Method and equipment for identifying lithology of surrounding rock of tunnel face

Abstract

The invention relates to the field of engineering geology, in particular to a tunnel face surrounding rock lithology identification method and equipment. The method comprises the following steps: cutting a to-be-identified tunnel face surrounding rock image according to a preset size, and outputting a sub-image set; sequentially inputting the sub-image set into a pre-constructed lithology identification model, and outputting an identification result corresponding to each sub-image; and the identification result corresponding to each sub-image is synthesized to obtain a corresponding lithology condition identification graph. According to the method, tunnel face surrounding rock is segmented from a whole image through an image segmentation algorithm (mainly, a tunnel lining area outside a tunnel face is cut off), the segmented tunnel face is cut into a plurality of sub-images with the same size, and a data set is constructed through the sub-images; lithology analysis is carried out through a pre-constructed lithology identification model; and comprehensively analyzing the tunnel face image based on the proposed partition identification comprehensive evaluation method, thereby realizing efficient and accurate identification of lithology of the tunnel face surrounding rock image.

Description

Technical Field

[0001] The invention relates to the field of engineering geology, and in particular to a method and equipment for identifying the lithology of surrounding rocks of a tunnel face. Background Art

[0002] Lithology identification of rocks has always been a very important and basic task in the field of engineering geology and mineral exploration. In the field of engineering geology, lithology identification involves the identification of unfavorable geological bodies in underground engineering and the prevention and control of disasters. Therefore, timely, efficient and accurate lithology identification technology is of great significance for engineering construction. The main methods of lithology identification include hand specimen identification, thin section identification, mineral testing, and some geophysical methods. These methods have their own value in various practical engineering applications. At present, the simplest and most direct methods are mainly hand specimen identification and rock thin section identification. Carrying out lithology identification requires high professional knowledge and geological experience of relevant personnel, and some methods are limited by experimental instruments and equipment. The required preprocessing is time-consuming and laborious, and it is difficult to identify lithology in a timely and accurate manner. Therefore, it is a very meaningful work to use new technical means to assist in lithology identification and reduce dependence on professional knowledge and geological experience.

[0003] At present, with the rapid development of computer vision technology and artificial intelligence technology, judging the category of objects through image recognition methods has become a hot research work. The rock's color, texture, and structural structure can be used to well judge its lithology. Color, texture, and structural structure can all show obvious characteristics in the image, so it is very feasible to apply computer vision technology to the task of lithology identification. For example, the patent with publication number CN116468939A discloses a method for intelligent classification of surrounding rocks of tunnel faces based on neural networks; the general process of identifying lithology through image recognition methods is: (1) Collect rock images, manually classify and annotate the images, and construct a rock image lithology data set; (2) Construct a deep learning model; (3) Conduct model training and evaluation based on the data set; (4) Input the rock image into the trained model to achieve lithology identification.

[0004] However, the research objects of rock identification through image recognition are basically concentrated on rock hand specimens, rock thin sections, drilling cores, and field rock outcrops. The most studied is the identification and classification of rock hand specimens. The main reason is that rock hand specimen images are easy to collect, have high image quality, and simple features. Data sets can be easily constructed without excessive image preprocessing technology. At present, there is no research related to the lithology identification of tunnel face surrounding rock images. Different from rock hand specimen images and rock thin section images, the characteristics of tunnel face surrounding rock images are that the rock image area of ​​the tunnel face is large, with an actual width of 10 to 18 meters and a height of 6 to 10 meters. Different lithology interlayers, different textures, structural structures, etc. are easy to appear in the same tunnel face surrounding rock image, which is easy to show heterogeneity. The closest research to the lithology identification of tunnel face surrounding rock is the lithology identification based on field rock outcrop images, followed by the lithology identification of drilling cores. The study on lithology recognition of outcrops in the field (Zhang Xialin et al., A method for lithology recognition of outcrops in the field) is mainly divided into two stages. The first stage is based on Fast RCNN target detection, that is, the target outcrops are framed and the lithology is not identified; the second stage is to use the ResNet + attention mechanism classification model to identify the lithology of the target outcrops. The main method of the study on drilling core recognition (Fatimah Alzubaidi et al., Automated lithology classification from drill core images using convolutional neural networks) is to cut the core images into small images of about 2*2cm to train the ResneXt50 model. In the prediction stage, the sliding window recognition method + smoothing function is used for comprehensive evaluation to eliminate the influence of tiny other lithology interlayers.

[0005] At present, the deep learning models used to build rock type recognition mainly include ResNet, VGG, Fast RCNN, Inception and other models. For model construction, some research will improve on the basis of the above original deep learning models. The main improvement methods include phased feature extraction, adding attention modules, global-local image feature fusion, image + data feature fusion, multi-image recognition comprehensive decision, etc. The evaluation of model performance mostly uses the confusion matrix method, which evaluates the overall recognition accuracy, specificity, recall rate of each model, and the recognition accuracy of each category to obtain the optimal rock type recognition model.

[0006] Since there is no research on lithology identification of tunnel face surrounding rock images, in order to fill the gap in lithology identification technology in this area, a method for lithology identification of tunnel face surrounding rock images is needed to solve the problem of heterogeneous lithology identification of tunnel face surrounding rock images and reduce the dependence of lithology identification on professional knowledge and geological experience in tunnel engineering rock mass evaluation. Summary of the invention

[0007] The purpose of the present invention is to overcome the problem of lack of lithology recognition of surrounding rock images of tunnel face in the prior art, and to provide a method and device for identifying lithology of surrounding rock of tunnel face.

[0008] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0009] A method for identifying the lithology of surrounding rocks of a tunnel face comprises the following steps:

[0010] a: Cut the surrounding rock image of the tunnel face to be identified according to the preset size and output a sub-image set;

[0011] b: inputting the sub-image set into a pre-built lithology recognition model in sequence, and the lithology recognition model outputs the recognition results corresponding to each sub-image; the recognition results include lithology classification and corresponding lithology probability;

[0012] c: synthesizing the identification results corresponding to each sub-image, generating a lithology identification map corresponding to the surrounding rock image of the tunnel face to be identified;

[0013] The pre-construction of the lithology identification model includes the following steps:

[0014] S1: Constructing tunnel face surrounding rock image dataset;

[0015] S2: Build a lithology recognition model, train the lithology recognition model through the tunnel face surrounding rock image data set, and output the model after the lithology recognition model converges; the lithology recognition model is a ResNet50 convolutional neural network.

[0016] As a preferred embodiment of the present invention, said c comprises the following steps:

[0017] c1: Fill each sub-image with different colors to represent different lithologies according to the lithology classification in the identification results corresponding to each sub-image;

[0018] c2: According to the lithology probability in the identification results corresponding to each sub-image, the transparency is used to represent the lithology probability and the transparency of each sub-image is adjusted;

[0019] c3: combining the adjusted sub-images in order and superimposing them on the original tunnel face image to generate a lithology identification image corresponding to the tunnel face surrounding rock image to be identified.

[0020] As a preferred embodiment of the present invention, the lithology classification includes mudstone, muddy sandstone, strongly weathered sandstone, moderately weathered sandstone, slightly weathered sandstone, strongly weathered limestone, moderately weathered limestone, slightly weathered limestone, shotcrete layer and black background.

[0021] As a preferred solution of the present invention, the cutting process includes automatic segmentation and image cutting;

[0022] The automatic segmentation uses an image segmentation model to perform image segmentation, so as to segment out a main image of the surrounding rock of the tunnel face;

[0023] The image cutting is used to cut the main image of the surrounding rock of the tunnel face into a plurality of sub-images of equal size according to a preset size.

[0024] As a preferred embodiment of the present invention, the image cutting in step a includes:

[0025] The main image of the surrounding rock of the tunnel face is cut according to preset sizes of a*a pixels, 2a*2a pixels, 4a*4a pixels, and 8a*8a pixels to obtain four sub-image sets;

[0026] At this time, in step b, the step of obtaining the identification results corresponding to each a*a sub-image includes:

[0027] Obtain the identification result of the sub-image corresponding to the sub-image in the sub-image set of a*a pixels, 2a*2a pixels, 4a*4a pixels, and 8a*8a pixels in the lithology identification model;

[0028] The various results are fused using the mean method, and the fused recognition result is used as the recognition result of the sub-image.

[0029] As a preferred solution of the present invention, the image segmentation model uses the U-Net convolutional neural network as the main framework, uses VGG16 as the feature extraction backbone network, and performs model training on it by segmenting and annotating the data set.

[0030] As a preferred embodiment of the present invention, S1 comprises the following steps:

[0031] S11: obtaining a tunnel face surrounding rock image, and manually marking the tunnel face surrounding rock image with corresponding lithology classification;

[0032] S12: automatically segmenting the tunnel face surrounding rock image data after manual annotation and classification using the image segmentation model;

[0033] S13: Cutting the tunnel face surrounding rock main image into a plurality of sub-images of equal size according to a preset size, and storing them as a tunnel face surrounding rock image dataset.

[0034] As a preferred solution of the present invention, the preset size of the image cutting is 224*224 pixels.

[0035] As a preferred solution of the present invention, the lithology identification model includes three feature extraction branches, a fully connected layer and a Softmax function;

[0036] The feature extraction branch includes a feature extraction layer, a convolution operation layer and an average pooling layer;

[0037] The convolution kernel sizes of the feature extraction layers in the three feature extraction branches are 7*7, 1*1 and 5*5 respectively.

[0038] A tunnel face surrounding rock lithology identification device comprises at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any of the methods described above.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention provides a method for identifying the lithology of surrounding rocks of a tunnel face. The surrounding rocks of the tunnel face are segmented from the entire image through an image segmentation algorithm (mainly by cropping out the tunnel lining area outside the tunnel face), and the segmented tunnel face is cut into multiple sub-images of equal size, and a data set is constructed with the sub-images; lithology analysis is performed through a pre-constructed lithology identification model; and a comprehensive analysis of the tunnel face image is performed based on a proposed partition identification comprehensive evaluation method, thereby achieving efficient and accurate identification of the lithology of the tunnel face surrounding rock image. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a schematic flow chart of a method for identifying the lithology of surrounding rocks at a tunnel face according to Embodiment 1 of the present invention;

[0042] Figure 2 A schematic diagram of an image of surrounding rock of a tunnel face taken in a method for identifying lithology of surrounding rock of a tunnel face according to Embodiment 2 of the present invention;

[0043] Figure 3 A schematic diagram of automatic segmentation of a tunnel face surrounding rock image in a tunnel face surrounding rock lithology identification method according to Embodiment 2 of the present invention;

[0044] Figure 4 It is a schematic diagram of the framework of the improved ResNet50 lithology identification model in the method for identifying lithology of surrounding rocks of a tunnel face described in Example 2 of the present invention;

[0045] Figure 5 This is a schematic diagram of the change in training accuracy before and after model improvement in a method for identifying lithology of surrounding rock at a tunnel face according to Example 2 of the present invention;

[0046] Figure 6 A schematic diagram of the change of training loss value before and after model improvement in a method for identifying lithology of surrounding rock of a tunnel face according to Example 2 of the present invention;

[0047] Figure 7 A schematic diagram of changes in the training accuracy of sub-images of various sizes in a method for identifying the lithology of surrounding rocks of a tunnel face according to Embodiment 3 of the present invention;

[0048] Figure 8 A schematic diagram of changes in sub-training loss values ​​of various sizes in a method for identifying lithology of surrounding rock at a tunnel face according to Example 3 of the present invention;

[0049] Fig. 9 A schematic diagram of tunnel face cutting in a method for identifying lithology of surrounding rocks of a tunnel face according to Embodiment 4 of the present invention;

[0050] Fig.10 This is a schematic diagram of the recognition effect before fusion in a method for identifying the lithology of surrounding rocks of a tunnel face according to Embodiment 4 of the present invention;

[0051] Fig.11 It is a schematic diagram of the recognition effect after fusion in a method for identifying the lithology of surrounding rocks of a tunnel face according to Embodiment 4 of the present invention;

[0052] Fig.12 It is a schematic diagram of the identification effect of the lithology zoning of the surrounding rock of the tunnel face in a method for identifying the lithology of the surrounding rock of the tunnel face described in Example 4 of the present invention;

[0053] Fig.13 This is a schematic structural diagram of a tunnel face surrounding rock lithology identification device described in Example 5 of the present invention, which utilizes a tunnel face surrounding rock lithology identification method described in any one of Examples 1-4. DETAILED DESCRIPTION

[0054] The present invention is further described in detail below in conjunction with test examples and specific implementation methods. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the content of the present invention belong to the scope of the present invention.

[0055] At present, the commonly used lithology identification methods are mainly hand specimen identification, thin section identification, mineral testing, etc., but they are generally limited by professional knowledge and test equipment, and are time-consuming and laborious. To solve the above problems, the present invention aims to make full use of the color, texture, structural structure and other characteristics of the tunnel face surrounding rock, and proposes a tunnel face surrounding rock image lithology identification method based on image recognition through computer vision technology, thereby breaking through the limitations of professional geological knowledge and test equipment.

[0056] In view of the characteristics of tunnel face surrounding rock images, namely, large image area, interlayers of different lithologies, different textures and structural structures, and existing research mainly focuses on rock hand specimens, rock thin sections, field rock outcrops, etc., there is no research related to lithology identification of tunnel face surrounding rock images. In view of this, the present invention provides a lithology identification method specifically for tunnel face surrounding rock images, including image segmentation, sub-image data set construction, deep learning model (improved ResNet50) training and other steps to solve the problem of heterogeneous lithology identification in the context of large-size images, and improve the efficiency and accuracy of tunnel face surrounding rock lithology identification.

[0057] Example 1

[0058] like Figure 1 As shown, a method for identifying the lithology of surrounding rock at a tunnel face comprises the following steps:

[0059] a: Cut the surrounding rock image of the tunnel face to be identified according to the preset size and output a sub-image set.

[0060] b: The sub-image sets are sequentially input into a pre-constructed lithology recognition model, and the lithology recognition model outputs the identification results corresponding to each sub-image; the identification results include lithology classification and corresponding lithology probability.

[0061] c: Summarize the identification results corresponding to each sub-image to generate a lithology identification map corresponding to the surrounding rock image of the tunnel face to be identified.

[0062] The pre-construction of the lithology identification model includes the following steps:

[0063] S1: Construct a tunnel face surrounding rock image dataset.

[0064] S2: Build a lithology recognition model, train the lithology recognition model through the tunnel face surrounding rock image data set, and output the model after the lithology recognition model converges; the lithology recognition model is a ResNet50 convolutional neural network.

[0065] Example 2

[0066] This embodiment is a specific implementation of the method for identifying the lithology of surrounding rocks of a tunnel face described in Embodiment 1, and includes the following steps:

[0067] a: Cut the surrounding rock image of the tunnel face to be identified according to the preset size and output a sub-image set. The cutting process includes automatic segmentation and image cutting, and further:

[0068] a1: The automatic segmentation uses an image segmentation model to perform image segmentation to segment the main body image of the surrounding rock of the tunnel face.

[0069] Since the images of the surrounding rock of the tunnel face are collected on-site at the tunnel project, after each tunnel excavation cycle, a smartphone or digital camera is used to shoot the tunnel face under sufficient lighting conditions. The images obtained are as follows: Figure 2 As shown in the figure. As shown in the figure, the image collected on site contains three areas, namely the face rock exposed by the excavation in front, the tunnel lining around the face, and the tunnel step surface below. The target area of ​​the study is the face rock, and other irrelevant areas need to be stripped out of the image. To address this problem, this application uses image segmentation technology based on an improved U-Net deep learning model. The image segmentation model uses the U-Net convolutional neural network as the main framework and VGG16 as the feature extraction backbone network. The model is trained by segmenting and annotating data sets. The tunnel face image segmentation model obtained after a large amount of image training has a good effect on the segmentation of the face rock. The automatic segmentation results of the face surrounding rock image are shown in the figure. Figure 3 shown.

[0070] The image cutting is used to cut the main image of the surrounding rock of the tunnel face into a plurality of sub-images of equal size according to a preset size.

[0071] b: The sub-image sets are sequentially input into a pre-constructed lithology recognition model, and the lithology recognition model outputs the identification results corresponding to each sub-image; the identification results include lithology classification and corresponding lithology probability.

[0072] Furthermore, the pre-construction of the lithology identification model includes the following steps:

[0073] S1: Construct a tunnel face surrounding rock image dataset.

[0074] S11: Obtain the tunnel face surrounding rock image, and manually annotate the tunnel face surrounding rock image with the corresponding lithology classification; this work is generally carried out by experienced geological experts to ensure the accuracy of the classification results. According to the tunnel face surrounding rock images collected in this embodiment, they are divided into 10 categories according to their characteristics, including mudstone, mud sandstone, strongly weathered sandstone, moderately weathered sandstone, slightly weathered sandstone, strongly weathered limestone, moderately weathered limestone, slightly weathered limestone, shotcrete layer, and black background. This constitutes the tunnel face surrounding rock image dataset of this embodiment, laying the foundation for the next step of model training.

[0075] S12: Automatically segmenting the tunnel face surrounding rock image data after manual annotation and classification using the image segmentation model.

[0076] S13: Cutting the tunnel face surrounding rock main image into a plurality of sub-images of equal size according to a preset size, and storing them as a tunnel face surrounding rock image dataset.

[0077] In tunnel engineering, the width of the tunnel face is generally between 10 and 18 meters, the height is generally between 6 and 10 meters, and the image pixel size is generally between 6000*4000 pixels and 3200*2400 pixels. Due to the large range of the tunnel face, different rock interlayers, different textures, structural structures, etc. may appear on the same tunnel face. In addition, part of the tunnel face rock mass may also be covered by concrete slurry during the lining shotcrete process, which has a great impact on the rock type identification results. To solve the above problems, this embodiment provides a solution to cut the tunnel face image obtained after annotation and segmentation into small sub-images of equal size according to a preset size.

[0078] S2: Build a lithology recognition model, train the lithology recognition model through the tunnel face surrounding rock image data set, and output the model after the lithology recognition model converges; the lithology recognition model is a ResNet50 convolutional neural network.

[0079] Since the lithology of the surrounding rock of the face is affected by factors such as color, texture, and structure, these factors have different scale characteristics. For example, the structural characteristics are generally reflected in the full scale, while the texture characteristics are smaller scales, and the color characteristics are generally reflected in the medium scale. The conventional ResNet50 convolutional neural network cannot extract the image features of the surrounding rock of the face from multiple scales. Therefore, this embodiment improves the ResNet50 convolutional neural network (see the model framework). Figure 4 shown):

[0080] ①In the first stage of the ResNet50 convolutional neural network, two branches are added to form a total of three feature extraction branches. The convolution kernel sizes of each branch are the original 7*7 convolution kernel, 1*1 convolution kernel and 5*5 convolution kernel, respectively. Different receptive field sizes are used to extract information of different scales. Among them, the branches with larger receptive fields can extract larger-scale structural features, the branches with smaller receptive fields can capture the surface texture features, and the medium receptive fields can retain a variety of medium-scale feature information such as color.

[0081] ②Then the same four-stage convolution operation is performed to obtain the average pooling features. Through the parallel multi-branch feature extraction network structure, the contribution of different scale features to lithology identification can be fully explored.

[0082] ③Finally, the three different scale features are fused through the fully connected layer, and the lithology classification is realized through the Softmax function, so that the accuracy of lithology recognition is significantly improved (such as Figure 5 and Figure 6 shown).

[0083] c: Summarize the identification results corresponding to each sub-image to generate a lithology identification map corresponding to the surrounding rock image of the tunnel face to be identified.

[0084] c1: Fill each sub-image with different colors to represent different lithologies according to the lithology classification in the identification results corresponding to each sub-image; the lithology classification includes mudstone, mud sandstone, strongly weathered sandstone, moderately weathered sandstone, slightly weathered sandstone, strongly weathered limestone, moderately weathered limestone, slightly weathered limestone, shotcrete layer and black background.

[0085] c2: According to the lithology probability in the identification results corresponding to each sub-image, the transparency of each sub-image is adjusted by using transparency to represent the size of the lithology probability; when the lithology probability ∈ [0.8,1], the transparency = 1; when the lithology probability ∈ [0.6,0.8), the transparency = 0.8; when the lithology probability ∈ [0.4,0.6), the transparency = 0.6; when the lithology probability ∈ [0.4,0], the transparency = 0.4.

[0086] c3: combining the adjusted sub-images in order and superimposing them on the original tunnel face image to generate a lithology identification image corresponding to the tunnel face surrounding rock image to be identified.

[0087] Example 3

[0088] This embodiment is a specific implementation method for determining the optimal cutting size of a sub-image, and the specific experimental scheme is as follows:

[0089] ① Scale the automatically segmented tunnel face image to a size of 3584*3584 pixels;

[0090] ② Cut the 3584*3584 pixel face image into sub-images of equal sizes of 896*896, 448*448, 224*224, and 120*120 pixels according to 4 rows and 4 columns, 8 rows and 8 columns, 16 rows and 16 columns, and 30 rows and 30 columns respectively;

[0091] ③ After cutting, the sub-images of each size are manually labeled and classified to form four data sets of different sizes;

[0092] ④ Use the conventional ResNet50 convolutional neural network to train the models for the four data sets;

[0093] ⑤ By analyzing the model training convergence and model evaluation (see Figure 7 and Figure 8), it is found that when the sub-image size is 224*224 pixels, the model performance reaches the optimal recognition effect. Under this size condition, the constructed sub-image dataset plays a key role in the subsequent model training and partition recognition. It can not only effectively process large-size face images, but also solve the problem of uneven lithology distribution, laying the foundation for accurate lithology recognition.

[0094] Example 4

[0095] The difference between this embodiment and the above embodiment is that this embodiment also introduces a decision-level fusion method in the lithology identification stage, and fuses the recognition results of sub-images of different sizes to obtain the lithology identification results of the specified area, further improving the accuracy of lithology identification. Specifically:

[0096] (1) The tunnel face images collected on site are segmented using an image segmentation model to remove irrelevant noise information, that is, the tunnel lining area images are removed, only the tunnel face area images are retained, and the images are transformed into a size of 3584*3584 pixels;

[0097] (2) Secondly, the tunnel face is cut into four images of different sizes according to the following four image cutting strategies:

[0098] ① The face image is evenly cut into 4 sub-images with a pixel size of 1792*1792 pixels in 2 rows and 2 columns, and each sub-image is positioned and numbered 1-1, 1-2, 2-1, 2-2 in order from left to right and from top to bottom;

[0099] ② The palm face image is evenly cut into 16 sub-images with a pixel size of 896*896 pixels in 4 rows and 4 columns, and each sub-image is positioned and numbered 1-1, 1-2, 1-3, 1-4, 2-1, 2-2, 2-3, 2-4, 3-1, 3-2, 3-3, 3-4 in order from left to right and from top to bottom;

[0100] ③ The palm face image is evenly cut into 8 rows and 8 columns, a total of 64 sub-images with a pixel size of 448*448 pixels, and each sub-image is positioned in sequence from left to right and from top to bottom and numbered as 1-1, 1-2, 1-3, 1-4, 1-5, 1-6, 1-7, 1-8, 2-1, 2-2, 2-3, 2-4, ..., 3-1, 3-2, 3-3, 3-4, ...;

[0101] ④ The palm face image is evenly cut into 16 rows and 16 columns, a total of 256 sub-images with a pixel size of 224*224 pixels, and each sub-image is positioned in sequence from left to right and from top to bottom and numbered as 1-1, 1-2, 1-3, 1-4, 1-5, 1-6, 1-7, 1-8, 2-1, 2-2, 2-3, 2-4, ..., 3-1, 3-2, 3-3, 3-4, ...;

[0102] (3) Sub-images of different sizes are sequentially input into the lithology recognition model for prediction. The specific prediction process is as follows:

[0103] ① Such as Fig. 9 As shown in the figure, at 224*224 pixels, the sub-image numbered 4-4 is recognized, and the recognition result is:

[0104]

[0105] Assume that the lithology identification result is P = (A, B, C, D, E, F, G, H, I, J), then the above result can be abbreviated as P 224 =(0,0,0,0,0,0,0,0,0.15,0.85);

[0106] ② Under 224*224 pixels, the image (1-1) under 1792*1792 pixels where the sub-image numbered 4-4 is located is recognized, and the result is P 1792 =(0.2, 0, 0, 0, 0, 0, 0, 0.07, 0.73);

[0107] ③ Under 224*224 pixels, the image (1-1) under 896*896 pixels where the sub-image numbered 4-4 is located is recognized, and the result is P 896 =(0.55, 0, 0, 0, 0, 0, 0, 0.05, 0.4);

[0108] ④ Under 224*224 pixels, the image (2-2) under 448*448 pixels where the sub-image numbered 4-4 is located is recognized, and the result is P 448 =(0,0,0,0,0,0,0,0,0.07,0.93).

[0109] (4) The above four recognition results are fused using the mean method, and the lithology probability distribution of the area numbered 4-4 under 224*224 pixels is obtained as P=((0+0.2+0.55+0) / 4,0,0,0,0,0,0,0,0,(0.15+0.07+0.05+0.07) / 4,(0.85+0.73+0.4+0.93) / 4)=(0.1875,0,0,0,0,0,0,0,0,0.085,0.7275). Through this decision-level fusion method, it can be seen that the probability that the 4-4 sub-image belongs to mudstone is the largest, which is 0.7275.

[0110] (5) Using the above method, the lithology recognition results of each 224*224 pixel sub-image are obtained in turn. The recognition results are further presented in a visual way. The specific process is as follows:

[0111] ① Set a blank white block for each 224*224 pixel sub-image;

[0112] ② Use different colors to represent different lithologies and fill in blank white blocks;

[0113] ③ Use transparency to indicate the size of lithology probability. The setting rule is: when lithology probability ∈ [0.8, 1], transparency = 1; when lithology probability ∈ [0.6, 0.8), transparency = 0.8; when lithology probability ∈ [0.4, 0.6), transparency = 0.6; when lithology probability ∈ [0.4, 0], transparency = 0.4;

[0114] ④ Combine the color blocks of each sub-image recognition result in order and superimpose them on the original tunnel face image. After using the recognition results of sub-images of different sizes for decision-level fusion and mapping them to a 224*224 pixel sub-image, the recognition result is closer to the actual lithology of the tunnel face, such as Fig.10 and 11 As shown in the figure, after the image recognition results are fused at the decision level, the strongly weathered sandstone and mudstone in the tunnel face can be more accurately identified and distinguished.

[0115] The method of the present invention solves the limitations of professional knowledge and test equipment in the current lithology identification method, as well as the problem of low recognition accuracy caused by the heterogeneity of the surrounding rock images of the tunnel face. By introducing computer vision technology, automatic segmentation of the tunnel face image is realized, a sub-image dataset of the surrounding rock image with a reasonable size is constructed, and a ResNet50 lithology identification model based on multi-scale feature fusion and lithology zoning identification of the tunnel face surrounding rock are proposed. Specifically, it has the following effects:

[0116] (1) Reduce the dependence on professional geological knowledge and test equipment: Through image segmentation and deep learning models, the demand for geological expertise and test equipment for rock type identification is reduced, allowing non-professionals to perform efficient rock type identification.

[0117] (2) Solve the problem of low accuracy in rock type recognition and heterogeneous rock type recognition: In view of the fact that the rock type of the tunnel face is affected by factors such as color, texture, and structural structure, as well as the problem of rock type heterogeneity that is common in the tunnel face, this patent provides a complete solution, including image segmentation, sub-image dataset construction, and an improved ResNet50 rock type recognition model, as well as regional recognition (the recognition effect of the rock type of the tunnel face is as follows: Fig.12 This improves the accurate identification of the heterogeneous lithology of the surrounding rock at the tunnel face.

[0118] Example 5

[0119] like Fig.13 As shown, a tunnel face surrounding rock lithology identification device includes at least one processor, a memory connected to the at least one processor, and at least one input / output interface connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a tunnel face surrounding rock lithology identification method described in the above embodiment. The input / output interface may include a display, a keyboard, a mouse, and a USB interface for inputting and outputting data.

[0120] Those skilled in the art can understand that: all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc. Various media that can store program codes.

[0121] When the above-mentioned integrated unit of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0122] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying the lithology of surrounding rock at a tunnel face, characterized in that: The following steps are involved: a: Cut the surrounding rock image of the tunnel face to be identified according to the preset size and output a sub-image set; b: inputting the sub-image set into a pre-built lithology recognition model in sequence, and the lithology recognition model outputs the recognition results corresponding to each sub-image; the recognition results include lithology classification and corresponding lithology probability; c: synthesizing the identification results corresponding to each sub-image, generating a lithology identification map corresponding to the surrounding rock image of the tunnel face to be identified; The pre-construction of the lithology identification model includes the following steps: S1: Constructing tunnel face surrounding rock image dataset; S2: Build a lithology recognition model, train the lithology recognition model through the tunnel face surrounding rock image data set, and output the model after the lithology recognition model converges; the lithology recognition model is a ResNet50 convolutional neural network.

2. A method for identifying the lithology of surrounding rock at a tunnel face according to claim 1, characterized in that: Said c comprises the following steps: c1: Fill each sub-image with different colors to represent different lithologies according to the lithology classification in the identification results corresponding to each sub-image; c2: According to the lithology probability in the identification results corresponding to each sub-image, the transparency is used to represent the lithology probability and the transparency of each sub-image is adjusted; c3: combining the adjusted sub-images in order and superimposing them on the original tunnel face image to generate a lithology identification image corresponding to the tunnel face surrounding rock image to be identified.

3. A method for identifying the lithology of surrounding rock at a tunnel face according to claim 2, characterized in that: The lithology classification includes mudstone, muddy sandstone, strongly weathered sandstone, moderately weathered sandstone, slightly weathered sandstone, strongly weathered limestone, moderately weathered limestone, slightly weathered limestone, shotcrete layer and black background.

4. A method for identifying the lithology of surrounding rock at a tunnel face according to claim 1, characterized in that: The cutting process includes automatic segmentation and image cutting; The automatic segmentation uses an image segmentation model to perform image segmentation, so as to segment out a main image of the surrounding rock of the tunnel face; The image cutting is used to cut the main image of the surrounding rock of the tunnel face into a plurality of sub-images of equal size according to a preset size.

5. A method for identifying the lithology of surrounding rock at a tunnel face according to claim 4, characterized in that: The image cutting in step a includes: The main image of the surrounding rock of the tunnel face is cut according to preset sizes of a*a pixels, 2a*2a pixels, 4a*4a pixels, and 8a*8a pixels to obtain four sub-image sets; At this time, in step b, the step of obtaining the identification results corresponding to each a*a sub-image includes: Obtain the identification result of the sub-image corresponding to the sub-image in the sub-image set of a*a pixels, 2a*2a pixels, 4a*4a pixels, and 8a*8a pixels in the lithology identification model; The various results are fused using the mean method, and the fused recognition result is used as the recognition result of the sub-image.

6. A method for identifying the lithology of surrounding rock at a tunnel face according to claim 4, characterized in that: The image segmentation model uses the U-Net convolutional neural network as the main framework, uses VGG16 as the feature extraction backbone network, and trains the model by segmenting and annotating the data set.

7. A method for identifying the lithology of surrounding rock at a tunnel face according to claim 6, characterized in that: The S1 comprises the following steps: S11: obtaining a tunnel face surrounding rock image, and manually marking the tunnel face surrounding rock image with corresponding lithology classification; S12: automatically segmenting the tunnel face surrounding rock image data after manual annotation and classification using the image segmentation model; S13: Cutting the tunnel face surrounding rock main image into a plurality of sub-images of equal size according to a preset size, and storing them as a tunnel face surrounding rock image dataset.

8. A method for identifying the lithology of surrounding rock at a tunnel face according to claim 4, characterized in that: The preset size of the image cutting is 224*224 pixels.

9. The method for identifying the lithology of surrounding rock at a tunnel face according to claim 1, characterized in that: The lithology recognition model includes three feature extraction branches, a fully connected layer and a Softmax function; The feature extraction branch includes a feature extraction layer, a convolution operation layer and an average pooling layer; The convolution kernel sizes of the feature extraction layers in the three feature extraction branches are 7*7, 1*1 and 5*5 respectively.

10. A tunnel face surrounding rock lithology identification device, characterized in that: It includes at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method described in any one of claims 1 to 9.

Images