Tunnel surrounding rock characteristic automatic identification method and system based on deep learning

Abstract

The invention discloses a tunnel surrounding rock characteristic automatic identification method and system based on deep learning, and belongs to the technical field of tunnel construction. The method comprises the following steps: collecting and preprocessing tunnel face image data to construct a data set; constructing an identification and classification model based on a YOLOv8 network and a DenseNet network; a feature region is obtained through YOLOv8 network detection, and the feature region is used as the input of a DenseNet network; training the recognition classification model by using the data set to obtain an optimization model; and acquiring a tunnel face image in real time, inputting the tunnel face image into the optimization model, and outputting a surrounding rock feature recognition result. According to the method, the surrounding rock characteristics in the tunnel face image can be quickly and accurately recognized, the surrounding rock recognition precision and efficiency are improved, and the limitation of manual judgment in the prior art is overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel construction, and more specifically to a method and system for automatically identifying tunnel surrounding rock features based on deep learning. Background Art

[0002] During tunnel construction, geological disasters such as water and mud bursts, landslides, and large deformations often occur, and these disasters are often closely related to the stability of the surrounding rock. Therefore, by using surrounding rock identification technology to obtain accurate information about the surrounding rock in real time, predict and prevent geological disasters, and avoid serious consequences such as the scrapping of construction machinery, it can ensure the smooth progress of tunnel engineering.

[0003] However, the existing tunnel surrounding rock identification technology mainly relies on manual experience and traditional methods, which usually involve on-site visual inspection and manual classification, and have the problems of strong subjectivity, low efficiency and large errors. With the complexity of the tunnel construction environment, traditional methods are often difficult to adapt to actual needs, especially when facing interference factors such as uneven lighting and dust, the accuracy and reliability of identification and classification are greatly reduced. Although some studies have applied machine vision and machine learning technologies to improve surrounding rock classification, these technologies are mostly focused on image feature extraction and basic classification, and there are still deficiencies in the automated and intelligent processing of surrounding rock features. In addition, the current technology also lacks the ability to dynamically update and optimize, which limits the adaptability of the model in practical applications.

[0004] Therefore, it is an urgent problem to be solved by technical personnel in this field to propose a method and system for automatic identification of tunnel surrounding rock features based on deep learning to further improve the accuracy and efficiency of surrounding rock identification. Summary of the invention

[0005] In view of this, the present invention provides a method and system for automatic identification of tunnel surrounding rock features based on deep learning, which can quickly and accurately identify surrounding rock features in tunnel face images, improve the accuracy and efficiency of surrounding rock identification, and overcome the limitations of manual judgment in the prior art.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] On the one hand, the present invention proposes a method for automatically identifying tunnel surrounding rock features based on deep learning, comprising the following steps:

[0008] Collect tunnel face image data and preprocess them to build a data set;

[0009] Building a recognition and classification model based on the YOLOv8 network and the DenseNet network; obtaining a feature region through detection by the YOLOv8 network, and using the feature region as an input of the DenseNet network;

[0010] Using the data set to train the recognition and classification model to obtain an optimized model;

[0011] The tunnel face image is acquired in real time and input into the optimization model, and the surrounding rock feature recognition result is output.

[0012] Preferably, before preprocessing, data enhancement processing is performed on the tunnel face image data.

[0013] Preferably, the total loss function l1 of the YOLOv8 network is constructed by combining the positioning loss, classification loss and confidence loss. The formula is as follows:

[0014] l1=λ coord +L loc +L conf +λ cls L cls ;

[0015] In the formula, λ coord is the first weight coefficient, L loc is the positioning loss, L conf is the confidence loss, λ cls is the second weight coefficient, L cls is the classification loss, λ coord and λ cls Used to adjust the losses of each part.

[0016] Preferably, the loss function L2 formula of the DenseNet network is as follows:

[0017]

[0018] Where N is the number of categories, yi is the true label of the i-th type of surrounding rock feature, and pi is the predicted probability of the i-th type of surrounding rock feature.

[0019] Preferably, the characteristic areas include fissure areas, weathering zone areas, and groundwater leakage areas;

[0020] The surrounding rock feature identification results include the degree of development of joints and fissures, the degree of weathering and the degree of groundwater development.

[0021] Preferably, the method further comprises:

[0022] The recognition process of the optimization model is visualized, and abnormal warnings are issued and the construction plan is adjusted according to the surrounding rock feature recognition results.

[0023] On the other hand, the present invention also proposes a tunnel surrounding rock feature automatic recognition system based on deep learning, which is used to implement the above-mentioned tunnel surrounding rock feature automatic recognition method based on deep learning, including:

[0024] The data processing unit collects and preprocesses the tunnel face image data to construct a data set;

[0025] A model building unit builds a recognition classification model based on a YOLOv8 network and a DenseNet network; a feature region is obtained by detecting the YOLOv8 network, and the feature region is used as an input of the DenseNet network;

[0026] A model training unit, using the data set to train the recognition and classification model to obtain an optimized model;

[0027] The recognition result output unit acquires the tunnel face image in real time, inputs it into the optimization model, and outputs the surrounding rock feature recognition result.

[0028] Preferably, the system further comprises:

[0029] The early warning display unit is used to visualize the recognition process of the optimization model, and to issue abnormal warnings and adjust the construction plan according to the surrounding rock feature recognition results.

[0030] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method and system for automatic identification of tunnel surrounding rock features based on deep learning. By collecting tunnel face image data and preprocessing it, a data set is constructed; a recognition and classification model is constructed based on the YOLOv8 network and the DenseNet network; the feature area is detected by the YOLOv8 network, and the feature area is used as the input of the DenseNet network; the recognition and classification model is trained using the data set to obtain an optimized model; the tunnel face image is input into the optimized model in real time, and the surrounding rock feature recognition result is output. The present invention can quickly and accurately identify the surrounding rock features in the tunnel face image, improve the accuracy and efficiency of surrounding rock identification, and overcome the limitations of manual judgment in the prior art. Geological anomalies are predicted through historical data and real-time data, and timely warnings and adjustment plans are provided to avoid accidents; unmanned monitoring of the entire construction cycle is realized, the burden on management personnel is reduced, and the overall management efficiency is improved. The present invention can also generate a visual surrounding rock grade classification report to help on-site engineers make decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0032] Figure 1 A flow chart of the method provided by the present invention;

[0033] Figure 2 The YOLOv8 network structure diagram provided by the present invention;

[0034] Figure 3 The DenseNet network structure diagram provided by the present invention;

[0035] Figure 4 A flow chart for building the recognition and classification model provided by the present invention;

[0036] Figure 5 This is a system architecture diagram provided by the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] On the one hand, the embodiment of the present invention proposes a method for automatically identifying tunnel surrounding rock features based on deep learning, such as Figure 1 As shown, the method comprises the following steps:

[0039] S1. Collect tunnel face image data and preprocess them to build a data set.

[0040] Before preprocessing, the tunnel face image data must be augmented to expand the data set, including image rotation, scaling, translation, flipping, and adding noise, in order to improve the robustness of the model.

[0041] The enhanced image is preprocessed, mainly including denoising and image enhancement. Denoising can use methods such as median filtering or Gaussian filtering, and image enhancement can use techniques such as histogram equalization to improve the quality of the image.

[0042] S2. Build a recognition and classification model based on the YOLOv8 network and the DenseNet network. The recognition and classification model building process is as follows: Figure 4 shown.

[0043] First, set the initial configuration files of the YOLOv8 and DenseNet algorithms, including network structure, training parameters, loss function, etc.

[0044] The network structure of YOLOv8 is as follows Figure 2As shown in the figure, it includes three layers: Backbone-Neck-Head. Backbone consists of multiple layers of convolution and downsampling operations, which are used to extract image features. Each layer extracts features layer by layer through convolution, batch normalization, and activation function (such as Leaky ReLU), and downsamples to reduce the spatial size of the feature map. Neck is the part that enhances feature fusion. It fuses multi-scale features (features of different layers) through a pyramid structure, so that the network can better detect objects of different sizes. Head is the final detection output layer, which is directly used to predict bounding box coordinates and category probabilities. It consists of a series of convolutional layers to separate the features of position prediction and category prediction.

[0045] In the hyperparameters, the learning rate is 0.001, the batch size is 16, and the total loss function L1 of YOLOv8 is composed of the following:

[0046] Localization Loss: The mean square error (MSE) is used to measure the localization error of the bounding box.

[0047] Classification Loss: Use cross entropy loss or focal loss to measure classification error.

[0048] Confidence Loss: Use mean square error (MSE) to measure the confidence loss of the bounding box.

[0049] Total loss function L1:

[0050] L1=λ coord +L loc +L conf +λ cls L cls

[0051] In the formula, λ coord is the first weight coefficient, L loc is the positioning loss, L conf is the confidence loss, λ cls is the second weight coefficient, L cls is the classification loss, λ coord and λ cls Used to adjust the losses of each part.

[0052] The network structure of DenseNet is as follows Figure 3As shown in the figure, DenseNet consists of three layers: Dense Block-Transition Layer-Growth Rate. In a Dense Block, each layer receives the feature maps output by all previous layers. Assuming there are n layers, the input of the lth layer is the output of all previous l-1 layers. Each layer usually consists of convolution (Conv) and nonlinear activation (such as ReLU). The Transition Layer is located between two Dense Blocks and is mainly used for downsampling. It reduces the spatial size of the feature map through convolution and Pooling operations. These layers usually also reduce the number of feature channels through 1x1 convolution to reduce the computational complexity of the model. The Growth Rate is the number of new feature channels output by each layer. A smaller growth rate helps the model stay small and avoid being too complex.

[0053] DenseNet extracts the surrounding rock area detected by YOLOv8, cuts out the detected surrounding rock area, and uses it as the input of DenseNet. Based on this pre-trained model, appropriate classification heads are added according to various classification standards of surrounding rock, so as to facilitate the input of high-dimensional data into the corresponding classification results. In the hyperparameters, the learning rate is set to 0.001, the batch size is set to 16, and the loss function is: Cross-Entropy Loss: For classification tasks, the loss is calculated by measuring the difference between the predicted probability distribution and the actual label distribution.

[0054] The total loss function L2 of DenseNet is as follows:

[0055]

[0056] Among them, N is the number of categories, yi is the true label of the i-th type of surrounding rock feature, and pi is the predicted probability of the i-th type of surrounding rock feature.

[0057] The YOLOv8 model is used to detect multiple features of the transmitted tunnel face image, quickly locate the main characteristic areas of the surrounding rock (such as cracks, weathering zones, groundwater leakage, etc.), and provide basic image data for subsequent feature classification. After YOLOv8 detects the characteristic areas, the image data of these areas can be input into the DenseNet model to classify and identify the characteristics of the surrounding rock joint and crack development degree, weathering degree, and groundwater development degree.

[0058] S3. Use the data set to train the recognition and classification model to obtain an optimized model.

[0059] The processed data set was divided into a training set and a validation set in a ratio of 8:2, and the recognition classification model was used for training and optimization, with an initial learning rate of 0.001 and a batch size of 16. The recognition accuracy of the model was improved by continuously adjusting the hyperparameters, and the optimized model was obtained.

[0060] S4. Obtain tunnel face images in real time and input them into the optimization model to output surrounding rock feature recognition results.

[0061] S5. Visualize the recognition process of the optimization model, issue abnormal warnings and adjust the construction plan based on the recognition results of surrounding rock characteristics.

[0062] On the other hand, the present invention also proposes a tunnel surrounding rock feature automatic recognition system based on deep learning, which is used to implement the above-mentioned tunnel surrounding rock feature automatic recognition method based on deep learning, such as Figure 5 As shown, the system includes:

[0063] The data processing unit collects and preprocesses the tunnel face image data to construct a data set;

[0064] The model building unit builds a recognition and classification model based on the YOLOv8 network and the DenseNet network; the feature area is obtained through the YOLOv8 network detection, and the feature area is used as the input of the DenseNet network;

[0065] The model training unit uses the data set to train the recognition and classification model to obtain an optimized model;

[0066] The recognition result output unit obtains the tunnel face image in real time, inputs the optimization model, and outputs the surrounding rock feature recognition result;

[0067] The early warning display unit is used to visualize the recognition process of the optimization model, and to issue abnormal warnings and adjust the construction plan based on the recognition results of the surrounding rock characteristics.

[0068] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0069] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for automatic identification of tunnel surrounding rock features based on deep learning, characterized in that: The following steps are involved: Collect tunnel face image data and preprocess them to build a data set; Building a recognition and classification model based on the YOLOv8 network and the DenseNet network; obtaining a feature region through detection by the YOLOv8 network, and using the feature region as an input of the DenseNet network; Using the data set to train the recognition and classification model to obtain an optimized model; The tunnel face image is acquired in real time and input into the optimization model, and the surrounding rock feature recognition result is output.

2. The method for automatic identification of tunnel surrounding rock features based on deep learning according to claim 1 is characterized in that: Before preprocessing, data enhancement processing is performed on the tunnel face image data.

3. The method for automatic identification of tunnel surrounding rock features based on deep learning according to claim 1 is characterized in that: The total loss function L1 of the YOLOv8 network is constructed by combining positioning loss, classification loss and confidence loss. The formula is as follows: L1−λ coord +L loc +L conf +λ cls L cls ; In the formula, λ coord is the first weight coefficient, L loc is the positioning loss, L conf is the confidence loss, λ cls is the second weight coefficient, L cls is the classification loss.

4. The method for automatic identification of tunnel surrounding rock features based on deep learning according to claim 1 is characterized in that: The loss function l2 formula of the DenseNet network is as follows: Where N is the number of categories, yi is the true label of the i-th type of surrounding rock feature, and pi is the predicted probability of the i-th type of surrounding rock feature.

5. The method for automatic identification of tunnel surrounding rock features based on deep learning according to claim 1, characterized in that: The characteristic areas include fissure areas, weathering belt areas, and groundwater leakage areas; The surrounding rock feature identification results include the degree of development of joints and fissures, the degree of weathering and the degree of groundwater development.

6. The method for automatic identification of tunnel surrounding rock features based on deep learning according to claim 1, characterized in that: Also includes: The recognition process of the optimization model is visualized, and abnormal warnings are issued and the construction plan is adjusted according to the surrounding rock feature recognition results.

7. A tunnel surrounding rock feature automatic recognition system based on deep learning, characterized in that: include: A data processing unit, used to collect tunnel face image data and perform preprocessing to construct a data set; A model building unit, used to build a recognition classification model based on a YOLOv8 network and a DenseNet network; obtain a feature area through detection by the YOLOv8 network, and use the feature area as an input of the DenseNet network; A model training unit, used to train the recognition and classification model using the data set to obtain an optimized model; The recognition result output unit is used to obtain the tunnel face image in real time, input it into the optimization model, and output the surrounding rock feature recognition result.

8. The deep learning-based automatic identification system for tunnel surrounding rock features according to claim 7 is characterized in that: Also includes: The early warning display unit is used to visualize the recognition process of the optimization model, and to issue abnormal warnings and adjust the construction plan according to the surrounding rock feature recognition results.

Images