Method for identifying fine-grained image of surrounding rock mass structure of tunnel face

Abstract

The invention provides a tunnel face surrounding rock mass structure fine-grained image recognition method. The method comprises the following steps of obtaining a tunnel face image and performing image preprocessing; zooming the tunnel face image to input a tunnel face region segmentation model; restoring a region segmentation result to an original size to obtain a tunnel face region in the original image, and performing tiling processing on the tunnel face region to obtain a tile-level tunnel face image; performing rock mass structure classification on the tile-level image, dividing the tile-level image into five types including a mosaic structure, a discrete structure, a layered structure, a complete structure and a fragmented structure, and obtaining a rock mass structure classification result of each tile; and the classification results are spliced according to the original positions of the tiles, a tunnel face rock mass structure distribution diagram is obtained, and a tunnel face rock mass structure distribution conclusion is obtained. The invention belongs to the technical field of geographic information systems, and particularly relates to a tunnel face surrounding rock mass structure fine-grained image recognition method. Therefore, the purpose of automatically identifying the tunnel face rock mass structure is achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of geographic information systems, and specifically refers to a method for fine-grained image recognition of surrounding rock mass structure of a tunnel face. Background Art

[0002] The amount of railway and highway construction in my country continues to grow, and the number and scale of tunnel projects are also increasing. The construction environment faced during construction is more complex, the geological conditions are more diverse, and the safety risks are higher. During the tunnel construction process, geological disasters such as landslides, sudden water gushing, and rock bursts occur from time to time, posing a serious threat to engineering construction. The face refers to the working face of the tunnel. As the project progresses, the geological surrounding rock information contained in the face is constantly revealed. It is an important basis for evaluating and classifying the surrounding rock, reflecting the safety status of the tunnel. Therefore, it is crucial to identify the surrounding rock conditions of the face in a timely and accurate manner. This is the key to ensuring safe and smooth construction.

[0003] However, the current recording of tunnel face information mainly relies on visual judgment, and tunnel engineers usually rely on their own experience and judgment to record the geological information of the tunnel face. Although this recording method can quickly obtain data, it has a low degree of intelligence, and the recording results are often affected by personal subjective factors, resulting in insufficient accuracy and reliability of the records.

[0004] Tunnel engineers usually only give a qualitative description of the rock mass structure at the tunnel face. However, the tunnel excavation area can reach more than 100 square meters, and the complex and diverse geological conditions are often uneven within such a large area. Combined with the actual situation of digital image shooting conditions and the growing development of computer vision, it is urgent to design a fine-grained automatic recognition method for the surrounding rock mass structure of the tunnel face image to improve the efficiency, accuracy and consistency of the geological risk assessment of the tunnel face. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method for fine-grained image recognition of the surrounding rock mass structure of a tunnel face, so as to achieve the purpose of automatically identifying the rock mass structure of the tunnel face.

[0006] A method for fine-grained image recognition of surrounding rock mass structure of a tunnel face comprises the following steps:

[0007] Step 1: Obtain tunnel face images and perform image preprocessing. Image quality is affected by shooting methods and lighting conditions, and there are problems such as image blur and uneven lighting. Image clarity is enhanced by histogram equalization and image sharpening.

[0008] Step 2: Scale the tunnel face image to input the face area segmentation model to obtain the face area and non-face area:

[0009] 2.1: The tunnel face image data is marked as the tunnel face area and the non-tunnel face area to obtain the tunnel face area segmentation dataset;

[0010] 2.2: Scale the preprocessed tunnel face image data. The purpose of this segmentation step is to identify the tunnel face area, so a large number of texture details of the tunnel face are not required. In order to speed up the processing speed of the segmentation model, the image data is scaled to 256×256 pixels;

[0011] 2.3: Establish a tunnel face region segmentation model, input the scaled tunnel face region segmentation data set into the model training, and optimize the model by adjusting the model parameters: obtain a model that can be used for tunnel face region segmentation;

[0012] Step 3: restore the result of the regional segmentation to the original size to obtain the tunnel face region in the original image, and perform tile processing on the tunnel face region to obtain a tile-level tunnel face image;

[0013] 3.1: The segmentation result of the tunnel face region obtained in step 2 is restored to the size of the original image through a restoration operation to obtain the tunnel face region in the original image;

[0014] 3.2: The tunnel face area in the image is tiled, and the original image is divided into M×N tiles of 299×299 pixels in order from bottom to top and from left to right, and the original position of the tile (m, n) is recorded. Since the edges of the original image are mostly non-tunnel face areas, the tiles with less than 299×299 pixels in the last row / column are discarded;

[0015] Step 4: Classify the rock structure of the tile-level image into five categories: mosaic structure, scattered structure, layered structure, complete structure, and fragmented structure. The rock structure classification results of each tile are obtained. The production process is as follows:

[0016] 4.1: Based on the tunnel face tile images obtained in step 3, the texture characteristics of different rock structures are classified by visual annotation to obtain tunnel face tile images containing five types of rock structures, and a tile-level tunnel face rock structure dataset is constructed;

[0017] 4.2: Establish a fine-grained tunnel face image rock mass structure classification model, input the tile-level tunnel face rock mass structure data set into the model training, optimize the model by adjusting the model parameters, and obtain a model that can be used for tile-level tunnel face rock mass structure classification;

[0018] Step 5: splice the tile-level rock mass structure classification results obtained in step 4 according to the original tile position (m, n) recorded in step 3 to obtain the face rock mass structure distribution map and draw conclusions on the face rock mass structure distribution.

[0019] Furthermore, in step 1, the acquisition of the tunnel face image during the tunnel construction process should be collected after the tunnel blasting and mucking process to ensure clear shooting. The image should cover the entire tunnel face, and the acquired image is used for the preparation of the data set.

[0020] Furthermore, the image quality in step 1 is affected by the shooting method and lighting conditions, and there is a problem of uneven lighting. When the lighting conditions are poor, the image appears to be overall or partially darkened. The image clarity is enhanced by histogram equalization.

[0021] Furthermore, the image quality in step 1 is affected by the shooting method and lighting conditions, and there is a problem of image blur. The visual effect of the image can be improved by image sharpening. Image sharpening can make the image look clearer by highlighting the high-frequency part of the image and enhancing the edge and details of the image.

[0022] Furthermore, in step 2, a segmentation model based on a U-Net convolutional neural network is used to extract the tunnel face area.

[0023] Furthermore, the tile determination standard in step 3 adopts the tunnel face area segmentation result obtained in step 2 as the standard, and the standard is as follows: if the number of pixels in the tunnel face area in the tile accounts for more than 90%, the tile is determined to be a tunnel face tile; if the number of pixels in the tunnel face area accounts for 0-90%, the tile is determined to be a tunnel face edge tile; if the number of pixels in the tunnel face area is 0, the tile is determined to be a non-tunnel face tile.

[0024] Furthermore, the classification model in step 4 adopts the Inception V3 deep convolutional neural network.

[0025] This solution has the following beneficial effects:

[0026] 1) This method realizes the automatic identification of the rock mass structure of the tunnel face. Compared with the traditional manual identification method, it has higher efficiency and accuracy. Through image processing and machine learning algorithms, it can quickly and accurately distinguish different types of rock mass structures from images, providing an important basis for subsequent surrounding rock condition assessment and support scheme design;

[0027] 2) This method can not only separate the tunnel face area from the background, but also classify the rock mass structure in the tunnel face area into five categories, which can more accurately reflect the actual situation of the surrounding rock;

[0028] 3) The final output of this method is the distribution map of the rock mass structure on the tunnel face, which is formed by splicing the classification results according to the original position, intuitively reflecting the spatial distribution of various rock mass structures on the tunnel face. This visual result form is easy for engineering and technical personnel to understand and use, and is more sophisticated than the traditional method of qualitatively describing the entire tunnel face;

[0029] Finally, the goal of fine-grained image recognition of the surrounding rock mass structure of the tunnel face was achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is an overall flow chart of a method for fine-grained image recognition of rock mass structure of surrounding rock mass of a tunnel face according to the present invention;

[0031] Figure 2 It is an overall flow chart of a method for fine-grained image recognition of rock mass structure of surrounding rock mass of a tunnel face according to the present invention;

[0032] Figure 3 A schematic flow chart of a tunnel face image preprocessing method of a general embodiment of a tunnel face surrounding rock mass structure fine-grained image recognition method of the present invention;

[0033] Figure 4 A schematic diagram of a histogram equalization process of an embodiment of a method for fine-grained image recognition of rock mass structure of surrounding rock mass of a tunnel face of the present invention;

[0034] Figure 5 The present invention is a schematic diagram of the tunnel face area segmentation process of an embodiment of a method for fine-grained image recognition of rock mass structure surrounding a tunnel face of the present invention. DETAILED DESCRIPTION

[0035] The following examples are further cited to illustrate the present invention in detail. The following examples are only used to further illustrate the present invention and cannot be understood as limiting the scope of protection of the present invention. Some non-essential improvements and adjustments made by those skilled in the art based on the above content of the present invention belong to the scope of protection of the present invention. The specific process parameters and the like in the following examples are also only examples within a suitable range, that is, those skilled in the art can make a selection within a suitable range through the description of this article, and are not limited to the specific values ​​​​exemplified below.

[0036] Embodiment 1:

[0037] This application proposes a method and device for fine-grained image recognition of the surrounding rock mass structure of a tunnel face, and the process is as follows: Figure 1 As shown, including:

[0038] Step 1: Obtain tunnel face images and perform image preprocessing. Since image quality is affected by shooting methods and lighting conditions, there are problems of image blur and uneven lighting. The image clarity is enhanced by histogram equalization and image sharpening.

[0039] 1.1: Obtain images of the tunnel face during the tunnel construction process. The images should be collected after the tunnel blasting and slag removal process. The images should be clear and cover the entire tunnel face. The acquired images will be used to prepare the data set.

[0040] 1.2: Image quality is affected by shooting methods and lighting conditions. There is a problem of uneven lighting. Especially when the lighting conditions are poor, the image will appear overall or partially dark. The image clarity is enhanced through histogram equalization.

[0041] Histogram equalization is an image enhancement method that can adjust the image by changing the grayscale histogram to make the pixel distribution in the image more uniform, enhance the image contrast, and make the details clearer;

[0042] The grayscale value of the original image is counted by pixel traversal method to obtain the grayscale histogram of the original image of the tunnel face. The number of pixels at each grayscale level in the histogram is accumulated to obtain the cumulative distribution function. The cumulative distribution function gives the proportion of pixels with grayscale values ​​less than or equal to a certain value in the grayscale value range to the total number of pixels. According to the cumulative distribution function, the grayscale value of each pixel in the original image is transformed according to the cumulative distribution function, and the mapping value of each pixel is calculated. These mapping values ​​are used to redistribute the pixel values ​​of the tunnel face, adjust the contrast of the image, replace the grayscale values ​​in the original image with the recalculated grayscale values, and generate a histogram-equalized image.

[0043] 1.3: Image quality is affected by shooting methods and lighting conditions, and there is a problem of image blur. Image sharpening can be used to improve the visual effect of the image;

[0044] Image sharpening can make the image look clearer by highlighting the high-frequency parts of the image, enhancing the edges and details of the image;

[0045] like Figure 4 As shown in the figure, the Sobel operator is used to calculate the gradient of the original image, and then the Laplace filter is used to filter the gradient image to enhance the high-frequency part of the image. The Gaussian filter is used to filter the gradient image after the first filtering to reduce the noise and low-frequency part of the image. Finally, the original image and the filtered image are weightedly superimposed to enhance the high-frequency part of the tunnel face image and obtain a sharpened image.

[0046] Step 2: Scale the tunnel face image to input the tunnel face area segmentation model to obtain the tunnel face area and the non-tunnel face area;

[0047] In order to realize the recognition of the tunnel face image, it is necessary to distinguish the tunnel face area from the invert arch, support and other areas in the image. Therefore, it is necessary to segment the image to obtain the tunnel face area range. The present invention uses a segmentation model based on a U-Net convolutional neural network to extract the tunnel face area. The specific steps are as follows: Figure 5 ,as follows:

[0048] 2.1: The tunnel face image data is marked as the tunnel face area and the non-tunnel face area to obtain the tunnel face area segmentation dataset;

[0049] 2.2: Scale the preprocessed tunnel face image data. The purpose of this segmentation step is to identify the tunnel face area, so a large number of texture details of the tunnel face are not required. In order to speed up the processing speed of the segmentation model, the image data is scaled to 256×256 pixels;

[0050] 2.3: Establish a tunnel face region segmentation model, input the scaled tunnel face region segmentation data set into the model training, and optimize the model by adjusting the model parameters, and finally obtain a model that can be used for tunnel face region segmentation;

[0051] The segmentation model uses a U-Net convolutional neural network, which consists of a contraction path and an expansion path, forming a U-shaped structure. The contraction path contains a series of convolutional layers, each of which is followed by a linear rectifier unit (ReLU) and a maximum pooling operation to gradually reduce the spatial resolution and increase the feature information, while the expansion path contains a large number of feature channels. Through continuous upsampling operations and connecting high-resolution features from the contraction path, the feature information and spatial information are fused to finally produce a segmentation result.

[0052] On this basis, the contrast loss function is added to solve the problems of class imbalance and detail loss in the segmentation task. The contrast loss function is a loss function used to measure the similarity or difference between two samples. Its design purpose is to enhance the similarity between similar samples and maximize the distance between dissimilar samples, thereby helping the model to better learn representations for specific tasks.

[0053] Step 3: restore the result of the regional segmentation to the original size to obtain the tunnel face region in the original image, and perform tile processing on the tunnel face region to obtain a tile-level tunnel face image;

[0054] In order to realize fine-grained recognition of tunnel face images, the present invention divides the tunnel face area in the image into fine-grained tile-level divisions based on the tunnel face spatial scale to obtain a tile-level tunnel face image. The specific steps are as follows:

[0055] 3.1: The segmentation result of the tunnel face region obtained in step 2 is restored to the size of the original image through a restoration operation, thereby obtaining the tunnel face region in the original image;

[0056] 3.2: Tile processing is performed on the palm face area in the image;

[0057] From bottom to top and from left to right, the original image is divided into M×N tiles of 299×299 pixels, and the original position of the tile (m, n) is recorded. Since the edges of the original image are mostly non-turbine face areas, the tiles with less than 299×299 pixels in the last row / column are discarded;

[0058] The tile is judged by the segmentation result of the face region obtained in step 2: if the number of pixels in the face region in the tile exceeds 90%, the tile is judged as a face tile; if the number of pixels in the face region is between 0-90%, the tile is judged as a face edge tile; if the number of pixels in the face region is 0, the tile is judged as a non-face tile;

[0059] Step 4: Establish a fine-grained image classification model for the rock mass structure of the tunnel face surrounding rock mass, classify the rock mass structure of the tile-level image, and obtain the rock mass structure classification results of each tile;

[0060] The present invention establishes a fine-grained face image rock mass structure classification model, and divides the tile-level images of the tunnel face into five categories: mosaic structure, scattered structure, layered structure, complete structure, and fragmented structure. The production process of the face surrounding rock mass structure classification model is as follows:

[0061] 4.1: Based on the tunnel face tile images obtained in step 3, the texture characteristics of different rock structures are classified by visual annotation to obtain tunnel face tile images containing five types of rock structures (mosaic structure, scattered structure, layered structure, complete structure, and fragmented structure), thereby constructing a tile-level tunnel face rock structure dataset;

[0062] 4.2: Establish a fine-grained tunnel face image rock mass structure classification model, input the tile-level tunnel face rock mass structure data set into the model training, and optimize the model by adjusting the model parameters, and finally obtain a model that can be used for tile-level tunnel face rock mass structure classification;

[0063] In order to improve the generalization ability of the model and increase the diversity of training samples, data enhancement is performed on the input tile-level tunnel face rock structure data, including horizontal and vertical flipping, random cropping, translation transformation and other operations of the sample data.

[0064] The classification model uses the Inception V3 deep convolutional neural network, which uses a series of structures called Inception modules. These modules are composed of convolution kernels of different sizes and pooling operations, which can capture multi-scale image features at the same time and improve the performance of the model. In addition, the model also introduces the concept of decomposed convolution, which decomposes large convolution kernels into multiple small convolution kernels to reduce the number of parameters and computational complexity of the model. After the last layer of convolution, the global average pooling layer takes the average of each channel feature map, and then connects the result to the fully connected layer, thereby reducing the risk of overfitting and reducing the number of parameters. In order to accelerate training convergence and prevent gradient disappearance, the model introduces an auxiliary classifier in the middle layer, which shares the loss function with the main classifier for joint training;

[0065] In order to enhance the feature extraction capability and performance of the model, the present invention uses the pre-trained model of Inception V3 to further fine-tune the training on the target task to utilize its rich feature representation learned on a large-scale dataset;

[0066] The present invention uses a ReLU activation function and a Softmax activation function. The ReLU activation function limits the input value to between 0 and positive infinity, and outputs zero when the input is a negative number, and outputs the input itself when the input is a positive number. The ReLU activation function helps to increase the sparsity of the network, thereby reducing the correlation between parameters and reducing the risk of overfitting. It can also solve the gradient vanishing problem, so that the network can converge faster, and help alleviate the training difficulties caused by the gradient vanishing. The Softmax activation function is often used in multi-classification tasks to convert the output of the network into a category probability distribution, so that the model can better perform classification predictions and has a clear probability explanation. By normalizing the output, the Softmax activation function helps the model learn the relationship between categories and improves the accuracy of the model in classification tasks;

[0067] Step 5: The tile-level rock mass structure classification results obtained in step 4 are spliced ​​according to the original tile positions (m, n) recorded in step 3 to obtain the face rock mass structure distribution map and draw conclusions about the face rock mass structure distribution.

[0068] The specific functions of each module of the method and device for fine-grained image recognition of rock mass structure surrounding a tunnel face disclosed in this embodiment can be referred to the aforementioned method embodiment, which will not be repeated here.

[0069] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0070] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0071] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0072] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0073] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0074] The above is a detailed introduction to a method for extracting crack features from a tunnel face based on an attention enhancement mechanism provided by the present application. This article uses specific examples to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and its core idea of ​​the present application. At the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for fine-grained image recognition of rock mass structure surrounding a tunnel face, characterized in that: The following steps are involved: Step 1: Obtain tunnel face images and perform image preprocessing. Image quality is affected by shooting methods and lighting conditions, and there are problems such as image blur and uneven lighting. Image clarity is enhanced by histogram equalization and image sharpening. Step 2: Scale the tunnel face image to input the face area segmentation model to obtain the face area and non-face area: 2.1: The tunnel face image data is marked as the tunnel face area and the non-tunnel face area to obtain the tunnel face area segmentation dataset; 2.2: Scale the pre-processed tunnel face image data to a size of 256×256 pixels; 2.3: Establish a tunnel face region segmentation model, input the scaled tunnel face region segmentation data set into the model training, and optimize the model by adjusting the model parameters: obtain a model that can be used for tunnel face region segmentation; Step 3: restore the result of the regional segmentation to the original size to obtain the tunnel face region in the original image, and perform tile processing on the tunnel face region to obtain a tile-level tunnel face image; 3.1: The segmentation result of the tunnel face region obtained in step 2 is restored to the size of the original image through a restoration operation to obtain the tunnel face region in the original image; 3.2: The palm face area in the image is tiled, and the original image is divided into M×N tiles of 299×299 pixels in order from bottom to top and from left to right. The original position of the tile (m, n) is recorded, and the tiles with less than 299×299 pixels in the last row / column are discarded; Step 4: Classify the rock structure of the tile-level image into five categories: mosaic structure, scattered structure, layered structure, complete structure, and fragmented structure. The rock structure classification results of each tile are obtained. The production process is as follows: 4.1: Based on the tunnel face tile images obtained in step 3, the texture characteristics of different rock structures are classified by visual annotation to obtain tunnel face tile images containing five types of rock structures, and a tile-level tunnel face rock structure dataset is constructed; 4.2: Establish a fine-grained tunnel face image rock mass structure classification model, input the tile-level tunnel face rock mass structure data set into the model training, optimize the model by adjusting the model parameters, and obtain a model that can be used for tile-level tunnel face rock mass structure classification; Step 5: splice the tile-level rock mass structure classification results obtained in step 4 according to the original tile position (m, n) recorded in step 3 to obtain the face rock mass structure distribution map and draw conclusions on the face rock mass structure distribution.

2. According to claim 1, a method for fine-grained image recognition of rock mass structure surrounding a tunnel face is characterized by: In step 1, the acquisition of tunnel face images during tunnel construction should be done after the tunnel blasting and slag removal process to ensure clear shooting. The images must cover the entire tunnel face, and the acquired images are used to prepare the data set.

3. According to claim 1, a method for fine-grained image recognition of surrounding rock mass structure of a tunnel face is characterized by: The image quality in step 1 is affected by the shooting method and lighting conditions, and there is a problem of uneven lighting. When the lighting conditions are poor, the image appears to be overall or partially darkened. The image clarity is enhanced by histogram equalization.

4. The method for fine-grained image recognition of rock mass structure surrounding a tunnel face according to claim 1, characterized in that: The image quality in step 1 is affected by the shooting method and lighting conditions, and there is a problem of image blur. The visual effect of the image is improved by image sharpening.

5. The method for fine-grained image recognition of rock mass structure surrounding a tunnel face according to claim 1, characterized in that: In step 2, a segmentation model based on a U-Net convolutional neural network is used to extract the tunnel face area.

6. The method for fine-grained image recognition of rock mass structure of surrounding rock mass at tunnel face according to claim 1, characterized in that: The tile determination standard in step 3 uses the tunnel face area segmentation result obtained in step 2 as the standard, and the standard is as follows: if the number of pixels in the tunnel face area in the tile accounts for more than 90%, the tile is determined to be a tunnel face tile; if the number of pixels in the tunnel face area accounts for 0-90%, the tile is determined to be a tunnel face edge tile; if the number of pixels in the tunnel face area is 0, the tile is determined to be a non-tunnel face tile.

7. The method for fine-grained image recognition of rock mass structure of surrounding rock mass at tunnel face according to claim 1, characterized in that: The classification model in step 4 adopts the InceptionV3 deep convolutional neural network.

Images