensions; This indicates the summation index; the sum of the feature weights of each dimension is 1.

3. The image primitive generation method for detecting hole processing damage in composite materials as described in claim 1, characterized in that, in Determining the initial seed reference point within each local sub-region involves determining the initial seed reference point within each local sub-region based on local stability constraints and spatial compactness constraints. For the Candidate pixels within a local sub-region Local stability constraints Represented as: in, This represents the local stability constraint of the candidate pixel; This represents the 3×3 effective neighborhood of a candidate pixel; and The fusion features of neighboring pixels and candidate pixels respectively represent the 1st, 2nd, and 3rd digits of the fusion features. One component; Represents the set of indices for all fused feature dimensions; The smaller the value, the more stable the fusion features within the neighborhood of the candidate pixel; Space compactness constraints Represented as: in, Indicates spatial compactness constraints; , () represents the row and column coordinates of the candidate pixel; , () represents the row and column coordinates of the geometric center of a local subregion; Indicates the scale of the reference area; The smaller the value, the closer the candidate pixel is to the center of the local sub-region; Traverse all candidate pixels that do not belong to the hole mask within each local sub-region, calculate local stability constraints and spatial compactness constraints respectively, and determine the initial seed reference point. : in, Indicates the first Initial seed reference point for each local sub-region; Indicates the first A local sub-region; Indicates the effective material area; This represents the regularization factor used to balance local stability constraints and spatial compactness constraints.

4. The image primitive generation method for detecting hole processing damage in composite materials as described in claim 1, characterized in that, screening... Seed expansion nodes are selected based on fusion feature similarity, spatial proximity, and local attribution constraints. Specifically, the selection of seed expansion nodes based on fusion feature similarity, spatial proximity, and local attribution constraints includes: For candidate pixels obtained by KD-Tree retrieval, calculate the comprehensive nearest neighbor distance between the candidate pixel and the initial seed reference point. : in, Indicates candidate pixels Compared with the initial seed reference point The combined proximity distance between them; and The first and second parts of the fused features of the candidate pixels and the initial seed reference point are respectively represented by the first and second parts of the fused features. One component; and These represent the spatial coordinates of the two respectively; Indicates the spatial constraint factor; Indicates the scale of the reference area; the combined proximity distance reflects both the differences in fusion characteristics and spatial distance; If a candidate pixel is located within the local sub-region and effective material region to which the initial seed reference point belongs, has not yet been assigned to another seed, and its comprehensive proximity distance is not greater than the preset search radius, then it is determined as a seed expansion node and assigned the same region label as the initial seed reference point.

5. The image primitive generation method for detecting hole processing damage in composite materials as described in claim 1, characterized in that, The edge connectivity relationships in the propagation graph are determined based on the differences in fusion features and spatial constraints. The edge connectivity relationships in the propagation graph are determined based on the differences in fusion features and spatial constraints, specifically including: For any two nodes that satisfy the local adjacency condition and Calculate the edge weights between the two: in, Represents a node and nodes Edge weights between them; and This represents the fused feature vector of the two nodes; and Represents the spatial coordinates of the two nodes; and These represent the scale parameters for feature differences and spatial differences, respectively. Indicates the scale of the reference area; This represents a positive constant to prevent edge weights from being zero. Indicates a spatial constraint switch; for two adjacent unassigned pixel nodes, let For the constrained connection edges between unassigned pixel nodes and the initial seed reference point or seed expansion node, let ,in This represents the spatial constraint factor.

6. The image primitive generation method for detecting hole processing damage in composite materials as described in claim 1, characterized in that, Using the initial seed reference point and seed expansion nodes as known constraint nodes, the region membership degree of unassigned pixel nodes is solved through propagation. Specifically, this includes: The initial seed reference point and its seed extension nodes are combined into a set of known constraint nodes, and each sub-extension node inherits the region label of the corresponding initial seed reference point. The graph Laplacian matrix is ​​calculated based on the edge weight matrix of the propagation graph, and the graph Laplacian matrix is ​​divided into blocks according to the order of known constrained nodes and unowned pixel nodes. The expression for solving the region membership degree of unassigned pixel nodes is: in, This represents the Laplacian submatrix corresponding to the unassigned pixel node; This represents the coupling submatrix between unassigned pixel nodes and known constrained nodes; A label indicator matrix representing known constraint nodes; This represents the membership matrix of the unassigned pixel node region to be solved.

7. The image primitive generation method for detecting hole processing damage in composite materials as described in claim 1, characterized in that, Based on the region membership degree, determine the region labels of unassigned pixel nodes, and output the region primitive generation results of the composite material hole processing exit region image, specifically including: For each unassigned pixel node, its region label is determined according to the principle of maximum membership. : in, Indicates unclaimed pixel nodes The area label, Represents a node For the Normalized membership degree of each region label This indicates the total number of region labels.

8. An image primitive generation system for detecting hole processing damage in composite materials, characterized in that, include: The image acquisition and preprocessing module is configured to: acquire an image of the hole processing exit area of ​​the composite material, and preprocess the image of the hole processing exit area to obtain an effective material area image; The multi-source feature fusion module is configured to: extract several features from the effective material region image, and fuse all the extracted features to obtain fused features; The seed reference point determination module is configured to divide the fused features into several local sub-regions and determine an initial seed reference point in each local sub-region. The seed expansion module is configured to perform nearest neighbor search centered on the initial seed reference point and filter seed expansion nodes; The sparse region propagation module is configured to: construct a sparse region propagation graph from unclaimed pixel nodes, initial seed reference points, and seed expansion nodes, and determine the edge connectivity relationships in the propagation graph; The region primitive output module is configured to: use the initial seed reference point and the seed expansion node as known constraint nodes, and solve the region membership degree propagation for unassigned pixel nodes. Based on the region membership degree, determine the region label of the unassigned pixel node, and output the region primitive generation result of the composite material hole processing exit region image.

9. An electronic device, characterized in that it comprises: Memory is used to store computer-readable instructions in a non-transitory manner. as well as Processor, for executing the computer-readable instructions, When the computer-readable instructions are executed by the processor, they perform the method described in any one of claims 1-6.

10. A storage medium, characterized in that, Non-transitory storage of computer-readable instructions, wherein when the non-transitory computer-readable instructions are executed by a computer, the method of any one of claims 1-6 is performed.

Images