An intelligent identification method, device and equipment for tunnel face rock and medium

Abstract

The application discloses a tunnel face rock intelligent identification method, device, equipment and medium, including the following steps: taking pictures of the tunnel construction site after blasting is completed; according to the pictures of the tunnel face, the corresponding mileage of the tunnel face rock is marked from the tunnel face sketch, longitudinal section drawing and engineering geological description related geological data; the marked tunnel face picture is image enhanced; a CCT rock intelligent identification classification model is constructed, the model is trained based on the enhanced picture, the final tunnel face rock intelligent identification model is obtained after the training is completed, and the intelligent identification of the tunnel face rock is completed by using the model. The CCT model of CNN and VIT in series is constructed, the local feature of the image is captured better by CNN, the global feature of the image is captured better by VIT, the series connection of the two can fully combine the global and local features of the tunnel face, and the feature loss is not easy to be caused.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel engineering, and particularly relates to a tunnel face rock intelligent identification method, device, equipment and medium. BACKGROUND

[0002] In recent years, the infrastructure industry in China has developed rapidly, and the area covered by the railway has become wider and wider. Especially when building railways in the southwest of China, many high mountains are encountered, so tunnels need to be built to facilitate travel. In the past decade, China has accumulated rich experience in tunnel construction and can overcome various complex and dangerous environments. Now that artificial intelligence is being used more and more widely, it is gradually matched with artificial intelligence technology in the field of tunnel construction to greatly improve efficiency and safety. In tunnel excavation, the identification of the face rock is extremely important, which involves the stability of the surrounding rock and the selection of support measures. The traditional method uses manual identification, which is highly experienced and has some misjudgments, resulting in some construction losses.

[0003] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0004] To solve the problems in the prior art, the present application provides a tunnel face rock intelligent identification method, device, equipment and medium, which solves the problems mentioned in the above background.

[0005] To achieve the above purpose, the present application provides the following technical scheme: a tunnel face rock intelligent identification method, comprising the following steps:

[0006] S1, taking a sample by photographing the face after the tunnel construction site blasting is completed;

[0007] S2, according to the face photograph taken in step S1, marking the corresponding mileage face rock from the face sketch, longitudinal section diagram and engineering geological explanation related geological data;

[0008] S3, image enhancement is performed on the marked face photograph;

[0009] S4, a CCT rock intelligent identification classification model is constructed, model training is performed based on the enhanced photograph, and after the training is completed, the final face rock intelligent identification model is obtained, and the intelligent identification of the face rock is completed by using the model.

[0010] Preferably, in step S1, the photographing and sampling of the face after the tunnel construction site blasting is completed comprises:

[0011]

[0011] After waiting for the completion of the blasting of the tunnel face, ventilation is performed for 20-30 minutes, and immediately after that, the equipment is carried to take photos. Before taking the photos, two 10000 Lux light sources are placed on both sides at a distance of 10-12 m from the tunnel face and 1.5 m from the center line of the tunnel. A smart phone with a pixel of ≥2000 million is used to take the photos, and in order to ensure the uniqueness of the pixel feature distribution of the photos, only one photo is taken at the same mileage.

[0012] Preferably, in step S2, the tunnel face lithology corresponding to the mileage is marked from the tunnel face sketch, the longitudinal section drawing and the engineering geological description related geological data, and the tunnel face lithology corresponding to the mileage is marked from the tunnel face sketch, the longitudinal section drawing and the engineering geological description related geological data.

[0013] The mileage information and the tunnel face lithology recorded on the geological sketch are collected from the tunnel face geological sketch, the tunnel longitudinal section drawing and the engineering geological description, and the photos at the same mileage are marked.

[0014] Preferably, in step S3, the image enhancement of the marked tunnel face photo includes: rotating the original image by 90°, 180° and 270° counterclockwise, flipping the original image up and down, and flipping the original image left and right.

[0015] Preferably, in step S4, the CCT lithology intelligent identification classification model is a CCT lithology intelligent identification classification model connected in series with CNN and VIT, and specifically includes: constructing a CCT lithology intelligent identification classification model connected in series with 2 layers of convolution layers and 2 layers of Transformer coding layers, and inputting the enhanced tunnel face image into the model for training.

[0016] Preferably, the CNN is used to extract various features of the original image, and through convolution operation, pooling operation, full connection operation and ReLu activation, the convolution value is finally flattened into a vector with a specific length.

[0017]

[0018] wherein, represents the pixel value of depth k at position (i, j) in the input feature map of the convolution layer, represents the pixel value of depth p at position (i+m, j+n) in the input data, represents the weight value of depth p at position (m, n) in the convolution kernel and depth k in the output feature map, and b k represents the bias term of depth k in the output feature map, and f represents the size of the convolution kernel.

[0019]

[0020] wherein, g represents the pooling function, denotes all values of the input tensor x within the pooling window;

[0021]

[0022]

[0023] wherein, denotes the output of the fully connected layer, W denotes the weight matrix, and b denotes the bias term.

[0024] Preferably, the VIT is used to encode the specific length sequence processed by the CNN, and the most core is the multi-head attention mechanism, which performs attention calculation on the input sequence, as shown in formula (5):

[0025]

[0026] wherein Q is a vector used to represent the specified position to be focused on, K is a vector used to represent the input sequence, V is an object to be weighted and averaged when calculating attention, d k is the dimension of the attention head.

[0027] In another aspect, to achieve the above object, the present application also provides the following technical scheme: an intelligent identification device for tunnel face rock, the device comprises:

[0028] The face shooting and sampling module shoots and samples the face after the blasting of the tunnel construction site;

[0029] The labeling module labels the face rock of the corresponding mileage from the face sketch, longitudinal section diagram and engineering geological explanation related geological data according to the face photos shot by the face shooting and sampling module;

[0030] The image enhancement module enhances the labeled face photos;

[0031] The intelligent identification model construction module constructs a CCT rock intelligent identification classification model, trains the model based on the enhanced photos, obtains the final face rock intelligent identification model after the training is completed, and completes the intelligent identification of the face rock by using the model.

[0032] In another aspect, to achieve the above object, the present application also provides the following technical scheme: an electronic device, comprising: a processor; and a memory for storing one or more programs;

[0033] When the one or more programs are executed by the processor, the processor performs the intelligent identification method.

[0034] In another aspect, to achieve the above object, the present application also provides the following technical solution: a computer readable storage medium, which stores a computer program, the computer program is executed by a processor to realize the intelligent identification method.

[0035] The present application has the advantages that: the present application adopts the high-definition working face photograph shooting mode, replaces the human eye with a machine, improves the identification efficiency and objectivity, and reduces the safety hazards of personnel identification. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 The present application has the advantages that: the present application adopts the high-definition working face photograph shooting mode, replaces the human eye with a machine, improves the identification efficiency and objectivity, and reduces the safety hazards of personnel identification.

[0037] Figure 2 The present application has the advantages that: the present application adopts the high-definition working face photograph shooting mode, replaces the human eye with a machine, improves the identification efficiency and objectivity, and reduces the safety hazards of personnel identification.

[0038] Figure 3 The present application has the advantages that: the present application adopts the high-definition working face photograph shooting mode, replaces the human eye with a machine, improves the identification efficiency and objectivity, and reduces the safety hazards of personnel identification.

[0039] Figure 4 The present application has the advantages that: the present application adopts the high-definition working face photograph shooting mode, replaces the human eye with a machine, improves the identification efficiency and objectivity, and reduces the safety hazards of personnel identification.

[0040] Figure 5 The present application has the advantages that: the present application adopts the high-definition working face photograph shooting mode, replaces the human eye with a machine, improves the identification efficiency and objectivity, and reduces the safety hazards of personnel identification.

[0041] Figure 6 The present application has the advantages that: the present application adopts the high-definition working face photograph shooting mode, replaces the human eye with a machine, improves the identification efficiency and objectivity, and reduces the safety hazards of personnel identification.

[0042] Figure 7 The present application has the advantages that: the present application adopts the high-definition working face photograph shooting mode, replaces the human eye with a machine, improves the identification efficiency and objectivity, and reduces the safety hazards of personnel identification.

[0043] In the figure, 110 is a working face shooting sampling module, 120 is a labeling module, 130 is an image enhancement module, 140 is an intelligent identification model construction module, 210 is a processor, and 220 is a memory. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0045] As Figure 1As shown, the present application particularly relates to an intelligent identification method for tunnel face rock, which comprises the following steps:

[0046] S1, taking pictures of the tunnel face after blasting: collecting the tunnel face image after blasting, carrying equipment into the tunnel to take pictures immediately after ventilation for 20-30 minutes after the completion of tunnel blasting, placing two 10000Lux diffuse light sources on both sides about 1.5m away from the tunnel center line at about 10-12m from the tunnel face before taking pictures, using a smart phone with a pixel of ≥20 million to take pictures, and in order to ensure the uniqueness of the pixel feature distribution of the picture, only one picture is taken at the same mileage, as shown in Figure 2

[0047] S2, based on the tunnel face image taken in S1, comparing the tunnel face geological sketch, tunnel longitudinal section drawing, engineering geological description and other geological data to manually label the rock nature of the tunnel face picture at the same mileage, as shown in Figure 3

[0048] S3, image enhancement: based on the tunnel face image labeled in S2, in order to preserve the original color and texture of the tunnel face image, only the rotation and flipping methods are used for image enhancement, which are 90° counterclockwise rotation, 180° rotation, 270° rotation, left-right flipping and up-down flipping, as shown in Figure 4

[0049] S4, based on the enhanced samples in S3, a Compact Convolutional Transformers (CCT) rock intelligent identification model of convolutional neural network (CNN) and visual Transformer (VIT) in series is constructed, as shown in Figure 5

[0050] The present application adopts CCT image classification algorithm to construct a series model of 2-layer CNN and 2-layer VIT, wherein CNN is used to extract various features of the original image, through convolution operation, pooling operation, full connection operation and ReLu activation, as formulas (1)-(4), and finally the convolution value is flattened into a vector with a specific length.

[0051]

[0052] Wherein, ​​​​represents the pixel value of depth k at position (i, j) in the input feature map of the convolutional layer, represents the pixel value of depth p at position (i+m, j+n) in the input data, represents the weight value of depth p at position (m, n) in the convolution kernel, and depth k in the output feature map, b k represents the bias term of depth k in the output feature map, and f represents the size of the convolution kernel;

[0053]

[0054] where g represents a pooling function (such as maximum pooling and average pooling), represents all values of the input tensor x within the pooling window.

[0055]

[0056] where, represents the output of the fully connected layer, W represents the weight matrix, and b represents the bias term.

[0057]

[0058] VIT is used to encode the specific length sequence processed by CNN, and the most core is the multi-head attention mechanism. The input sequence is subjected to attention calculation, as shown in formula (5).

[0059]

[0060] where Q is a vector used to represent a specified position to be focused on, K is a vector used to represent an input sequence, V is an object to be weighted and averaged when calculating attention, d k is the dimension of the attention head.

[0061] The enhanced image is input into the CCT model for training, so that the model fully extracts and learns the feature differences between different lithologies. After the model is fully trained and calculated for multiple times, a final intelligent identification model of the tunnel face lithology is obtained, and the intelligent identification of the tunnel face lithology is completed by using the final intelligent identification model of the tunnel face lithology.

[0062] Based on the same inventive concept as the above method embodiments, the embodiments of the present application also provide an intelligent identification device for tunnel face lithology. The device can realize the functions provided by the above method embodiments, as shown in Figure 6 The device comprises:

[0063] The tunnel face shooting and sampling module 110 shoots and samples the tunnel face after blasting at the tunnel construction site;

[0064] The labeling module 120: according to the tunnel face photos taken by the tunnel face shooting sampling module, the tunnel face lithology of the corresponding mileage is labeled from the tunnel face sketch, the longitudinal section drawing and the engineering geological description related geological data;

[0065] The image enhancement module 130: the labeled tunnel face photos are subjected to image enhancement;

[0066] The intelligent identification model construction module 140: a CCT lithology intelligent identification classification model is constructed, the model training is carried out based on the enhanced photos, the final tunnel face lithology intelligent identification model is obtained after the training is completed, and the intelligent identification of the tunnel face lithology is completed by using the model.

[0067] Based on the same inventive concept as the above method embodiment, the embodiment of the present application also provides an electronic device, as shown in the figure, Figure 7 The device includes a processor 210 and a memory 220 for storing one or more programs.

[0068] When the one or more programs are executed by the processor 210, the processor performs the intelligent identification method.

[0069] The intelligent identification method includes:

[0070] The tunnel face after the blasting of the tunnel construction site is completed is photographed and sampled;

[0071] According to the tunnel face photos taken in the above step, the tunnel face lithology of the corresponding mileage is labeled from the tunnel face sketch, the longitudinal section drawing and the engineering geological description related geological data;

[0072] The labeled tunnel face photos are subjected to image enhancement;

[0073] A CCT lithology intelligent identification classification model is constructed, the model training is carried out based on the enhanced photos, the final tunnel face lithology intelligent identification model is obtained after the training is completed, and the intelligent identification of the tunnel face lithology is completed by using the model.

[0074] Based on the same inventive concept as the above method embodiment, the embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by the processor 210 to realize the intelligent identification method.

[0075] The tunnel face after the blasting of the tunnel construction site is completed is photographed and sampled;

[0076] According to the tunnel face photos taken in the above step, the tunnel face lithology of the corresponding mileage is labeled from the tunnel face sketch, the longitudinal section drawing and the engineering geological description related geological data;

[0077] The labeled tunnel face photos are subjected to image enhancement;

[0078] The CCT lithology intelligent identification classification model is constructed, model training is carried out based on the enhanced photos, the final intelligent identification model of the working face lithology is obtained after the training is completed, and the intelligent identification of the working face lithology is completed by using the model.

[0079] The CCT model in series connection of CNN and VIT is constructed, the local feature of the image is better captured by the CNN, the global feature of the image is better captured by the VIT, the full and local features of the working face are fully combined by connecting the two, and feature loss is not caused.

[0080] Although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art can modify the technical solutions recorded in the foregoing embodiments or make equivalent replacement for part of the technical features, and any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for intelligently identifying the rock mass of a tunnel face, characterized in that, It comprises the following steps: S1, taking a sample of the tunnel face after blasting at the tunnel construction site; S2, according to the tunnel face photo taken in step S1, marking the tunnel face lithology corresponding to the mileage from the tunnel face sketch, longitudinal section diagram and engineering geology description related geological data; S3, image enhancement of the marked tunnel face photo: based on the marked tunnel face image, the original color and texture of the tunnel face image are retained, and the tunnel face image is enhanced by rotation and flipping, which respectively includes counterclockwise 90° rotation, 180° rotation, 270° rotation, left-right flipping and up-down flipping; S4, constructing a CCT lithology intelligent identification classification model, training the model based on the enhanced photo, obtaining the final tunnel face lithology intelligent identification model after training, and completing the intelligent identification of the tunnel face lithology by using the model; The CCT lithology intelligent identification classification model constructed is a CCT lithology intelligent identification classification model connected with CNN and VIT, which specifically includes: constructing a CCT lithology intelligent identification classification model connected with 2 convolution layers and 2 Transformer coding layers, and inputting the enhanced tunnel face image into the model for training; The CNN is used to extract various features of the original image, and through convolution operation, pooling operation, full connection operation and ReLu activation, the convolution value is finally flattened into a vector with a specific length, as shown in formulas (1)-(4); wherein, represents a pixel value at position (i, j) with depth k in the input feature map of the convolution layer, represents a pixel value at position (i+m, j+n) with depth p in the input data, represents a weight value at position (m, n) with depth p in the convolution kernel and depth k in the output feature map, b k represents a bias term at depth k in the output feature map, and f represents the size of the convolution kernel. where g denotes a pooling function, denotes an input tensor all values within the pooling window; wherein, represents the output of the fully connected layer, W represents the weight matrix, and b represents the bias term. The VIT is used to encode the specific length sequence processed by the CNN, and the most core is the multi-head attention mechanism, which performs attention calculation on the input sequence, as shown in formula (5): where Q is a vector for indicating a position to be focused on, K is a vector for indicating an input sequence, V is an object to be weightedly averaged when attention is calculated, d k is a dimension of an attention head.

2. The intelligent identification method of the tunnel face rock mass according to claim 1, characterized in that: In step S1, the tunnel face after blasting at the tunnel construction site is taken and sampled, which comprises: After waiting for 20-30 minutes after the tunnel face blasting is completed, immediately take a photo with the equipment, place two 10000Lux diffuse light sources at a distance of 10-12m from the tunnel face and 1.5m from the tunnel center line on both sides before taking the photo, use a smart phone with a pixel of ≥2000 million to take a photo, and in order to ensure the uniqueness of the pixel feature distribution of the photo, only one photo is taken at the same mileage. 3.The intelligent identification method of tunnel face rock mass according to claim 1, characterized in that: In step S2, the tunnel face lithology corresponding to the mileage is marked from the tunnel face sketch, longitudinal section diagram and engineering geology description related geological data, which comprises: Collect the tunnel face geological sketch, tunnel longitudinal section diagram and engineering geological description of the same tunnel, read the mileage information and tunnel face lithology recorded on the geological sketch, and mark the photo at the same mileage.

4. The intelligent identification device of the tunnel face rock mass according to any one of claims 1-3, characterized in that: The device comprises: A tunnel face photographing and sampling module (110) for photographing and sampling the tunnel face after blasting at the tunnel construction site; A marking module (120) for marking the tunnel face lithology corresponding to the mileage from the tunnel face sketch, longitudinal section diagram and engineering geology description related geological data according to the tunnel face photo taken by the tunnel face photographing and sampling module; An image enhancement module (130) for image enhancement of the marked tunnel face photo; The intelligent identification model construction module (140) is configured to construct a CCT lithology intelligent identification classification model, train the model based on the enhanced photos, obtain a final tunnel face lithology intelligent identification model after the training is completed, and complete intelligent identification of the tunnel face lithology by using the model.

5. An electronic device, comprising: The electronic device includes a processor (210) and a memory (220) for storing one or more programs; The one or more programs, when executed by the processor (210), cause the processor to perform the intelligent identification method of any one of claims 1-3.

6. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program, when executed by the processor (210), implements the intelligent identification method of any one of claims 1-3.

Images