Tunnel face geological condition description method based on image recognition

Abstract

The invention provides a tunnel face geological condition description method based on image recognition, and belongs to the technical field of geological detection and neural networks. The method comprises the following steps: firstly, obtaining a trained convolutional neural network model; wherein in the preprocessing process before neural network training, an edge detection technology is added; then acquiring rock types corresponding to all points of the tunnel face through image and video information of the tunnel face acquired in real time; and finally, describing the geological classification and distribution of the tunnel face in real time by adopting a grid analysis method for the geological distribution type of the tunnel face and combining a description method of geological engineering literatures. The edge processing of the image information obtained by the method is more accurate, so that the classification of the convolutional neural network is more accurate; meanwhile, a grid analysis method is adopted to process the tunnel face, the accurate distribution area and condition of each rock are obtained, and the geological distribution condition of the tunnel face can be described more accurately.

Description

Technical Field

[0001] The present invention relates to geological exploration and neural network technologies, and more particularly, to a method for describing the geological conditions of a tunnel face based on image recognition. Background Art

[0002] Describing the geological distribution of a tunnel face involves several key technologies. First, geological engineering is the foundation, which includes a detailed analysis and evaluation of the rock and soil masses around the tunnel. Through geological exploration and stratigraphic analysis, geological engineering determines the type, thickness, and stability of rock layers, as well as possible geological structures such as fault zones and contact relationships between rock layers. Second, geological hazard and prevention technologies are crucial for assessing geological risks that the tunnel area may face, such as landslides, groundwater gushing, and rock layer fractures. In addition, groundwater engineering technologies are used to analyze the groundwater level, hydrogeological characteristics, and their impact on tunnel stability. By comprehensively using geological exploration and detection technologies such as ground-penetrating radar and core drilling, accurate geological data can be obtained to support the decision-making process of tunnel design. Finally, tunnel design and construction technologies combine the above geological information to determine the optimal tunnel alignment, cross-sectional shape, and support measures to ensure the safety, reliability, and economy of the tunnel project.

[0003] As can be seen from the above background art, it is crucial to determine the geological distribution of the tunnel face itself. In the past, when determining the geological distribution of the tunnel face itself, technologies such as geological exploration, vibration wave detection, and neural network analysis were often used. Among them, neural network analysis technology is often used to analyze the geological distribution in geological images. However, due to the differences between various neural network technologies and the uncertainty of edge division, when analyzing images through neural network analysis technology, the judged geological types may be incorrect.

[0004] Therefore, adopting a method to make the edge analysis of the neural network more accurate, and thus making the pattern classification judgment of the neural network for geological images more accurate, will be beneficial to the accurate description of the geological distribution of the tunnel face. Summary of the Invention

[0005] The object of the present invention is to overcome the above problems existing in the prior art and greatly improve its technical effect on the basis of the original technology. First, the present invention provides a method for describing the geological conditions of a tunnel face based on image recognition, which includes:

[0006] (1) Training a deep learning model for geological classification of tunnel faces; training the face images and video information inside the tunnel collected in the past through convolutional neural network technology to obtain a trained convolutional neural network model; the training process includes: data collection, data preprocessing, feature extraction, and parameter optimization; among them, edge detection technology is added during the data preprocessing process to make the material boundaries obtained by deep learning technology more accurate;

[0007] (2) Determining the geological distribution of the tunnel face; obtaining the image information and video information of the tunnel face in real time, and inputting the image information and video information into the trained convolutional neural network model to determine the geological distribution of the tunnel face in real time;

[0008] (3) Describing the geological distribution of the tunnel face; dividing the tunnel face into multiple small blocks by using the grid analysis method, observing the geological distribution in each small block according to the geological distribution of the tunnel face determined in (2), and determining the geological distribution type of the tunnel face; and calculating the distribution area of each rock mass type of the tunnel face through summation, and combining the description methods in geological engineering literature to describe the geological classification and distribution of the tunnel face in real time.

[0009] Furthermore, the deep learning model for training the geological classification of the tunnel face includes: a convolutional neural network is a deep learning model mainly used to process and analyze image and video data information, used to extract the features of image and video information and classify the objects in the image and video information; data collection obtains the image information and video information of the tunnel face through a laser scanner and a high-definition camera; data preprocessing refers to operations including denoising, image enhancement, cropping, and standardization on the collected image information and video information to ensure the quality and consistency of the data information; feature extraction refers to extracting features from the image and video information through convolutional operations, and the extracted features represent the edges, textures, and shapes in the image and video information; parameter optimization is achieved through training data and the backpropagation algorithm, and the main purpose is to adjust the weights and biases of the model to minimize the loss function, so as to achieve a good prediction effect; the trained convolutional neural network model includes: determining whether the convolutional neural network is trained by including reaching the training cycle number and minimizing the loss function, and calling the trained convolutional neural network the trained convolutional neural network model.

[0010] Furthermore, the edge detection technology added during the data preprocessing process includes: Due to the complexity of geology, when only using convolutional neural network technology to process the image and video information of the tunnel face, the target edge division is often inaccurate. Therefore, during the data preprocessing process, the edge detection technology is added to first divide the target edges of the image and video information, and then use the divided image and video information as the input of the convolutional neural network. The convolutional neural network directly uses the target edge division in the preprocessing process to extract features of the targets within each edge to obtain target features. To ensure the accuracy of the edge detection technology in dividing the target edges in the image and video information during preprocessing, multiple edge processing technologies are used to process the image and video information of the same tunnel face simultaneously, and observe whether the edge divisions of the same target are exactly the same. If they are exactly the same, it is determined that the edge division of the target is correct; if they are not exactly the same, the edge division of the target is determined according to the segmentation and splicing method. The segmentation and splicing method includes two parts: segmentation and splicing. Among them, segmentation includes: centering on the target, dividing the edge of the target into multiple edge segments according to the azimuth, counting the divisions of the edge segments of different edge detection technologies in the same azimuth, and counting whether the divisions of the same edge segment in the same azimuth are not less than 2 / 3 of the total number of all edge detection technologies. If not less than, it is determined that this division is the correct edge division of the target corresponding azimuth; if less than, continue to increase the number of edge detections until the divisions of the same edge segment in the corresponding azimuth are not less than 2 / 3 of the total number of all edge detection technologies, and determine that this division is the correct edge division of the target corresponding azimuth. Splicing includes: obtaining the correct edge divisions of all different azimuths of the divided target, and splicing all the obtained correct edge divisions together to form the overall edge division of the target.

[0011] Furthermore, the real-time determination of the geological distribution of the tunnel face includes: the rock type, rock strength, and rock structure of the corresponding point of the tunnel face.

[0012] Furthermore, the observation of the geological distribution in each small block to determine the geological distribution type of the tunnel face includes: dividing the geological distribution type of the tunnel face into chaotic distribution and flat distribution. The chaotic distribution refers to counting the number of more than one type of geology in each small block. If the number of small blocks with more than one type of geology is not less than half of the total number of small blocks, it is defined as chaotic distribution; if the number of small blocks with more than one type of geology is less than half of the total number of small blocks, it is defined as flat distribution.

[0013] Furthermore, the calculation of the distribution area of each rock mass type of the tunnel face by summation includes: The calculation formula for obtaining the distribution area of each rock mass type of the tunnel face by summation is:

[0014]

[0015] Among them, H j represents the distribution area of the j-th type of rock mass, i represents the i-th small block, n represents that the mesh analysis method divides into n small blocks in total, and h i represents the area where the j-th type of rock mass is distributed in the i-th small block.

[0016] The beneficial effects of the present invention are as follows:

[0017] The present invention provides a method for describing the geological conditions of a tunnel face based on image recognition. First, the convolutional neural network is trained through the image information and video information of the tunnel face collected in the past to obtain a trained convolutional neural network model. Among them, edge detection technology is added in the preprocessing process before neural network training. Subsequently, the images and video information of the tunnel face collected in real time are input into the trained convolutional neural network model to obtain the rock types corresponding to each point of the tunnel face. Finally, the mesh analysis method is used for the geological distribution types of the tunnel face, and combined with the description method in geological engineering literature, the geological classification and distribution of the tunnel face are described in real time. The edge processing of the image information obtained by this method will be more accurate, thereby making the convolutional neural network classification more accurate. At the same time, the mesh analysis method is used to process the tunnel face to obtain accurate distribution areas and conditions of each rock, making the description of the geological distribution of the tunnel face more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 : Flowchart of a method for describing the geological conditions of a tunnel face based on image recognition according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0019] The following detailed description of specific embodiments of the present invention is provided with reference to the accompanying drawings; it should be understood that the specific embodiments given here are only for the purpose of illustrating and explaining the present invention and should not be used to limit the present invention.

[0020] It should be noted that many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention may have other embodiments and variations, and therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0021] Such as Figure 1As shown, it is a flowchart of a method for describing the geological conditions of a tunnel face based on image recognition according to an embodiment of the present invention. The flowchart includes: Step S101, training a deep learning model for tunnel face geological classification; training the face image information and video information inside the tunnel collected in the past through convolutional neural network technology to obtain a trained convolutional neural network model; the training process includes: data collection, data preprocessing, feature extraction, and parameter optimization; among them, edge detection technology is added during the data preprocessing process, making the material boundaries obtained by deep learning technology more accurate; Step S102, determining the geological distribution of the tunnel face; obtaining the image information and video information of the tunnel face in real time, inputting the image information and video information into the trained convolutional neural network model, and determining the geological distribution of the tunnel face in real time; Step S103, describing the geological distribution of the tunnel face; using the grid analysis method to divide the tunnel face into multiple small blocks, observing the geological distribution in each small block according to the geological distribution of the tunnel face determined in Step S102, and determining the geological distribution type of the tunnel face; and obtaining the distribution area of each rock type of the tunnel face through summation calculation, and describing the geological classification and distribution of the tunnel face in real time in combination with the description method in geological engineering literature.

[0022] Specifically, the present invention first trains a convolutional neural network through the face image information and video information collected in the past to obtain a trained convolutional neural network model; among them, edge detection technology is added during the preprocessing process before neural network training; subsequently, through the image and video information of the tunnel face collected in real time, input into the trained convolutional neural network model, the rock types corresponding to each point of the tunnel face are obtained; finally, the grid analysis method is used for the geological distribution type of the tunnel face, and the geological classification and distribution of the tunnel face are described in real time in combination with the description method in geological engineering literature.

[0023] Step S101, training a deep learning model for tunnel face geological classification; training the face image information and video information inside the tunnel collected in the past through convolutional neural network technology to obtain a trained convolutional neural network model; the training process includes: data collection, data preprocessing, feature extraction, and parameter optimization; among them, edge detection technology is added during the data preprocessing process, making the material boundaries obtained by deep learning technology more accurate.

[0024] Specifically, a convolutional neural network is a deep learning model mainly used to process and analyze image and video data information, extract features of image and video information, and classify objects in image and video information; data collection obtains image and video information of the tunnel face through a laser scanner and a high-definition camera; data preprocessing refers to operations including denoising, image enhancement, cropping, and standardization on the collected image and video information to ensure the quality and consistency of data information; feature extraction refers to extracting features from image and video information through convolutional operations, and the extracted features represent edges, textures, and shapes in image and video information; parameter optimization is achieved through training data and the backpropagation algorithm, and the main purpose is to adjust the weights and biases of the model to minimize the loss function, thereby achieving a good prediction effect; the trained convolutional neural network model includes: determining whether the convolutional neural network is trained by including reaching the number of training cycles and minimizing the loss function, and calling the trained convolutional neural network the trained convolutional neural network model.

[0025] In the above embodiment, specifically, due to the complexity of geology, when only using convolutional neural network technology to process the image and video information of the tunnel face, the situation of inaccurate target edge division often occurs. Therefore, in the data preprocessing process, edge detection technology is added to first divide the target edges of the image and video information, and then use the divided image and video information as the input of the convolutional neural network. The convolutional neural network directly uses the division of the target edges in the preprocessing process, and then extracts features of the targets within each edge to obtain target features; to ensure the accuracy of the edge detection technology in the preprocessing for dividing the target edges in the image and video information, multiple edge processing technologies are used to process the image and video information of the same tunnel face simultaneously, and observe whether the edge divisions of the same target are completely consistent. If they are completely consistent, it is determined that the edge division of the target is correct; if they are not completely consistent, the edge division of the target is determined according to the segmentation and splicing method; the segmentation and splicing method includes two parts: segmentation and splicing; among them, segmentation includes: centering on the target, dividing the edge of the target into multiple edge segments according to the azimuth, counting the divisions of the edge segments of different edge detection technologies in the same azimuth, and counting whether the division of the same edge segment in the same azimuth is not less than 2 / 3 of the number of all edge detection technologies. If it is not less than, it is determined that the division is the correct edge division of the target corresponding azimuth; if it is less than, continue to increase the number of edge detections until the division of the same edge segment in the corresponding azimuth is not less than 2 / 3 of the number of all edge detection technologies, and determine that the division is the correct edge division of the target corresponding azimuth; splicing includes: obtaining the correct edge divisions of all different azimuths of the divided target, and splicing all the obtained correct edge divisions together to form the overall edge division of the target.

[0026] Step S102, determination of the geological distribution of the tunnel face; obtain the image information and video information of the tunnel face in real time, input the image information and video information into the trained convolutional neural network model, and determine the geological distribution of the tunnel face in real time.

[0027] Specifically, determining the geological distribution of the tunnel face in real time includes: the rock type, rock strength, and rock structure of the corresponding points on the tunnel face, etc.

[0028] Step S103, description of the geological distribution of the tunnel face; divide the tunnel face into multiple small blocks by using the grid analysis method, observe the geological distribution in each small block according to the geological distribution of the tunnel face determined in Step S102, and determine the geological distribution type of the tunnel face; and obtain the distribution area of each rock mass type of the tunnel face through summation calculation, and describe the geological classification and distribution of the tunnel face in real time in combination with the description method in geological engineering literature.

[0029] Specifically, the method for observing the geological distribution in each small block and determining the geological distribution type of the tunnel face: divide the geological distribution type of the tunnel face into chaotic distribution and flat distribution; the chaotic distribution refers to counting the number of more than one type of geology in each small block. If the number of small blocks with more than one type of geology is not less than half of the total number of small blocks, it is defined as chaotic distribution; if the number of small blocks with more than one type of geology is less than half of the total number of small blocks, it is defined as flat distribution.

[0030] In the above embodiment, specifically, the method for obtaining the distribution area of each rock mass type of the tunnel face through summation calculation: the calculation formula for obtaining the distribution area of each rock mass type of the tunnel face through summation calculation is:

[0031]

[0032] Among them, H j represents the distribution area of the jth type of rock mass, i represents the ith small block, n represents the total number of small blocks divided by the grid analysis method, and h i represents the area where the jth type of rock mass is distributed in the ith small block.

[0033] To more clearly explain the above embodiments, the following will be described in conjunction with a specific case: In the Madong Iron Project, the Shuangwendan No. 2 Tunnel has created five of the most remarkable projects throughout the project line and in Malaysia, namely the longest single-hole double-track, the longest drill-and-blast method construction, the longest auxiliary adit, the longest single-hole double-track ballastless track, and the longest reverse slope drainage railway tunnel; the construction organization and arrangement units of the Shuangwendan No. 2 Tunnel are numerous, there are many tunnel working faces, and the management span is large. It is necessary to form a systematic tunnel mechanized rapid construction technology to significantly improve the project construction efficiency and reduce the project construction cost on the premise of ensuring the project safety and quality; therefore, the real-time monitoring and description of the tunnel geological distribution are crucial, and the tunnel real-time monitoring technology is applied to the Shuangwendan No. 2 Tunnel in the Madong Iron Project; First, the convolutional neural network technology is used to train the face image information and video information of the strata distribution of projects similar to the Shuangwendan No. 2 Tunnel in the Madong Iron Project collected in the past to obtain a trained convolutional neural network model; and edge detection technology is added during the preprocessing of the data before convolutional neural network processing, making the material boundaries obtained by the deep learning technology more accurate; Subsequently, the face image information and video information of the Shuangwendan No. 2 Tunnel are obtained in real time, and the obtained image information and video information are input into the trained convolutional neural network model to determine the geological distribution of the face of the Shuangwendan No. 2 Tunnel in real time; Finally, the grid analysis method is used to divide the face of the Shuangwendan No. 2 Tunnel into multiple small pieces, and according to the determined geological distribution of the face of the Shuangwendan No. 2 Tunnel, observe the geological distribution in each small piece to determine the geological distribution type of the face of the Shuangwendan No. 2 Tunnel, and determine whether the distribution type of the Shuangwendan No. 2 Tunnel is chaotic distribution or flat distribution; and calculate the distribution area of each rock mass type of the face of the Shuangwendan No. 2 Tunnel through summation, and describe the geological classification and distribution of the face of the Shuangwendan No. 2 Tunnel in real time in combination with the description method in geological engineering literature.

[0034] It should be understood that the above embodiments are one or more embodiments of the present invention, and there are many other embodiments and their deformations based on the present invention; when ordinary technicians in this industry do not make pioneering innovations, the deformations and modifications made through the present invention all fall within the protection scope of the present invention.

Claims

1. A method for describing geological conditions of a tunnel face based on image recognition, characterized in that: The method comprises: 1) Training a deep learning model for geological classification of tunnel faces; training previously collected tunnel face image information and video information using convolutional neural network technology to obtain a trained convolutional neural network model; the training process includes: data collection, data preprocessing, feature extraction and parameter optimization; wherein edge detection technology is added in the data preprocessing process to make the material boundary obtained by the deep learning technology more accurate; 2) Determine the geological distribution of the tunnel face: obtain image information and video information of the tunnel face in real time, input the image information and video information into the trained convolutional neural network model, and determine the geological distribution of the tunnel face in real time; 3) Description of the geological distribution of the tunnel face; the grid analysis method is used to divide the tunnel face into multiple small blocks. According to the geological distribution of the tunnel face determined in 2), the geological distribution in each small block is observed to determine the geological distribution type of the tunnel face; and the distribution area of ​​each rock type of the tunnel face is obtained by summation calculation, and the geological classification and distribution of the tunnel face are described in real time in combination with the description method of geological engineering literature.

2. The method for describing geological conditions of a tunnel face based on image recognition according to claim 1, characterized in that: The deep learning model for training tunnel face geological classification includes: a convolutional neural network is a deep learning model mainly used for processing and analyzing image and video data information, and is used for extracting features of image and video information and classifying objects in the image and video information; data collection includes obtaining image information and video information of the tunnel face through a laser scanner and a high-definition camera; data preprocessing refers to performing operations including denoising, image enhancement, cropping and standardization on the collected image information and video information to ensure the quality and consistency of the data information; feature extraction refers to extracting features from image and video information through convolution operations, and the extracted features represent edges, textures and shapes in the image and video information; parameter optimization is achieved through training data and back propagation algorithms, and the main purpose is to adjust the weights and biases of the model so that it can minimize the loss function, thereby achieving a good prediction effect; the trained convolutional neural network model includes: determining whether the convolutional neural network is trained by reaching the number of training cycles and minimizing the loss function, and the trained convolutional neural network is called a trained convolutional neural network model.

3. The method for describing geological conditions of a tunnel face based on image recognition according to claim 1, characterized in that: The edge detection technology is added in the data preprocessing process, including: due to the complexity of geology, only the convolutional neural network technology is used to process the image and video information of the tunnel face, which often results in inaccurate target edge division. Therefore, the edge detection technology is added in the data preprocessing process to first divide the target edges of the image and video information, and then the divided image and video information is used as the input of the convolutional neural network. The convolutional neural network directly uses the target edge division in the preprocessing process, and then extracts features of the target within each edge to obtain target features; in order to ensure the accuracy of the edge detection technology in the preprocessing for the target edge division in the image and video information, a variety of edge processing technologies are used to process the image and video information of the same tunnel face at the same time to observe whether the edge division of the same target is completely consistent. If it is completely consistent, the edge division of the target is determined. The division is correct; if not completely consistent, the edge division of the target is determined according to the segmentation and splicing method; the segmentation and splicing method includes two parts: segmentation and splicing; wherein, segmentation includes: taking the target as the center, dividing the edge of the target into multiple edge segments according to the orientation, counting the divisions of the edge segments of different edge detection technologies in the same orientation, counting whether the divisions of the same edge segments in the same orientation are not less than 2 / 3 of the number of all edge detection technologies, if not less than, determining the division as the correct edge division of the corresponding orientation of the target; if less than, continuing to increase the number of edge detections until the divisions of the same edge segments in the corresponding orientation are not less than 2 / 3 of the number of all edge detection technologies, and determining the division as the correct edge division of the corresponding orientation of the target; splicing includes: obtaining the correct edge divisions of all different orientations of the divided target, and splicing all the obtained correct edge divisions together to form the overall edge division of the target.

4. The method for describing geological conditions of a tunnel face based on image recognition according to claim 3, characterized in that: The real-time determination of the geological distribution of the tunnel face includes: rock type, rock strength and rock structure at the corresponding point of the tunnel face.

5. The method for describing geological conditions of a tunnel face based on image recognition according to claim 1, characterized in that: The observing of the geological distribution in each small block and determining the geological distribution type of the tunnel face includes: dividing the geological distribution type of the tunnel face into chaotic distribution and flat distribution; the chaotic distribution refers to counting the number of more than one type of geological pattern in each small block, if the number of small blocks with more than one type of geological pattern is not less than half of the total number of small blocks, it is defined as chaotic distribution; if the number of small blocks with more than one type of geological pattern is less than half of the total number of small blocks, it is defined as flat distribution.

6. The method for describing geological conditions of a tunnel face based on image recognition according to claim 1, characterized in that: The method of obtaining the distribution area of ​​each rock mass type on the tunnel face by summing up includes: the calculation formula for obtaining the distribution area of ​​each rock mass type on the tunnel face by summing up is: Among them, H j represents the distribution area of ​​the jth rock mass type, i represents the ith small block, n represents the number of small blocks divided by the grid analysis method, and h i It represents the distribution area of ​​the j-th rock type in the ith small block.

Images