CN-122023403-A - Intelligent bridge structure damage identification method and system based on machine vision

CN122023403ACN 122023403 ACN122023403 ACN 122023403ACN-122023403-A

Abstract

The invention discloses an intelligent recognition method and system for bridge structure damage based on machine vision, in particular to the technical field of image recognition, which are used for solving the problems that the existing method is difficult to deal with biological shielding and discontinuous distribution of bridge damage areas in mountain areas, so that recognition is missed and misjudgment is caused; the method is realized by acquiring a bridge surface image sequence, analyzing correlative mutation points of the epiphyte covering local density and the surface geometric curvature to judge a shielding region, further coupling and analyzing the biological growth direction in the shielding region and the bridge structure stress distribution direction to determine a suspected damage guide region, analyzing multi-scale fractal dimension sequence mutation inflection points of the guide region image texture to adaptively determine a characteristic space scale and stripping biological shielding to obtain a damage candidate region, constructing a space relation diagram for the candidate region to correlate discontinuous characteristics and form a complete damage region, and finally identifying the damage type based on the characteristics of the complete region.

Inventors

Rao Lingli
LI FENG
LONG JIANXU
TANG HONG
HAO ZENGTAO
WANG YI
LI BIN
He Feixue
LI QUANXIU
ZHANG YA

Assignees

贵州交通职业大学

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. The intelligent bridge structure damage identification method based on machine vision is characterized by comprising the following steps of: S1, acquiring an image sequence of the bridge structure surface in a mountain area; S2, calculating local density of the epiphyte coverage of the bridge structure surface of the mountain area based on the image sequence, determining geometric curvature of the bridge structure surface of the mountain area, and judging whether an epiphyte shielding area exists in the image sequence by analyzing correlation mutation points between the local density and the geometric curvature; S3, when the periphyton shielding area exists, determining a suspected damage guide area by analyzing the coupling relation between the growth direction of the periphyton in the periphyton shielding area and the stress distribution direction of the bridge structure; s4, determining a characteristic space scale by analyzing a mutation inflection point of a multi-scale fractal dimension sequence of a texture of an image sequence corresponding to the suspected damage guide region, and stripping the periphyton shielding region under the characteristic space scale to obtain a damage body candidate region; s5, associating discontinuous distribution damage features with the damage body candidate region to form a complete damage region; S6, identifying the damage type of the bridge structure based on the complete damage area.
2. The intelligent recognition method for bridge structural damage based on machine vision according to claim 1, wherein S1 comprises: Acquiring a group of digital images with overlapping areas on the bridge structure surface of the mountain area along a preset inspection path through a stereoscopic vision system carried on the mobile platform; And synchronously recording shooting position and attitude parameters of each frame of digital image to form an image sequence.
3. The intelligent recognition method for bridge structural damage based on machine vision according to claim 1, wherein S2 comprises: carrying out vegetation region segmentation based on a color space threshold on a single frame digital image in an image sequence to obtain a vegetation region; in the vegetation region, counting the vegetation pixel ratio in a sliding window mode, and generating a partial density map covered by the periphyton; Meanwhile, based on the image sequence, shooting positions and gesture parameters, a triangular mesh model of the bridge structure surface of the mountain area is obtained through three-dimensional reconstruction, gaussian curvatures at all vertexes of the triangular mesh model are calculated, and a geometric curvature distribution map is generated; on a local density map and a geometric curvature distribution map which are spatially registered, calculating a pearson correlation coefficient of a local density value and a geometric curvature value through a sliding window, and judging a window center point of which the absolute value of the pearson correlation coefficient is lower than a preset correlation coefficient threshold value as a correlation mutation point; if the correlative mutation points exist, judging that the parasitic biological shielding region exists in the image sequence.
4. The intelligent recognition method for damage to bridge structures based on machine vision according to claim 1, wherein S3 comprises: Extracting a main direction of image textures as a growth direction of the periphyton based on the gray level co-occurrence matrix in the periphyton shielding area; acquiring a stress distribution direction of a bridge structure obtained in advance through finite element analysis based on a bridge design model and a standard load working condition; after spatial registration, calculating a measure of directional consistency between the growth direction and the stress distribution direction; And determining a local area with the direction consistency metric exceeding a preset consistency threshold value as a suspected damage guide area in the periphyton shielding area.
5. The intelligent recognition method for bridge structural damage based on machine vision according to claim 1, wherein S4 comprises: extracting gray images from an image sequence corresponding to the suspected damage guide area; Carrying out multi-scale Gaussian pyramid decomposition on the gray level image, and calculating the fractal dimension of the image based on a box counting method under each scale to obtain a multi-scale fractal dimension sequence; applying a variable point detection algorithm to the multi-scale fractal dimension sequence, identifying a mutation inflection point with obvious change of the fractal dimension value in the multi-scale fractal dimension sequence, and determining a scale parameter corresponding to the mutation inflection point as a characteristic space scale; and under the characteristic space scale, carrying out morphological open operation on the suspected injury guiding region by using structural elements constructed based on scale parameters, and separating the periphyton shielding region to obtain an injury body candidate region.
6. The intelligent recognition method for bridge structural damage based on machine vision according to claim 5 is characterized in that multi-scale Gaussian pyramid decomposition is carried out on gray images, fractal dimensions of the images are calculated based on a box counting method under each scale to obtain a multi-scale fractal dimension sequence, the method comprises the steps of taking an original gray image as a reference layer, sequentially carrying out Gaussian blur and fixed multiplying power downsampling to construct an image Gaussian pyramid containing multiple scales, using a series of grid coverage images with incremental side lengths for each scale image in the pyramid, counting the number of grids containing target textures under the side length of each grid, fitting linear relations between the side length logarithm of the grid and the grid number logarithm, wherein the slope is the fractal dimension under the corresponding scale, and collecting fractal dimension values under all scales to form the multi-scale fractal dimension sequence.
7. The intelligent recognition method for bridge structural damage based on machine vision according to claim 1, wherein S5 comprises: extracting centroid coordinates of each damaged body candidate region and the main axis direction of the region; Calculating Euclidean distance and relative azimuth angle between any two damaged body candidate areas based on centroid coordinates; Constructing a space relation diagram taking the consistency of a distance threshold value and an azimuth angle as a constraint, and judging that the candidate region pairs of the damaged body meeting the constraint are associated; And carrying out region clustering based on the spatial relation diagram, and merging the related damage body candidate regions to form a complete damage region.
8. The intelligent recognition method for the bridge structure damage based on the machine vision, which is characterized by constructing a spatial relationship diagram taking a distance threshold value and azimuth angle consistency as constraints, judging that the damage body candidate region pairs meeting the constraints are related, wherein the method comprises the steps of calculating centroid coordinates of each damage body candidate region and a main axis direction of a minimum circumscribed rectangle, calculating Euclidean distances between the region pairs based on the centroid coordinates, calculating direction included angles between the region pairs based on the main axis directions, judging that the corresponding region pairs meet the azimuth angle consistency constraints when the Euclidean distances between the region pairs are smaller than a preset distance threshold value and the direction included angles are smaller than a preset angle threshold value, and adding all the region pairs meeting the constraints into the spatial relationship diagram as edges so as to finish the related judgment.
9. The intelligent recognition method of bridge structural damage based on machine vision according to claim 1, wherein S6 comprises: extracting a damage characteristic vector composed of the area, perimeter and rectangle degree of the region and texture contrast based on a gray level co-occurrence matrix from the complete damage region; Inputting the damage characteristic vector into a support vector machine classification model which is obtained through sample training in advance; calculating the matching degree between the damage characteristic vector and a preset typical damage type characteristic mode through a support vector machine classification model; and outputting the damage type of the bridge structure corresponding to the complete damage area according to the characteristic mode of the typical damage type with the highest matching degree.
10. The intelligent recognition system for the damage of the bridge structure based on the machine vision is used for realizing the intelligent recognition method for the damage of the bridge structure based on the machine vision, and is characterized by comprising the following steps: the image acquisition module is used for acquiring an image sequence of the bridge structure surface in the mountain area; the shielding judging module is used for calculating the local density of the epiphyte coverage of the bridge structure surface of the mountain area based on the image sequence, determining the geometric curvature of the bridge structure surface of the mountain area, and judging whether a epiphyte shielding area exists in the image sequence by analyzing the correlation mutation points between the local density and the geometric curvature; The area determining module is used for determining a suspected damage guide area by analyzing the coupling relation between the growth direction of the periphyton in the periphyton shielding area and the stress distribution direction of the bridge structure when the periphyton shielding area exists; The candidate extraction module is used for determining a characteristic space scale by analyzing abrupt inflection points of a multi-scale fractal dimension sequence of textures of an image sequence corresponding to the suspected damage guide region, and stripping the periphyton shielding region under the characteristic space scale to obtain a damage body candidate region; The region complement module is used for associating discontinuous distribution of damage characteristics to the candidate region of the damage body to form a complete damage region; And the type identification module is used for identifying the damage type of the bridge structure based on the complete damage area.

Description

Intelligent bridge structure damage identification method and system based on machine vision Technical Field The invention relates to the technical field of image recognition, in particular to an intelligent recognition method and system for bridge structural damage based on machine vision. Background In the field of bridge structure safety operation and maintenance, the machine vision technology has the advantages of non-contact, high efficiency and batch acquisition and analysis, and is particularly suitable for the damage identification work of the bridge structure in the inspection scene of the mountain bridge. The bridge in the mountain area is influenced by the terrain, climate and load conditions, the structural damage type is complex, the damp and multi-bacteria environment easily causes the damaged area to breed by moss, lichen and other by-products, and meanwhile, complex damage is easily formed under the complex load effect, and the characteristics provide higher requirements for the applicability of the machine vision recognition technology. The existing intelligent recognition method for the damage of the bridge structure based on the machine vision generally extracts damage visual characteristics after preprocessing the image by collecting the bridge structure image, and further completes damage recognition, and the core of the intelligent recognition method depends on capturing and analyzing clear visual characteristics of a damage area, so that the intelligent recognition method has mature application in bridge scenes with relatively simple environments such as plain. In the existing bridge structure damage identification, the situations of multi-layer vision shielding and compound damage discontinuous distribution of a mountain bridge damage area are difficult to effectively cope with, the visual characteristics of a shielding object and a damage body cannot be accurately stripped, and effective association analysis cannot be carried out on the damage characteristics of the discontinuous distribution, so that the problems of missed judgment and misjudgment in the damage identification are caused, and the accurate requirements of the mountain bridge structure damage identification are difficult to meet. Disclosure of Invention In order to overcome the defects in the prior art, the invention provides an intelligent recognition method and system for damage of a bridge structure based on machine vision, which are used for solving the problems in the prior art. In order to achieve the above purpose, the present invention provides the following technical solutions: The intelligent bridge structure damage identification method based on machine vision comprises the following steps: S1, acquiring an image sequence of the bridge structure surface in a mountain area; S2, calculating local density of the epiphyte coverage of the bridge structure surface of the mountain area based on the image sequence, determining geometric curvature of the bridge structure surface of the mountain area, and judging whether an epiphyte shielding area exists in the image sequence by analyzing correlation mutation points between the local density and the geometric curvature; S3, when the periphyton shielding area exists, determining a suspected damage guide area by analyzing the coupling relation between the growth direction of the periphyton in the periphyton shielding area and the stress distribution direction of the bridge structure; s4, determining a characteristic space scale by analyzing a mutation inflection point of a multi-scale fractal dimension sequence of a texture of an image sequence corresponding to the suspected damage guide region, and stripping the periphyton shielding region under the characteristic space scale to obtain a damage body candidate region; s5, associating discontinuous distribution damage features with the damage body candidate region to form a complete damage region; S6, identifying the damage type of the bridge structure based on the complete damage area. Further, S1 includes: Acquiring a group of digital images with overlapping areas on the bridge structure surface of the mountain area along a preset inspection path through a stereoscopic vision system carried on the mobile platform; And synchronously recording shooting position and attitude parameters of each frame of digital image to form an image sequence. Further, S2 includes: carrying out vegetation region segmentation based on a color space threshold on a single frame digital image in an image sequence to obtain a vegetation region; in the vegetation region, counting the vegetation pixel ratio in a sliding window mode, and generating a partial density map covered by the periphyton; Meanwhile, based on the image sequence, shooting positions and gesture parameters, a triangular mesh model of the bridge structure surface of the mountain area is obtained through three-dimensional reconstruction, gaussian curvatures at all vertexes of t