CN111402227B - Bridge crack detection method - Google Patents

Abstract

The invention discloses a bridge crack detection method, belongs to the technical field of bridge detection, and aims to improve the detection accuracy and efficiency of bridge crack detection. The method includes the following steps: performing crack segmentation on a group of collected bridge images; and detecting and classifying bridge cracks by using a pre-built bridge crack classification model according to the crack segmentation results. The segmentation processing of bridge cracks adopts an improved GAC algorithm model to segment the visible cracks in the bottom image of the bridge captured by the high-definition camera of the UAV; the construction of the bridge crack classification model adopts the deep learning method, and designs a The integrated neural network model is used for the identification of bridges; the three-dimensional reconstruction of bridge cracks and the detection of crack information use a moving cube algorithm to determine the number, average width, geometric properties of cracks and their spatial relationship with the whole, so that professionals can Qualitative or quantitative analysis of cracks. The present invention realizes the use of deep learning-based computer detection technology to solve construction problems such as corresponding crack detection.

Description

A method of bridge crack detection

technical field

The invention relates to a bridge crack detection method, which belongs to the technical field of bridge detection.

Background technique

At present, bridge crack detection and maintenance mainly rely on manual detection. Manual detection methods are time-consuming and require a lot of human, material and financial resources, not only the detection accuracy is low, and the human influence factor is large, but in many cases it is impossible to visually detect cracks due to the inaccessibility of the area or the microscopic size of the cracks.

SUMMARY OF THE INVENTION

In order to overcome the deficiencies in the prior art, the present invention provides a bridge crack detection method, which can improve the bridge crack detection efficiency and accuracy.

In order to achieve the above object, adopt the following technical scheme to realize during the present invention:

A method for detecting bridge cracks, the method comprising the steps of:

Perform crack segmentation on a set of collected bridge images;

According to the crack segmentation results, a pre-built bridge crack classification model is used to detect and classify bridge cracks.

Further, the method for performing crack segmentation on the collected bridge image includes the following steps:

A spectral clustering algorithm based on Nystrom approximation theory is used to segment the cracks in each bridge image as the initial contour;

Along the upward direction, map the segmented cracks to the next bridge image in turn, as the initial outline of the cracks in the bridge image, and use the improved GAC model to complete the segmentation of each crack until the end of all bridge image segmentation;

Along the downward direction, map the cracks segmented along the upward direction to the next bridge image in turn, as the initial outline of the cracks in the bridge image, and use the improved GAC model to complete the segmentation of each crack until all bridges are Image segmentation ends.

Further, the method for completing the segmentation of each fracture by using the improved GAC model includes the following steps:

Calculate the grayscale mean and grayscale standard deviation of each segmented crack region, and use the grayscale mean and grayscale standard deviation as the grayscale similarity information of each crack region;

Construct the grayscale similarity information item according to the grayscale similarity information;

The GAC model is improved by adding the gray-level similarity information term as an external energy term to the energy functional of the GAC model.

Further, the construction method of the bridge crack classification model includes the following steps:

Collection of original bridge images with various types of cracks;

Perform crack marking and crack segmentation on the collected original bridge images to construct a sample dataset of bridge cracks;

Combined with the global features of the bridge, an 8-layer deep convolutional neural network is initially constructed, including one input layer, three convolution layers, three pooling layers, two fully connected layers and one output layer, using a softmax classifier;

The constructed deep convolutional neural network is trained and tested using the sample data set to determine the structure and parameters of the bridge crack classification model.

Further, data cleaning is performed on the original bridge images to remove unlabeled bridge images.

Further, the method further includes: before performing crack segmentation, converting the bridge image into a grayscale image, and performing enhancement and normalization processing on the grayscale image.

Further, the method also includes:

According to the crack segmentation results, 3D reconstruction of bridge cracks is carried out, so as to obtain the 3D visualization effect of cracks.

Further, a moving cube algorithm is used to perform three-dimensional reconstruction of bridge cracks and crack information detection.

Further, during crack segmentation, the local extrema corresponding to noise and irregular details are eliminated.

Further, the types of cracks include: plastic cracks, shrinkage cracks, radial cracks in arch bridges, longitudinal cracks in the gap between boxes under the abdomen, pier cap cracks, and arch foot cracks.

Compared with the prior art, the present invention at least has the following beneficial effects:

Using the bridge crack classification model to detect and classify bridge cracks can significantly improve the accuracy and efficiency of bridge crack detection, and realize the use of deep learning-based computer detection technology to solve construction problems such as crack detection.

Description of drawings

1 is a flowchart of a method for detecting bridge cracks provided according to an embodiment of the present invention;

2 is a flowchart of a method for dividing a bridge crack according to an embodiment of the present invention;

3 is a flowchart of a method for constructing a bridge crack classification model provided according to an embodiment of the present invention;

FIG. 4 is a frame structure diagram of an algorithm platform suitable for the method of the present invention provided according to an embodiment of the present invention.

Detailed ways

The embodiment of the present invention includes four parts: a method for segmenting and processing bridge cracks, establishing a bridge crack classification model, three-dimensional reconstruction of bridge cracks and crack information detection, and building a bridge crack detection platform. The segmentation processing of bridge cracks adopts an improved GAC algorithm model to segment the visible cracks in the bottom image of the bridge captured by the high-definition camera of the UAV; the construction of the bridge crack classification model adopts the deep learning method, and designs a The integrated neural network model is used for the identification of bridges; the three-dimensional reconstruction of bridge cracks and the detection of crack information use a moving cube algorithm to determine the number, average width, geometric properties of cracks and their spatial relationship with the whole, so that professionals can Qualitative or quantitative analysis of cracks; construction of bridge crack detection platforms, including high-performance building image processing platforms featuring deep learning bridge crack detection technology. Based on this platform, key issues such as bridge crack detection, processing and analysis are studied. The research of the present invention is carried out from the above four aspects, and realizes the use of deep learning-based computer detection technology to solve construction problems such as corresponding crack detection.

The present invention will be further described below in conjunction with the accompanying drawings. The following examples are only used to illustrate the technical solutions of the present invention more clearly, and cannot be used to limit the protection scope of the present invention.

As shown in FIG. 1, the bridge crack detection method provided by the embodiment of the present invention includes the following steps:

A set of bridge images was captured using a drone high-definition camera. During the flight of the drone, the safe distance between different drones and the detection site is controlled according to the difference of the detection object: the bridge piers and towers are generally controlled at about 5 meters, and the complex terrain such as cables and steel components is generally controlled at about 10 meters. , the specific safety distance needs to be determined according to the site conditions; generally, the light-facing surface of the detection object is selected for image acquisition to avoid backlight and improve the success rate of data acquisition; generally, detection is performed in a sunny and windless environment to minimize the impact of environmental factors and reduce Safety risks in the detection process; the time interval for drone shooting is not more than 3s; the distance between the height of the drone and the wall beam has no obvious change during the shooting time, and the angle difference between adjacent pictures is ignored.

Step 1: Segment the bridge cracks on a set of bridge images:

Step1: Use the spectral clustering algorithm (Spectral Clustering algorithm based on Nystrom, SCN) based on Nystrom approximation theory to segment the visible cracks in each image as the initial contour;

Step2: Along the upward direction, map the segmented visible cracks to the next image in turn, as the initial outline of the cracks in the image, and then use the improved GAC model to complete the segmentation of each crack until all image segmentation ends ;

Step3: Along the downward direction, map the segmented visible cracks to the next image in turn, as the initial contour of the crack in the C image, and then use the improved GAC model to complete the segmentation of each crack until all images are Split ends.

Since the shooting time between adjacent pictures in each group of bridge images is short, and the gap between adjacent positions and time is small, the corresponding grayscale information between adjacent pictures changes slowly. Therefore, the gray level information of the segmented cracks can be used as similarity information to guide the segmentation of unsegmented cracks, which helps to improve the segmentation accuracy and efficiency of bridge cracks to a certain extent. In the embodiment of the present invention, the concrete method that adopts the improved GAC model to complete the segmentation of each crack is:

It is proposed to first calculate the gray mean and gray standard deviation of the segmented crack area, take the gray mean and gray standard deviation as the gray similarity information of the crack area, and construct the gray similarity according to the gray similarity information. information term; then, this term is added to the energy functional of the GAC model as an external energy term, thereby enabling improvements to the GAC model.

Step 2: Use the pre-built bridge crack classification model to classify and detect bridge cracks;

The specific method of constructing the bridge crack classification model is as follows:

(1) Data collection: Collect and organize 3000 original bridge images, and select 400 bridge images of plastic cracks, shrinkage cracks, radial cracks in arch bridges, longitudinal cracks in the gap between the lower abdominal boxes, pier cap cracks, and arch foot cracks according to the markers. ; 1800 cases are used as training data and 1200 cases are used as test data.

(2) Image preprocessing: data cleaning is performed on all the collected data, unlabeled data is removed, the collected images are converted into grayscale images, and images with weak contrast are enhanced, and then normalized to the same size of experimental data;

(3) Crack segmentation processing: Using the above crack segmentation processing method, the cracks in the image are segmented and extracted, and the bridge crack sample data set is constructed for DCNN training and testing;

(4) Construction of DCNN: A deep convolutional neural network with 8 layers (excluding input and output layers) is initially constructed by combining the global features of the bridge, including one input layer, three convolution layers, three pooling layers, and two full layers. Connection layer and one output layer, using softmax classifier;

(5) Discussion on different model parameters of the same model structure: For the bridge global feature sample space set, the influence of input images with different spatial resolutions and different iteration times on the DCNN recognition rate and training time was discussed;

(6) Discussion on different model structures: On the basis of the initially constructed 8-layer network structure and global features, by changing the size of the convolution kernel, the number of feature maps and the number of network layers, different model structures are used to identify bridge cracks;

(7) Comparative analysis of different optimization algorithms: After selecting the appropriate model structure, compare and analyze the pooling method (mean sampling and maximum sampling), activation function (Sigmoid function and Re LU function), and training algorithm (batch gradient descent method and band The influence of the gradient descent method with elastic momentum) on the recognition results;

(8) Decision evaluation: Through comparative experiments, the parameters and structures of different models are analyzed and discussed to provide a reference for constructing a suitable deep convolutional neural network for the computer-aided detection of bridge cracks, so as to improve the recognition performance and enhance the network robustness. and generalization ability.

Step 3: 3D reconstruction of bridge cracks and crack information detection;

The volume dataset of images consists of a series of 2D slices. Assuming that the resolution of each slice is M×N, and the number of slices (including virtual slices) is L, then these slices form a spatial discrete data field with a resolution of M×N×L. This data field can be regarded as the result obtained by sampling the continuous function f(x, y, z) in the x, y, and z directions at certain intervals. If the volume data is regarded as a sampling set of a certain object property in a spatial region, and the value at a non-sampling point is estimated by the interpolation of its adjacent sampling points, then the set of points with a certain same value in the spatial region will be form an isosurface. Since the gray values of different cracks are different in the image, when the appropriate value is selected to define the isosurface, the three-dimensional reconstruction of different cracks can be realized. In the embodiment of the present invention, the three-dimensional reconstruction of bridge cracks and the detection of crack information use a moving cube algorithm to determine the number, average width, geometric properties of cracks and their spatial relationship with the whole, so that professionals can qualitatively or quantitatively determine cracks. analyze.

As shown in FIG. 4 , an embodiment of the present invention also provides a bridge crack detection platform, which can be used for the aforementioned bridge crack detection method. The platform’s powerful computing and storage capabilities can not only meet the needs of bridge crack research, but also It can provide technical support for other image applications. Based on this platform, we will carry out research featuring bridge crack research, and achieve and maintain the international leading level in some aspects, so as to promote my country's leap-forward development in the field of crack detection. The overall framework of the algorithm platform is divided into three layers: the bottom layer, the middle layer, and the application layer.

(1) Bottom layer

The bottom layer is mainly built using mature open source algorithm libraries such as ITK, VTK, and FSL. It mainly retains ITK's existing basic algorithms such as image reading and writing, filtering, registration, segmentation, and format conversion. For VTK, because VTK is a software package developed for the general visualization field, not only for the field of crack detection, some of the algorithms and data structures are not required for crack detection, and these huge and complicated algorithms and data structures. Greatly increases the difficulty of users. Therefore, the functions of surface rendering, mesh rendering and volume rendering commonly used in 3D display of crack detection are mainly reserved in this algorithm platform. The underlying algorithms also include FSL and self-developed algorithms. In addition, if there are other excellent crack detection processing and analysis algorithm libraries, they can also be integrated into the underlying algorithm.

(2) Intermediate layer

Based on the underlying algorithm, some commonly used basic algorithms are connected, combined, and encapsulated into relatively complete and practical application functions. The middle layer summarizes the underlying algorithm into five parts, which are format conversion, filtering, registration, segmentation and display of crack images. The middle layer reserves a unified interface for the same type of algorithms to facilitate the application layer invocation.

(3) Application layer

In the application layer, the executable program provided by the middle layer is recombined using the graphical interactive interface to form three modules with interactive functions, which are format conversion, registration and segmentation, and display interaction modules, which are convenient for users to use.

As will be appreciated by those skilled in the art, the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

The above are only the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the technical principle of the present invention, several improvements and modifications can also be made. These improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (6)

1. A bridge crack detection method is characterized by comprising the following steps:

carrying out crack segmentation on the collected group of bridge images;

detecting and classifying the bridge cracks by adopting a pre-constructed bridge crack classification model according to the crack segmentation result;

according to the crack segmentation result, performing three-dimensional reconstruction on the bridge crack so as to obtain the three-dimensional visualization effect of the crack;

performing three-dimensional reconstruction and crack information detection on the bridge cracks by adopting a moving cube algorithm;

the method for carrying out crack segmentation on the acquired bridge image comprises the following steps:

dividing cracks in each bridge image to serve as initial contours by adopting a spectral clustering algorithm promoted based on Nystrom approximation theory;

sequentially mapping the divided cracks to the next bridge image along the upward direction to serve as the initial contour of the crack in the bridge image, and finishing the division of each crack by adopting an improved GAC model until the division of all bridge images is finished;

sequentially mapping the cracks segmented along the upward direction to the next bridge image along the downward direction to serve as initial contours of the cracks in the bridge image, and completing segmentation of each crack by adopting an improved GAC model until all bridge images are segmented;

the method for completing the segmentation of each crack by adopting the improved GAC model comprises the following steps:

calculating the gray level mean value and the gray level standard deviation of each divided crack area, and taking the gray level mean value and the gray level standard deviation as the gray level similarity information of each crack area;

constructing a gray level similarity information item according to the gray level similarity information;

and adding the gray level similarity information item as an external energy item to an energy functional of the GAC model, thereby improving the GAC model.

2. The bridge crack detection method of claim 1, wherein the bridge crack classification model construction method comprises the following steps:

collecting original bridge images containing various cracks;

carrying out crack marking and crack segmentation on the collected original bridge image to construct a sample data set of bridge cracks;

initially constructing 8 layers of deep convolutional neural networks by combining global characteristics of a bridge, wherein each deep convolutional neural network comprises a first input layer, a third convolutional layer, a third pooling layer, a second full-connection layer and an output layer, and a softmax classifier is adopted;

and training and testing the constructed deep convolution neural network by adopting the sample data set so as to determine the structure and parameters of the bridge crack classification model.

3. The bridge crack detection method of claim 2, wherein the original bridge image is subjected to data cleaning to remove unmarked bridge images.

4. The bridge crack detection method of claim 1 or 3, further comprising: before crack segmentation is carried out, the bridge image is converted into a gray image, and enhancement and normalization processing are carried out on the gray image.

5. The bridge crack detection method of claim 1 wherein local extrema corresponding to noise and irregular detail are eliminated during crack segmentation.

6. The bridge crack detection method of claim 1, wherein the types of cracks comprise: plastic cracks, shrinkage cracks, arch bridge radial cracks, longitudinal cracks at the gap between the underbelly boxes, pier cap cracks and arch foot cracks.

Images