Concrete structure crack image detection method and system based on deep learning

Abstract

The invention discloses a concrete structure crack image detection method and system based on deep learning. The method comprises the following steps: acquiring a multi-view image; performing spatial alignment according to the multi-view image to obtain a matched image sequence; performing image fusion according to the matched image sequence to obtain a fused image; performing semantic segmentation based on transfer learning according to the fused image and a preset neural network model to obtain a crack mask; according to the crack mask and the fusion image, skeleton extraction is carried out, and a preliminary crack network topology is obtained; optimizing the preliminary fracture network topology to obtain a fracture network topology; carrying out finite element analysis according to preset material attributes, a stress state and a concrete structure to obtain a stress distribution cloud picture; and according to the stress distribution cloud picture and the fracture network topology, carrying out space matching to obtain a stress fracture correlation model. The method can analyze the dynamic change of the concrete crack.

Description

Technical Field

[0001] The present invention relates to the technical field of image detection, and particularly to a method and system for detecting concrete structure crack images based on deep learning. Background Art

[0002] Currently, concrete structures are widely used in construction projects. However, during their service life, cracks may occur due to natural factors (such as temperature, humidity, freeze-thaw cycles) and load effects (such as gravity, wind load). These cracks not only affect the appearance of the structure but may also lead to a decrease in load-bearing capacity and durability, thus posing potential safety hazards. Therefore, timely and accurate detection and assessment of the state of concrete cracks are of great significance for ensuring the safety of buildings, extending their service life, and reducing maintenance costs.

[0003] In the prior art, ultrasonic testing is a non-destructive testing method that determines the presence and depth of cracks by measuring the propagation characteristics of ultrasonic waves in concrete. The specific principle is to utilize the changes in the propagation speed, attenuation degree, and reflection signal of ultrasonic waves in concrete to identify cracks. When ultrasonic waves encounter cracks, reflection and scattering occur, resulting in an extended propagation time and a weakened signal intensity. By analyzing these changes, the location and depth of the cracks can be determined.

[0004] However, ultrasonic testing still faces many challenges in practical applications. The dynamic change characteristics of cracks and their correlation with the internal stress distribution further increase the difficulty of continuous monitoring and comprehensive analysis. Therefore, ultrasonic testing cannot achieve information fusion of cracks and stress distribution, and thus cannot analyze the dynamic changes of concrete cracks. In summary, there is a lack of models and methods for analyzing the dynamic changes of concrete cracks in the prior art. Summary of the Invention

[0005] The present invention provides a method and system for detecting concrete structure crack images based on deep learning to analyze the dynamic changes of concrete cracks.

[0006] In a first aspect, to solve the above technical problems, the present invention provides a method for detecting concrete structure crack images based on deep learning, including: Obtaining multi-view images; Performing spatial alignment according to the multi-view images to obtain a sequence of matching images; Performing image fusion according to the sequence of matching images to obtain a fused image; Performing semantic segmentation based on transfer learning according to the fused image and a preset neural network model to obtain a crack mask; Performing skeleton extraction according to the crack mask and the fused image to obtain a preliminary crack network topology; Optimize the preliminary crack network topology to obtain the crack network topology; Conduct finite element analysis based on the preset material properties, stress states, and concrete structures to obtain a stress distribution nephogram; Perform spatial matching based on the stress distribution nephogram and the crack network topology to obtain a stress-crack association model.

[0007] In an alternative implementation, perform spatial alignment based on the multi-view images to obtain a sequence of matching images, including: Establish an image acquisition coordinate system based on the multi-view images to obtain image coordinates; Perform a spatial transformation based on the image coordinates to obtain a spatial transformation matrix; Perform resampling based on the spatial transformation matrix and the multi-view images to obtain an image with a unified view; Perform pixel interpolation based on the image with a unified view to obtain a complete image; Perform image denoising based on the complete image to obtain a low-noise image; Perform edge detection based on the low-noise image to obtain a sequence of matching images.

[0008] In an alternative implementation, perform image fusion based on the sequence of matching images to obtain a fused image, including: Perform multi-scale decomposition based on the sequence of matching images to obtain multi-scale feature images; Perform pixel fusion based on the multi-scale feature images and the preset weight rules to obtain a preliminary fused image; Perform contrast enhancement based on the preliminary fused image to obtain an image with enhanced contrast; Perform unsharp masking based on the image with enhanced contrast to obtain a clear image; Perform edge detection based on the clear image to obtain a crack contour; Perform image overlay based on the crack contour and the clear image to obtain a fused image.

[0009] In an alternative implementation, perform skeleton extraction based on the crack mask and the fused image to obtain a preliminary crack network topology, including: Perform mask skeleton extraction based on the crack mask and the fused image to obtain an extracted image; Perform morphological filtering based on the extracted image to obtain a filtered image; Perform connection analysis based on the filtered image to obtain a preliminary crack network topology.

[0010] In an alternative embodiment, optimizing the preliminary crack network topology to obtain a crack network topology includes: Performing pruning optimization based on the preliminary crack network topology to obtain a redundant-free topology; Performing skeleton segmentation based on the redundant-free topology to obtain a backbone skeleton line; Performing branch detection based on the backbone skeleton line and the redundant-free topology to obtain branch positions; Performing geometric calculations based on the branch positions to obtain branch lengths and angles; Performing information fusion based on the branch lengths and angles and the redundant-free topology to obtain a crack network topology.

[0011] In an alternative embodiment, performing finite element analysis based on preset material properties, stress states, and concrete structures to obtain a stress distribution contour map includes: Performing finite element mesh division on the concrete structure based on the preset material properties and the stress state to obtain a finite element model; Performing stress calculations based on the finite element model to obtain a stress distribution contour map.

[0012] In an alternative embodiment, performing spatial matching based on the stress distribution contour map and the crack network topology to obtain a stress-crack correlation model includes: Performing feature extraction based on the crack network topology to obtain crack features; Constructing a stress-crack correlation model based on the crack features and the stress distribution contour map to obtain a stress-crack correlation model.

[0013] In a second aspect, the present invention provides a concrete structure crack image detection system based on deep learning, including: An input module for acquiring multi-view images; An alignment module for performing spatial alignment based on the multi-view images to obtain a sequence of matching images; A fusion module for performing image fusion based on the sequence of matching images to obtain a fused image; A mask generation module for performing semantic segmentation based on the fused image and a preset neural network model through transfer learning to obtain a crack mask; A topology generation module for performing skeleton extraction based on the crack mask and the fused image to obtain a preliminary crack network topology; A topology optimization module for optimizing the preliminary crack network topology to obtain a crack network topology; A stress calculation module, which is used to perform finite element analysis based on preset material properties, stress states, and concrete structures to obtain a stress distribution contour map; An association construction module, which is used to perform spatial matching based on the stress distribution contour map and the crack network topology to obtain a stress-crack association model.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention adopts a method combining multi-view image acquisition and a pre-trained neural network model. It acquires a sequence of concrete surface images through multi-view industrial cameras, and uses image spatial alignment technology to form a consistent image sequence, and then performs fusion processing to generate a high-resolution fused image, providing a high-quality image data basis for subsequent analysis.

[0015] (2) The present invention combines a pre-trained neural network model and uses transfer learning technology to perform semantic segmentation on the fused image to extract a crack mask, providing a reliable basis for subsequent crack network topology extraction. On this basis, skeleton extraction is performed based on the crack mask and the fused image to construct a preliminary crack network topology structure, and it is optimized through graph theory methods to obtain an accurate crack network topology, enabling it to truly reflect the dynamic changes of cracks.

[0016] (3) The present invention performs finite element analysis according to preset material properties and stress states to generate a stress distribution contour map, providing an important basis for understanding the mechanical mechanism of crack dynamic changes. By performing spatial matching between the stress distribution contour map and the crack network topology, a stress-crack association model is constructed to establish the association between crack dynamic changes and their internal stress distribution, realizing a comprehensive analysis of crack dynamic changes.

[0017] (4) The present invention realizes efficient and accurate analysis of the dynamic changes of concrete cracks through the combination of multi-view image acquisition and a pre-trained neural network model. The spatial alignment and fusion processing of multi-view images improve the integrity of crack information, the semantic segmentation function of the pre-trained neural network model improves the accuracy of crack recognition, skeleton extraction and graph theory optimization further refine the topology of the crack network, and the spatial matching between finite element analysis and the crack network topology clearly presents the mechanical mechanism of crack dynamic changes. Description of the Drawings

[0018] Figure 1 is a schematic flowchart of a method for detecting concrete structure crack images based on deep learning provided by an embodiment of the present invention; Figure 2 is a schematic structural diagram of a system for detecting concrete structure crack images based on deep learning provided by an embodiment of the present invention. Detailed Embodiments

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] Referring to Figure 1 , the embodiment of the present invention provides a method for detecting concrete structure crack images based on deep learning, including the following steps: S11, obtaining multi-view images; S12, performing spatial alignment according to the multi-view images to obtain a sequence of matching images; S13, performing image fusion according to the sequence of matching images to obtain a fused image; S14, performing semantic segmentation based on transfer learning according to the fused image and a preset neural network model to obtain a crack mask; S15, performing skeleton extraction according to the crack mask and the fused image to obtain a preliminary crack network topology; S16, optimizing the preliminary crack network topology to obtain a crack network topology; S17, performing finite element analysis according to preset material properties, stress states, and concrete structures to obtain a stress distribution nephogram; S18, performing spatial matching according to the stress distribution nephogram and the crack network topology to obtain a stress-crack association model.

[0021] In step S11, it is necessary to obtain multi-view images. It should be noted that the acquisition of multi-view images is completed by a high-resolution industrial camera, and the camera needs to have sufficient resolution and frame rate to ensure that the details of the concrete surface are clearly visible. Exemplarily, a camera that can collect thirty frames per second and has a resolution of up to twenty million pixels can be selected. Of course, according to different application scenarios and user requirements, cameras with other parameters can also be selected. The pre-trained neural network model has achieved basic functions and has good robustness through a large amount of data training in advance. Exemplarily, the deep residual network is adopted in the present invention. Of course, according to different user requirements and application scenarios, other pre-trained neural network models can also be selected.

[0022] In step S12, performing spatial alignment according to the multi-view images to obtain a sequence of matching images includes: establishing an image acquisition coordinate system according to the multi-view images to obtain image coordinates; performing a spatial transformation according to the image coordinates to obtain a spatial transformation matrix; Resampling is performed according to the spatial transformation matrix and the multi-view images to obtain a unified-view image; Pixel interpolation is performed according to the unified-view image to obtain a complete image; Image denoising is performed according to the complete image to obtain a low-noise image; Edge detection is performed according to the low-noise image to obtain a sequence of matching images.

[0023] It should be noted that in step S12, the spatial alignment of multi-view images is achieved through a series of image processing operations, and finally a sequence of matching images is obtained. First, the establishment of the image acquisition coordinate system is the basis for the spatial alignment of multi-view images. A certain fixed point on the concrete surface is selected as the origin to establish a three-dimensional rectangular coordinate system. The establishment of this coordinate system enables subsequent spatial transformation calculations to accurately align images from different perspectives. Exemplarily, the origin of the coordinate system can be set as the center point of the concrete surface or a certain significant feature point, such as the starting position of a crack or the center of a hole. Of course, according to different application scenarios and user requirements, other representative positions can also be selected, and the present invention does not limit this. Second, the calculation of the spatial transformation matrix is achieved through a feature point matching method. Specifically, significant features such as corner points and edges in the multi-view images are extracted, and the corresponding relationships between adjacent images are calculated using these feature points. For example, natural features such as cracks and holes on the concrete surface can be used as matching points, and the transformation parameters are optimized and solved by combining the least squares method, thereby achieving accurate registration. This process ensures that images from different perspectives can be spatially aligned, providing a basis for subsequent image fusion and analysis.

[0024] Next, resample the multi-view images according to the spatial transformation matrix to obtain unified-view images. During the resampling process, pixel interpolation needs to be considered to ensure that the images remain clear and complete after transformation. Exemplarily, bilinear interpolation is applicable to general cases and can improve the operation efficiency while maintaining the image quality; for occasions that require higher precision, bicubic interpolation can be used. Although the computational amount increases, a smoother transition effect can be obtained. Of course, according to specific application scenarios and user requirements, other interpolation methods can also be selected, and the present invention does not limit this. Subsequently, perform pixel interpolation on the resampled unified-view images to obtain complete images. This process further fills in the pixel missing due to the view transformation and ensures the integrity of the images. The interpolation operation is based on the gray values of the surrounding pixels, and by calculating the estimated values of the missing pixels, the images are made more visually continuous and complete. Next, perform denoising processing on the complete images to obtain low-noise images. The purpose of image denoising is to eliminate the random noise introduced during the acquisition and processing, thereby improving the image quality. Exemplarily, methods such as median filtering or Gaussian filtering can be selected. Taking median filtering as an example, a five-by-five pixel filtering window is selected, and the median of the surrounding area is taken to replace the original value for each pixel point, which can both retain the edge features and suppress the noise. Of course, according to the different types of noise and image features, other denoising algorithms can also be selected, and the present invention does not limit this. Finally, perform edge detection on the low-noise images to obtain image feature points. Exemplarily, the present invention selects to extract significant features such as corners and edges in the images. Then perform image matching through the obtained image feature points to obtain a sequence of matching images. Edge detection is a key step in identifying defects such as cracks and spalling on the concrete surface. Exemplarily, Sobel or Canny operators can be used for edge detection. Taking crack detection as an example, first perform image grayscale processing, set an appropriate detection threshold, and extract the crack contours that form an obvious contrast with the background. Through this process, defect features such as cracks can be clearly identified and extracted, providing an important basis for subsequent crack analysis and evaluation.

[0025] In step S13, according to the sequence of matching images, perform image fusion to obtain a fused image, including: According to the sequence of matching images, perform multi-scale decomposition to obtain multi-scale feature images; According to the multi-scale feature images and a preset weight rule, perform pixel fusion to obtain a preliminary fused image; According to the preliminary fused image, perform contrast enhancement to obtain an image with enhanced contrast; According to the image with enhanced contrast, perform unsharp masking processing to obtain a clear image; According to the clear image, perform edge detection to obtain crack contours; Based on the crack profile and the clear image, image superposition is performed to obtain a fused image.

[0026] It should be noted that in step S13, first, the multi-scale decomposition is achieved by performing wavelet transform on the sequence of matching images. This algorithm can effectively capture the crack features at different scales on the concrete surface. The wavelet transform decomposes the image into high-frequency and low-frequency components. The high-frequency components contain the detailed information of the cracks, while the low-frequency components retain the overall structure of the cracks. For example, fine cracks are more obvious at higher decomposition levels, while wide cracks are better retained in the low-frequency components. Exemplarily, for cracks with a width of 0.5 mm, the best feature performance can be obtained at the third-level decomposition. Of course, according to different application scenarios and user requirements, other decomposition levels can also be selected, and the present invention does not limit this. Secondly, pixel fusion is based on the multi-scale feature images and a preset weight rule. The design of the weighted fusion rule needs to consider the importance of different scale features and assign weights according to the significant features of the cracks. Higher weights are given to regions with clear edges, while the weight values are reduced for blurred regions. Exemplarily, the weight of regions with a significant feature value exceeding 0.8 can be set to 0.6, while the weight of regions with a feature value lower than 0.3 can be set to 0.2. In this way, the preliminary fused image can better retain the key information of the cracks while suppressing noise and irrelevant details.

[0027] Subsequently, exemplarily, in the present invention, the contrast enhancement processing is achieved by the histogram equalization method, which can expand the gray-scale dynamic range of the image, thereby improving the crack recognition effect. In practical applications, expanding the contrast of the original image from the range of 50 to 200 to the range of 10 to 240 can significantly improve the visibility of the crack area. The image after contrast enhancement can more clearly show the difference between the crack and the background, providing a better basis for subsequent processing. Then, the unsharp masking processing highlights the high-frequency details in the image by subtracting the image processed by Gaussian blur from the original image. Selecting appropriate Gaussian kernel size and weight coefficient has a significant impact on the result. Exemplarily, in the present invention, for the concrete surface image, selecting a 5×5 Gaussian kernel and a weight coefficient of 0.7 can achieve better results. The clear image after unsharp masking processing can more clearly show the details of the crack while reducing the interference of noise. Then, edge detection is achieved by the Sobel or Canny operator for extracting the crack contour. The Sobel operator can effectively detect the crack edges in the vertical and horizontal directions, while the Canny operator is suitable for the detection of oblique cracks. In terms of threshold selection, the double-threshold method is adopted, and the high threshold is set to twice the low threshold, which can effectively suppress false edges. The extracted crack contour provides accurate crack position information for subsequent image superposition. Finally, image superposition is the process of combining the crack contour with the clear image. The natural fusion of the contour and the original image is achieved through transparency adjustment. Exemplarily, in the present invention, the crack contour is superimposed on the optimized image with a transparency of 30%, which not only preserves the details of the original image but also highlights the crack position information. In addition, by color-coding to label cracks of different widths, the severity of the cracks can be intuitively reflected, providing a basis for subsequent damage assessment. Exemplarily, the area where the crack width is less than 0.5 mm can be labeled yellow, and the area where the width is greater than 0.5 mm can be labeled red. Of course, according to different application scenarios and user requirements, other color-coding methods can also be selected, and the present invention does not limit this.

[0028] In step S14, it is necessary to perform semantic segmentation based on transfer learning according to the fused image and a preset neural network model to obtain a crack mask; It should be noted that in step S14, semantic segmentation is performed based on transfer learning by fusing an image and a preset neural network model to obtain a crack mask. First, the fused image is obtained through a series of processes such as spatially aligning and image fusing of multi-view images. This process ensures the integrity and accuracy of crack information in the image, providing a high-quality data basis for subsequent semantic segmentation. Second, the preset neural network model is the core tool for semantic segmentation based on transfer learning. Transfer learning utilizes a pre-trained model. Exemplarily, in the present invention, a deep residual network is selected and fine-tuned to adapt to the characteristics of concrete cracks, thereby improving the accuracy and efficiency of the model. Of course, according to different application scenarios and user requirements, other types of pre-trained models can also be selected, and the present invention does not limit this. Then, semantic segmentation is to perform pixel-level classification on the fused image based on the neural network model of transfer learning to extract the mask of the crack region. This process performs classification prediction on each pixel point through the forward propagation calculation of the network model to generate a crack mask. Exemplarily, the output of the semantic segmentation model is a mask with the same size as the input image, where the pixel values of the crack region are marked with specific class labels, and the non-crack regions are marked as background or other classes. Finally, the obtained crack mask provides a basis for further crack analysis. Through the crack mask, the location, shape, and distribution of cracks can be clearly identified, providing important data support for subsequent steps such as skeleton extraction and crack network topology construction.

[0029] In step S15, based on the crack mask and the fused image, skeleton extraction is performed to obtain a preliminary crack network topology, including: Based on the crack mask and the fused image, mask skeleton extraction is performed to obtain an extraction image; Based on the extraction image, morphological filtering is performed to obtain a filtered image; Based on the filtered image, connection analysis is performed to obtain a preliminary crack network topology.

[0030] It should be noted that in step S15, first, the mask skeleton extraction is based on the crack mask and the fused image. Its purpose is to extract the center line of the crack from the crack mask to characterize the spatial distribution characteristics of the crack. The skeleton extraction technology uses the distance transformation method. By calculating the shortest distance from each pixel point to the boundary, the center line of the crack is determined. Exemplarily, in the present invention, the threshold of the distance transformation is set to half of the crack width to ensure that the extracted skeleton line can accurately reflect the center position of the crack. Of course, according to different application scenarios and user requirements, other skeleton extraction methods or threshold parameters can also be selected, and the present invention does not limit this. Secondly, morphological filtering is an important step in processing the extracted image. Its purpose is to remove noise and irrelevant details while retaining the main structure of the crack. Morphological filtering adopts a combination of opening operation and closing operation and is achieved by setting an appropriate size of the structural element. Exemplarily, the size of the structural element can be set to 3×3 pixels. By continuously performing opening and closing operations, small branches and breakpoints can be successfully removed, and the main crack structure can be retained. Of course, according to the characteristics of the crack and the noise level of the image, the size and shape of the structural element can also be adjusted, and the present invention does not limit this.

[0031] Next, the connection analysis is based on the filtered image. Its purpose is to construct the topological structure of the crack network. This process is based on graph theory methods. The intersection points of the cracks are defined as nodes, the connecting lines are defined as edges, and a complete connection relationship matrix is established through the depth-first search algorithm. Exemplarily, the node spacing can be distributed in the range of 5 to 20 cm, and the direction angles are mainly concentrated in two directions of 45 degrees and 90 degrees. This connection analysis method can not only clearly show the distribution law of the cracks, but also provide important topological information for subsequent crack propagation analysis. Of course, according to the complexity of the crack network and the analysis requirements, the definitions of nodes and edges can also be adjusted, and the present invention does not limit this. Finally, through the acquisition and processing operations of the above features, step S15 realizes the conversion from the crack mask and the fused image to the preliminary crack network topology. The preliminary crack network topology not only retains the center line information of the cracks, but also constructs the topological relationship between the cracks through connection analysis. This topological structure provides an important data basis for subsequent crack propagation analysis, stress correlation analysis, and the formulation of repair plans.

[0032] In step S16, the preliminary crack network topology is optimized to obtain the crack network topology, including: According to the preliminary crack network topology, pruning optimization is performed to obtain a redundant-free topology; According to the redundant-free topology, skeleton segmentation is performed to obtain the main skeleton line; According to the main skeleton line and the redundant-free topology, branch detection is performed to obtain the branch positions; Perform geometric calculations based on the described branch positions to obtain branch lengths and angles. Perform information fusion based on the branch lengths and angles and the redundancy-removed topology to obtain a fracture network topology.

[0033] It should be noted that in step S16, first, pruning optimization is performed based on the preliminary fracture network topology with the aim of removing redundant connections and retaining key fracture paths. This process is achieved through the minimum spanning tree algorithm. Exemplarily, the Kruskal algorithm is adopted in the present invention, which can effectively remove redundant branches in the fracture network while retaining the connectivity and integrity of the network. Of course, according to different application scenarios and user requirements, other graph theory optimization algorithms can also be selected, and the present invention does not limit this. Secondly, skeleton segmentation is performed based on the redundancy-removed topology with the aim of extracting the main skeleton line. This process is achieved through a thinning algorithm. Exemplarily, the Zhang's thinning method is adopted in the present invention, which can simplify the fracture network into a single-pixel-width skeleton line while keeping the topological structure of the fractures unchanged. For example, for a fracture with uneven width, the skeleton line obtained after thinning can accurately reflect the trend and morphological characteristics of the fracture. Exemplarily, the number of iterations in the present invention is set to 10,000 times. Of course, the number of iterations of the thinning algorithm can be adjusted according to the complexity of the fractures to ensure the accuracy and integrity of the skeleton line. Then, branch detection is performed based on the main skeleton line and the redundancy-removed topology with the aim of identifying the branch positions. This process is achieved by analyzing the connectivity of pixel points on the skeleton line, and the specific steps are as follows: First, traverse each pixel point on the skeleton line; second, for each pixel point, count the number of connected pixel points in its neighborhood; finally, when it is determined that the neighborhood connectivity number of a certain pixel point is greater than a preset threshold, it is determined that this point is a branch point and its position information is recorded. Exemplarily, the present invention uses an 8-neighborhood as the statistical range, and when the neighborhood connectivity number of a pixel point is greater than 2, it is considered that this point is a branch point. In this way, the positions of all branches can be accurately located, providing basic data for subsequent geometric calculations and information fusion.

[0034] Subsequently, geometric calculations are performed based on the branch positions with the aim of obtaining the lengths and angles of the branches. The branch length is calculated by the Euclidean distance from the branch point to the branch end, and the branch angle is calculated by the vector included angle. Finally, information fusion is performed based on the branch lengths, angles, and the redundancy-removed topology with the aim of obtaining a complete fracture network topology. This process is achieved by integrating geometric parameters with the topological structure, constructing a dataset containing features such as the number of branches, average length, and main angle distribution, and forming a complete fracture network topology.

[0035] In step S17, perform finite element analysis based on the preset material properties, stress states, and concrete structures to obtain a stress distribution nephogram, including: According to the preset material properties and the stress state, perform finite element mesh division on the concrete structure to obtain a finite element model; According to the finite element model, perform stress calculation to obtain a stress distribution nephogram.

[0036] It should be noted that in step S17, finite element analysis is performed based on the preset material properties and stress state to generate a stress distribution nephogram. First, the finite element mesh division is carried out according to the preset material properties and stress state, and its purpose is to divide the concrete structure into several small elements for accurate numerical calculation. The quality of the mesh division directly affects the accuracy of the finite element analysis. Exemplarily, for a common concrete beam structure, tetrahedral or hexahedral elements can be used for division, and the element size is controlled at the centimeter level. Exemplarily, for a concrete beam with a length of 6 meters and a cross-section of 400 mm × 400 mm, the initial mesh size can be set to 50 mm. Of course, according to different application scenarios and user requirements, other element types can also be selected or the element size can be adjusted, and the present invention does not limit this. Then, assign the preset material properties to each element. Exemplarily, the present invention uses elastic modulus, Poisson's ratio, compressive strength, etc., and constructs a mechanical model in combination with the stress state of each element.

[0037] Subsequently, the stress calculation is based on the finite element model, and the stress distribution inside the structure is solved by numerical methods. Exemplarily, the stress nephogram obtained through finite element analysis can intuitively display the stress distribution inside the structure. Exemplarily, taking a simply supported beam as an example, the lower edge of the mid-span is the tensile area, and the stress can reach 2.5 MPa, and the upper edge is the compressive area, and the stress can reach 12 MPa. The stress concentration areas appear at the load application points, supports, and section mutation positions. For the mesh refinement of the stress concentration area, the original element size can be halved. For example, when the stress at the support exceeds 20 MPa, the mesh size of this area is refined from 50 mm to 25 mm, and recalculation can obtain a more accurate stress distribution. Finally, the generation of the stress distribution nephogram provides an important basis for structural optimization. Exemplarily, if it is found that the stress in some areas is too large, it can be optimized by increasing the cross-sectional size, adding reinforcement bars or setting stiffeners, etc. For example, when the mid-span bending moment stress of the beam exceeds the allowable value, the cross-sectional height can be increased from 400 mm to 500 mm, or the reinforcement ratio can be increased from 0.8% to 1.2%. In addition, in actual engineering, factors such as temperature stress and shrinkage stress also need to be considered for concrete components. For example, for a large-volume concrete structure, it is necessary to analyze the temperature stress caused by the hydration heat during the pouring process. At this time, a heat conduction analysis module can be added to the finite element model, the cement hydration heat release curve can be set, the temperature field distribution can be calculated, and then the temperature stress can be obtained.

[0038] In step S18, based on the stress distribution contour map and the fracture network topology, spatial matching is performed to obtain a stress-fracture correlation model, including: According to the fracture network topology, feature extraction is performed to obtain fracture features; Based on the fracture features and the stress distribution contour map, a stress-fracture correlation model is constructed to obtain a stress-fracture correlation model.

[0039] It should be noted that in step S18, by performing spatial matching between the stress distribution contour map and the fracture network topology, a stress-fracture correlation model is constructed, thereby realizing the quantitative analysis and prediction of the fracture development law. First, the fracture feature extraction is based on the fracture network topology, aiming to extract the key geometric characteristics of the fractures from the topological structure, such as the number of branches, average length, main angle distribution and other features. These features form a feature vector, providing basic data for the subsequent stress-fracture correlation analysis. Of course, according to different application scenarios and user requirements, other fracture features can also be extracted, such as the number of branches and angle distribution of the fractures, which are not limited in this invention. Second, the stress distribution contour map is obtained through finite element analysis. Exemplarily, for concrete with a strength grade of C30, its compressive strength is 30 MPa, so the stress value in the high-stress area reaches more than 21 MPa. The stress distribution contour map intuitively reflects the stress state inside the structure, providing a visual basis for the correlation between fracture features and stress.

[0040] Next, the construction of the stress-crack correlation model is achieved by spatially matching the crack characteristics with the stress distribution contour map. Specifically, after spatially matching the crack network topology with the stress distribution contour map, the spatial coincidence degree between the stress area and the crack network can be calculated. Exemplarily, the areas with a coincidence degree exceeding 80% are often the high-incidence areas of crack propagation. This spatial matching method can quantitatively analyze the relationship between the stress concentration area and the crack distribution, providing a basis for predicting the crack propagation direction and speed. In the judgment of the crack propagation direction, the analysis is based on the principal stress direction. When the stress component in a certain area exceeds 60% of the concrete tensile strength, this direction will become the preferred direction of crack propagation. Exemplarily, in the analysis of floor slab cracks, it is found that in the area where the stress component reaches 4 MPa, the cracks will extend along this direction. The calculation of the crack propagation speed needs to consider multiple factors, such as the angle between the crack propagation direction and the main direction of the existing cracks. Exemplarily, when the angle is less than 15 degrees, the propagation speed reaches twice the normal speed. Exemplarily, the prediction model established by the support vector regression algorithm of the present invention shows that under the action of the dead load, the crack propagates 0.1 mm per day. Of course, according to the specific structure and load conditions, the prediction model of the crack propagation speed can be adjusted, and the present invention does not limit this. Finally, the dynamic update of the topology map information is based on the crack propagation speed. Exemplarily, for a specific shear wall member, when it is found that the crack propagation speed in a certain area exceeds 0.5 mm per day, this area will be marked as a high-risk area. This dynamic update mechanism can timely reflect the evolution process of the structural damage state, providing an important reference for structural maintenance. By establishing a dynamic correlation model between stress and cracks, the law of crack development can be grasped.

[0041] In summary, in order to realize the analysis of the dynamic changes of concrete cracks, this solution adopts a method combining multi-view image acquisition and a pre-trained neural network model. First, an image sequence of the concrete surface is obtained through multi-view industrial cameras, and the images from different views are matched using the spatial alignment technology of the images to form a consistent image sequence. Then, the fused image sequence is processed by fusion to generate a high-resolution fused image. On this basis, combined with the pre-trained neural network model, the fused image is semantically segmented using transfer learning technology to extract the crack mask. Subsequently, based on the crack mask and the fused image, the skeleton is extracted to construct a preliminary crack network topology, and it is optimized by graph theory methods to finally obtain an accurate crack network topology. In addition, according to the preset material properties and stress states, finite element analysis is performed to generate a stress distribution contour map. Finally, the stress distribution contour map is spatially matched with the crack network topology to construct a stress-crack correlation model.

[0042] In a dynamic monitoring environment, the morphology and orientation of concrete cracks change with time, loading conditions, and environmental factors. Therefore, continuous monitoring of cracks is required to capture their dynamic changes. Through multi-view image acquisition, the morphology and orientation of cracks can be comprehensively captured from different angles, solving the problem that it is difficult to fully present crack information from a single view. The spatial alignment and fusion processing of images further improve the integrity and accuracy of crack information. The introduction of a pre-trained neural network model enables semantic segmentation to efficiently and accurately identify crack masks, providing a reliable basis for subsequent crack network topology extraction. The skeleton extraction and graph theory optimization steps further refine the topology of the crack network, enabling it to more realistically reflect the dynamic changes of cracks. The stress distribution contour map generated by finite element analysis provides an important basis for understanding the mechanical mechanism of crack dynamic changes. By spatially matching the stress distribution contour map with the crack network topology, the correlation between crack dynamic changes and their internal stress distribution can be established, thus achieving a comprehensive analysis of crack dynamic changes.

[0043] This solution combines multi-view image acquisition and a pre-trained neural network model to achieve efficient and accurate analysis of the dynamic changes of concrete cracks. The spatial alignment and fusion processing of multi-view images improve the integrity of crack information, the semantic segmentation function of the pre-trained neural network model enhances the accuracy of crack identification, and the skeleton extraction and graph theory optimization further refine the topology of the crack network. The spatial matching of finite element analysis and the crack network topology enables the mechanical mechanism of crack dynamic changes to be clearly presented.

[0044] Referring to Figure 2 , an embodiment of the present invention provides a concrete structure crack image detection device based on deep learning, including: An input module for acquiring multi-view images; An alignment module for performing spatial alignment based on the multi-view images to obtain a sequence of matching images; A fusion module for performing image fusion based on the sequence of matching images to obtain a fused image; A mask generation module for performing semantic segmentation based on the fused image and a preset neural network model through transfer learning to obtain a crack mask; A topology generation module for performing skeleton extraction based on the crack mask and the fused image to obtain a preliminary crack network topology; A topology optimization module for optimizing the preliminary crack network topology to obtain a crack network topology; A stress calculation module for performing finite element analysis based on preset material properties, loading conditions, and concrete structures to obtain a stress distribution contour map; An association construction module, configured to perform spatial matching according to the stress distribution nephogram and the crack network topology to obtain a stress-crack association model.

[0045] It should be noted that the concrete structure crack image detection device based on deep learning provided in the embodiments of the present invention is used to execute all the process steps of the concrete structure crack image detection method based on deep learning in the above embodiments. The working principles and beneficial effects of the two correspond one by one, so they will not be elaborated here.

[0046] Embodiments of the present invention also provide an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above embodiments of the concrete structure crack image detection method based on deep learning are implemented, such as Figure 1 the step S11 shown. Alternatively, when the processor executes the computer program, the functions of each module / unit in the above device embodiments are implemented.

[0047] Exemplarily, the computer program can be divided into one or more modules / units. The one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the electronic device.

[0048] The electronic device can be a computing device such as a desktop computer, a notebook, a palm computer, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device. It may include more or fewer components than the above, or combine some components, or different components. For example, the electronic device may further include input / output devices, network access devices, a bus, etc.

[0049] The so-called processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the electronic device, and connects various parts of the entire electronic device through various interfaces and circuits.

[0050] The memory can be used to store the computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory, the processor realizes various functions of the electronic device. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, at least one magnetic disk storage device, flash device, or other volatile solid-state storage devices.

[0051] Among them, if the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0052] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative effort.

[0053] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only for the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting concrete structure crack images based on deep learning, characterized in that, Including: Obtain multi-view images; Perform spatial alignment based on the multi-view images to obtain a sequence of matching images; Perform image fusion based on the sequence of matching images to obtain a fused image; Based on the fused image and a preset neural network model, perform semantic segmentation based on transfer learning to obtain a crack mask; Based on the crack mask and the fused image, perform skeleton extraction to obtain a preliminary crack network topology; Optimize the preliminary crack network topology to obtain a crack network topology; Perform finite element analysis based on preset material properties, stress states, and concrete structures to obtain a stress distribution contour map; Based on the stress distribution contour map and the crack network topology, perform spatial matching to obtain a stress-crack correlation model.

2. The method for detecting concrete structure crack images based on deep learning according to claim 1, wherein, The performing spatial alignment based on the multi-view images to obtain a sequence of matching images includes: Based on the multi-view images, establish an image acquisition coordinate system to obtain image coordinates; Based on the image coordinates, perform a spatial transformation to obtain a spatial transformation matrix; Based on the spatial transformation matrix and the multi-view images, perform resampling to obtain uniformly-viewed images; Based on the uniformly-viewed images, perform pixel interpolation to obtain complete images; Based on the complete images, perform image denoising to obtain low-noise images; Based on the low-noise images, perform edge detection to obtain a sequence of matching images.

3. The method for detecting concrete structure crack images based on deep learning according to claim 1, characterized in that The performing image fusion based on the sequence of matching images to obtain a fused image includes: Based on the sequence of matching images, perform multi-scale decomposition to obtain multi-scale feature images; Based on the multi-scale feature images and a preset weight rule, perform pixel fusion to obtain a preliminary fused image; Based on the preliminary fused image, perform contrast enhancement to obtain an image with enhanced contrast; Based on the image with enhanced contrast, perform unsharp masking to obtain a clear image; Based on the clear image, perform edge detection to obtain a crack contour; Based on the crack contour and the clear image, perform image overlay to obtain a fused image.

4. The method for detecting concrete structure crack images based on deep learning according to claim 1, characterized in that The performing skeleton extraction based on the crack mask and the fused image to obtain a preliminary crack network topology includes: Based on the crack mask and the fused image, perform masked skeleton extraction to obtain an extracted image; Based on the extracted image, perform morphological filtering to obtain a filtered image; Based on the filtered image, perform connection analysis to obtain a preliminary crack network topology.

5. The method for detecting concrete structure crack images based on deep learning according to claim 1, wherein, The optimizing the preliminary crack network topology to obtain a crack network topology includes: Based on the preliminary crack network topology, perform pruning optimization to obtain a redundant-free topology; Based on the redundant-free topology, perform skeleton segmentation to obtain main skeleton lines; Based on the main skeleton lines and the redundant-free topology, perform branch detection to obtain branch positions; Based on the branch positions, perform geometric calculations to obtain branch lengths and angles; Based on the branch lengths and angles and the redundant-free topology, perform information fusion to obtain a crack network topology.

6. The method for detecting cracks in concrete structure images based on deep learning according to claim 1, characterized in that The performing finite element analysis based on preset material properties, stress states, and concrete structures to obtain a stress distribution contour map includes: Perform finite element mesh division on the concrete structure according to the preset material properties and the stress state to obtain a finite element model; Perform stress calculation according to the finite element model to obtain a stress distribution nephogram.

7. The method for detecting concrete structure crack images based on deep learning according to claim 1, characterized in that Perform spatial matching according to the stress distribution nephogram and the crack network topology to obtain a stress-crack correlation model, including: Perform feature extraction according to the crack network topology to obtain crack features; Construct a stress-crack correlation model according to the crack features and the stress distribution nephogram to obtain a stress-crack correlation model.

8. A concrete structure crack image detection system based on deep learning, characterized in that, Including: An input module for acquiring multi-view images; An alignment module for performing spatial alignment according to the multi-view images to obtain a sequence of matching images; A fusion module for performing image fusion according to the sequence of matching images to obtain a fused image; A mask generation module for performing semantic segmentation based on transfer learning according to the fused image and a preset neural network model to obtain a crack mask; A topology generation module for performing skeleton extraction according to the crack mask and the fused image to obtain a preliminary crack network topology; A topology optimization module for optimizing the preliminary crack network topology to obtain a crack network topology; A stress calculation module for performing finite element analysis according to the preset material properties, stress state and concrete structure to obtain a stress distribution nephogram; An association construction module for performing spatial matching according to the stress distribution nephogram and the crack network topology to obtain a stress-crack correlation model.

9. An electronic device, characterized in that, Including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the deep learning-based concrete structure crack image detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the deep learning-based concrete structure crack image detection method according to any one of claims 1 to 7.

Images