CN-121563989-B - Tunnel lining surface crack detection method and system

CN121563989BCN 121563989 BCN121563989 BCN 121563989BCN-121563989-B

Abstract

The invention provides a method and a system for detecting cracks on the surface of a tunnel lining, which relate to the technical field of tunnel engineering and comprise the steps of obtaining lining surface images and crack marks of detected tunnels and lining surface images of tunnels to be detected; the method comprises the steps of carrying out feature extraction on lining surface images based on a visual cell mechanism to obtain sensing feature images of detected tunnels and tunnels to be detected, carrying out embedding mapping and contrast learning constraint on the sensing feature images of the detected tunnels and crack labels to obtain crack embedding spaces, mapping the sensing feature images of the tunnels to be detected to the crack embedding spaces, carrying out distribution alignment on the detected tunnel embedding features and the tunnel embedding features to be detected in the crack embedding spaces through a countermeasure mechanism, and generating lining surface crack detection results of the tunnels to be detected through the aligned crack embedding spaces. The invention solves the problem that the existing detection method is difficult to stably distinguish between the crack and the non-crack due to inconsistent crack characteristic distribution under different tunnel scenes.

Inventors

CUI ZHENXI
ZHANG BIAO
LUAN HUIJIE
Peng Faxin
CHEN MINGYI
FAN QIULIN
CAO YUBIN
DENG XIANGHUI
LIU MINGLIAN
PU ZHIFENG
DING XIAO
HUANG BIN
LI ZHAOKUI

Assignees

中铁十八局集团第五工程有限公司
西安工业大学

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (8)

1. The tunnel lining surface crack detection method is characterized by comprising the following steps of: constructing surface image data of a cross-tunnel lining, wherein the surface image data comprises lining surface images and crack marks of detected tunnels and lining surface images of tunnels to be detected; Feature extraction is carried out on the lining surface image based on a visual cell mechanism to obtain a perception feature map of the detected tunnel and the tunnel to be detected, and the feature extraction method comprises the following steps: A group of filter kernels with different directions and scales are adopted to carry out convolution operation on the lining surface image, and a direction selective response of the lining surface image is generated; Introducing a maximum competitive aggregation mechanism, and performing cross-scale and cross-direction aggregation on the direction selective response to generate a comprehensive crack response; calculating local neighborhood response of the corresponding pixel through the comprehensive crack response, and constructing a local inhibition item through a difference relation between the comprehensive crack response and the local neighborhood response; generating a perception feature map of each lining surface image through the comprehensive crack response and the local inhibition item; performing embedding mapping and contrast learning constraint through the perceived feature map and the crack mark of the detected tunnel, and constructing to obtain a crack embedding space; mapping a perception feature map of a tunnel to be detected to the crack embedding space, and carrying out distribution alignment on the detected tunnel embedding feature and the tunnel embedding feature to be detected in the crack embedding space through a countermeasure mechanism; and generating a lining surface crack detection result of the tunnel to be detected through the aligned crack embedding space.
2. The method for detecting the surface cracks of the tunnel lining according to claim 1, wherein the construction of the crack embedding space by embedding mapping and contrast learning constraint through the perceived feature map and the crack labeling of the detected tunnel comprises the following steps: The method comprises the steps of enabling a perception feature image of a detected tunnel and a crack mark to be aligned on a pixel layer plane, and constructing a perception feature sample set, wherein the perception feature sample set comprises pixel-level perception feature vectors and corresponding crack class labels; performing embedding mapping on the pixel-level perception feature vector through an embedding mapping function to obtain a corresponding embedding representation; constructing a corresponding embedded representation sample set through the embedded representation and the corresponding crack class label; constructing a sample pair set based on the crack class labels and the embedded representation sample set; constructing contrast learning loss through a sample pair set; updating parameters of the embedded mapping function by minimizing contrast learning loss; And re-performing embedding mapping through the embedding mapping function after updating the parameters to obtain a crack embedding space, wherein the crack embedding space comprises the detected tunnel embedding characteristics and the corresponding crack type labels.
3. The method for detecting a crack on a surface of a tunnel liner according to claim 1, wherein mapping the perceived feature map of the tunnel to be detected to the crack embedding space and aligning the detected tunnel embedding feature and the tunnel embedding feature to be detected in the crack embedding space by a countermeasure mechanism comprises: After pixelating the perception feature map of the tunnel to be detected, carrying out embedding mapping through an embedding mapping function to obtain the embedding feature of the tunnel to be detected; constructing an embedded space discriminator; Inputting the detected tunnel embedding characteristics and the tunnel embedding characteristics to be detected into an embedding space discriminator to obtain the confidence of the tunnel; Respectively calculating the compaction-separation indexes of the detected tunnel and the tunnel to be detected; calculating class center weighting countermeasures through the confidence coefficient of the belonging tunnel, the detected tunnel and the compact-separation index of the tunnel to be detected; updating parameters of the embedded mapping function through class center weighting against loss; and re-performing embedding mapping through the embedding mapping function after updating the parameters to obtain the tunnel embedding characteristics to be detected, which are aligned with the detected tunnel embedding characteristics.
4. A tunnel lining surface crack detection method as claimed in claim 3, wherein the calculating of the compaction-separation index for the detected tunnel and the tunnel to be detected, respectively, comprises: calculating an embedded class center for each crack class through the detected tunnel embedded characteristics; calculating the intra-class compactness and the inter-class separation of the detected tunnel through the embedded class center; Calculating a compaction-separation index of the detected tunnel through the intra-class compactness and the inter-class separation; calculating the Euclidean distance between the tunnel embedding characteristics to be detected and the center of each embedding class; And projecting the embedded features of the tunnel to be detected to the nearest embedded class center through the Euclidean distance, and calculating the compact-separation index of the tunnel to be detected based on the relative distance relation between the embedded class center and other embedded class centers.
5. The method for detecting the surface crack of the tunnel lining according to claim 1, wherein the step of generating the detection result of the surface crack of the lining of the tunnel to be detected by the aligned crack embedding space comprises the following steps: Inputting the tunnel embedding characteristics to be detected in the aligned crack embedding space into an embedding classifier to obtain the crack class prediction probability of each pixel; acquiring an embedded class center through the aligned crack embedded space; adjusting the crack type prediction probability through the embedded type center and the distribution information of the embedded characteristics of the tunnel to be detected; Constructing a crack candidate region based on the adjusted crack class prediction probability, and applying length constraint, width constraint and linearity constraint to the crack candidate region to obtain a crack probability map; and acquiring a detection result of the surface crack of the lining of the tunnel to be detected through the crack probability map.
6. A tunnel liner surface crack detection system, comprising: The construction module is used for constructing surface image data of a cross-tunnel lining, wherein the surface image data comprises lining surface images and crack marks of detected tunnels and lining surface images of tunnels to be detected; The extraction module is used for extracting the characteristics of the lining surface image based on a visual cell mechanism to obtain a perception characteristic diagram of the detected tunnel and the tunnel to be detected; the mapping module is used for carrying out embedding mapping and contrast learning constraint through the perceived feature map and the crack mark of the detected tunnel, and constructing to obtain a crack embedding space; the alignment module is used for mapping the perception feature map of the tunnel to be detected to the crack embedding space and carrying out distribution alignment on the detected tunnel embedding features and the tunnel embedding features to be detected in the crack embedding space through a countermeasure mechanism; The detection module is used for generating a detection result of the lining surface crack of the tunnel to be detected through the aligned crack embedding space; the extraction module comprises: The convolution unit is used for carrying out convolution operation on the lining surface image by adopting a group of filter kernels with different directions and scales to generate a direction selective response of the lining surface image; the aggregation unit is used for introducing a maximum competitive aggregation mechanism, and carrying out cross-scale and cross-direction aggregation on the direction selective response to generate a comprehensive crack response; The first calculation unit is used for calculating local neighborhood response of the corresponding pixel through the comprehensive crack response and constructing a local inhibition item through a difference relation between the comprehensive crack response and the local neighborhood response; And the combination unit is used for generating a perception characteristic map of each lining surface image through the comprehensive crack response and the local inhibition item.
7. The tunnel liner surface crack detection system of claim 6, wherein the mapping module comprises: the alignment unit is used for aligning the perception feature map of the detected tunnel with the crack mark on the pixel layer surface to construct a perception feature sample set, and the perception feature sample set comprises pixel-level perception feature vectors and corresponding crack type labels; the first mapping unit is used for carrying out embedding mapping on the pixel-level perception feature vector through an embedding mapping function to obtain a corresponding embedding representation; the first construction unit is used for constructing a corresponding embedded representation sample set through the embedded representation and the corresponding crack class label; A second construction unit for constructing a sample pair set based on the crack class label and the embedded representation sample set; a third construction unit for constructing a contrast learning loss through the sample pair set; the first updating unit is used for updating the parameters of the embedded mapping function by minimizing the contrast learning loss; and the second mapping unit is used for carrying out embedding mapping again through the embedding mapping function after updating the parameters to obtain a crack embedding space, wherein the crack embedding space comprises the detected tunnel embedding characteristics and the corresponding crack type labels.
8. The tunnel liner surface crack detection system of claim 6, wherein the alignment module comprises: The third mapping unit is used for carrying out embedding mapping through an embedding mapping function after pixelating the perception feature map of the tunnel to be detected to obtain the embedding feature of the tunnel to be detected; A fourth construction unit for constructing an embedding space discriminator; The judging unit is used for inputting the detected tunnel embedding characteristics and the tunnel embedding characteristics to be detected into the embedding space discriminator to obtain the confidence of the tunnel; a second calculation unit for calculating a compaction-separation index of the detected tunnel and the tunnel to be detected, respectively; a third calculation unit for calculating a class center weighted countermeasures loss by the belonging tunnel confidence, the detected tunnel, and the compact-separate index of the tunnel to be detected; A second updating unit for updating the parameters of the embedded mapping function against loss through class center weighting; And the fourth mapping unit is used for carrying out embedding mapping again through the embedding mapping function after updating the parameters to obtain the tunnel embedding characteristics to be detected, which are aligned with the detected tunnel embedding characteristics.

Description

Tunnel lining surface crack detection method and system Technical Field The invention relates to the technical field of tunnel engineering, in particular to a tunnel lining surface crack detection method and system. Background The tunnel lining is used as an important stress and protection component of a tunnel structure, and the surface crack is an important representation form reflecting the stress state, construction quality and service performance degradation of the lining. Potential safety hazards such as water seepage, steel bar corrosion, structural bearing capacity reduction and the like can be caused by the generation and development of cracks, so that the surface cracks of the tunnel lining are timely and accurately detected and identified, and the method has important significance for safety evaluation and maintenance decision of the tunnel structure. At present, the detection of the surface cracks of the tunnel lining mainly adopts a detection method of manual inspection, crack identification based on visible light images and three-dimensional laser or geometric information. The manual inspection mode relies on experience of inspection personnel, is strong in subjectivity and low in efficiency, and has high operation safety risk in a long tunnel or an operation tunnel, so that the requirements of refinement and long-term monitoring are difficult to meet. Crack detection methods based on image processing or deep learning, while increasing the level of automation, still face many challenges in practical applications. On one hand, complicated background characteristics such as construction joints, template joints, water seepage traces, surface pollution and the like exist on the surface of the tunnel lining, the appearance and the appearance are similar to those of cracks in height, false detection is easy to cause, on the other hand, the illumination conditions inside the tunnel are complicated and uneven, and different tunnels have obvious differences in lining materials, construction processes and service environments, so that the generalization capability of the existing model is insufficient when the existing model is applied across tunnels and working conditions. In addition, the existing deep learning method generally depends on a large number of high-quality labeling samples, and the manual labeling cost of tunnel lining cracks, particularly fine cracks and early cracks, is high and poor in consistency, so that further improvement of the model performance is limited. Meanwhile, part of methods focus on pixel-level or local texture features, and lack of effective characterization of essential features such as continuity and direction consistency of a crack structure, so that a crack detection result is discontinuous, and accurate quantification of key parameters such as crack length, trend and the like is affected. Therefore, how to fully utilize the structural characteristics of cracks in a complex tunnel environment, reduce the dependence on large-scale labeling data, and enhance the adaptability of a model to different illumination conditions, different lining materials and different tunnel working conditions becomes a technical problem to be solved in the field of tunnel lining surface crack detection. Disclosure of Invention The invention aims to provide a tunnel lining surface crack detection method and system so as to solve the problems. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: in a first aspect, the application provides a method for detecting a surface crack of a tunnel liner, which comprises the following steps: constructing surface image data of a cross-tunnel lining, wherein the surface image data comprises lining surface images and crack marks of detected tunnels and lining surface images of tunnels to be detected; Feature extraction is carried out on the lining surface image based on a visual cell mechanism, so as to obtain a perception feature map of the detected tunnel and the tunnel to be detected; performing embedding mapping and contrast learning constraint through the perceived feature map and the crack mark of the detected tunnel, and constructing to obtain a crack embedding space; mapping a perception feature map of a tunnel to be detected to the crack embedding space, and carrying out distribution alignment on the detected tunnel embedding feature and the tunnel embedding feature to be detected in the crack embedding space through a countermeasure mechanism; and generating a lining surface crack detection result of the tunnel to be detected through the aligned crack embedding space. In a second aspect, the present application further provides a tunnel liner surface crack detection system, including: The construction module is used for constructing surface image data of a cross-tunnel lining, wherein the surface image data comprises lining surface images and crack marks of detected tunnels and lining surface imag