Method and system for quickly positioning and identifying fine cracks of tunnel lining

Abstract

The invention discloses a method and a system for quickly positioning and identifying fine cracks of a tunnel lining, and relates to the technical field of tunnel disease detection. The method comprises the following steps: acquiring a to-be-identified tunnel lining image; screening the tunnel lining image by using a rapid discrimination method for the suspected pixel points to obtain a candidate tunnel lining image; constructing a hybrid model with an attention mechanism based on a convolutional neural network and a state space algorithm, and training the hybrid model to obtain a tunnel lining fine crack detection model; and detecting and positioning the candidate tunnel lining image by using the tunnel lining fine crack detection model to obtain a crack positioning identification result. According to the method, the collected images are quickly distinguished, the detection time is greatly shortened, and the candidate images obtained by distinguishing are subjected to more detailed crack detection, so that the accuracy of crack detection is ensured, and the detection efficiency is integrally improved.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel disease detection, and in particular to a method and system for quickly locating and identifying fine cracks in a tunnel lining. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Tunnels play an irreplaceable role in the overall transportation network. Tunnel disease detection is an important part of ensuring the safety and stability of tunnel structures. The main tunnel diseases include: lining cracks, tunnel water damage, tunnel frost damage, earthquake damage and fire. Among them, lining cracks are one of the more typical and common tunnel diseases, and the manifestation is mostly lining cracks. Therefore, tunnel lining crack detection is one of the most difficult projects in the quality inspection of transportation projects. Existing tunnel lining inspections are mostly carried out by manual inspections using aerial vehicles. The vehicles move at extremely slow speeds. The inspectors observe the tunnel lining with their naked eyes, take pictures of the diseases, and measure and record the location, shape, size and other data of the diseases, which is inefficient. Partially automated tunnel lining disease detection methods are mostly tunnel inspection vehicles. Industrial cameras are used to take pictures of the tunnel lining and collect disease information. The detection efficiency is relatively high and the image resolution is high. However, photos with cracks account for a small number of photos. It will take up a lot of resources to perform detailed inspections on each photo. In addition, conventional image recognition algorithms have problems such as inaccurate recognition in the rapid identification of lining cracks and fine cracks, so they cannot meet the needs of accurate detection of tunnel lining cracks.

[0004] In summary, there are at least the following problems in the prior art: the existing tunnel lining crack detection method has low detection efficiency and inaccurate identification of fine cracks. Therefore, how to ensure detection accuracy and shorten detection time in tunnel lining crack detection has become an urgent problem to be solved in the prior art. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a method and system for quickly locating and identifying fine cracks in tunnel linings, which can quickly identify the collected images, greatly shorten the detection time, and then perform more detailed crack detection on the identified candidate images, thereby ensuring the accuracy of crack detection and improving the overall detection efficiency.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions: A first aspect of the present invention provides a method for quickly locating and identifying fine cracks in a tunnel lining, comprising the following steps: Acquire a tunnel lining image to be identified; The tunnel lining image is screened by using a fast discrimination method for questionable pixels to obtain a candidate tunnel lining image; A hybrid model with attention mechanism is constructed based on convolutional neural network and state space algorithm, and the hybrid model is trained to obtain a tunnel lining fine crack detection model. The tunnel lining subtle crack detection model is used to detect and locate the candidate tunnel lining images to obtain the crack location and recognition results.

[0007] Furthermore, the specific steps of using the rapid discrimination method for questionable pixels to screen the tunnel lining image are as follows: Preprocessing the tunnel lining image to be identified; Set the threshold of the percentage of questionable pixels in the overall pixel count of the image; Compare the proportion of questionable pixels in the preprocessed tunnel lining image with the threshold; Tunnel lining images exceeding the threshold are screened out as candidate tunnel lining images.

[0008] Furthermore, the specific steps of preprocessing the tunnel lining image to be identified are: By performing one or more operations of grayscale processing, image smoothing, logarithmic transformation, bilateral filtering, edge detection, and closed operation processing on the tunnel lining image to be identified, cracks existing in the tunnel lining image to be identified are marked.

[0009] Furthermore, the hybrid model consists of the Unet framework combined with the visual state space block in the Mamba algorithm and incorporates the attention mechanism.

[0010] Furthermore, the specific steps for training the hybrid model are: Obtain known samples including healthy data and crack data, and divide the samples into a training set and a test set; Use the training set to train the constructed hybrid model; The trained hybrid model is tested using the test set.

[0011] The second aspect of the present invention provides a method system for quickly locating and identifying fine cracks in a tunnel lining, comprising: A data acquisition module is configured to acquire a tunnel lining image to be identified; A fast discrimination module is configured to screen the tunnel lining image using a fast discrimination method for questionable pixels to obtain a candidate tunnel lining image; A model building module is configured to build a hybrid model with an attention mechanism based on a convolutional neural network and a state space algorithm, and train the hybrid model to obtain a tunnel lining fine crack detection model; The crack identification module is configured to detect and locate the candidate tunnel lining image using the tunnel lining fine crack detection model to obtain a crack location identification result.

[0012] Furthermore, the fast identification module includes: A preprocessing module is configured to preprocess the tunnel lining image to be identified; An image screening module is configured to set a threshold value of the proportion of questionable pixels in the overall pixels of the image; The proportion of questionable pixels in the preprocessed tunnel lining image is compared with the threshold; tunnel lining images exceeding the threshold are screened out as candidate tunnel lining images.

[0013] Furthermore, the preprocessing module is configured as follows: By performing one or more operations of grayscale processing, image smoothing, logarithmic transformation, bilateral filtering, edge detection, and closed operation processing on the tunnel lining image to be identified, cracks existing in the tunnel lining image to be identified are marked.

[0014] Furthermore, the hybrid model consists of the Unet framework combined with the visual state space block in the Mamba algorithm and incorporates the attention mechanism.

[0015] Furthermore, the model building module is also configured to: Obtain known samples including healthy data and crack data, and divide the samples into a training set and a test set; Use the training set to train the constructed hybrid model; The trained hybrid model is tested using the test set.

[0016] One or more of the above technical solutions have the following beneficial effects: The present invention discloses a method and system for quickly locating and identifying fine cracks in tunnel linings. By pre-processing and judging the collected tunnel health data and crack data, and using a fast discrimination method to perform preliminary crack identification on the image, the selection of crack data images can be quickly completed, and the crack data can be used to train a proposed fine crack locating and identifying model, thereby improving the model's recognition speed for cracks and the accuracy of fine crack identification.

[0017] The tunnel lining fine crack detection system proposed in the present invention integrates the visual state space (VSS) block and the AG (attention gate) unit into the Unet recognition framework, which can improve the overall operation efficiency and resource utilization of the algorithm, improve the recognition effect of fine cracks, and reduce the invalid resource occupation during crack recognition.

[0018] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0020] Figure 1 This is a flow chart of a method for quickly locating and identifying fine cracks in a tunnel lining in Embodiment 1 of the present invention; Figure 2 This is a flow chart of a rapid identification method in Embodiment 1 of the present invention; Figure 3 Schematic diagram of the structure of the tunnel lining fine crack detection model in Example 1 of the present invention.

[0021] Figure 4 This is a schematic diagram of an original image in Embodiment 1 of the present invention; Figure 5 Schematic diagram of a preprocessed image in Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0022] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0023] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "include" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or their combinations; Embodiment 1: Embodiment 1 of the present invention provides a method for quickly locating and identifying fine cracks in a tunnel lining. Figure 1 As shown, it includes starting to input an image, using a tunnel crack rapid identification method to screen it, then using a tunnel lining fine crack detection model to locate and identify the cracks, and then displaying the crack location identification results and recording and saving the data.

[0024] The specific steps include: Step 1: Obtain the tunnel lining image to be identified.

[0025] Step 2: Use the fast discrimination method for questionable pixels to screen the tunnel lining image and obtain the candidate tunnel lining image: Step 2.1: Preprocess the tunnel lining image to be identified.

[0026] Among them, cracks existing in the tunnel lining image to be identified are marked by performing one or more operations of grayscale processing, image smoothing, logarithmic transformation, bilateral filtering, edge detection, and closed operation processing on the tunnel lining image to be identified.

[0027] In a specific implementation, in view of the fact that crack images account for a relatively low proportion in the captured images, the probability of identifying healthy images is reduced, which can reduce resource usage and improve recognition speed. First, the image is quickly identified for cracks to determine whether there are cracks. Since the image data volume is very large, the images in the data set must first be preprocessed to reduce the amount of useless information, highlight the target information, and further reduce the amount of calculation in the image processing process. In this embodiment, the crack image is preprocessed, such as Figure 2 As shown in the figure, the input image is processed in sequence by grayscale processing, image smoothing, logarithmic transformation, bilateral filtering, edge detection and closing operation: (1) Image grayscale. Convert a three-channel image into a single-channel image. Grayscale reduces the amount of image data. The grayscale image retains the same pixel brightness information as the color image, and has only a single channel. The amount of computation required to process the grayscale image is greatly reduced, which improves the processing speed while retaining most of the image information.

[0028] Grayscale formula: .

[0029] in, GRAY Represents the grayscale result, R, G, B Represents the red, green, and blue channel values ​​respectively.

[0030] (2) Contrast enhancement. Due to factors such as the image acquisition device and the acquisition environment, the contrast between the cracks and the background of the initially acquired image is low, so the contrast needs to be enhanced. In order to enhance the contrast between the cracks and the background, this embodiment compares the effects of histogram equalization, piecewise linear method, and gamma correction processing, and finally selects gamma correction for contrast enhancement processing.

[0031] Gamma correction is a technique in digital image processing that is used to adjust the contrast and brightness of an image. It is based on the observation that the human eye's perception of brightness is not linear, but follows a nonlinear relationship. This nonlinear relationship can be described by the Gamma value.

[0032] The basic idea of ​​gamma correction is to compensate for the brightness characteristics of the display device through a power law conversion. When an image is sent from a computer to a display device (such as a monitor or projector), it usually undergoes a nonlinear transformation.

[0033] The basic formula for gamma correction can be described as: .

[0034] Among them, c is a positive number, generally 1, r is the pixel value of the input image (normalized value), s is the pixel value of the output image (also normalized value), is the Gamma value. Sometimes, considering the offset, the expression is written as:

[0035] in It is the offset, and its parameters are the same as above.

[0036] (3) Filtering and denoising. The road surface itself has various irregular noises, and the camera will introduce some new noises in the process of acquiring images. A large amount of noise greatly increases the difficulty of detecting and extracting cracks. It is necessary to reduce the noise interference in the image through filtering algorithms. The effects of mean filtering, median filtering, Gaussian filtering and bilateral filtering on image processing were compared. Finally, combined with the characteristics of road noise and the characteristics of various filtering algorithms, denoising was completed by first performing median filtering and then bilateral filtering, reducing the difficulty of detecting the presence of cracks and extracting cracks.

[0037] Median filtering is a nonlinear digital filter technology that is often used to remove noise from images or other signals. It is a common noise reduction algorithm for processing salt and pepper noise. Median filtering sets the grayscale value of each pixel to the median of the grayscale values ​​of all pixels in a certain area window of the point. While using the neighborhood average method to remove noise, the boundary is also blurred. Median filtering is a nonlinear image processing method that can take into account the retention of boundary information while removing noise. Select a window W containing an odd number of points, scan this window on the image, arrange the pixels contained in the window in ascending or descending order of grayscale, and take the grayscale value in the middle to replace the grayscale value of the point. Bilateral filtering is also a nonlinear filtering method. It is a compromise between the spatial proximity and pixel value similarity of the image, and takes into account the spatial information and grayscale similarity to achieve the purpose of edge preservation and denoising. Compared with Gaussian filtering, bilateral filtering has an additional Gaussian variance sigma-d. It is a Gaussian filtering function based on spatial distribution. Therefore, near the edge, the pixels farther away will not affect the pixel values ​​on the edge too much, thus ensuring the preservation of the pixel values ​​near the edge. Therefore, bilateral filtering can ensure certain edge information while filtering, providing better edge preservation characteristics than Gaussian filtering.

[0038] (4) The Canny edge detection algorithm is used to detect the edge of the image. The main principles of the Canny edge detection algorithm are as follows: ① Noise suppression (noise removal): Since edge detection is easily affected by noise, before edge detection, a Gaussian filter is generally used to smooth the image (usually with a 5*5 kernel) to reduce the details of the edge. The influence of noise. The Gaussian filter can effectively blur the image while retaining the details of the edge.

[0039] ② Gradient calculation: Use the Sobel operator (Sobel operator is a linear filter that can detect edges in an image) to calculate the first-order derivative (image gradient) of the image in the horizontal and vertical directions. and Find the gradient and direction of the boundary, the formula is as follows: , .

[0040] Among them, Gx and Gy represent the images after horizontal and vertical edge detection respectively, G represents the size of the gradient, is an angle, indicating the direction of the gradient.

[0041] The direction of the gradient is always perpendicular to the boundary. Gradient directions are classified into four categories: vertical, horizontal, and two diagonal.

[0042] ③ Non-maximum suppression: After obtaining the direction and magnitude of the gradient, the entire image should be scanned to remove those points that are not on the boundary. Each pixel is checked to see if the gradient of this point is the largest among the surrounding points with the same gradient direction (the pixel with the largest gradient is selected as the edge point, thereby suppressing the non-edge point).

[0043] ④Dual threshold detection (hysteresis threshold): In order to determine which boundaries are real boundaries, two thresholds minVal and maxVal need to be set. When the grayscale gradient of the image is higher than maxVal, it is considered to be a real boundary, and those below minval will be discarded. If it is between the two, it depends on whether this point is connected to a point that is determined to be a real boundary point. If it is, it is considered to be a boundary point, and if not, it is discarded.

[0044] ⑤Edge connection: By iteratively visiting weak edge pixels and connecting them with their adjacent strong edge pixels, the true edge is finally determined.

[0045] (5) Finally, the edge image is morphologically closed using the morphological closing operator.

[0046] Closing is an operation in image morphology that combines dilation and erosion operations. The principle of closing is to first perform an erosion operation on the image and then a dilation operation. This process can eliminate small holes in the image, fill small breaks, smooth the boundaries of objects, and connect narrow gaps.

[0047] The closing operation can be expressed mathematically as: .

[0048] in, is the closing function, A is the input image, B is the structural element, represents the corrosion operation, Represents an expansion operation.

[0049] After image preprocessing, a feature detection method is created. This embodiment uses the ORB algorithm for feature detection. Then, the program uses the ORB algorithm to detect and calculate the key points and descriptors of the image, and draws the feature image of the key points.

[0050] ORB first looks for special areas in the image, called keypoints. Keypoints are small areas of the image that stand out, such as corners, and have a characteristic where the pixel values ​​change dramatically from light to dark. ORB then calculates a corresponding feature vector for each keypoint. The feature vector created by the ORB algorithm contains only 1s and 0s, called a binary feature vector. The order of 1s and 0s varies depending on the specific keypoint and the pixel area around it. The vector represents the intensity pattern around the keypoint, so multiple feature vectors can be used to identify larger areas or even specific objects in the image.

[0051] After feature detection, a feature image is obtained which marks the suspicious pixels.

[0052] Finally, the program saves the output feature image as a file and uses the plot function to display the original image and the output image. The original image is as follows: Figure 4 As shown, the output image is Figure 5 shown.

[0053] Step 2.2: Set a threshold value for the percentage of questionable pixels in the overall pixel points of the image. In this embodiment, K is used to represent the environmental threshold, which represents the percentage of quality and safety standards corresponding to different detection environments and standards, and is determined according to different situations.

[0054] Step 2.3: Compare the percentage of questionable pixels in the preprocessed tunnel lining image with the threshold.

[0055] like Figure 2As shown in the figure, the definition formula for calculating the crack discrimination rate is as follows: α=D / A×100%.

[0056] Among them, D is the number of questionable pixels in the generated image; A is the number of pixels in the entire image, and α is the discrimination percentage parameter according to standards such as tunnel construction and building construction.

[0057] Step 2.4: Filter out tunnel lining images exceeding the threshold as candidate tunnel lining images.

[0058] If α≥K, that is, the discrimination percentage parameter exceeds the environmental threshold K, the image is identified as an image that needs to be focused on and the corresponding image data is put into the data set to be detected. Otherwise, the data is removed from the data set to be detected.

[0059] Step 3: Based on the convolutional neural network and state space algorithm, a hybrid model with an attention mechanism is constructed, and the hybrid model is trained to obtain a tunnel lining fine crack detection model. Based on the rapid discrimination method, this part further identifies and quickly locates the cracks in the images with cracks that are quickly detected.

[0060] Step 3.1: Construct a hybrid model with attention mechanism based on convolutional neural network and state space algorithm. Figure 3 As shown in the figure, the hybrid model consists of the Unet framework combined with the visual state space (VSS) block in the Mamba algorithm and the attention mechanism (AG).

[0061] Based on the research of Swin-Unet, this embodiment replaces the existing Swin Transformer block with the VSS (visual state space) block in the Mamba model, which can improve the overall operation efficiency and resource utilization of the algorithm. At the same time, consider adding an AG (attention gate) unit to the Unet structure, including weighting of features of different dimensions, activation functions, etc., so as to realize the analysis of features of different dimensions and improve the recognition effect of subtle cracks.

[0062] The two main features of the Unet network structure are: U-shaped network structure and Skip Connection. Unet combines shallow features and deep features through a skip-connected U-shaped network structure for the final semantic segmentation map generation. Unlike FCN, UNet combines shallow features and deep features in a splicing manner. FCN combines shallow features and deep features in an additive manner. The U-shaped network architecture can more fully integrate shallow features and deep features, which is the main reason why UNet performs better than FCN. Shallow feature maps tend to express basic feature units such as points, lines, and edge contours, and contain more spatial information. Deep feature maps tend to express the semantic information of images, contain less spatial information, and have more semantic features.

[0063] The backbone of UNet is divided into two symmetrical parts: the left side is the feature extraction network (encoder), in which the original input image is downsampled four times through convolution-max pooling to obtain a four-level feature map; the right side is the feature fusion network (decoder), in which the feature maps of each level are fused with the feature maps obtained by deconvolution through jump connection, and the last layer predicts the semantic map by calculating the loss with the label.

[0064] The VSS network block is mainly based on the visual Mamba. Specifically, the traditional SSM is a linear time-invariant system, which is constructed by hidden states h(t) ∈ R N , maps x(t) ∈ R to y(t) ∈ R, where x(t) and y(t) are both uncertain functions. Given A ∈ C N ×N as evolution parameters, B, C ∈ C N as the projection parameter of the state size N, and the skip connection D ∈ C 1 The model can be formulated as linear ordinary differential equations (ODEs) as follows: .

[0065] Where h(t) is the hidden state function and h'(t) is the derivative of h(t) with respect to time t.

[0066] Visual Mamba further introduces a Cross Scan Module (CSM) and then integrates the convolution operation into the block. In the VSS block, the input features first pass through a linear embedding layer and then split into two paths. One branch undergoes depthwise convolution and SiLU activation, then enters the SS2D module and merges with the other branch after SiLU activation after layer normalization. Instead of using positional embedding like typical visual transformers, this VSS block chooses a streamlined structure without an MLP stage, allowing denser blocks to be stacked within the same depth budget.

[0067] CBAM module: CBAM is a feed-forward convolutional neural network attention module that sequentially infers attention maps along the channel and spatial dimensions, and then multiplies the attention maps by the input feature maps for adaptive feature refinement. Its computational overhead is negligible and it can be seamlessly integrated into any CNN architecture.

[0068] The specific algorithm framework CBAM module consists of two attention modules: Channel Attention Module and Spatial Attention Module. Channel attention uses global average pooling and global maximum pooling to obtain global statistics of each channel respectively, and learns the weight of the channel through two fully connected layers. Then, the two processed results are added together, and each channel is scaled by normalizing the weights to between 0 and 1 using the Sigmoid function. Finally, the scaled channel features are multiplied with the original features to produce features with enhanced channel importance. The spatial attention module uses maximum pooling and average pooling to obtain the maximum and average values ​​of each spatial position. Specifically, since multiple channels are generated after convolution, the spatial attention in CBAM performs maximum pooling and average pooling operations on the channel of each feature point. After obtaining two matrices, the two matrices are concatenated, and the weight of each spatial position is learned through a convolution layer and a Sigmoid function. Finally, the weights are applied to each spatial position on the feature map to produce features with enhanced spatial importance.

[0069] Based on the above research, the VSS (visual state space) block in the Mamba model is integrated into the Unet model to improve the overall operation efficiency and resource utilization of the algorithm. At the same time, the CBAM module is considered to be added to the Unet structure, including the weighting of features of different dimensions, activation functions, etc., so as to realize the analysis of features of different dimensions and improve the recognition effect of fine cracks.

[0070] Step 3.2: Train the hybrid model.

[0071] Step 3.2.1: Obtain known samples including healthy data and crack data, and divide the samples into training set and test set.

[0072] Step 3.2.2: Use the training set to train the constructed hybrid model.

[0073] Step 3.2.3: Use the test set to test the trained hybrid model.

[0074] Step 4: Use the tunnel lining fine crack detection model to detect and locate the candidate tunnel lining image to obtain the crack location and recognition results.

[0075] Specifically, Figure 3 As shown in the figure, the original two-dimensional RGB image is first input into the Patch Partition module, which divides the input image into patches, with each 4x4 adjacent pixels as a patch, and then flattens them in the channel direction, and converts each patch into an embedding vector as the input of the model. Then, the linear transformations such as linear correlation and basis transformation are performed on the channel data of each pixel through the Linear Embedding layer. Then, the encoder composed of VSS blocks and the like is passed through the encoder, which consists of four stages. At the end of the last three stages, patch merging operations are applied to reduce the height and width of the features while increasing the number of channels. Similarly, the decoder consists of four stages. At the beginning of the first three stages, patch expansion operations are used to reduce the number of feature channels and increase the height and width. After the decoder, the input image is divided into small patches, which are then linearly projected and converted into feature vectors to achieve the output of the picture. The decoder consists of VSS blocks and patch expansion layers, following the style of the encoder, so that the output feature size is exactly the same, thereby enhancing the spatial details lost during downsampling through skip connections. In the process of skip connection, the CBAM attention module is innovatively used. After two different pooling methods, global maximum pooling and global average pooling, two 1×1×C channel features are obtained. Then the two channel features are sent to a shared two-layer neural network composed of a multi-layer perceptron and a hidden layer, and the activation function is ReLU. After the elements of the two output features are multiplied and added, the required channel attention map can be obtained through the Sigmoid activation function, and the importance of the features is enhanced by weights to improve recognition accuracy. In the encoder and decoder, 2 VSS blocks are used in each stage, and the pre-trained VMamba-Tiny is loaded in the encoder.

[0076] Embodiment 2: Embodiment 2 of the present invention provides a method system for quickly locating and identifying fine cracks in a tunnel lining, comprising: The data acquisition module is configured to acquire the tunnel lining image to be identified.

[0077] The fast discrimination module is configured to screen the tunnel lining image by using a fast discrimination method for questionable pixels to obtain a candidate tunnel lining image.

[0078] The model building module is configured to build a hybrid model with an attention mechanism based on a convolutional neural network and a state space algorithm, and train the hybrid model to obtain a tunnel lining fine crack detection model, wherein the hybrid model is composed of a Unet framework combined with a visual state space block in the Mamba algorithm and an attention mechanism.

[0079] The crack identification module is configured to detect and locate the candidate tunnel lining image using the tunnel lining fine crack detection model to obtain a crack location identification result.

[0080] The fast identification module includes: A preprocessing module is configured to preprocess the tunnel lining image to be identified; An image screening module is configured to set a threshold value of the proportion of questionable pixels in the overall pixels of the image; The proportion of questionable pixels in the preprocessed tunnel lining image is compared with the threshold; tunnel lining images exceeding the threshold are screened out as candidate tunnel lining images.

[0081] The preprocessing module is also configured to: By performing one or more operations of grayscale processing, image smoothing, logarithmic transformation, bilateral filtering, edge detection, and closed operation processing on the tunnel lining image to be identified, cracks existing in the tunnel lining image to be identified are marked.

[0082] The model building module is also configured to: Obtain known samples including healthy data and crack data, and divide the samples into a training set and a test set; Use the training set to train the constructed hybrid model; The trained hybrid model is tested using the test set.

[0083] The steps involved in the above embodiment 2 correspond to those in the method embodiment 1. For the specific implementation method, please refer to the relevant description part of the embodiment 1.

[0084] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0085] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for quickly locating and identifying fine cracks in tunnel lining, characterized in that: The following steps are involved: Acquire a tunnel lining image to be identified; The tunnel lining image is screened by using a fast discrimination method for questionable pixels to obtain a candidate tunnel lining image; A hybrid model with attention mechanism is constructed based on convolutional neural network and state space algorithm, and the hybrid model is trained to obtain a tunnel lining fine crack detection model. The tunnel lining subtle crack detection model is used to detect and locate the candidate tunnel lining images to obtain the crack location and recognition results.

2. The method for rapid location and identification of fine cracks in tunnel lining according to claim 1, characterized in that: The specific steps of screening tunnel lining images using the rapid discrimination method for questionable pixels are as follows: Preprocessing the tunnel lining image to be identified; Set the threshold of the percentage of questionable pixels in the overall pixel count of the image; Compare the proportion of questionable pixels in the preprocessed tunnel lining image with the threshold; Tunnel lining images exceeding the threshold are screened out as candidate tunnel lining images.

3. The method for rapid location and identification of fine cracks in tunnel lining according to claim 2, characterized in that: The specific steps for preprocessing the tunnel lining image to be identified are: By performing one or more operations of grayscale processing, image smoothing, logarithmic transformation, bilateral filtering, edge detection, and closed operation processing on the tunnel lining image to be identified, cracks existing in the tunnel lining image to be identified are marked.

4. The method for rapid location and identification of fine cracks in tunnel lining according to claim 1, characterized in that: The hybrid model consists of the Unet framework combined with the visual state space block in the Mamba algorithm and incorporates the attention mechanism.

5. The method for rapid location and identification of fine cracks in tunnel lining according to claim 1, characterized in that: The specific steps for training the hybrid model are: Obtain known samples including healthy data and crack data, and divide the samples into a training set and a test set; Use the training set to train the constructed hybrid model; The trained hybrid model is tested using the test set.

6. A method system for quickly locating and identifying fine cracks in tunnel lining, characterized in that: include: A data acquisition module is configured to acquire a tunnel lining image to be identified; A fast discrimination module is configured to screen the tunnel lining image using a fast discrimination method for questionable pixels to obtain a candidate tunnel lining image; A model building module is configured to build a hybrid model with an attention mechanism based on a convolutional neural network and a state space algorithm, and train the hybrid model to obtain a tunnel lining fine crack detection model; The crack identification module is configured to detect and locate the candidate tunnel lining image using the tunnel lining fine crack detection model to obtain a crack location identification result.

7. The method system for quickly locating and identifying fine cracks in tunnel lining according to claim 6, characterized in that: The fast identification module includes: A preprocessing module is configured to preprocess the tunnel lining image to be identified; An image screening module is configured to set a threshold value of the proportion of questionable pixels in the overall pixels of the image; The proportion of questionable pixels in the preprocessed tunnel lining image is compared with the threshold; tunnel lining images exceeding the threshold are screened out as candidate tunnel lining images.

8. The method system for quickly locating and identifying fine cracks in tunnel lining according to claim 7, characterized in that: The preprocessing module is also configured to: By performing one or more operations of grayscale processing, image smoothing, logarithmic transformation, bilateral filtering, edge detection, and closed operation processing on the tunnel lining image to be identified, cracks existing in the tunnel lining image to be identified are marked.

9. The method system for quickly locating and identifying fine cracks in tunnel lining according to claim 6, characterized in that: The hybrid model consists of the Unet framework combined with the visual state space block in the Mamba algorithm and incorporates the attention mechanism.

10. The method system for quickly locating and identifying fine cracks in tunnel lining according to claim 6, characterized in that: The model building module is also configured to: Obtain known samples including healthy data and crack data, and divide the samples into a training set and a test set; Use the training set to train the constructed hybrid model; The trained hybrid model is tested using the test set.

Images