A visual detection and quality evaluation method for small aircraft structures

Abstract

The embodiment of the application discloses a kind of visual detection and quality evaluation method for aircraft small structure, comprising: S1: obtain three-dimensional point cloud data, and the three-dimensional point cloud data is preprocessed;S2: carry out multi-angle two-dimensional image acquisition;S3: the two-dimensional image texture obtained in S2 is mapped to three-dimensional model surface, and three-dimensional model including surface information is generated;S4: extracting geometric feature and surface texture feature from three-dimensional model;S5: fusion and comparative analysis are carried out to identify potential abnormalities and defects;S6: output specific abnormal type and position;S7: according to the abnormal detection result of S6, comprehensive quality evaluation is carried out, and quality evaluation report is generated.The technical scheme provided by the embodiment of the application, by fusing three-dimensional scanning and high-resolution two-dimensional image technology and using classification model based on machine learning, the comprehensiveness, accuracy and efficiency of detection are significantly improved, so that production quality and safety are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine vision and automated detection, and particularly relates to a visual detection and quality evaluation method for small aircraft structural parts. BACKGROUND

[0002] Small aircraft structural parts are an indispensable important part in the research and production of aircraft, mainly used for the connection between different parts in the aircraft component assembly process, playing a role in transmitting and distributing loads between different parts. They must meet extremely high quality standards to meet the quality and safety of the aircraft.

[0003] The traditional quality detection method for small aircraft structural parts mainly relies on manual inspection or simple mechanical measurement. These detection methods not only consume a long time and are low in efficiency, but also have obvious limitations in detection accuracy and repeatability. With the development of technology, although automated detection technology based on two-dimensional images has been introduced, these detection methods often fall short when dealing with complex three-dimensional geometry and subtle surface texture, making it difficult to accurately identify small defects and internal structural abnormalities. SUMMARY

[0004] The purpose of the present application is to solve the above technical problems. The embodiments of the present application provide a visual detection and quality evaluation method for small aircraft structural parts to solve the problems of existing detection methods for small aircraft structural parts, such as obvious limitations in detection accuracy and repeatability, and difficulty in accurately identifying small defects and internal structural abnormalities when dealing with complex three-dimensional geometry and subtle surface texture.

[0005] The technical scheme of the present application: the embodiments of the present application provide a visual detection and quality evaluation method for small aircraft structural parts, comprising:

[0006] S1: using a three-dimensional scanner to scan the small aircraft structural part in all directions, obtaining three-dimensional point cloud data of the outer surface of the small aircraft structural part, and preprocessing the three-dimensional point cloud data;

[0007] S2: using a high-resolution camera to collect two-dimensional images of the small aircraft structural part from multiple angles;

[0008] S3: using the preprocessed three-dimensional point cloud data to construct a three-dimensional model, and mapping the texture of the two-dimensional images obtained in S2 to the surface of the three-dimensional model to generate a three-dimensional model including surface texture information;

[0009] S4: using a three-dimensional feature extraction algorithm to extract geometric features and surface texture information from the three-dimensional model, including the shape, size, surface texture information and subtle defects of the small structural part;

[0010] S5: In a multi-modal data fusion manner, the geometric features and surface texture information extracted in S4 are fused and compared with standard samples to identify potential abnormalities and defects;

[0011] S6: Using a pre-trained classification model, the comparative data after fusion in S5 are analyzed to output abnormality detection results, including the abnormality type and abnormality position of the aircraft small structure;

[0012] S7: Comprehensive quality assessment is performed on the abnormality detection results of S6 to generate a quality assessment report and propose optimization suggestions based on the detection results.

[0013] Optionally, in the visual detection and quality assessment method for aircraft small structure as described above, the three-dimensional scanner for scanning is pre-positioned in the working area of the aircraft small structure, and S1 comprises:

[0014] S11: Initial calibration of the three-dimensional scanner, including calibration of the position, angle and focal length of the scanner, so that the scanning area of the calibrated three-dimensional scanner covers the overall outer surface of the aircraft small structure;

[0015] S12: Control the three-dimensional scanner to scan the aircraft small structure from multiple perspectives according to the predetermined path and angle, and obtain three-dimensional point cloud data corresponding to each perspective;

[0016] S13: Splice and fuse the three-dimensional point cloud data corresponding to each perspective obtained in S12, and use the Iterative Closest Point algorithm ICP to align multiple sets of three-dimensional point cloud data in S13 to generate a three-dimensional point cloud data set;

[0017] S14: Denoising processing of the spliced three-dimensional point cloud data set, using a statistical filtering algorithm to remove noise points in the three-dimensional point cloud data set, the statistical filtering algorithm identifies and removes abnormal points by calculating the neighborhood statistical characteristics of each point;

[0018] S15: Identify and remove redundant points and isolated points in the three-dimensional point cloud data set by using a density clustering algorithm; the density clustering algorithm detects and removes redundant points with insufficient density by defining a density threshold and a minimum number of neighbors;

[0019] S16: Global alignment of the three-dimensional point cloud data set after denoising and redundancy removal using a global least squares fitting algorithm.

[0020] Optionally, in the visual detection and quality assessment method for aircraft small structure as described above, the high-resolution camera is pre-positioned in the working area of the aircraft small structure, and S2 comprises:

[0021] S21: initial calibration of the high-resolution camera, including calibration of the position, angle and focal length of the camera, so that the camera covers all surface areas of the small structural part;

[0022] S22: design a multi-angle shooting path, determine the specific shooting points of the camera at different positions and angles, the shooting path includes at least 8 positions at different angles;

[0023] S23: set the image acquisition parameters of the camera, including exposure time, aperture size and sensitivity;

[0024] S24: according to the planned shooting path, control the camera to collect multi-angle images of the small structural part of the aircraft at the predetermined position and angle, obtain the full two-dimensional image data of the small structural part of the aircraft, and at least 3 images are taken at each position and each angle;

[0025] S25: correct the distortion in the image using image distortion correction algorithm for the collected multi-position and multi-angle images, and perform color correction and contrast enhancement;

[0026] S26: splice and fuse the two-dimensional images collected at multiple positions and multiple angles, use image splicing algorithm to splice the images seamlessly into a complete image covering the entire surface of the small structural part, and the overlapping part between the images in the splicing process is not less than 30%.

[0027] Optionally, in the visual detection and quality evaluation method for the small structural part of the aircraft as described above, the S3 comprises:

[0028] S31: using the pre-processed three-dimensional point cloud data obtained in S1, a preliminary three-dimensional model is constructed by Delaunay triangulation algorithm, and a grid model containing the geometric shape of the small structural part of the aircraft is generated;

[0029] S32: register the two-dimensional image obtained in S2 with the preliminary three-dimensional model, including: using SIFT algorithm to identify corresponding feature points in the two-dimensional image and the three-dimensional model, and aligning the feature points through the camera projection matrix;

[0030] S33: texture mapping of the registered two-dimensional image, mapping the surface texture information in the two-dimensional image to the outer surface of the three-dimensional model; wherein the texture mapping uses perspective projection texture mapping algorithm to project the texture coordinates corresponding to each two-dimensional image pixel onto the surface of the three-dimensional model;

[0031] S34: texture fusion of the two-dimensional images obtained at multiple positions and multiple angles, using weighted average fusion algorithm to fuse the texture of the overlapping area, and in the fusion process, the pixels of each overlapping area are weighted and averaged;

[0032] S35: Detail enhancement is performed on the texture-mapped and fused three-dimensional model, specifically, a high dynamic range enhancement algorithm is used to adjust the brightness and contrast of the texture.

[0033] Optionally, in the method for visual inspection and quality evaluation of the small structure of the aircraft as described above, the S4 comprises:

[0034] S41: Geometric features are extracted from the three-dimensional model, including: using a surface fitting algorithm to fit each surface of the three-dimensional model to obtain the geometric shape of the small structure of the aircraft, the surface fitting algorithm is to minimize the error between the fitted surface and the three-dimensional point cloud data by least squares method, the calculation formula is:

[0035]

[0036] wherein z i is a point in the point cloud data, f(x i , y i ) is a fitted surface function, (x i , y i ) is the coordinate of the point cloud data, and n1 is the number of points.

[0037] S42: Dimension measurement is performed on the three-dimensional model, including: using a distance calculation algorithm to measure the dimensions of each part of the small structure of the aircraft, the distance calculation algorithm is used to calculate the Euclidean distance between two points in the three-dimensional model, the calculation formula is:

[0038]

[0039] wherein (x1, y1, z1) and (x2, y2, z2) are the coordinates of two points in the three-dimensional model, and d is the distance between the two points.

[0040] S43: Surface texture information of the three-dimensional model is extracted, including: using a normal vector analysis algorithm to calculate the normal vector of each point of the three-dimensional model, the normal vector is defined as a vector perpendicular to the surface, the calculation formula is:

[0041]

[0042] wherein N is the normal vector, v1 and v2 are two non-parallel vectors on the surface, × represents the cross product of vectors, and |·| represents the length of the vector.

[0043] S44: Fine defects in the three-dimensional model are identified, including: using a Gaussian curvature and mean curvature analysis algorithm to calculate the curvature of the surface of the three-dimensional model to identify fine defects such as depressions, protrusions and cracks on the surface of the small structure of the aircraft.

[0044] Optionally, in the method for visual detection and quality evaluation of small aircraft structures as described above, the Gaussian curvature and mean curvature analysis algorithm in S44 is respectively:

[0045]

[0046] wherein κ1 and κ2 are principal curvatures, and are partial derivatives of the surface in x and y directions, K is Gaussian curvature, and H is mean curvature.

[0047] Optionally, in the method for visual detection and quality evaluation of small aircraft structures as described above, S5 comprises:

[0048] S51: Obtain the geometric feature and surface texture information of the standard sample, including the shape, size, and surface texture information of the small aircraft structure, and take it as the reference data;

[0049] S52: Compare and analyze the geometric feature extracted in S4 with the geometric feature of the standard sample, and use a shape matching algorithm to evaluate the consistency of the geometric feature by calculating the similarity between the two, and the formula of the shape matching algorithm is:

[0050]

[0051] wherein A and B represent the geometric feature point sets of the small aircraft structure and the standard sample respectively, n2 is the number of geometric feature points, and ∥A a -B b ∥ is the Euclidean distance between the geometric feature points A a and B b of the small aircraft structure and the standard sample;

[0052] S53: Compare and analyze the surface texture information extracted in S4 with the surface texture information of the standard sample, and use a texture similarity algorithm to evaluate the consistency of the surface texture information by calculating the similarity between the two, and the formula of the texture similarity algorithm is:

[0053]

[0054] wherein T1 and T2 represent the texture images of the small aircraft structure and the standard sample respectively, m is the height of the texture image, n3 is the width of the texture image, and |T1(v,c)-T2(v,c)| is the absolute difference value between the corresponding pixel points of the small aircraft structure and the standard sample;

[0055] S54: Fuse the geometric feature and surface texture information, and use a weighted fusion algorithm to combine the geometric feature similarity and the texture information similarity to form a comprehensive similarity score, and the formula of the weighted fusion algorithm is:

[0056] Fused Similarity=w g ·Geometry Similarity

[0057] +w t ·Texture Similarity;

[0058] wherein, w g and w t are weight coefficients of the geometric features and the surface texture information, Geometry Similarity is the geometric feature similarity, and Texture Similarity is the surface texture information similarity.

[0059] S55: When the comprehensive similarity score is lower than the preset threshold, marking as a potential abnormal and defect region based on the preset threshold.

[0060] Optionally, in the method for visual detection and quality evaluation of the small structure of the aircraft as described above, the S6 comprises:

[0061] S61: The pre-trained classification model is a classification model based on an extreme learning machine.

[0062] S62: Using the geometric features and the surface texture information in the historical detection data and the standard sample as training data, input weights and bias are randomly generated by the extreme learning machine, and output weights are determined, so as to complete the training of the classification model.

[0063] S63: The comprehensive similarity score after the multi-modal data fusion in S5 is input into the trained extreme learning machine classification model, and is analyzed and processed to generate a classification result.

[0064] S64: According to the output result of the classification model, the category to which the input data belongs is determined, including normal, surface defect, and geometric deformation. The extreme learning machine calculates the classification probability of the input data, maps it to the predefined abnormal category, and outputs the final abnormal type and abnormal feature point.

[0065] S65: The position of the abnormality is determined by analyzing the abnormal feature point output in the classification result, including: in the three-dimensional model, according to the spatial coordinates of the abnormal feature point, the specific position of the abnormal feature point on the small structure of the aircraft is located in combination with the geometric features and the surface texture information, and corresponding marking information is generated.

[0066] Optionally, in the method for visual detection and quality evaluation of the small structure of the aircraft as described above, the S65 specifically comprises:

[0067] S651: Spatial coordinate determination, extract the spatial coordinates (x i , y i , z i ) of the abnormal feature points o identified by the classification model, and record the specific positions of the abnormal feature points in the three-dimensional model;

[0068] S652: Geometric feature combination, according to the spatial coordinates of the abnormal feature points, use geometric features to locate the abnormal feature points, including calculating the distance between the abnormal feature points and the surrounding geometric feature points;

[0069] S653: Coordinate determination of two-dimensional graph, map the spatial coordinates of the abnormal feature points to the texture coordinates (u d , v d ) of the two-dimensional image, and use the perspective projection formula for conversion, the expression is:

[0070]

[0071] Where (F x , F y ) is the focal length of the camera, and (C x , C y ) is the principal point coordinate of the camera;

[0072] S654: Surface texture information combination, use the texture coordinates (u d , v d ) to extract the corresponding texture information in the two-dimensional image, calculate the similarity between the abnormal feature points o and the surrounding texture feature points, and the calculation formula is:

[0073]

[0074] Where S oe is the similarity between the abnormal feature points o and the texture feature points e, T f (k) and T f (k) are the e-th texture value of the abnormal feature points o and the texture feature points e, respectively, and n4 is the number of texture feature points;

[0075] S655: Mark abnormal feature points, including: combine geometric features and surface texture information, and determine the final marking value of the abnormal feature points i by weighted average of the geometric distance and texture similarity of the abnormal feature points, the calculation formula is:

[0076] M o = W h ·min k (d op )+W t ·min k (S oe );

[0077] wherein M o is the final marking value of the abnormal feature point o, W h and W t are the weight coefficients of the geometric feature and the surface texture information, min k (d op ) is the minimum distance between the abnormal feature point o and the geometric feature point p, min k (S oe ) is the minimum similarity between the abnormal feature point and the texture feature point;

[0078] S656: generating marking information, converting the final determined marking value M o into description information of the abnormal type and position, and marking on the three-dimensional model and the two-dimensional image to generate marking information including the spatial coordinates of the abnormal feature point o, the abnormal type, and the abnormal degree; the marking format is: marking information = {coordinates: (x c , y c , z c ), type: abnormal type, degree: M o}.

[0079] Optionally, in the method for visual detection and quality evaluation of the small structure of the aircraft as described above, the S7 comprises:

[0080] S71: summarizing the abnormal detection results, collecting and sorting the abnormal types and abnormal positions output by the classification model in S6, including the spatial coordinates of each abnormal feature point, the abnormal type, and the abnormal degree, to form a preliminary detection result data set;

[0081] S72: comprehensively analyzing the detected abnormal data, including calculating the data, distribution position, and type proportion of the abnormal feature points, specifically using a statistical analysis method to evaluate the severity of each type of abnormality and its distribution in the whole aircraft small structure, and the statistical analysis formula is:

[0082]

[0083] wherein P(type) is the severity of the corresponding type of abnormality, n5 is the number of abnormal feature points of the corresponding type, and n6 is the total number of all abnormal feature points;

[0084] S73: formulating a comprehensive quality evaluation standard according to the industry standard and the production requirement, and determining the geometric tolerance, surface quality, and defect allowable range;

[0085] S74: comparing the severity of each abnormal point with the evaluation standard, and calculating the score of each abnormal feature point; wherein for each abnormal type, the score of the abnormal feature point is calculated using the following scoring formula:

[0086]

[0087] wherein q r is the rth quality evaluation index score, and the maximum allowed value is the pre-set quality evaluation standard;

[0088] S75: Perform weighted average on the scores q r of all abnormal points to calculate the comprehensive quality evaluation score of the small structure of the aircraft, and the quality evaluation standard formula used is:

[0089]

[0090] wherein Q is the comprehensive quality evaluation score, q r is the rth quality evaluation index score, w is the weight coefficient of the index, and m1 is the number of evaluation indexes;

[0091] S76: Quality evaluation report generation, according to the comprehensive quality evaluation standard, quantitatively score and grade the detection results, generate a quality evaluation report, and the report content includes the basic information of the detection object, the detection method, the detection result, the comprehensive evaluation score, the quality grade and the specific abnormal description;

[0092] S77: Based on the quality evaluation report, propose optimization suggestions for the detection results, including improving the design, optimizing the manufacturing process and strengthening the quality control.

[0093] The beneficial effects of the present application are as follows:

[0094] On the one hand, by fusing three-dimensional scanning and high-resolution two-dimensional image technology, comprehensive and accurate detection of small structure can be achieved, this multi-modal fusion not only enhances the recognition ability of surface and internal structure abnormalities of small structure, but also through detailed surface texture analysis, can accurately locate the micro-defects and potential structure problems, in addition, the method also improves the image quality through high dynamic range image and advanced image processing technology, ensures the accuracy and reliability of the detection data.

[0095] On the other hand, by using a classification model based on machine learning, not only the efficiency and accuracy of abnormal detection are improved, but also the new detection situation can be automatically learned and adapted based on historical data, this intelligent analysis method reduces the demand for human operation, reduces the detection cost, and improves the automation level of production process, the generation of comprehensive quality evaluation report and the proposal of optimization suggestions further help enterprises to monitor and improve production process, so as to improve product quality while ensuring aviation safety. BRIEF DESCRIPTION OF DRAWINGS

[0096] The accompanying drawings are used to provide a further understanding of the technical solutions of the present application, and constitute a part of the specification, and are used to explain the technical solutions of the present application together with the embodiments of the present application, and do not constitute a limitation on the technical solutions of the present application.

[0097] Figure 1 A flowchart of a visual detection and quality evaluation method for small aircraft structural parts provided by an embodiment of the present application is shown in the figure.

[0098] Figure 2 A flowchart of acquiring three-dimensional point cloud data in the visual detection and quality evaluation method for small aircraft structural parts provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0099] To make the objectives, technical solutions and advantages of the present application clearer, the embodiments of the present application will be described in detail below with reference to the accompanying drawings. It should be explained that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other at will.

[0100] As explained in the above background, the importance of small aircraft structural parts in aircraft development and the necessity of quality detection of the small aircraft structural parts have been explained. In view of the existing detection methods for small aircraft structural parts, there are obvious problems of limitation in detection accuracy and repeatability, and it is difficult to accurately identify small defects and internal structure abnormalities in three-dimensional geometry and fine surface texture.

[0101] The existing detection methods cannot effectively integrate three-dimensional geometric information and two-dimensional image texture information, resulting in lack of depth and accuracy in comprehensive quality evaluation of small aircraft structural parts. This technical limitation is particularly prominent in the detection process of small defects on the surface of small structural parts and internal structure abnormalities, for example, it is difficult to accurately capture small cracks or depressions on the complex curved surface of small structural parts using only two-dimensional image technology, and it is also difficult to effectively evaluate the details and texture of the surface material by relying only on three-dimensional scanning data. In addition, the low efficiency of the existing detection methods in data processing and abnormal positioning also seriously affects the production efficiency and cost control.

[0102] In view of the above problems, an embodiment of the present application provides a visual detection and quality evaluation method for small aircraft structural parts, which develops a new technology capable of efficiently integrating three-dimensional scanning and two-dimensional image data, and automatically performing depth analysis and quality evaluation. This has important practical significance for improving the manufacturing and detection standards of small aircraft structural parts and ensuring aviation safety.

[0103] The following specific embodiments provided by the present application can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0104] The accompanying drawings are used to provide a further understanding of the technical solutions of the present application, and constitute a part of the specification, and are used to explain the technical solutions of the present application together with the embodiments of the present application, and do not constitute a limitation on the technical solutions of the present application.Figure 1 The flowchart of the method for visual inspection and quality evaluation of the small structure of the aircraft provided by the embodiment of the present application is shown in FIG. 1. Figure 1 As shown in the figure, the method for visual inspection and quality evaluation of the small structure of the aircraft provided by the embodiment of the present application comprises the following steps:

[0105] S1: A three-dimensional scanner is used to scan the small structure of the aircraft in all directions to obtain three-dimensional point cloud data of the outer surface of the small structure of the aircraft, and the three-dimensional point cloud data is preprocessed;

[0106] S2: A high-resolution camera is used to collect two-dimensional images of the small structure of the aircraft from multiple angles;

[0107] S3: The three-dimensional model is constructed using the preprocessed three-dimensional point cloud data, and the texture of the two-dimensional image obtained in S2 is mapped to the surface of the three-dimensional model to generate a three-dimensional model including surface texture information;

[0108] S4: A three-dimensional feature extraction algorithm is used to extract geometric features and surface texture information from the three-dimensional model, and the extraction content includes the shape, size, surface texture information and subtle defects of the small structure;

[0109] S5: A multi-modal data fusion method is used to fuse and compare the geometric features and surface texture information extracted in S4 with standard samples to identify potential abnormalities and defects;

[0110] S6: A pre-trained classification model is used to analyze the fused comparison data in S5 to output an abnormality detection result, including the abnormality type and abnormality position of the small structure of the aircraft;

[0111] S7: The abnormality detection result of S6 is comprehensively quality evaluated to generate a quality evaluation report and propose optimization suggestions based on the detection result.

[0112] In one implementation manner of the embodiment of the present application, the three-dimensional scanner for scanning can be pre-positioned in the working area of the small structure of the aircraft; specifically, the three-dimensional scanner for scanning is pre-positioned in the working area of the small structure of the aircraft, and the three-dimensional scanner is initially calibrated, including calibrating the position, angle and focal length of the scanner, so that the scanning area of the calibrated three-dimensional scanner covers the entire outer surface of the small structure of the aircraft. Figure 2 The flowchart of the method for visual inspection and quality evaluation of the small structure of the aircraft provided by the embodiment of the present application is shown in FIG. 1.

[0113] S11: initial calibration of the three-dimensional scanner, including calibration of the position, angle and focal length of the scanner, so that the scanning area of the calibrated three-dimensional scanner covers the entire outer surface of the small aircraft structure;

[0114] S12: controlling the three-dimensional scanner to scan the small aircraft structure from multiple perspectives according to a predetermined path and angle, and obtaining three-dimensional point cloud data corresponding to each perspective;

[0115] S13: splicing and fusing the three-dimensional point cloud data corresponding to each perspective obtained in S12;

[0116] In this step, the Iterative Closest Point (ICP) algorithm can be used to align multiple sets of three-dimensional point cloud data to generate a three-dimensional point cloud data set;

[0117] S14: denoising the spliced three-dimensional point cloud data set, using a statistical filtering algorithm to remove noise points in the three-dimensional point cloud data set, the statistical filtering algorithm identifies and removes abnormal points by calculating the neighborhood statistical characteristics of each point;

[0118] S15: identifying and removing redundant points and isolated points in the three-dimensional point cloud data set by using a density clustering algorithm (such as the DBSCAN algorithm);

[0119] In this step, the density clustering algorithm detects and removes redundant points with insufficient density by defining a density threshold and a minimum number of neighbors;

[0120] S16: using a global least squares fitting algorithm to globally align the three-dimensional point cloud data set after denoising and redundancy removal, to ensure the overall consistency and accuracy of the three-dimensional point cloud data, the global least squares fitting algorithm minimizes the squared error between the point clouds to achieve global alignment.

[0121] In this step, the specific steps of global alignment are as follows:

[0122] Define error function: the global least squares fitting algorithm first defines an error function that quantifies the difference between two sets of point cloud data, and selects a squared error function to quantify the Euclidean distance between the point clouds when performing global alignment; let the error function be E, the expression is: Where p i is a point in the source point cloud, q i is the corresponding point in the target point cloud, T is a transformation function including translation, rotation, etc.

[0123] Finding the optimal transformation: the goal of the algorithm is to find a transformation T that minimizes the error function E, which involves solving for the rotation matrix R and the translation vector t. To minimize the error function, the derivatives of R and t are taken and set to zero. Solving this optimization problem requires iterative methods such as the Iterative Closest Point algorithm or nonlinear minimization techniques;

[0124] Iterative solution: in practical applications, to improve the alignment accuracy, iterative methods are used to gradually optimize R and t. In each iteration, the position of the point cloud is updated based on the current transformation estimate, and then the error function is recalculated and the transformation parameters are updated;

[0125] Finally: the iteration continues until the transformation converges or the preset number of iterations is reached. The final R and t will be used to align all the point cloud data obtained from scanning to a global coordinate.

[0126] In this implementation, by performing the above S11 to S16, the all-around high-precision scanning of the small aircraft structure can be realized, and the acquired three-dimensional point cloud data can be comprehensively preprocessed. The specific implementation of the steps of positioning and calibration, multi-view scanning, point cloud data splicing and fusion, denoising processing, redundant point removal and point cloud data alignment processing, etc. ensures the high quality and high precision of the point cloud data, and provides a reliable data basis for subsequent three-dimensional model reconstruction and feature extraction.

[0127] In an implementation manner of the embodiment of the application, a high-resolution camera can be pre-positioned in the working area of the small aircraft structure; the implementation process of S2 can include:

[0128] S21: initial calibration of the high-resolution camera pre-positioned in the working area of the small aircraft structure, including calibration of the position, angle and focal length of the camera, so that the camera covers all surface areas of the small structure;

[0129] In this step, the camera covers all surface areas of the small structure, and the camera should be 10-30 cm away from the small structure, and the angle should be adjusted to 45 degrees to 90 degrees with the normal of the small structure surface to ensure the best imaging effect.

[0130] S22: design a multi-angle shooting path to determine the specific shooting points of the camera at different positions and angles to ensure that all surface details of the small structure can be captured comprehensively. The shooting path includes at least 8 positions at different angles to ensure uniform spacing between each angle and cover all surfaces of the small structure;

[0131] S23: set the image acquisition parameters of the camera, including exposure time, aperture size and sensitivity;

[0132] This step requires ensuring that clear and detailed images can be obtained under different lighting conditions. The specific parameters can be set as follows: exposure time 1 / 100s to 1 / 1000s, aperture size f / 2.8 to f / 8, and sensitivity ISO 100 to ISO 800.

[0133] S24: According to the planned shooting path, control the camera to capture multi-angle images of the small structure of the aircraft at predetermined positions and angles, and obtain full two-dimensional image data of the small structure of the aircraft. At least 3 images are taken at each position and each angle to ensure that the obtained images are comprehensive and clear.

[0134] S25: Use image distortion correction algorithm (such as distortion correction algorithm based on calibration board) to correct the distortion in the collected multi-position and multi-angle images, and perform color correction and contrast enhancement to improve the quality and consistency of the images.

[0135] S26: Stitch and fuse the two-dimensional images collected at multiple positions and multiple angles, and use image stitching algorithm (such as SIFT feature point matching algorithm) to seamlessly stitch the images into a complete image covering the entire surface of the small structure. The overlapping part between images in the stitching process is not less than 30% to ensure the quality and accuracy of the stitching.

[0136] In this implementation, through the above steps S21 to S26, full-range high-resolution image acquisition of the small structure of the aircraft can be realized, and comprehensive processing of the obtained image data can be performed. The specific implementation of the steps of positioning and calibration, multi-angle shooting path planning, image acquisition parameter setting, multi-angle image acquisition, image correction and processing, and image stitching and fusion ensures the high quality and high precision of the image data, providing a reliable image basis for subsequent three-dimensional model reconstruction and feature extraction.

[0137] In one implementation of an embodiment of the present application, the implementation process of S3 can include:

[0138] S31: Use the pre-processed three-dimensional point cloud data obtained in S1 to construct a preliminary three-dimensional model by Delaunay triangulation algorithm to generate a mesh model containing the geometric shape of the small structure of the aircraft.

[0139] S32: Register the two-dimensional image obtained in S2 with the preliminary three-dimensional model. This step includes: using SIFT algorithm to identify corresponding feature points in the two-dimensional image and the three-dimensional model, and aligning the feature points through the camera projection matrix.

[0140] The SIFT feature point detection formula in this step is:

[0141] D(x, σ) = (L(x, kσ) - L(x, σ)) · G(x, σ).

[0142] wherein x represents a point in the image, σ represents a scale parameter, L represents an image scale space, k is a scale factor, D(x, σ) represents a difference of feature points in the scale space, L(x, σ) represents a Gaussian blur result of the image at scale σ, and G(x, σ) represents a Gaussian kernel function at scale σ.

[0143] S33: Texture mapping is performed on the registered two-dimensional images, and surface texture information in the two-dimensional images is mapped to the outer surface of the three-dimensional model; wherein a perspective projection texture mapping algorithm is used for texture mapping, and texture coordinates corresponding to each two-dimensional image pixel are projected onto the surface of the three-dimensional model; the perspective projection formula in this step is:

[0144]

[0145] wherein u and v represent coordinates in the two-dimensional image, x, y, and z represent coordinates in the three-dimensional model, and f is a focal length.

[0146] S34: Texture fusion is performed on the two-dimensional images obtained at multiple positions and angles, a weighted average fusion algorithm is used to fuse the texture in the overlapping regions, and in the fusion process, weighted average processing is performed on each pixel in the overlapping regions.

[0147] The formula for weighted average processing in this step is:

[0148] wherein I f represents a pixel value after fusion, I i represents a pixel value of the i-th image, w i represents a weight of the i-th image, and n represents the number of images.

[0149] S35: Detail enhancement is performed on the three-dimensional model after texture mapping and fusion, and specifically, a high dynamic range (HDR) enhancement algorithm is used to adjust the brightness and contrast of the texture, so as to ensure the details and realism of the texture.

[0150] The pixel value I e after enhancement in this step is obtained by logarithmic transformation and contrast adjustment from the original pixel value I o , and the calculation formula is:

[0151] I e = log(1+I o )·α;

[0152] wherein I e represents a pixel value after enhancement, I o represents an original pixel value, α represents a contrast adjustment coefficient, and log represents a natural logarithm function.

[0153] In the implementation, by performing the steps S31-S35, the two-dimensional image texture can be accurately mapped to the three-dimensional model surface, the high-precision three-dimensional model containing surface information is generated, the specific implementation of the steps of Delaunay triangulation, SIFT algorithm for image registration, perspective projection texture mapping, multi-image weighted average fusion, HDR detail enhancement, etc. is ensured, the high quality and high precision of the three-dimensional model are ensured, and reliable data foundation is provided for subsequent feature extraction and quality evaluation. The method not only improves the precision of three-dimensional reconstruction, but also significantly improves the details and realism of the model.

[0154] In an implementation of the embodiment of the present application, the implementation process of S4 can include:

[0155] S41: extracting geometric features from the three-dimensional model, including: using a surface fitting algorithm to fit each surface of the three-dimensional model to obtain the geometric shape of the small aircraft structure, the surface fitting algorithm being to minimize the error between the fitting surface and the three-dimensional point cloud data by least squares method, and the calculation formula being:

[0156]

[0157] wherein z i is a point in the point cloud data, f(x i , y i ) is a fitting surface function, (x i , y i ) is a coordinate of the point cloud data, and n1 is the number of points;

[0158] S42: measuring the size of the three-dimensional model, including: using a distance calculation algorithm to measure the size of each part of the small aircraft structure, the distance calculation algorithm being used to calculate the Euclidean distance between two points in the three-dimensional model, and the calculation formula being:

[0159]

[0160] wherein (x1, y1, z1) and (x2, y2, z2) are coordinates of two points in the three-dimensional model, and d is the distance between the two points;

[0161] S43: extracting surface texture information of the three-dimensional model, including: using a normal vector analysis algorithm to calculate the normal vector of each point of the three-dimensional model, the normal vector being defined as a vector perpendicular to the surface, and the calculation formula being:

[0162]

[0163] wherein N is the normal vector, v1 and v2 are two non-parallel vectors on the surface, × represents vector cross product, and |·| represents the length of the vector;

[0164] S44: Identifying subtle defects in the three-dimensional model, including: using Gaussian curvature and mean curvature analysis algorithm to calculate the curvature of the surface of the three-dimensional model, identifying the subtle defects of the concave, convex and crack on the surface of the small structure of the aircraft.

[0165] In this implementation, by performing the above steps S41 to S44, the geometric features and surface texture features of the small structure of the aircraft can be accurately extracted from the three-dimensional model, including the shape, size, surface texture and subtle defects of the small structure of the aircraft. The specific implementation of the steps such as surface fitting algorithm, distance calculation algorithm, normal vector analysis algorithm, Gaussian curvature and mean curvature analysis algorithm ensures high quality and high precision of feature extraction, and provides a reliable data basis for subsequent multi-modal data fusion and quality assessment.

[0166] Further, in an embodiment of the implementation, the Gaussian curvature and mean curvature analysis algorithm in S44 above is respectively:

[0167]

[0168] Wherein, κ1 and κ2 are principal curvatures, and are the partial derivatives of the surface in x and y directions, K is the Gaussian curvature, and H is the mean curvature.

[0169] In an implementation of an embodiment of the present application, the implementation process of S5 above can include:

[0170] S51: Obtain the geometric features and surface texture information of the standard sample, including the shape, size and surface texture information of the small structure of the aircraft, and use it as the reference data;

[0171] S52: Compare and analyze the geometric features extracted in S4 with the geometric features of the standard sample, and use a shape matching algorithm to evaluate the consistency of the geometric features by calculating the similarity between the two, and the formula of the shape matching algorithm is:

[0172]

[0173] Wherein, A and B represent the geometric feature point sets of the small structure of the aircraft and the standard sample respectively, n2 is the number of geometric feature points, and ∥A a -B b ∥ is the Euclidean distance between the geometric feature points A a and B b of the small structure of the aircraft and the standard sample;

[0174] S53: Compare the surface texture information extracted in S4 with the surface texture information of the standard sample, adopt a texture similarity algorithm, and evaluate the consistency of the surface texture information by calculating the similarity between the surface texture information, and the formula of the texture similarity algorithm is:

[0175]

[0176] wherein T1 and T2 respectively represent the texture images of the small aircraft structure and the standard sample, m is the height of the texture image, n3 is the width of the texture image, and |T1(v, c)-T2(v, c)| is the absolute difference value between the corresponding pixel points of the small aircraft structure and the standard sample;

[0177] S54: Fuse the geometric features and the surface texture information, adopt a weighted fusion algorithm, combine the geometric feature similarity and the texture information similarity to form a comprehensive similarity score;

[0178] The formula of the weighted fusion algorithm in this step is:

[0179] Fused Similarity=w g ·Geometry Similarity+w t ·Texture Similarity;

[0180] wherein w g and w t are the weight coefficients of the geometric features and the surface texture information, Geometry Similarity is the geometric feature similarity, and Texture Similarity is the surface texture information similarity;

[0181] S55: Based on the preset setting threshold, when the comprehensive similarity score is lower than the setting threshold, mark as a potential abnormal and defect area.

[0182] In this implementation, by executing the above steps S51 to S55, the fusion and comparison analysis of the geometric features and the surface texture information extracted from S4 and the standard sample can be realized, the specific implementation of the shape matching algorithm, the texture similarity algorithm, and the multi-modal data fusion algorithm, etc. steps ensure the high quality and high precision of the abnormal and defect identification, and provide a reliable data basis for the subsequent quality evaluation and optimization suggestion.

[0183] In an implementation of an embodiment of the present application, the implementation process of S6 described above can include:

[0184] S61: The pre-trained classification model is an Extreme Learning Machine (ELM) based classification model, which has fast learning and good generalization performance, and is suitable for processing high-dimensional feature data after multi-modal fusion;

[0185] S62: Use the geometric features and surface texture information in the historical detection data and standard samples as training data, randomly generate input weights and biases by Extreme Learning Machine, and determine output weights to complete the classification model training;

[0186] The specific training process in this step includes data preprocessing, feature standardization and classification model parameter optimization to ensure the high precision and stability of the classification model.

[0187] S63: Input the comprehensive similarity score of the multi-modal data fusion in S5 into the trained Extreme Learning Machine classification model for analysis and processing. After the model receives the input data, it undergoes nonlinear transformation of the hidden layer nodes and linear combination of the output layer to generate the classification result;

[0188] S64: According to the output result of the classification model, determine the category to which the input data (geometric features and surface texture information) belongs, including normal, surface defect and geometric deformation; the Extreme Learning Machine calculates the classification probability of the input data and maps it to the predefined abnormal category to output the final abnormal type and abnormal feature point;

[0189] S65: Determine the position of the abnormality by analyzing the abnormal feature points output in the classification result, including: in the three-dimensional model, according to the spatial coordinates of the abnormal feature points, combined with the geometric features and surface texture information, locate the specific position of the abnormal feature points on the small structure of the aircraft and generate corresponding marking information.

[0190] In this implementation, by executing the above steps S61 to S65, the pre-trained Extreme Learning Machine based classification model is used to analyze the fused comparison data in S5, which can quickly and accurately output the specific abnormal type and position. The specific implementation of classification model selection, model training, feature data input, abnormal type identification and abnormal position positioning ensures the high quality and high precision of abnormal detection and positioning, and provides a reliable basis for subsequent comprehensive quality evaluation and optimization suggestions.

[0191] Further, in one embodiment of the implementation, the implementation process of S65 above specifically includes:

[0192] S651: Spatial coordinate determination, determine the spatial coordinates (x i , y i , z i) extraction, record the specific location of abnormal feature points in the three-dimensional model;

[0193] S652: geometric feature combination, according to the spatial coordinates of the abnormal feature points, the abnormal feature points are positioned by using the geometric features, which specifically includes calculating the distance between the abnormal feature points and the surrounding geometric feature points, and the calculation formula is:

[0194]

[0195] Where d op is the Euclidean distance between the abnormal point o and the geometric feature point p, (x o , y o , z o ) and (x p , y p , z p ) are the spatial coordinates of the abnormal point and the geometric feature point, respectively;

[0196] S653: two-dimensional graph coordinate determination, the spatial coordinates of the abnormal feature points are mapped to the texture coordinates (u d , v d ) of the two-dimensional image, and the conversion is performed by using the perspective projection formula, and the expression is:

[0197]

[0198] Where (F x , F y ) is the focal length of the camera, and (C x , C y ) is the principal point coordinate of the camera;

[0199] S654: surface texture information combination, the corresponding texture information is extracted in the two-dimensional image by using the texture coordinates (u d , v d ), and the similarity between the abnormal feature point o and the surrounding texture feature points is calculated, and the calculation formula is:

[0200]

[0201] Where S oe is the similarity between the abnormal feature point o and the texture feature point e, T f (k) and T f (k) are the e-th texture value of the abnormal feature point o and the texture feature point e, respectively, and n4 is the number of texture feature points;

[0202] S655: marking abnormal feature points, including: combining the geometric features and the surface texture information, determining the final marking value of the abnormal feature point i by weighted average of the geometric distance and the texture similarity of the abnormal feature point, and the calculation formula is:

[0203] M o = W h ·min k (d op ) + W t ·min k (S oe );

[0204] Wherein, M o is the final marking value of the abnormal feature point o, W h and W t are weight coefficients of geometric features and surface texture information respectively, min k (d op ) is the minimum distance between the abnormal feature point o and the geometric feature point p, min k (S oe ) is the minimum similarity between the abnormal feature point and the texture feature point;

[0205] S656: Generate marking information, convert the final determined marking value M o into description information of abnormal type and position, and mark on the three-dimensional model and the two-dimensional image to generate marking information including the spatial coordinates of the abnormal feature point o, the abnormal type and the abnormal degree; the marking format is: marking information = {coordinates: (x c , y c , z c ), type: abnormal type, degree: M o}.

[0206] In this embodiment, by performing the above steps S651 to S656, the specific position of the abnormal point on the small structural part can be accurately positioned according to the spatial coordinates of the abnormal feature point, combined with the geometric features and the surface texture features, and the corresponding marking information is generated. The specific implementation of the steps of geometric distance calculation, perspective projection conversion, texture similarity calculation, weighted average marking, etc. ensures the high quality and high precision of the abnormal point positioning and marking, and provides a reliable data basis for comprehensive quality evaluation.

[0207] In an implementation manner of the embodiment of the application, the implementation process of S7 can include:

[0208] S71: Abnormal detection result is summarized, the abnormal type and the abnormal position output by the classification model in S6 are collected and arranged, including the spatial coordinates of each abnormal feature point, the abnormal type, the abnormal degree, to form a preliminary detection result data set;

[0209] S72: Comprehensive analysis of detected abnormal data, including calculation of abnormal feature point data, distribution position and type proportion, specific statistical analysis method is adopted to evaluate the severity of each type of abnormality and its distribution in the overall aircraft small structure, the statistical analysis formula is:

[0210]

[0211] Wherein, P(type) is the severity of the corresponding type of abnormality, n5 is the number of abnormal feature points of the corresponding type, n6 is the total number of all abnormal feature points;

[0212] S73: According to industry standards and production requirements, formulate comprehensive quality evaluation standard, determine the allowable range of geometric tolerance, surface quality and defect;

[0213] S74: Compare the severity of each abnormal point with the evaluation standard, and calculate the score of each abnormal feature point; wherein, for each abnormal type, the following scoring formula is used to calculate the score of abnormal feature point:

[0214]

[0215] Wherein, q r rth quality evaluation index score, the maximum allowable value is the pre-set quality evaluation standard;

[0216] S75: Weighted average of the scores of all abnormal points q r The comprehensive quality evaluation score of the aircraft small structure is calculated, and the quality evaluation standard formula used is:

[0217]

[0218] Wherein, Q is the comprehensive quality evaluation score, q r rth quality evaluation index score, w is the weight coefficient of the index, m1 is the number of evaluation indexes;

[0219] S76: Quality evaluation report generation, according to the comprehensive quality evaluation standard, the detection results are quantitatively scored and graded, and the quality evaluation report is generated, the report content includes the basic information of the detection object, the detection method, the detection result, the comprehensive evaluation score, the quality grade and the specific abnormality description.

[0220] In this step, the format of the quality evaluation report is as follows:

[0221] Detection object: aircraft small structure number;

[0222] Detection method: combination of three-dimensional scanning and visual detection;

[0223] Detection results: including detailed information of various abnormal points;

[0224] Comprehensive evaluation score: Q;

[0225] Quantity level: divided into different levels according to Q;

[0226] Abnormal description: specific abnormal type, location and degree.

[0227] S77: Based on the quality evaluation report, optimization suggestions for the detection results are proposed, including improving design, optimizing manufacturing process and strengthening quality control.

[0228] In this step, the optimization suggestion making process includes, for example:

[0229] S771, identify the main abnormal type and location, analyze the possible causes and effects according to the main abnormal points in the detection results;

[0230] S772, develop targeted measures such as adjusting design parameters, improving manufacturing process, strengthening material detection, etc;

[0231] S773, propose specific optimization suggestions and improvement schemes, including implementation steps, expected effects and precautions;

[0232] In this implementation, by executing the above steps S71 to S75, the comprehensive quality evaluation can be carried out according to the abnormal detection results of S6, the detailed quality evaluation report can be generated, and the optimization suggestions based on the detection results can be proposed. The specific implementation of the steps of abnormal detection result summary, comprehensive analysis, quality evaluation standard formulation, quality evaluation report generation and optimization suggestion formulation ensures the high quality and high precision of quality evaluation and optimization suggestion, and provides a reliable basis for improving the manufacturing quality and reliability of small structural parts of aircraft.

[0233] The visual detection and quality evaluation method for small structural parts of aircraft provided by the embodiment of the application has the following beneficial effects:

[0234] On the one hand, by fusing three-dimensional scanning and high-resolution two-dimensional image technology, comprehensive and accurate detection of small structural parts can be realized. This multi-modal fusion not only enhances the recognition ability of surface and internal structure abnormalities of small structural parts, but also accurately locates micro-defects and potential structural problems through detailed surface texture analysis. In addition, the method also significantly improves the image quality through high dynamic range image and advanced image processing technology, ensuring the accuracy and reliability of the detection data.

[0235] On the other hand, by adopting a machine learning-based classification model, not only the efficiency and accuracy of anomaly detection are improved, but also new detection situations can be automatically learned and adapted based on historical data. This intelligent analysis method reduces the need for manual operation, reduces detection costs, and improves the automation level of the production process. The generation of comprehensive quality evaluation reports and the proposal of optimization suggestions further help enterprises to monitor and improve the production process, thereby improving product quality while ensuring aviation safety.

[0236] The present application encompasses any substitutions, modifications, equivalent methods and solutions made on the essence and scope of the present application. In order to make the public have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can be fully understood without these details by those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.

[0237] Although the embodiments of the present application are disclosed as above, the content is only the embodiment adopted for the convenience of understanding the present application, and is not used to limit the present application. Any person skilled in the art of the present application can make any modification and change in the form and details without departing from the spirit and scope of the present application disclosed, but the patent protection scope of the present application shall be subject to the scope defined by the appended claims.

Claims

1. A method for visual inspection and quality assessment of small aircraft structures, characterized in that, Comprise; S1: using a three-dimensional scanner to scan the aircraft small structure in all directions, obtaining the three-dimensional point cloud data of the outer surface of the aircraft small structure, and preprocessing the three-dimensional point cloud data; S2: using a high-resolution camera to collect two-dimensional images of the aircraft small structure from multiple angles; S3: using the preprocessed three-dimensional point cloud data to construct a three-dimensional model, and mapping the texture of the two-dimensional images obtained in S2 to the surface of the three-dimensional model to generate a three-dimensional model including surface texture information; S4: using a three-dimensional feature extraction algorithm to extract geometric features and surface texture information from the three-dimensional model, including the shape, size, surface texture information and subtle defects of the small structure; S5: using a multi-modal data fusion method, the geometric features and surface texture information extracted in S4 are fused and compared with the standard sample to identify potential abnormalities and defects; S6: using a pre-trained classification model to analyze the fused comparison data in S5, outputting an abnormality detection result, including the abnormal type and location of the aircraft small structure; S7: comprehensive quality evaluation of the abnormality detection result of S6, generating a quality evaluation report, and putting forward optimization suggestions based on the detection result; Wherein, the S5 comprises: S51: obtaining the geometric features and surface texture information of the standard sample, including the shape, size, surface texture information of the aircraft small structure, which is used as the reference data; S52: comparing the geometric features extracted in S4 with the geometric features of the standard sample, and using a shape matching algorithm to evaluate the consistency of the geometric features by calculating the similarity between the two, the shape matching algorithm formula is: ； wherein, and respectively denote the sets of geometric feature points of the aircraft small structure and the standard sample, is the number of geometric feature points, is the Euclidean distance between the geometric feature points of the aircraft small structure and the standard sample, and respectively denote the sets of geometric feature points of the aircraft small structure and the standard sample, S53: comparing the surface texture information extracted in S4 with the surface texture information of the standard sample, using a texture similarity algorithm to evaluate the consistency of the surface texture information by calculating the similarity between the two, the texture similarity algorithm formula is: ； wherein, and respectively represent the texture images of the small structure of the airplane and the standard sample, is the height of the texture image, is the width of the texture image, is the absolute difference between the corresponding pixel points of the small structure of the airplane and the standard sample; S54: fuse the geometric features and surface texture information, use weighted fusion algorithm, combine the geometric feature similarity and texture information similarity to form a comprehensive similarity score, the weighted fusion algorithm formula is: Fused Similarity Geometry Similarity Texture Similarity; wherein, and are weight coefficients for the geometric feature and the surface texture information, respectively, Geometry Similarity is the geometric feature similarity, and Texture Similarity is the surface texture information similarity. S55: based on the preset threshold, when the comprehensive similarity score is lower than the set threshold, it is marked as a potential abnormal and defect area.

2. The method for visual inspection and quality assessment of small aircraft structures according to claim 1, characterized in that, The three-dimensional scanner for scanning is pre-positioned in the working area of the aircraft small structure, and the S1 comprises: S11: initial calibration of the three-dimensional scanner, including calibration of the position, angle and focal length of the scanner, so that the scanning area of the calibrated three-dimensional scanner covers the whole outer surface of the aircraft small structure; S12: control the three-dimensional scanner to scan the aircraft small structure from multiple angles according to the predetermined path and angle, and obtain the three-dimensional point cloud data corresponding to each angle; S13: splice and fuse the three-dimensional point cloud data corresponding to each angle obtained in S12, and use the iterative closest point algorithm ICP in S13 to align multiple sets of three-dimensional point cloud data to generate a three-dimensional point cloud data set; S14: Denoising the spliced three-dimensional point cloud dataset, using a statistical filtering algorithm to remove noise points in the three-dimensional point cloud dataset, the statistical filtering algorithm identifies and removes abnormal points by calculating the neighborhood statistical characteristics of each point; S15: Identify and remove redundant points and isolated points in the three-dimensional point cloud dataset by density clustering algorithm; the density clustering algorithm detects and deletes redundant points with insufficient density by defining a density threshold and a minimum number of neighbors; S16: Global alignment of the three-dimensional point cloud dataset after denoising and redundancy removal using global least squares fitting algorithm.

3. The method for visual inspection and quality assessment of small aircraft structures according to claim 1, characterized in that, The high-resolution camera is pre-positioned in the working area of the small structure of the aircraft, and S2 includes: S21: Initial calibration of the high-resolution camera, including calibration of the position, angle and focal length of the camera, so that the camera covers all surface areas of the small structure; S22: Design a multi-angle shooting path to determine the specific shooting points of the camera at different positions and angles, and the shooting path includes at least 8 positions at different angles; S23: Set the image acquisition parameters of the camera, including exposure time, aperture size and sensitivity; S24: Control the camera to collect multi-angle images of the small structure of the aircraft at the predetermined positions and angles according to the planned shooting path, and obtain omnidirectional two-dimensional image data of the small structure of the aircraft, at least 3 images are taken at each position and each angle; S25: Correct the distortion in the image using an image distortion correction algorithm, and perform color correction and contrast enhancement on the collected multi-position and multi-angle images; S26: Splice and fuse the two-dimensional images collected at multiple positions and multiple angles, use image splicing algorithm to splice the images seamlessly into a complete image covering the entire surface of the small structure, and the overlapping part between the images in the splicing process is not less than 30%.

4. The method for visual inspection and quality assessment of small aircraft structures according to claim 1, characterized in that, S3 includes: S31: Use the preprocessed three-dimensional point cloud data obtained in S1 to construct a preliminary three-dimensional model by Delaunay triangulation algorithm to generate a grid model containing the geometric shape of the small structure of the aircraft; S32: Register the two-dimensional images obtained in S2 with the preliminary three-dimensional model, including: using SIFT algorithm to identify corresponding feature points in the two-dimensional images and the three-dimensional model, and aligning the feature points through the camera projection matrix; S33: Texture mapping of the registered two-dimensional images, mapping the surface texture information in the two-dimensional images to the outer surface of the three-dimensional model; wherein the texture mapping uses a perspective projection texture mapping algorithm to project the texture coordinates corresponding to each two-dimensional image pixel onto the surface of the three-dimensional model; S34: Texture fusion of the two-dimensional images obtained at multiple positions and multiple angles, using a weighted average fusion algorithm to fuse the textures in the overlapping areas, and during the fusion process, the pixels in each overlapping area are processed by weighted average; S35: Detail enhancement of the three-dimensional model after texture mapping and fusion, specifically using a high dynamic range enhancement algorithm to adjust the brightness and contrast of the texture.

5. The method for visual inspection and quality assessment of small aircraft structures according to claim 1, characterized in that, S4 includes: S41: Extracting geometric features from the three-dimensional model, including: using a surface fitting algorithm to fit each surface of the three-dimensional model, obtaining the geometric shape of the aircraft small structure, the surface fitting algorithm is to minimize the error between the fitting surface and the three-dimensional point cloud data by least squares method, the calculation formula is: ； wherein, is a point in the point cloud data, is a fitted surface function, is a coordinate of the point cloud data, is a number of points; S42: Dimensional measurement of the three-dimensional model, including: using distance calculation algorithm to measure the size of each part of the aircraft small structure, the distance calculation algorithm is used to calculate the Euclidean distance between two points in the three-dimensional model, the calculation formula is: ； wherein, and are coordinates of two points in a three-dimensional model, is the distance between the two points; S43: Extracting surface texture information of the three-dimensional model, including: using normal vector analysis algorithm to calculate the normal vector of each point of the three-dimensional model, the normal vector is defined as the vector perpendicular to the surface, the calculation formula is: ； wherein is a normal vector, and are two non-parallel vectors on the surface, denotes the vector cross product, denotes the modulus of a vector; S44: Identifying subtle defects in the three-dimensional model, including: using Gaussian curvature and mean curvature analysis algorithm to calculate the curvature of the surface of the three-dimensional model, identifying subtle defects such as concave, convex and crack on the surface of the aircraft small structure.

6. The method for visual inspection and quality assessment of small aircraft structures according to claim 5, characterized in that, The Gaussian curvature and mean curvature analysis algorithm in S44 are respectively: ； ； wherein and is the principal curvature, and are the partial derivatives of the surface in and directions, is the Gaussian curvature, is the mean curvature.

7. The method for visual inspection and quality assessment of small aircraft structures according to claim 1, characterized in that, The S6 includes: S61: The pre-trained classification model is a classification model based on extreme learning machine; S62: Using the geometric features and surface texture information in the historical detection data and standard samples as training data, generating input weight and bias randomly by extreme learning machine, and determining output weight, completing classification model training; S63: Input the comprehensive similarity score after multi-modal data fusion in S5 into the trained extreme learning machine classification model for analysis and processing, and generate classification results; S64: According to the output results of the classification model, determine the category to which the input data belongs, including normal, surface defect and geometric deformation, the extreme learning machine calculates the classification probability of the input data, and maps it to the pre-defined abnormal category, and outputs the final abnormal type and abnormal feature point; S65: Determine the position of the abnormality by analyzing the abnormal feature points output in the classification results, including: in the three-dimensional model, according to the spatial coordinates of the abnormal feature points, combining the geometric features and surface texture information, positioning the specific position of the abnormal feature points on the aircraft small structure, and generating corresponding marking information.

8. The method for visual inspection and quality assessment of small aircraft structures according to claim 7, characterized in that, The S65 specifically includes: S651: Space coordinate determination, space coordinates of the abnormal feature point o identified by the classification model Extraction is performed, and the specific position of the abnormal feature point in the three-dimensional model is recorded; S652: Geometric feature combination, according to the spatial coordinates of the abnormal feature points, using geometric features to locate the abnormal feature points, including calculating the distance between the abnormal feature points and the surrounding geometric feature points; S653: Coordinate determination of two-dimensional figure, mapping the spatial coordinates of the abnormal feature points to the texture coordinates of the two-dimensional image The conversion is performed by using the perspective projection formula, and the expression is: ； wherein, is the camera focal length, is the principal point coordinate of the camera; S654: Surface texture information combination, use texture coordinates In the two-dimensional image, the corresponding texture information is extracted, and the similarity of the abnormal feature point o and the surrounding texture feature points is calculated. The calculation formula is: ； wherein, is a similarity between the abnormal feature point o and the texture feature point , and are the first texture values of the abnormal feature point o and the texture feature point , is the number of the texture feature points; S655: Marking abnormal feature points, including: combining geometric features and surface texture information, determining the final marking value of abnormal feature point i by weighted average of geometric distance and texture similarity of abnormal feature points, the calculation formula is: ； wherein, is the final label value of the outlier feature point, is the final label value of the outlier feature point, and are the weight coefficients of the geometric feature and the surface texture information, respectively, is the minimum distance between the outlier feature point o and the geometric feature point p, is the minimum similarity between the outlier feature point and the texture feature point; S656: Generate label information, and determine the final label value Convert the description information of the abnormal type and position into a label information including the spatial coordinates of the abnormal feature point o, the abnormal type, and the abnormal degree; the label format is: label information Coordinates: Type: abnormal type, degree: .

9. The method for visual inspection and quality assessment of small aircraft structures according to claim 1, characterized in that, The S7 includes: S71: Summarizing the abnormal detection results, collecting and organizing the abnormal types and abnormal positions output by the classification model in S6, including the spatial coordinates of each abnormal feature point, abnormal type, abnormal degree, forming a preliminary detection result data set; S72: Comprehensive analysis of detected abnormal data, including calculation of abnormal feature point data, distribution position and type proportion, specific statistical analysis method is adopted to evaluate the severity of each type of abnormality and its distribution in the overall small structure of the aircraft, the statistical analysis formula is: ； wherein, is the severity of the corresponding type of anomaly, is the number of anomaly feature points of the corresponding type, is the total number of all anomaly feature points; S73: According to industry standards and production requirements, formulate comprehensive quality evaluation standard, determine the allowable range of geometric tolerance, surface quality and defect; S74: Compare the severity of each abnormal point with the evaluation standard, calculate the score of each abnormal feature point; for each abnormal type, the following scoring formula is used to calculate the score of the abnormal feature point: ； wherein, is the first item quality evaluation index score, the maximum allowed value is the pre-set quality evaluation standard; S75: Calculate the score of all abnormal points The integrated quality evaluation score of the small structure of the aircraft is calculated by weighted average of the parts, and the quality evaluation standard formula used is: ； wherein, is the overall quality assessment score, is the score of the quality assessment indicator, is the weight coefficient of the indicator, is the number of assessment indicators; S76: Quality evaluation report generation, according to the comprehensive quality evaluation standard, the detection result is quantitatively scored and graded, and the quality evaluation report is generated, the report content includes the basic information of the detection object, the detection method, the detection result, the comprehensive evaluation score, the quality grade and the specific abnormal description; S77: Based on the quality evaluation report, optimization suggestions are put forward for the detection results, including improving design, optimizing manufacturing process and strengthening quality control.

Images