Aircraft skin damage identification method, device and equipment, storage medium and product

Abstract

The embodiment of the invention provides an aircraft skin damage identification method and device, equipment, a storage medium and a product. The method comprises the following steps: acquiring an initial point cloud of an aircraft skin surface through structured light equipment, and performing data preprocessing on the initial point cloud to obtain a preliminary correction point cloud; calculating reflection intensity distribution of each area on the surface of the aircraft skin based on the preliminary correction point cloud, and determining a compensated point cloud; extracting geometric features from the compensated point cloud, and performing denoising processing on the compensated point cloud to obtain a denoised point cloud; constructing a three-dimensional grid model representing a three-dimensional structure of the aircraft skin surface based on the de-noised point cloud, and determining a damage candidate area of the aircraft skin surface; obtaining a point cloud subset associated with the damage candidate area and carrying out damage type classification on the point cloud subset to obtain a classification result of damage types; according to a classification result, generating a damage labeling graph; and based on the damage marking graph, calibrating the damage size parameter to generate a final damage report.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of aerospace, and particularly relates to a method, device, equipment, storage medium and product for identifying damage of an aircraft skin based on structured light. BACKGROUND

[0002] The aircraft skin is a core structure on the outside of an aircraft. Micro-damage such as surface cracks and dents on the aircraft skin may expand due to airflow and pressure during flight, causing structural failure and directly affecting flight safety. The detection efficiency determines the aircraft maintenance cycle, and inefficient detection increases the time of parking and the operating cost. Therefore, high-precision and high-efficiency detection is a core requirement of aviation maintenance.

[0003] Currently, there are two main detection technologies, namely, traditional manual detection and structured light detection technology. The traditional manual detection relies on an operator to check point by point with a gauge, which is time-consuming and affected by the experience and fatigue state of the operator, and the consistency and accuracy of detection are difficult to guarantee. The structured light detection technology is a non-contact optical technology that acquires three-dimensional point cloud data of the skin surface by projecting a specific light pattern and capturing deformation. Although this technology solves the problem of low efficiency of manual detection, it has technical problems of being easily disturbed by the reflection of the skin and inefficient processing of distorted point clouds. On the one hand, the skin is usually made of high-reflective aluminum alloy, and the reflection of environmental light or the structured light itself will mix into the collected signal, causing the point cloud coordinates of the structured light detection to deviate from the true position and failing to reflect the actual shape of the skin. The point cloud is the basis for damage identification, and distortion directly destroys the accuracy. On the other hand, for distorted point clouds, the existing technology lacks effective processing methods, neither removes invalid data points caused by reflection and equipment errors, nor relies on manual assistance for damage classification and size quantification, and cannot achieve micron-level high-precision measurement. The above two problems make it difficult for the existing skin damage detection technology to balance “anti-reflection interference” and “high-precision automatic identification”, and cannot meet the actual needs of aviation maintenance for safety and efficiency. Therefore, there is an urgent need for a technical solution that can systematically solve the above problems. SUMMARY

[0004] The present application provides a method, device, equipment, storage medium and product for identifying damage of an aircraft skin, which can improve the accuracy of identifying damage of an aircraft skin.

[0005] In a first aspect, the present application provides a method for identifying damage of an aircraft skin, which comprises the following steps: An initial point cloud of the surface of the aircraft skin is acquired by a structured light device, and data preprocessing is performed on the initial point cloud to obtain a preliminary corrected point cloud; Based on the preliminary corrected point cloud, the reflection intensity distribution of each region of the surface of the aircraft skin is calculated, and a compensated point cloud is determined based on the calculated reflection intensity and a preset reflection intensity threshold value; extract geometric features from the compensated point cloud, and perform denoising processing on the compensated point cloud based on the geometric features by using a spatial domain filtering method to obtain a denoised point cloud; construct a three-dimensional mesh model representing a three-dimensional structure of the aircraft skin surface based on the denoised point cloud, and determine a damage candidate area of the aircraft skin surface from the three-dimensional mesh model by using a point cloud local geometric feature analysis method; obtain a point cloud subset associated with the damage candidate area, and classify the point cloud subset into a damage type by using a deep learning model to obtain a classification result of the damage type; According to the classification result, the boundary information of the damage candidate area is fused to generate a damage annotation map; and Based on the damage annotation map, an error compensation model constructed by using an optimization algorithm based on parameter fitting is used to calibrate the damage size parameter to generate a final damage report.

[0006] In a second aspect, the present application provides an aircraft skin damage identification device, the device comprising: A point cloud acquisition and preprocessing module is configured to acquire an initial point cloud of an aircraft skin surface by using a structured light device, and to perform data preprocessing on the initial point cloud to obtain a preliminary corrected point cloud. A reflection intensity analysis and compensation module is configured to calculate the reflection intensity distribution of each region of the aircraft skin surface based on the preliminary corrected point cloud, and to determine a compensated point cloud based on the calculated reflection intensity and a preset reflection intensity threshold. A point cloud denoising module is configured to extract geometric features from the compensated point cloud, and to perform denoising processing on the compensated point cloud based on the geometric features by using a spatial domain filtering method to obtain a denoised point cloud. A damage area positioning module is configured to construct a three-dimensional mesh model representing a three-dimensional structure of the aircraft skin surface based on the denoised point cloud, and to determine a damage candidate area of the aircraft skin surface from the three-dimensional mesh model by using a point cloud local geometric feature analysis method. A damage classification module is configured to obtain a point cloud subset associated with the damage candidate area, and to classify the point cloud subset into a damage type by using a deep learning model to obtain a classification result of the damage type. A damage annotation generation module is configured to fuse the boundary information of the damage candidate area according to the classification result to generate a damage annotation map. A damage report generation module is configured to calibrate the damage size parameter by using an error compensation model constructed by using an optimization algorithm based on parameter fitting based on the damage annotation map to generate a final damage report.

[0007] In a third aspect, the present application provides an aircraft skin damage identification device, the device comprising: a processor, and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the aircraft skin damage identification method as described above.

[0008] In a fourth aspect, the present application provides a computer storage medium, the computer storage medium storing computer program instructions, the computer program instructions being executed by a processor to implement the aircraft skin damage identification method as described above.

[0009] In a fifth aspect, the present application provides a computer program product comprising a computer program, the computer program being executed by a processor to implement the aircraft skin damage identification method as described above.

[0010] The aircraft skin damage identification method, device, equipment, storage medium and product implemented by the present application can solve the data quality problem by data preprocessing (for example, polarization filtering) combined with dynamic acquisition parameter adjustment strategy (for example, adaptive light compensation) in view of the high reflectivity and high precision requirements of the aircraft skin, improve the damage identification precision by three-dimensional geometric feature extraction combined with deep learning, and realize size quantization calibration by an error compensation model (for example, least square method iterative optimization) constructed by an optimization algorithm based on parameter fitting, forming a complete damage identification scheme suitable for the aviation scene, effectively making up for the shortcomings of the prior art in the field of aircraft skin damage detection, and further improving the aircraft skin damage identification precision. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the embodiments of the present application will be briefly introduced below. Those skilled in the art can obtain other drawings according to these drawings without creating any creative labor.

[0012] Figure 1 A flowchart of the aircraft skin damage identification method according to the embodiments of the present application is shown; Figure 2 A flowchart of the method for obtaining the compensated point cloud according to the embodiments of the present application is shown; Figure 3 A structural diagram of the aircraft skin damage identification device according to the embodiments of the present application is shown; Figure 4 A hardware structural diagram of the aircraft skin damage identification device according to the embodiments of the present application is shown. DETAILED DESCRIPTION

[0013] The features and exemplary embodiments of the various aspects of the present application will be described in detail below with reference to the drawings. For the purpose of clarity, the description is divided into the following sections: technical scheme, technical effects, and specific embodiments. The technical scheme section describes the technical solutions of the present application. The technical effects section describes the technical effects of the present application. The specific embodiments section describes the specific embodiments of the present application. The purpose of the above sections is to provide a better understanding of the present application. The specific embodiments described herein are merely intended to explain the present application, and are not intended to limit the present application. The present application can be implemented without some of the specific details described below. The following description of the embodiments is merely intended to provide a better understanding of the present application by showing examples of the present application.

[0014] It should be noted that, in this document, relational terms such as first and second, and the like, are used solely to distinguish one entity or action from another entity or action, without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, an element defined by the phrase "comprises a... " does not exclude the existence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0015] To solve the problems in the prior art, the embodiments of the present application provide an aircraft skin damage identification method, device, equipment, storage medium and product.

[0016] The aircraft skin damage identification method provided by the embodiments of the present application will be introduced first.

[0017] Figure 1 A flowchart of the aircraft skin damage identification method according to the embodiments of the present application is shown. As shown in Figure 1 The method can include the following steps: Step S102: Obtain the initial point cloud of the aircraft skin surface through the structured light device, and perform data preprocessing on the initial point cloud to obtain the preliminary corrected point cloud. In an embodiment, the initial point cloud data is obtained by scanning the skin surface through the structured light device, and the reflected light is filtered through polarization filtering to preliminarily reduce the interference of reflected light, thereby obtaining the preliminary corrected point cloud.

[0018] A digital micromirror array of a structured light device projects a coded stripe pattern onto the skin surface. A binocular camera simultaneously acquires images of the deformed stripes. The 3D coordinates of each pixel are calculated based on the mapping relationship between the stripe phase offset and depth value, yielding raw point cloud data including reflective noise. For this raw point cloud data, orthogonal polarizers are installed in front of the binocular camera lens to acquire the light intensity values ​​of the horizontal and vertical polarization components. The polarization degree value of each sampling point is calculated. When the polarization degree value exceeds a preset polarization threshold, the point is identified as a specular reflection point and marked, resulting in a polarization-marked point cloud. Based on the spatial distribution of non-reflective points in the polarization-marked point cloud, the average Euclidean distance between each point and its k nearest neighbors is calculated. When the average distance is greater than a preset outlier threshold, the outlier point is removed. For the void regions formed after removing reflective and outlier points, interpolation is performed using a weighted average of the depth values ​​of neighboring non-reflective points to fill the voids, thus determining the preliminary corrected point cloud.

[0019] Specifically, in one embodiment, the structured light device uses a digital micromirror array as the projection unit. This array contains 1024×768 micromirrors, each with independently controllable deflection angles. A coded fringe pattern combining Gray code and phase shift is generated by controlling the micromirror deflection state. The Gray code sequence determines the number of fringe periods, and the phase shift sequence achieves sub-pixel precision positioning. The binocular camera uses an industrial-grade CMOS sensor with a resolution of 2048×1536 pixels and a frame rate of 60fps. The baseline distance between the two cameras is set between 200mm and 300mm. The fringe phase shift is calculated using a four-step phase shift method. Four fringe images with a phase difference of π / 2 are projected, and the gray values ​​of the four images are used to determine the fringe pattern. , , , Calculate the package phase φ = arctan[( - ) / ( - The absolute phase value is obtained after the combination of the Gray code unwrapping, and the phase value is converted into the actual depth coordinate through the pre-calibrated phase-depth mapping table. The polar line constraint is used to further correct the depth value by the binocular vision system to eliminate the system error of monocular measurement. The polarizer is a linear polarizer, and the polarization directions are set as two orthogonal directions of horizontal and vertical. The specular reflection light on the skin surface has specific polarization characteristics, and the polarization degree of the diffuse reflection light is low. The polarization degree is calculated by measuring the light intensity of the two polarization directions. When the skin is made of aluminum alloy, the polarization degree of the specular reflection area is usually greater than 0.6, and the polarization degree of the normal diffuse reflection area is less than 0.3, so the reflection points can be effectively distinguished according to the polarization degree threshold. In the k-nearest neighbor algorithm, the k value is adaptively adjusted according to the point cloud density, and the number of neighborhood points is usually 8 to 12. The statistical method is used for outlier judgment, and the distance from the point to the neighborhood center of gravity is calculated. When the distance is more than 2.5 times the standard deviation of the average neighborhood distance, it is determined as an outlier. The radial basis function interpolation method is used for hole filling. The non-reflective points on the hole boundary are used as control points to construct a radial basis function network, and the interpolation coefficients are obtained by solving a linear equation set, so as to reconstruct the depth value of the hole area. The curvature constraint is introduced in the interpolation process to ensure that the filled surface is geometrically continuous with the original skin surface.

[0020] Through the above steps, the high light interference of the skin is eliminated. The polarization filtering + adaptive light compensation combined scheme is designed to distinguish the specular reflection points by using the orthogonal polarizer, to adjust the exposure time and light source power according to the reflection intensity, to control the point cloud distortion rate within the threshold, and to adapt to the complex light environment of the hangar.

[0021] Step S104: Based on the preliminary corrected point cloud, the reflection intensity distribution of each region of the aircraft skin surface is calculated, and the compensated point cloud is determined based on the calculated reflection intensity and the preset reflection intensity threshold. In one embodiment, the surface reflection intensity distribution is calculated according to the preliminary corrected point cloud. If the reflection intensity exceeds the preset reflection intensity threshold, the adaptive light compensation algorithm based on the reflection intensity distribution difference is used to adjust the exposure time and light source power to obtain the compensated point cloud. If the reflection intensity does not exceed the preset reflection intensity threshold, the preliminary corrected point cloud is directly used as the compensated point cloud.

[0022] Reference Figure 2FIG. 2 shows a flow diagram of a method for obtaining a compensated point cloud according to an embodiment of the present application, including: step S202: obtaining optical properties of sampling points. Specifically, obtaining the original gray value corresponding to each sampling point in the preliminary corrected point cloud on the camera imaging plane. Step S204: constructing a reflection intensity distribution map based on the optical properties of the sampling points. In one embodiment, based on a preset camera response function, the original gray value (for example, an eight-bit gray value) is mapped to a linear light intensity value; then, combined with the angle (for example, the cosine value of the angle) between the surface normal vector at each sampling point and the incident direction of the structured light, the surface reflection intensity coefficient is calculated; finally, based on the reflection intensity coefficients of all sampling points, the reflection intensity distribution map of the aircraft skin surface is constructed. Step S206: calculating the average reflection intensity of each local region of the aircraft skin surface based on the reflection intensity distribution map. Step S208: dividing the reflection intensity region based on the average reflection intensity. Comparing the average reflection intensity with a preset reflection intensity threshold, and dividing the aircraft skin surface into different reflection intensity regions according to the comparison result. In one embodiment, when the average reflection intensity exceeds the preset reflection intensity threshold, the aircraft skin surface is divided into, for example, a strong reflection region, a medium reflection region and a weak reflection region. Step S210: determining the optimal set of acquisition parameters for different reflection intensity regions. For different reflection intensity regions divided, a high dynamic range imaging strategy is adopted, and multiple sets of different exposure time and light source power combinations are set for each region; in one embodiment, the exposure time is distributed in logarithmic intervals from the preset shortest time to the longest time, and the light source power is linearly increased from the lowest power to the rated power, for each set of parameters, the corresponding image frame is collected and weighted fusion analysis is performed to determine the optimal acquisition parameters corresponding to each reflection intensity region. Step S212: using the determined optimal acquisition parameters to perform data reacquisition and data fusion on different reflection intensity regions. According to the optimal acquisition parameters determined for each region, the strong reflection region, the weak reflection region, etc. are re-scanned and collected; the data at the region boundary caused by parameter difference is processed by interpolation compensation to eliminate the discontinuity of the splicing boundary, and finally the compensated point cloud is obtained. In one embodiment, the strong reflection region adopts a short exposure low power parameter set, the weak reflection region adopts a long exposure high power parameter set, the transition zone at the region boundary is processed by bilinear interpolation to eliminate the gray level mutation of the splicing boundary, and the compensated point cloud is obtained.

[0023] Specifically, in one embodiment, the camera response function is obtained by pre-calibration, using a standard gray card to capture multiple sets of images under different lighting conditions, establishing a mapping relationship between gray value and actual light intensity. The function usually presents nonlinear characteristics, with a larger slope in the low gray interval and a tendency to saturation in the high gray interval. For the 0 to 255 gray range of eight-bit images, the response function maps it to the normalized 0 to 1 light intensity value interval, realizing accurate conversion from image space to photometric space. The calculation of surface reflection intensity coefficient needs to consider the geometric configuration of the light source and the reflection characteristics of the material. According to the Lambert reflection law, the reflection intensity is proportional to the cosine of the incident angle, but the skin aluminum alloy surface has both diffuse reflection and specular reflection characteristics. By calculating the dot product of each point cloud normal vector and the incident direction of the light source, the cosine value is obtained, and then multiplied by the normalized light intensity value corresponding to the point to get the reflection intensity coefficient. Combining the spatial coordinates and the reflection intensity set, the reflection intensity distribution map containing position and optical information is formed. The reflection region division adopts an adaptive threshold method based on statistical distribution. First, calculate the histogram distribution of the reflection intensity of the entire skin surface, identify three distribution peaks through Gaussian mixture model fitting, respectively corresponding to weak reflection, medium reflection and strong reflection regions. When the average reflection intensity exceeds the preset threshold, it indicates that there is a large area of strong reflection that needs to be compensated. The threshold is usually set to 0.7 of the normalized intensity, which is obtained through a large number of experimental statistics, and can effectively distinguish between situations that need to be compensated and do not need to be compensated. The parameter setting of high dynamic range imaging follows the logarithmic interval principle, ensuring that the entire dynamic range is covered. The exposure time sequence increases by a power of 2, starting from the shortest exposure time, doubling each time, until the longest exposure time is reached. The light source power increases linearly, starting from the lowest power and increasing by a fixed step to the rated power. This combination can obtain a sequence of images with different exposure amounts, providing rich brightness level information for subsequent fusion. Each set of parameters corresponds to the acquisition of one frame of image, forming an exposure sequence image set. The weighted fusion process uses a weight distribution strategy based on pixel reliability. For each pixel position, calculate its signal-to-noise ratio in different exposure images, pixels with high signal-to-noise ratio are given a larger weight, and pixels close to saturation or underexposure have a reduced weight. The weight function adopts a Gaussian distribution, with the maximum weight at the middle gray level and gradually decreasing at both ends. The fused high dynamic range image is obtained by weighted averaging, which contains complete detail information from dark to bright. According to the brightness characteristics of different reflection regions, the optimal exposure parameter combination of each region is extracted from the fused image. When reacquiring in different regions, the spatial continuity of the parameters needs to be considered. The strong reflection area uses short exposure and low power parameters to reduce the number of saturated pixels and preserve high light details; the weak reflection area uses long exposure and high power parameters to improve the dark signal intensity and enhance the visibility of details; the medium reflection area uses a compromise parameter to balance the performance of bright and dark parts. An overlapping transition zone is set between adjacent regions, with a width of 10% to 20% of a single acquisition field of view, and the acquisition parameters are gradually adjusted in the transition zone to achieve smooth transition of the parameters.The bilinear interpolation processing is a key step to eliminate the stitching marks. In the transition zone at the regional boundary, the gray value of each pixel is determined by the acquisition results of the adjacent two regions. The interpolation weight is calculated according to the distance of the pixel to the regional boundary, and the closer the distance is, the greater the weight is. Through this gradual transition, the gray level mutation caused by different exposure parameters is eliminated, and the visually continuous point cloud data is obtained. When detecting the wing skin of a certain type of aircraft, due to the large curvature of the wing leading edge, the reflection characteristics of different positions are obviously different. The top of the leading edge is close to the vertical incidence, and the reflection intensity is the highest, which is easy to overexpose; while the side area has a large incidence angle, and the reflection intensity is reduced. Through the above adaptive compensation method, the appropriate acquisition parameters can be automatically selected for different regions to ensure the uniformity of the detection data quality of the entire wing surface. Further, the quality evaluation of the compensated point cloud is realized by calculating two indexes of point cloud density uniformity and gray distribution range. The density uniformity reflects the sampling sufficiency of different regions, and the gray distribution range reflects the utilization rate of dynamic range. After adaptive light compensation, the standard deviation of point cloud density is reduced, the effective gray range is improved, and the data basis for subsequent damage identification is significantly improved.

[0024] Step S106: extracting geometric features from the compensated point cloud, and performing denoising processing on the compensated point cloud based on the geometric features by using a spatial domain filtering method to obtain a denoised point cloud. In an embodiment, a geometric feature vector is extracted from the compensated point cloud, a two-dimensional Gaussian filtering algorithm with a preset standard deviation range is used to remove noise interference, and a denoised point cloud is determined.

[0025] For each sampling point in the compensated point cloud, k nearest neighbor points within a preset radius range are searched to form a neighborhood point set, a covariance matrix is constructed for the neighborhood point set and eigenvalue decomposition is performed, the eigenvector corresponding to the minimum eigenvalue is taken as the normal vector of the sampling point, the normal vector of the adjacent point is made to be less than 90 degrees by propagating the normal vector direction, and a directional consistent normal vector field is obtained. According to the normal vector field, a quadratic polynomial surface z=ax2+bxy+cy2+dx+ey+f is fitted in the neighborhood of each sampling point, the coefficients a to f are solved by the least square method, the second-order partial derivative is calculated according to the surface equation to form a Hessian matrix, and the eigenvalues of the Hessian matrix are the principal curvatures and the secondary curvatures. At the same time, the distance standard deviation of the point to the centroid of the neighborhood point set is calculated as the neighborhood point distance deviation, and a geometric feature vector containing the normal vector, the principal curvature value and the distance deviation is obtained. For each component of the geometric feature vector, a corresponding feature map is constructed, a two-dimensional Gaussian filtering kernel is used for convolution operation on the feature map, and the standard deviation of the filtering kernel is adaptively adjusted within a preset range according to the local density of the point cloud. If the amplitude of the feature value change of a certain point before and after filtering exceeds a preset threshold, the point is marked as a noise point and removed from the point cloud, and the remaining points constitute a denoised point cloud.

[0026] Specifically, in one embodiment, k-nearest neighbor search is implemented by building KD-tree data structure of point cloud for fast query. For a million-level point cloud data typical of a skin surface, KD-tree can reduce the complexity of neighborhood search from O(n) to O(log n). The selection of k value is related to the density of point cloud, k takes 8 to 12 in dense area, and k takes 15 to 20 in sparse area, which ensures that there are enough neighborhood points participating in feature calculation. The construction of covariance matrix is based on the spatial distribution characteristics of neighborhood points. After the neighborhood point set is centralized, a 3x3 covariance matrix is calculated, and the elements are the covariances of each coordinate component. Three eigenvalues and corresponding eigenvectors are obtained by Jacobi iteration method or QR decomposition. The eigenvector corresponding to the minimum eigenvalue is the normal vector of the point, which reflects the minimum change direction of the neighborhood point set. The normal vector is oriented by a propagation algorithm, starting from the seed point, gradually adjusting the normal vector direction of adjacent points, so that the included angle is less than 90 degrees. Quadratic polynomial surface fitting uses least squares method to solve overdetermined equation set. For n points in the neighborhood , , , construct an n x 6 coefficient matrix A, where each row is [ ², , ², , , 1], and obtain the polynomial coefficients by solving the normal equation Ax= b. The Hessian matrix is composed of second-order partial derivatives ²z x²、 ²z y、 ²z y², and its eigenvalues and ​​The principal curvature is the main curvature, which reflects the bending degree of the skin surface. The damage area usually shows abnormal curvature, such as negative principal curvature at the dent, and sharp change of curvature at the crack edge. The construction of the two-dimensional Gaussian filter kernel considers the anisotropic distribution of the point cloud. The filter kernel size is adaptively adjusted according to the local point density. A 3x3 kernel is used in dense areas, and the size is expanded to 5x5 or 7x7 in sparse areas. The standard deviation σ is dynamically adjusted within a preset range, usually 1.5 to 2.5 times the local average point spacing. Feature maps are constructed for the normal vector, principal curvature and distance deviation, and independent filtering is performed. Multi-feature comprehensive evaluation is used to determine noise points. When the angle between the normal vector of a point and the average normal vector of the neighborhood exceeds 30 degrees, or the principal curvature value exceeds 3 times the standard deviation of the neighborhood average, or the distance deviation exceeds the threshold, the point is marked as a potential noise point. If more than two of the three features are abnormal, the point is confirmed as a noise point and removed, achieving a robust denoising effect.

[0027] Step S108: Based on the denoised point cloud, a three-dimensional mesh model representing the three-dimensional structure of the aircraft skin surface is constructed, and a damage candidate area of the aircraft skin surface is determined from the three-dimensional mesh model by a point cloud local geometric feature analysis method. In one embodiment, a three-dimensional mesh model is constructed for the denoised point cloud, and a curvature analysis algorithm based on point cloud neighborhood is used to detect potential damage areas and determine damage candidate areas.

[0028] A three-dimensional mesh is constructed for the denoised point cloud using the Delaunay triangulation algorithm. Each point in the point cloud is inserted as a vertex by the point-by-point insertion method. According to the criterion that no other vertex is contained in the incircle of any triangle, adjacent vertices are connected to form triangular facets, and a three-dimensional mesh model containing vertex, edge and facet topological relationships is obtained. According to the three-dimensional mesh model, all adjacent vertices directly connected to each vertex are searched, and a quadratic surface of these adjacent vertices is fitted by the least squares method. The principal curvature and the secondary curvature of the surface at the vertex are calculated. A curvature feature map is constructed according to the absolute values of the Gaussian curvature and the mean curvature. For the curvature feature map, when the absolute value of the Gaussian curvature or the absolute value of the mean curvature of a vertex exceeds a preset threshold, the vertex is marked as a damage seed point. From the seed point, the adjacent vertices are expanded. If the curvature values of the adjacent vertices and the seed point differ by less than a preset tolerance, they are included in the same area. Through iterative expansion, multiple connected areas are formed, and damage candidate areas are determined.

[0029] Specifically, in one embodiment, the Delaunay triangulation is implemented using an incremental insertion algorithm. A convex hull containing all the point cloud is initially constructed, and then interior points are inserted one by one. For each inserted point, a triangle containing the point is found, the triangle is deleted and a new triangle is formed by connecting the new point. The Delaunay property is maintained by edge flip operations, i.e. no other vertex is contained in the circumcircle of any triangle, which guarantees the quality of the triangular mesh and avoids the occurrence of excessively narrow triangles. The calculation of curvature features is based on local surface fitting. For each vertex in the mesh, a ring of adjacent vertices directly connected to it is collected, usually containing 6 to 8 vertices. The coordinates of these vertices are transformed into a local coordinate system with the target vertex as the origin and the normal vector as the z-axis, and then a quadratic polynomial surface z=ax2+2bxy+cy2is fitted. The coefficients a, b, and c are solved by the least squares method, forming a 2x2 shape operator matrix, the eigenvalues of which are the principal curvatures and . The Gaussian curvature K= x reflects the intrinsic bending degree of the surface, and the mean curvature H=( + ) / 2 represents the extrinsic bending of the surface. It should be noted that the selection of damage seed points uses a double threshold judgment. When the absolute value of the Gaussian curvature exceeds 3 times the standard deviation of the normal curvature range of the skin, or the absolute value of the mean curvature exceeds the preset threshold, the vertex is marked as a potential seed point. For pit-type damage, the Gaussian curvature is negative and has a large absolute value; for protrusion-type damage, the Gaussian curvature is positive; and the edge of a crack exhibits a sharp change in mean curvature. By way of example, the region growing process uses a priority queue to manage the vertices to be processed. Starting from the seed point, all its adjacent vertices are added to the queue, and the vertices are sorted according to the curvature similarity. The first vertex in the queue is taken out, and the difference between its curvature value and the curvature of the seed point is calculated. If the difference is less than 20% of the preset tolerance range, the vertex is merged into the current region, and its unvisited adjacent vertices are added to the queue. This process is repeated until the queue is empty, forming a connected damage candidate region. Preferably, the formed candidate region is post-processed and optimized. The area, perimeter, and compactness of each candidate region are calculated, and regions with too small an area may be noise-induced false damage and are removed. For a narrow candidate region, the main direction is identified by skeleton extraction, and it is determined whether it is a crack-type damage. Through this multi-level screening, the accuracy of damage candidate region identification is improved.

[0030] The above steps achieve automatic high-precision processing of point clouds. Using two-dimensional Gaussian filtering and damage detection algorithm based on neighborhood principal curvature, a three-dimensional mesh is constructed by Delaunay triangulation, and a damage candidate region is located by combining curvature thresholding, achieving automatic noise removal and damage localization with reduced positioning error.

[0031] Step S110: obtaining a point cloud subset associated with the damage candidate region, and performing damage type classification on the point cloud subset by using a deep learning model to obtain a classification result of the damage type. In an embodiment, a point cloud subset in the damage candidate region is obtained, a deep learning model is used to classify the damage type, and a classification result is obtained.

[0032] The bounding box coordinates of the damage candidate region are obtained, a point cloud subset containing the damage region and its surrounding neighborhood is cropped from the original point cloud according to the bounding box, voxelization processing is performed on the point cloud subset, the point cloud is converted into a three-dimensional voxel grid according to a preset resolution, each voxel records the density and normal vector distribution information of the points inside it, and a voxelization feature tensor is obtained. For the voxelization feature tensor, data augmentation is performed through rotation, mirroring and scaling, a three-dimensional convolutional neural network is used to extract multi-scale spatial features, convolutional layers extract edge contour features from low layers, local shape patterns from middle layers, and global semantic features from high layers, feature pyramids are used to fuse feature maps of different levels to obtain multi-scale feature representations. According to the multi-scale feature representations, full connection layers are used to map to a damage category space, a softmax function is used to calculate the probability distribution of each damage category, the categories include cracks, pits, corrosion, scratches and bulges, the category with the highest probability is selected as the predicted category, and its confidence score is recorded to obtain a preliminary classification result. For the preliminary classification result, if the confidence score is lower than a preset threshold, geometric feature descriptors of the candidate region are extracted, including surface roughness, depth distribution histogram and shape compactness, a pre-trained support vector machine is used for secondary discrimination, and the discrimination results of deep learning and geometric features are fused to determine the final classification result.

[0033] Specifically, in one embodiment, the point cloud cropping employs an axis-aligned bounding box method to achieve accurate region extraction. According to the boundary vertex coordinates of the damage candidate region, the minimum and maximum coordinate values are calculated to form a three-dimensional bounding box. To preserve the context information of the damage edge, the dimensions of the bounding box are expanded by 10% to 15%, ensuring that the cropped point cloud subset contains complete damage features and their transition regions. For typical skin damage, the cropped point cloud subset usually contains 5000 to 20000 points, which not only guarantees the integrity of local details but also controls the computational complexity of subsequent processing. Specifically, the voxelization process is a key process for converting irregular point clouds into regular three-dimensional grids. The voxel resolution is set to 2mm x 2mm x 2mm, which can capture millimeter-level damage details. For each voxel, the number of points contained inside is counted as the density value, and the average normal vector of these points is calculated as the directional feature. Empty voxels are assigned a value of zero, and occupied voxels are normalized to the 0 to 1 interval according to the point density. In this way, sparse point cloud data is converted into dense 32x32x32 or 64x64x64 three-dimensional tensor representation, suitable for convolutional neural network processing. The voxelization process preserves the spatial topology and local geometric features of the point cloud, providing a standardized input format for deep learning. It should be noted that data augmentation strategies are crucial for improving model generalization. Rotation augmentation is performed in three axes, with rotation angles randomly sampled within the range of -30 degrees to 30 degrees; mirror augmentation is performed by flipping along the main plane; scale transformation adjusts within the range of 0.8 to 1.2 times, simulating the scanning effect at different distances. In addition, random point dropping and Gaussian noise injection are also introduced to simulate data loss and measurement errors in actual detection. Each training sample produces 8 to 10 variants by combining different augmentation operations, greatly expanding the diversity of the training data set.The full connection layer maps the 256-dimensional feature vector to the 5-dimensional class space, and calculates the posterior probability of each class by the softmax function. The training process uses the cross-entropy loss function, and uses the stochastic gradient descent optimizer with momentum, with an initial learning rate of 0.01, which is attenuated to 0.1 of the original every 30 epochs. In one possible implementation, the confidence threshold is determined by the validation set statistical method. Collect the confidence scores of all prediction results on the validation set, and draw the confidence-accuracy curve. When the confidence is less than 0.7, the classification accuracy decreases significantly, so 0.7 is set as the threshold to trigger secondary discrimination. For low-confidence samples, additional geometric feature descriptors are extracted for auxiliary discrimination. It can be understood that the calculation of the geometric feature descriptor is based on the statistical characteristics of the damage area. The surface roughness is obtained by calculating the standard deviation of the distance from the point to the fitting plane; the depth distribution histogram quantizes the depth value into 20 intervals, and counts the point distribution of each interval; the shape compactness is defined as the cubic root of the volume to surface area ratio, reflecting the regularity of the damage. These features form a 15-dimensional vector, which is input into the pre-trained support vector machine for classification. The support vector machine uses a radial basis kernel function, and the kernel parameter and penalty coefficient are optimized by grid search. Further, the fusion discrimination strategy comprehensively considers the classification results of deep learning and geometric features. When the two methods are consistent, the class is directly output; when the predictions are inconsistent, the final class is determined by weighted voting according to the respective confidence. The weight of the deep learning result is set to 0.6, and the weight of the geometric feature result is set to 0.4, which is obtained through a large number of experiments. Through this multi-model fusion mechanism, the classification accuracy is improved, especially for the corrosion type damage with fuzzy boundary, the recognition accuracy is improved obviously. For example, when a suspected damage on the fuselage skin is detected, the system first discriminates it as a pit with a confidence of 0.65 through the deep learning model. Since the confidence is lower than the threshold, geometric feature analysis is triggered, and it is found that the surface roughness of the area is abnormally high, and the depth distribution presents a multi-peak feature, which is more consistent with the characteristics of corrosion damage. Finally, the system outputs the corrosion damage judgment and marks the area that needs to be checked.

[0034] Step S112: according to the classification result, the boundary information of the damage candidate area is fused to generate a damage annotation map. In one embodiment, according to the classification result, the boundary information of the damage candidate area is fused, if the boundary curvature change rate is higher than the change threshold, it is marked as a crack damage, if the boundary curvature change rate is not higher than the change threshold, it is marked according to the classification result of the deep learning model respectively as other damage types, and a damage annotation map is obtained.

[0035] According to the classification result, a boundary contour point sequence of the damage candidate area is extracted, a contour simplification algorithm is used to retain key feature points, an angle change formed by each point in the feature point sequence and its adjacent points before and after is calculated, a local curvature is obtained by dividing the angle change value by the arc length between the points, a curvature change rate is obtained by dividing the difference value of the curvatures of the adjacent points by the distance between the points, and a boundary curvature change rate sequence is constructed. For the boundary curvature change rate sequence, if the change rate values of a plurality of continuous points in the sequence exceed a preset change threshold value, the damage area is re-labeled as a crack damage, if the change threshold value is not continuously exceeded, the original classification result output by the deep learning model is maintained, and a corrected damage category is obtained. According to the corrected damage category, a preset color coding value is assigned to each damage type, the point cloud in the damage area is marked according to the corresponding color value, the marking result is superimposed on the three-dimensional coordinate space of the original skin point cloud through color rendering, and a damage annotation diagram that can distinguish different damage types is formed.

[0036] Specifically, in one embodiment, the boundary contour extraction is implemented by an eight-neighborhood tracing algorithm. Starting from an arbitrary boundary point of the damage region, the neighboring boundary points are searched in a clockwise or counterclockwise direction to form an ordered sequence of contour points. The contour simplification adopts a recursive segmentation method, sets a distance threshold of 0.5 mm, retains feature points with obvious curvature changes, and eliminates redundant intermediate points. The simplified contour not only maintains the original shape characteristics, but also reduces the subsequent calculation amount. Specifically, the calculation of local curvature is based on the geometric relationship of discrete points. For a point Pi on the contour, two vectors are formed with its adjacent points Pi-1 and Pi+1 before and after it. The curvature κ is calculated by the relationship between the vector angle θ and the arc length s, i.e. κ = θ / s. The curvature change rate is obtained by the curvature difference Δκ between the adjacent two points divided by the distance Δs between the points, i.e. dκ / ds = Δκ / Δs. This discrete calculation method is suitable for processing point cloud data and can effectively capture the geometric variation characteristics of the contour. The crack-type damage usually exhibits a sharp change in curvature at the boundary, while the boundaries of other types of damage are relatively smooth. It should be noted that the determination of continuous over-threshold value adopts a sliding window detection method. The window size is set to 5 to 7 consecutive points, and when more than 80% of the points in the window have a curvature change rate exceeding the preset threshold, it is considered that there is a continuous over-threshold value. The preset threshold is obtained by statistical analysis of a large number of samples, and is usually set to 3 times the standard deviation of the mean normal curvature change rate. This determination method can not only identify the linear extension characteristics of cracks, but also avoid misjudgment caused by individual noise points. Exemplarily, color coding is implemented in the HSV color space to achieve differentiated display. The crack damage is assigned a hue value of 0 degrees corresponding to the red color system, the pit damage is assigned a hue value of 240 degrees corresponding to the blue color system, the corrosion damage is assigned a hue value of 60 degrees corresponding to the yellow color system, the scratch damage is assigned a hue value of 120 degrees corresponding to the green color system, and the bulge damage is assigned a hue value of 300 degrees corresponding to the purple color system. The saturation is set to the maximum value to ensure bright colors, and the brightness is adjusted according to the damage depth, with a lower brightness for a greater depth. Preferably, the rendering of the damage annotation map adopts point cloud coloring technology. For each point in a damage region, a corresponding RGB color value is assigned according to the type of damage it belongs to. The undamaged area remains displayed in the original grayscale, forming a clear visual contrast. Three-dimensional rendering is achieved through OpenGL or DirectX graphics interface, supporting interactive operations such as rotation, scaling, and translation, facilitating observation of damage distribution from different angles. The rendering result can be exported as a standard PLY or OBJ format file, containing vertex coordinates and color information.

[0037] Step S114: Based on the damage annotation map, an error compensation model constructed by a parameter fitting-based optimization algorithm is used to calibrate the damage size parameters to generate a final damage report. In one embodiment, the damage size parameters including length, width, and depth are quantified from the damage annotation map, and an iterative optimization algorithm based on least squares method is used to calibrate the measurement error to determine the final damage report.

[0038] Three-dimensional point cloud coordinates of each damage area are extracted from the damage annotation map, principal component analysis is used to determine the principal axis direction of the damage, the maximum span of the point cloud along the principal axis direction is calculated as the damage length, the maximum span perpendicular to the principal axis is calculated as the width, and the maximum vertical distance of the point cloud to the surrounding undamaged area fitting plane is calculated as the depth, to obtain the initial size parameters. A measurement error model is established for the initial size parameters, the linear relationship between the multiple measurement values and the standard reference values is fitted by the least squares method, the residual value of each measurement point is calculated, the regression coefficient is adjusted according to the residual size, and the iterative calculation is performed until the residual sum of squares converges to below a preset threshold, to obtain the corrected size value. According to the corrected size value, a damage detection report is prepared, the position coordinates, damage type, length, width and depth values of each damage are recorded, the damage grade is determined according to the comparison between the damage size and the preset grading standard, and the final damage report containing all the damage information is formed.

[0039] Specifically, in practical applications, principal component analysis realizes principal axis extraction by constructing the covariance matrix of the point cloud of the damage area. The three-dimensional coordinates of all points in the damage area are normalized to the centroid coordinate system, a 3x3 covariance matrix is calculated, and three orthogonal principal directions are obtained by eigenvalue decomposition. The eigenvector corresponding to the largest eigenvalue is the principal axis direction of the damage, which is usually consistent with the extension direction of the damage. Project all points along the principal axis, and the difference between the maximum and minimum values of the projection is the damage length. Specifically, the determination of the reference plane uses a robust fitting method. Select the undamaged point cloud within a radius of 2 to 3 times the damage size around the damage area, use the random sample consensus algorithm to remove outliers, and fit the plane with the remaining points. The fitting plane equation is ax+by+cz+d=0, where the coefficients are solved by the least squares method. The damage depth is defined as the maximum perpendicular distance from the points in the damage area to the reference plane, with positive values indicating protrusions and negative values indicating depressions. This local reference plane method can adapt to the curvature changes of the skin and improve the accuracy of depth measurement. It should be noted that the measurement error model considers both systematic and random errors. Systematic errors mainly come from the calibration deviation of the scanning device and environmental factors, which are linear deviations between the measured values and the reference values; random errors are caused by measurement noise and follow a normal distribution. By collecting multiple measurement data of the same damage, a linear regression model y=ax+b+e is established, where y is the measured value, x is the reference value, a and b are the regression coefficients, and e is the random error term. Exemplarily, the iterative optimization process uses the weighted least squares method. In the initial iteration, all measurement points have equal weights, and the residual error ri=yi-axi-b is calculated. According to the residual error, the weight wi=1 / (1+|ri| / s) is updated, where s is the residual standard deviation. The regression coefficients are recalculated using the new weights, and the process is repeated until the coefficients of the adjacent two iterations change by less than a preset threshold, usually set to 0.001. After 3 to 5 iterations, the sum of squared residuals usually converges to a stable value. Preferably, the damage level determination uses a four-level classification standard. Level one damage is a superficial damage with a depth of less than 0.5 mm; level two damage has a depth of 0.5 mm to 2 mm; level three damage has a depth of 2 mm to 5 mm; and level four damage has a depth exceeding 5 mm or penetrates the skin. The report also records the area of the damage, which is obtained by calculating the convex hull of the point cloud in the damage area, providing a comprehensive quantitative basis for maintenance decisions.

[0040] The above steps establish a digitalized collaborative link for detection and repair. The output includes a standardized report containing "damage location + type + size + level", which supports direct docking with the China Eastern Airlines hangar MRO system, SAP system, and digital twin module, realizes real-time synchronization of detection data, and reduces data entry time.

[0041] Therefore, the scheme of the application adapts to the multi-scene collaborative demand of the machine library. It ensures the technical compatibility of the machine library intelligent device, such as supporting handheld mobile devices to view the detection results on site, and damage data triggering AGV scheduling, while meeting the requirements of machine library explosion-proof and anti-electromagnetic interference.

[0042] The scheme of the application can solve the data quality problem through "polarization filtering + adaptive light compensation" for the high-reflectivity and high-precision requirements of the aircraft skin, improve the damage identification accuracy through "three-dimensional geometric feature extraction + deep learning", and realize size quantization calibration through "least squares iterative optimization", forming a complete damage identification scheme that adapts to the aviation scene, effectively making up for the shortcomings of the prior art in the field of aircraft skin damage detection, and further improving the aircraft skin damage identification accuracy.

[0043] Specifically, the above-mentioned manner of the application has the following technical advantages: The anti-reflective capability is more suitable for the aviation scene: compared with the detection technology without anti-reflective design, the application adds polarization filtering and adaptive light compensation, which can improve the skin reflection point recognition rate and the point cloud efficiency; compared with the image-ultrasound fusion technology, the device is more portable and suitable for the narrow maintenance space of the machine library, without relying on ultrasonic equipment and eliminating reflection through a pure optical solution.

[0044] The damage positioning accuracy is higher: compared with the detection technology relying on two-dimensional feature binaryzation, the application can locate 0.1mm-level micro-cracks through three-dimensional point cloud curvature analysis, with high candidate area recognition accuracy, far exceeding the macro-damage recognition rate of the previous technology; compared with the image-ultrasound fusion technology, three-dimensional grid modeling combined with principal curvature analysis can directly reflect the skin surface morphology, reducing the damage positioning error and meeting the micron-level requirement of aviation maintenance.

[0045] The classification and quantization are more intelligent and accurate: the application uses three-dimensional convolutional neural network CNN + secondary discrimination to improve the damage classification accuracy, especially for the recognition accuracy of fuzzy boundary corrosion damage; at the same time, the least squares iterative optimization is added, the size measurement error is reduced, and it can be directly used for aviation material selection.

[0046] The digital collaborative capability is stronger: the report of the application can be directly connected to the machine library MRO system, automatically associated with the work card maintenance requirements, and synchronized to the digital twin module to realize virtual damage mapping; it supports real-time data retrieval by handheld mobile devices, and maintenance personnel do not need to go back and forth to the control room, so that the single-aircraft detection time is shortened from 4 hours to 1.5 hours, which meets the efficient and paperless operation and maintenance goal of the machine library.

[0047] Figure 3 The structure of the aircraft skin damage identification device 300 according to the embodiment of the application is shown in the structural schematic diagram, which includes the following modules: The point cloud acquisition and preprocessing module 302 is configured to acquire an initial point cloud of the aircraft skin surface by using a structured light device, and perform data preprocessing on the initial point cloud to obtain a preliminary corrected point cloud.

[0048] The reflection intensity analysis and compensation module 304 is configured to calculate a reflection intensity distribution of each region of the aircraft skin surface based on the preliminary corrected point cloud, and determine a compensated point cloud based on the calculated reflection intensity and a preset reflection intensity threshold.

[0049] The point cloud denoising module 306 is configured to extract geometric features from the compensated point cloud, and perform denoising processing on the compensated point cloud based on the geometric features by using a spatial domain filtering method to obtain a denoised point cloud.

[0050] The damage region positioning module 308 is configured to construct a three-dimensional mesh model representing a three-dimensional structure of the aircraft skin surface based on the denoised point cloud, and determine a damage candidate region of the aircraft skin surface from the three-dimensional mesh model by using a point cloud local geometric feature analysis method.

[0051] The damage classification module 310 is configured to acquire a point cloud subset associated with the damage candidate region, and perform damage type classification on the point cloud subset by using a deep learning model to obtain a classification result of the damage type.

[0052] The damage annotation generation module 312 is configured to generate a damage annotation map according to the classification result and by fusing boundary information of the damage candidate region.

[0053] The damage report generation module 314 is configured to calibrate a damage size parameter by using an error compensation model constructed based on a parameter fitting optimization algorithm based on the damage annotation map to generate a final damage report.

[0054] Figure 4 A hardware structure schematic diagram of an aircraft skin damage identification device 400 according to an embodiment of the present application is shown.

[0055] The aircraft skin damage identification device 400 can include a processor 402 and a memory 404 storing computer program instructions.

[0056] Specifically, the processor 402 can include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or can be configured as one or more integrated circuits implementing the embodiments of the present application.

[0057] The memory 404 can include mass storage for data or instructions. As an example and not by way of limitation, the memory 404 can include a Hard Disk Drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a Universal Serial Bus (USB) drive or a combination of two or more of these. In one example, the memory 404 can include removable or non-removable (or fixed) media, where the memory 404 is nonvolatile solid-state memory. The memory 404 can be internal or external to the integrated gateway disaster recovery appliance.

[0058] The memory 404 can include read-only memory (ROM), random-access memory (RAM), magnetic disk storage mediums, optical storage mediums, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Thus, in general, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software that, when executed (by one or more processors), is operable to

[0059] The processor 402 implements the … method in the embodiments shown by reading and executing computer program instructions stored in the memory 404. Figure 1 The … method in the embodiments shown.

[0060] In one example, the xx appliance also includes a communication interface 406 and a bus 410. Where, as shown, the processor 402, the memory 404, the communication interface 406 are connected and communicate with each other through the bus 410. Figure 4

[0061] The communication interface 406 is mainly used to realize the communication between the modules, devices, units and / or appliances in the embodiments of the present application.

[0062] ​Bus 410 includes a hardware, software, or both that couples components of the online data traffic metering device to each other. As an example without limitation, bus can include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or another suitable bus or a combination of two or more of these. Where suitable, bus 410 can include one or more buses. Although specific busses are described and illustrated in this embodiment, this application contemplates any suitable bus or interconnect.

[0063] In addition, in combination with the aircraft skin damage identification method in the above embodiments, the embodiments of the present application can provide a computer storage medium to implement. The computer storage medium has computer program instructions stored thereon; the computer program instructions are executed by a processor to implement any one of the aircraft skin damage identification methods in the above embodiments.

[0064] The embodiments of the present application also provide a computer program product, comprising a computer program, the computer program is executed by a processor to implement any one of the aircraft skin damage identification methods in the above embodiments.

[0065] It needs to be clear that the present application is not limited to the specific configurations and processes described above and shown in the drawings. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between steps, after understanding the spirit of the present application.

[0066] The functions indicated in the structural block diagrams described above can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, and the like. When implemented in software, the elements of the present application are program or code segments that are used to perform the required tasks. The program or code segments can be stored in a machine-readable medium, or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link. The "machine-readable medium" can include any medium that can store or transfer information. Examples of the machine-readable medium include an electronic circuit, a semiconductor memory device, a read-only memory (ROM), a flash memory, an erasable read-only memory (EROM), a floppy diskette, a compact disk read-only memory (CD-ROM), an optical disk, a hard disk, a fiber optic medium, a radio frequency (RF) link, and the like. The code segments can be downloaded via computer networks such as the Internet, an intranet, and the like.

[0067] It is also noted that the examples mentioned in the present application describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the steps mentioned in the examples, that is, the steps can be performed in the order mentioned in the examples, or in an order different from the examples, or several steps can be performed simultaneously.

[0068] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions can also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other processing device to operate in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0069] The above merely describes a specific implementation of the present application. Those skilled in the art can clearly understand the specific working processes of the system, modules and units described above for the convenience and brevity of description, and can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein again. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application.

Claims

1. A method of aircraft skin damage detection, characterized in that, The method comprises the following steps: An initial point cloud of the aircraft skin surface is acquired by a structured light device, and data preprocessing is performed on the initial point cloud to obtain a preliminary corrected point cloud; Based on the preliminary corrected point cloud, the reflection intensity distribution of each region of the aircraft skin surface is calculated, and a compensated point cloud is determined based on the calculated reflection intensity and a preset reflection intensity threshold; Geometric features are extracted from the compensated point cloud, and a spatial domain filtering method is used to denoise the compensated point cloud based on the geometric features to obtain a denoised point cloud; A three-dimensional mesh model representing the three-dimensional structure of the aircraft skin surface is constructed based on the denoised point cloud, and a point cloud local geometric feature analysis method is used to determine the damage candidate area of the aircraft skin surface from the three-dimensional mesh model; A point cloud subset associated with the damage candidate area is acquired, a deep learning model is used to classify the point cloud subset to obtain a classification result of the damage type; According to the classification result, the boundary information of the damage candidate area is fused to generate a damage annotation map; and Based on the damage annotation map, an error compensation model constructed by a parameter fitting-based optimization algorithm is used to calibrate the damage size parameters to generate a final damage report.

2. The aircraft skin damage detection method of claim 1, wherein The data preprocessing of the initial point cloud to obtain the preliminary corrected point cloud comprises: Calculate the polarization feature of each sampling point in the initial point cloud; and Based on the comparison result of the polarization feature and the preset polarization feature threshold, correct the initial point cloud to obtain the preliminary corrected point cloud.

3. The aircraft skin damage detection method of claim 2, wherein The calculation of the polarization feature of each sampling point in the initial point cloud comprises: Obtain the vertical polarization component intensity and the horizontal polarization component intensity of each sampling point; and Based on the vertical polarization component intensity and the horizontal polarization component intensity, calculate the polarization degree of each sampling point.

4. The aircraft skin damage detection method of claim 3, wherein, The vertical polarization component intensity and the horizontal polarization component intensity of each sampling point are collected by a binocular camera of the structured light device with an installed orthogonal polarizer.

5. The aircraft skin damage detection method of claim 1, wherein, The calculation of the reflection intensity distribution of each region of the aircraft skin surface comprises: Obtain the optical property information of each sampling point in the preliminary corrected point cloud; and Convert the optical property information into reflection intensity values, and based on the reflection intensity values of all sampling points, construct a reflection intensity distribution map of the aircraft skin surface.

6. The aircraft skin damage detection method of claim 5, wherein, The optical property information at least includes a gray value.

7. The aircraft skin damage identification method according to any one of claims 1 to 4, characterized in that, The determination of the compensated point cloud based on the calculated reflection intensity and the preset reflection intensity threshold comprises: In response to the calculated reflection intensity exceeding the preset reflection intensity threshold, a dynamic acquisition parameter adjustment strategy is applied to obtain the compensated point cloud; and In response to the calculated reflection intensity not exceeding the preset reflection intensity threshold, the preliminary corrected point cloud is taken as the compensated point cloud.

8. The aircraft skin damage detection method of claim 7, wherein, The dynamic acquisition parameter adjustment strategy comprises: Divide the aircraft skin surface into multiple reflection intensity regions according to the reflection intensity distribution; Configure multiple sets of acquisition parameters for different reflection intensity regions, and determine the optimal acquisition parameters of each region through a weighted fusion strategy; and Based on the optimal acquisition parameters, data reacquisition and data fusion are performed on the corresponding reflection intensity region to generate the compensated point cloud.

9. The aircraft skin damage detection method according to any one of claims 1 to 4, characterized in that, The spatial domain filtering method comprises: Adaptively configuring parameters of a filtering kernel based on spatial distribution characteristics of the compensated point cloud.

10. The aircraft skin damage detection method according to any one of claims 1 to 4, characterized in that, The spatial domain filtering comprises Gaussian filtering.

11. The aircraft skin damage detection method according to any one of claims 1 to 4, characterized in that, Determining the damage candidate region of the aircraft skin surface from the three-dimensional mesh model through a point cloud local geometric feature analysis method comprises: For each vertex of the three-dimensional mesh model, fitting a quadric surface of its adjacent vertices and calculating the Gaussian curvature and the mean curvature at the vertex; Marking the vertex whose absolute value of the Gaussian curvature or the absolute value of the mean curvature exceeds a corresponding preset threshold as a damage seed point; Starting from each damage seed point, iteratively expanding to its adjacent vertices, incorporating the vertices with a curvature value difference less than a preset tolerance into the same region, and finally generating a connected damage candidate region.

12. The aircraft skin damage detection method according to any one of claims 1 to 4, characterized in that, Classifying the damage type of the point cloud subset using a deep learning model to obtain a classification result of the damage type comprises: Obtaining a probability distribution of each damage category output by the deep learning model; Determining a preliminary classification result and its confidence based on the probability distribution; and In response to the confidence being lower than a preset confidence threshold, extracting auxiliary geometric features of the point cloud subset, and verifying the preliminary classification result using a discriminant model to determine the classification result.

13. The aircraft skin damage detection method of claim 12, wherein, The discriminant model is a pre-trained support vector machine.

14. The aircraft skin damage detection method according to any one of claims 1 to 4, characterized in that, Fusing boundary information of the damage candidate region according to the classification result to generate a damage annotation map comprises: Analyzing the geometric feature change rate of the damage candidate region boundary; Comparing the geometric feature change rate with a preset change rate threshold; and According to the comparison result and the classification result, marking the damage candidate region to generate the damage annotation map.

15. The aircraft skin damage detection method according to any one of claims 1 to 4, characterized in that, The damage size parameters include the length, width and depth of the damage.

16. The aircraft skin damage detection method according to any one of claims 1 to 4, characterized in that, Based on the damage annotation map, using an error compensation model constructed by a parameter fitting-based optimization algorithm to calibrate the damage size parameters to generate a final damage report comprises: Extracting three-dimensional point cloud coordinates of each damage candidate region from the damage annotation map to obtain initial damage size parameters; Establishing the error compensation model constructed based on the least square method to correct the initial damage size parameters to obtain corrected damage size parameters; and Based on the corrected damage size parameters, determining the damage grade and generating the final damage report.

17. An aircraft skin damage detection apparatus, characterized by, The device comprises: A point cloud acquisition and preprocessing module for acquiring an initial point cloud of an aircraft skin surface through a structured light device and performing data preprocessing on the initial point cloud to obtain a preliminary corrected point cloud; A reflection intensity analysis and compensation module for calculating the reflection intensity distribution of each region of the aircraft skin surface based on the preliminary corrected point cloud, and determining a compensated point cloud based on the calculated reflection intensity and a preset reflection intensity threshold; A point cloud denoising module for extracting geometric features from the compensated point cloud and performing denoising processing on the compensated point cloud based on the geometric features using a spatial domain filtering method to obtain a denoised point cloud; The damage area positioning module is configured to construct a three-dimensional mesh model representing a three-dimensional structure of the aircraft skin surface based on the denoised point cloud, and determine a candidate damage area of the aircraft skin surface from the three-dimensional mesh model by using a point cloud local geometric feature analysis method. The damage classification module is configured to obtain a point cloud subset associated with the candidate damage area, and perform damage type classification on the point cloud subset by using a deep learning model to obtain a classification result of the damage type. The damage label generation module is configured to generate a damage label map according to the classification result and by fusing boundary information of the candidate damage area. The damage report generation module is configured to calibrate the damage size parameter by using an error compensation model constructed based on a parameter fitting optimization algorithm based on the damage label map to generate a final damage report.

18. An aircraft skin damage detection apparatus, characterized by, The device comprises a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the aircraft skin damage identification method according to any one of claims 1-16.

19. A computer-readable storage medium, characterized in that, The computer storage medium stores computer program instructions, and the computer program instructions are executed by a processor to implement the aircraft skin damage identification method according to any one of claims 1-16.

20. A computer program product, characterised in that, The computer program is executed by a processor to implement the aircraft skin damage identification method according to any one of claims 1-16.

Images