Search

CN-122023377-A - Automatic detection method for quasi-convex quadrilateral airplane cover plate image

CN122023377ACN 122023377 ACN122023377 ACN 122023377ACN-122023377-A

Abstract

The invention discloses an automatic detection method for a quasi-convex quadrilateral airplane cover plate image, which breaks through the dependence of a traditional algorithm on a rigid straight line by defining the concept and the judgment standard of the quasi-convex quadrilateral, realizes contour noise reduction by combining Fourier transformation, eliminates noise interference while retaining core characteristics, identifies quasi-straight line segments and fits based on curvature analysis to construct control points along with quadrilateral acquisition, corrects distortion by perspective transformation, and finally completes cover plate detection by combining template matching to form a complete technical link of contour extraction-noise reduction-judgment-segmentation fitting-distortion correction-matching detection. The invention can realize the rapid, accurate and automatic detection of the aircraft cover plate and is suitable for the actual application scene of the aircraft maintenance site.

Inventors

  • YUE WANDE
  • LIU ZHOUZHOU
  • CAO JINGXIN

Assignees

  • 西安航空学院

Dates

Publication Date
20260512
Application Date
20260212

Claims (10)

  1. 1. The automatic detection method for the quasi-convex quadrilateral airplane cover plate image is characterized by comprising the following steps of: step 1, preprocessing an original image of an airplane cover plate, and extracting an image contour; step 2, carrying out Fourier transform noise reduction treatment on the image contour; step3, judging quasi-convex quadrangles; Step 4, calculating the curvature of the contour and identifying the segmentation points; Step 5, screening and fitting the quasi-straight line segment; Step 6, constructing a quadrangle and acquiring a control point; Step 7, perspective distortion correction; And 8, detecting template matching.
  2. 2. The method for automatically detecting the cover plate image of the quasi-convex quadrilateral airplane according to claim 1, wherein the step 1 is specifically: preprocessing an original image of an aircraft cover plate by adopting a Canny edge detection algorithm: firstly, removing image noise through Gaussian filtering, then setting a double threshold value to extract edge contours, and finally obtaining an initial contour point set { (x 0 ,y 0 ),(x 1 ,y 1 ),…,(x K−1 ,y K−1 ) }, wherein K is the total point number of the contours.
  3. 3. The method for automatically detecting the cover plate image of the quasi-convex quadrilateral airplane according to claim 2 is characterized in that the standard deviation sigma of Gaussian filtering is 1.4, the middle-high threshold value of the double threshold value is 60, and the low threshold value is 30.
  4. 4. The method for automatically detecting the cover plate image of the quasi-convex quadrilateral airplane according to claim 3, wherein the step 2 is specifically: Step 2-1: the complex of contour coordinates, namely converting the initial contour point set into one-dimensional complex signals: s(k)=x(k)+j⋅y(k),k=0,1,...,K−1; Wherein j is an imaginary unit; Converting the two-dimensional contour shape information into a one-dimensional signal; step 2-2, solving coefficients by discrete Fourier transform; Carrying out one-dimensional Discrete Fourier Transform (DFT) on the complex signal, and solving a Fourier coefficient a (u), wherein a (0) corresponds to the position of the centroid of the profile, u=1, 2,3 corresponds to the global scale feature, u is more than or equal to 4 corresponds to the core shape feature, and u is approximately equal to K/2 corresponds to the surface noise; Step 2-3, frequency component screening; Removing direct current component and very low frequency component interference, selecting a frequency parameter d, and reserving symmetrically distributed medium and low frequency component intervals [0, d-1] and [ K-d, K-1] to realize noise filtering, wherein the frequency parameter d is 14; step 2-4, recovering the profile by inverse Fourier transform; Performing inverse Fourier transform on the screened frequency components, and separating a real part and an imaginary part to obtain a noise-reduced contour coordinate; step 2-5, verifying energy retention rate; The energy retention rate is controlled to be 99%, so that noise reduction is ensured while profile characteristics are retained.
  5. 5. The method for automatically detecting the cover plate image of the quasi-convex quadrilateral airplane according to claim 4, wherein the step 3 is specifically: The quasi-convex quadrangle is defined as that after four longest quasi-straight line sections extracted from an image contour are subjected to straight line fitting, if four fitting straight lines can enclose a convex quadrangle, the graph is a quasi-convex quadrangle; The image contour is composed of four quasi-straight line sections, the lengths of the neighboring quasi-straight line sections are connected through arc sections, the lengths of the quasi-straight line sections are non-zero and are larger than those of the neighboring arc sections, the curvature radius Ri of the quasi-straight line sections is larger than that of neighboring round corners and is 3 times larger than that of the neighboring arc sections, the curvature radius Rn of the quasi-straight line sections meets Rn and is smaller than or equal to 0.3×min (Ri, ri), wherein min (Ri, ri) represents the minimum value of the curvature radius of the quasi-straight line sections and the curvature radius of the corresponding neighboring round corners, the ratio of the longest side to the shortest side of a fitting quadrangle is smaller than or equal to 4, and the four inner angles of the fitting quadrangle are all within the range of (30 degrees and 150 degrees).
  6. 6. The method for automatically detecting the cover plate image of the quasi-convex quadrilateral airplane according to claim 5, wherein the step 4 is specifically: Calculating the curvature of each point of the profile by adopting a normalized inner product formula, wherein P is a current target point, P −n 、P +n is adjacent profile points of which the two sides of the target point are separated by n points, and n=5; The segmentation point judgment comprises the steps of setting a curvature threshold value to be-0.99, setting a point with the curvature value smaller than or equal to the threshold value as a quasi-straight line segment point, setting a point with the curvature value larger than the threshold value as an arc line segment point, setting a point with a preamble neighborhood as an arc line segment point and a subsequent neighborhood as a quasi-straight line segment point as a quasi-straight line segment starting point, and setting a point with the curvature value smaller than or equal to the threshold value as a quasi-straight line segment end point.
  7. 7. The method for automatically detecting the cover plate image of the quasi-convex quadrilateral airplane according to claim 6, wherein the step 5 is specifically: Sorting the identified quasi-straight line segments according to the descending order of length, selecting the first four longest quasi-straight line segments as core characteristic line segments, carrying out straight line fitting on each quasi-straight line segment by adopting a least square method to obtain four fitting straight lines L 1 、L 2 、L 3 、L 4 , sorting the identified quasi-straight line segments according to the descending order of length if the number of the identified quasi-straight line segments exceeds 4, then taking the first 4 segments, carrying out secondary smoothing on the outline after Fourier noise reduction by adopting a 5-point moving average filter if the number of the identified quasi-straight line segments is less than 4, re-executing curvature calculation and segmentation point identification in the step 4, judging that the outline is a non-quasi-convex quadrilateral if the 4 effective quasi-straight line segments still cannot be obtained after secondary treatment, directly removing, and terminating the subsequent flow.
  8. 8. The method for automatically detecting the cover plate image of the quasi-convex quadrilateral airplane according to claim 7, wherein the step 6 is specifically: the four fitting straight lines are intersected pairwise to form a convex quadrangle, namely an accompanying quadrangle, the intersection point of the adjacent straight lines L i and L i+1 is the vertex Pi of the accompanying quadrangle, the four vertexes are used as control points for perspective distortion correction, i=1, 2,3 and 4, and when i=4, L i+1 = L 5 = L 1 .
  9. 9. The method for automatically detecting the cover plate image of the quasi-convex quadrilateral airplane according to claim 8, wherein the step 7 is specifically: And (3) based on four control points, realizing distortion correction of the quasi-convex quadrilateral outline by adopting perspective transformation, setting the coordinates of any point P in the distorted image as (x, y), setting the coordinates of a corresponding point P 'in the corrected orthographic projection image as (x', y '), establishing an equation set through four pairs of control points, namely P 1 ~P 4 in the distorted image and a standard vertex P 1 ′~P 4 ' in the orthographic projection image, solving perspective transformation coefficients, and substituting the perspective transformation coefficients into a transformation formula to obtain the orthographic projection outline.
  10. 10. The method for automatically detecting the cover plate image of the quasi-convex quadrilateral airplane according to claim 9, wherein the step 8 is specifically: Step 8-1, constructing a template library, namely collecting front projection images of various standard aircraft cover plates, extracting contours, storing the contours into the template contour library after Fourier noise reduction and linear fitting pretreatment, and establishing a corresponding relation between the contours and the cover plate types; Step 8-2, feature matching, namely adopting Hu invariant moment as a core feature of the contour matching, and calculating the similarity between the contour to be detected and the template contour through MATCHSHAPE functions in OpenCV, wherein the core feature refers to rotation, translation and scaling invariance; and 8-3, judging the result, namely setting the similarity threshold value to be 0.1, judging that the matching is successful if the similarity is smaller than the threshold value, outputting the corresponding cover plate model, and otherwise, sending an abnormal prompt.

Description

Automatic detection method for quasi-convex quadrilateral airplane cover plate image Technical Field The invention belongs to the technical field of aircraft maintenance detection and computer image processing, and particularly relates to an automatic detection method for a quasi-convex quadrilateral aircraft cover plate image. Background The air transportation safety is a core premise of the development of civil aviation industry, the aircraft needs to regularly carry out mandatory airworthiness inspection such as C inspection, D inspection and the like, and tens to hundreds of various cover plates need to be opened for each definite inspection so as to finish the overhaul of internal components. At present, the detection work of aircraft apron mainly relies on maintenance engineer manual work to accomplish, and not only intensity of labour is big, detection efficiency is low, still has the potential safety hazard such as the apron omission installation, misloading that lead to because of the human negligence, and many aviation maintenance enterprises in China have taken place the incident that the apron detection mistake causes. Along with the development of image processing and computer vision technology, the method is increasingly widely applied in the field of industrial detection, but the technology in the automatic detection direction of the aircraft cover plate is relatively deficient, and the prior related technical scheme has obvious limitations: 1. The detection method based on the deep learning is highly dependent on a large-scale labeling data set and high-performance GPU hardware support, and most of aircraft maintenance sites are resource-limited environments, so that the deployment requirements of the aircraft maintenance sites are difficult to meet; 2. The defects of the Hough transform method are that the traditional Hough Transform (HT) and an improved algorithm thereof are based on rigid straight line assumption, the round corner structure commonly existing in an airplane cover plate and the perspective distortion problem generated in the shooting process cannot be effectively adapted, the algorithm time complexity is as high as O (N2), and the instantaneity is difficult to ensure; 3. The existing contour noise reduction method is easy to excessively smooth contour details or retain a large amount of noise, so that the subsequent shape recognition accuracy is reduced. The aircraft cover plate has the characteristics of being provided with round corners, being easily affected by perspective distortion, being easily affected by stains and illumination interference and the like, and the comprehensive requirements of the aircraft maintenance site on detection adaptability, precision, instantaneity and deployment cost cannot be met in the prior art. Disclosure of Invention In order to overcome the defects of the prior art, the invention provides an automatic detection method for a quasi-convex quadrilateral airplane cover plate image, which breaks through the dependence of a traditional algorithm on a rigid straight line by defining the concept and the judgment standard of the quasi-convex quadrilateral, realizes contour noise reduction by combining Fourier transformation, eliminates noise interference while retaining core characteristics, identifies quasi-straight line segments based on curvature analysis and fits, constructs an accompanying quadrilateral acquisition control point, corrects distortion by perspective transformation, and finally completes cover plate detection by combining template matching to form a complete technical link of contour extraction-noise reduction-judgment-segmentation fitting-distortion correction-matching detection. The invention can realize the rapid, accurate and automatic detection of the aircraft cover plate and is suitable for the actual application scene of the aircraft maintenance site. The technical scheme adopted for solving the technical problems is as follows: step 1, preprocessing an original image of an airplane cover plate, and extracting an image contour; step 2, carrying out Fourier transform noise reduction treatment on the image contour; step3, judging quasi-convex quadrangles; Step 4, calculating the curvature of the contour and identifying the segmentation points; Step 5, screening and fitting the quasi-straight line segment; Step 6, constructing a quadrangle and acquiring a control point; Step 7, perspective distortion correction; And 8, detecting template matching. Preferably, the step 1 specifically includes: preprocessing an original image of an aircraft cover plate by adopting a Canny edge detection algorithm: firstly, removing image noise through Gaussian filtering, then setting a double threshold value to extract edge contours, and finally obtaining an initial contour point set { (x 0,y0),(x1,y1),…,(xK−1,yK−1) }, wherein K is the total point number of the contours. Preferably, the standard deviation sigma of the Gaussian filter is 1.4