CN-121999284-A - Intelligent concrete crack identification and damage quantitative evaluation method and system

Abstract

The application discloses an intelligent concrete crack identification and damage quantitative evaluation method and system, and relates to the field of computer vision. The method comprises the steps of obtaining a crack image of a concrete structure, carrying out feature extraction and pixel level segmentation on the crack image based on a double-branch network, obtaining a crack pixel mask, wherein the double-branch network comprises a space branch, a semantic branch and a feature fusion module of coordinate perception, carrying out geometric quantification on the crack pixel mask, obtaining quantification data, and grading the health state of the concrete structure according to the quantification data. According to the method, the crack continuity detection and the fine quantitative evaluation effect can be realized through the design of the double-branch network.

Inventors

• LIN CHANG
• CHANG XIAONING
• YANG YONGQIN
• PAN LISHA
• XU SHUYING

Assignees

• 海南大学

Dates

Publication Date

20260508

Application Date

20260120

Claims (10)

1. 1. The intelligent concrete crack identification and damage quantitative evaluation method is characterized by comprising the following steps of: acquiring a crack image of a concrete structure; Performing feature extraction and pixel level segmentation on the crack image based on a double-branch network to obtain a crack pixel mask, wherein the double-branch network comprises a space branch, a semantic branch and a feature fusion module for coordinate perception; geometrically quantizing the crack pixel mask to obtain quantized data; And grading the health state of the concrete structure according to the quantized data.
2. 2. The method of claim 1, wherein the semantic branches comprise a lightweight MobileNetV and a multi-level attention refinement module to obtain multi-scale semantic features.
3. 3. The method of claim 2, wherein the spatial branch comprises a plurality of continuous coordinate convolution modules, a multi-scale main path feature is output, the last-stage coordinate convolution module is connected with a spatial enhancement module, the spatial enhancement module comprises a residual block, the main path feature output by the last-stage coordinate convolution module is subjected to feature extraction, the spatial attention path feature is obtained through a spatial attention mechanism, and the main path feature output by the last-stage coordinate convolution module and the spatial attention path feature are subjected to weighted fusion, so that deep priori features are obtained.
4. 4. The method of claim 3, wherein the feature fusion module performs feature fusion on the semantic features and the main path features/the deep priori features of the corresponding scale through residual connection and SE attention mechanism to obtain fusion features.
5. 5. The method according to claim 3, wherein the feature fusion modules include a plurality of feature fusion modules, and each feature fusion module fuses the output of the semantic branch, the output of the spatial branch, and the output of the feature fusion module at the previous stage of the corresponding hierarchy to obtain the fused feature; The first-stage feature fusion module fuses the deep priori features, the main path features output by the penultimate-stage coordinate convolution module and the semantic features output by the last-stage attention refinement module to obtain the output of the first-stage feature fusion module.
6. 6. The method of claim 5, wherein the method further comprises: For the output of each feature fusion module, respectively acquiring mixed loss including a Dice loss, a cross entropy loss and a Focal loss; And obtaining total loss according to the mixing loss corresponding to each characteristic fusion module.
7. 7. The utility model provides a concrete crack intelligent identification and damage quantitative evaluation system which characterized in that, the system includes: the image acquisition module is used for acquiring a crack image of the concrete structure; The image processing module is used for carrying out feature extraction and pixel level segmentation on the crack image based on a double-branch network to obtain a crack pixel mask, wherein the double-branch network comprises a space branch, a semantic branch and a feature fusion module for coordinate perception; The quantization module is used for geometrically quantizing the crack pixel mask to obtain quantized data; and the evaluation module is used for grading the health state of the concrete structure according to the quantitative data.
8. 8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 6 when the computer program is executed.
9. 9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.
10. 10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.

Description

Intelligent concrete crack identification and damage quantitative evaluation method and system Technical Field The application relates to the technical field of computer vision, in particular to an intelligent concrete crack identification and damage quantitative evaluation method and system. Background With the acceleration of the urban development, a large number of concrete buildings and infrastructures adopting earthquake-resistant designs are put into long-term use, such as bridges, dams, etc. using high-ductility fiber-reinforced cement-based composite materials (ECC, ENGINEERED CEMENTITIOUS COMPOSITES). These structures inevitably develop cracks under long term effects of vehicle loading, environmental erosion, microseismic accumulation, and the like. The existing crack detection and evaluation technology has the following main problems: 1. The crack characteristics of the earthquake-resistant materials are difficult to identify, and many novel earthquake-resistant concrete materials (stress-HARDENING MATERIALS) have the characteristic of multi-crack development, namely a large number of dense micro cracks can be generated to dissipate energy. The prior art is difficult to distinguish the normal micro cracks from harmful macro cracks caused by fatigue or damage, and false alarm or missing alarm is easy to cause. 2. The daily inspection data volume is large, the efficiency is low, and the daily inspection of long-distance highways or large bridges is carried out to generate massive image data. The manual interpretation efficiency is low, the traditional large model has high computational complexity, and the large model is difficult to be deployed at the edge end of an unmanned plane or a patrol car for real-time processing. 3. The complex service environment is disturbed, namely in a daily use scene, oil stains, tire marks, water stains or shadow changes are often attached to the surface of the concrete, and the traditional edge detection algorithm has poor anti-interference capability. 4. The method lacks a fine quantification means, namely the existing method stays at qualitative discovery of cracks, and lacks high-precision quantitative calculation of the maximum width and the average width of the cracks. For the earthquake-resistant structure, the width of the crack is a core basis for judging whether rust prevention, grouting repair or structural reinforcement is needed. Disclosure of Invention Accordingly, it is necessary to provide a method and a system for intelligent recognition and quantitative damage assessment of concrete cracks with high accuracy and light weight. In a first aspect, the application provides an intelligent concrete crack identification and damage quantitative evaluation method. The method comprises the following steps: acquiring a crack image of a concrete structure; Performing feature extraction and pixel level segmentation on the crack image based on a double-branch network to obtain a crack pixel mask, wherein the double-branch network comprises a space branch, a semantic branch and a feature fusion module for coordinate perception; Geometrically quantizing the crack pixel mask to obtain quantized data; And grading the health state of the concrete structure according to the quantized data. In one embodiment, the semantic branches include a lightweight MobileNetV and a multi-level attention refinement module that obtain multi-scale semantic features. In one embodiment, the spatial branch comprises a plurality of continuous coordinate convolution modules, a multi-scale main path feature is output, the final-stage coordinate convolution module is connected with a spatial enhancement module, the spatial enhancement module comprises a residual block, the main path feature output by the final-stage coordinate convolution module is subjected to feature extraction, the spatial attention path feature is obtained through a spatial attention mechanism, and the main path feature and the spatial attention path feature output by the penultimate-stage coordinate convolution module are subjected to weighted fusion, so that deep priori features are obtained. In one embodiment, the feature fusion module performs feature fusion on the semantic features of the corresponding scale and the main path features/deep priori features through residual connection and an SE attention mechanism to obtain fusion features. In one embodiment, the feature fusion modules comprise a plurality of feature fusion modules, and each feature fusion module fuses the output of the semantic branch, the output of the space branch and the output of the previous-stage feature fusion module of the corresponding level to obtain fusion features; The first-stage feature fusion module fuses the deep priori features, the main path features output by the penultimate coordinate convolution module and the semantic features output by the final-stage attention refinement module to obtain the output of the first-stage feature fus