Method and system for identifying multi-element geological information of tunnel face based on convolutional neural network

Abstract

The invention discloses a tunnel face multi-element geological information identification method and system based on a convolutional neural network, and relates to the technical field of tunnel engineering, and the method comprises the steps: inputting an obtained tunnel face image into a tunnel face geological information identification model, outputting the identification results of the three kinds of geological information of a tunnel face, and obtaining the identification result of the three kinds of geological information; the underground water type, the rock mass structure type and the rock mass weathering degree grade of the tunnel face are output; wherein the construction of the recognition model comprises the following steps: acquiring tunnel face image data on site, constructing a data set, and dividing the data set into a training set and a test set; on the basis of the training set, training of various classification models based on different convolutional neural networks is carried out on the geological information of each tunnel face; based on the test set, comprehensively evaluating each classification model according to model evaluation indexes, and selecting and constructing an optimal classification model; and constructing the three optimal classification models in parallel to obtain an identification model. According to the method, parallel, rapid and accurate intelligent identification and classification of the multi-element geological information of the tunnel face can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering, and in particular to a method and system for identifying multivariate geological information of a tunnel face based on a convolutional neural network. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Geological cataloging is an investigation work that records geological information and surrounding rock conditions at the newly excavated tunnel face. It can effectively obtain the geological conditions of the current tunnel face, so as to evaluate the geological disasters that may occur during tunnel construction, adjust the construction plan in time, and ensure the safe operation of the tunnel. As underground engineering construction continues to shift to complex areas in the central and western regions, the construction process continues to face extreme construction environments such as high cold, high ground temperature, rock burst risks, and high-intensity earthquake zones. Disasters such as surrounding rock collapse and sudden water inrush occur frequently. The status of geological cataloging in tunnel construction is becoming increasingly prominent.

[0004] At present, the traditional geological cataloging work of tunnel faces is mainly based on the on-site experience of engineering geologists. However, this method has certain disadvantages: on the one hand, it is difficult to objectively and accurately obtain geological information of the face due to its low efficiency, incomplete information collection, and great subjective influence of personnel; on the other hand, the harsh tunnel construction environment, thick dust and low light conditions are very inconvenient for geological cataloging personnel to conduct on-site analysis and measurement. At the same time, disasters such as surrounding rock collapse and sudden water inrush in the tunnel always threaten the safety of the staff. On this basis, the tunnel construction process is closely connected, and the time window from tunnel slag discharge to frame erection is short. Traditional geological cataloging work is difficult to meet this time requirement. Therefore, it is particularly important to improve the accuracy and efficiency of geological cataloging work without affecting the construction progress.

[0005] In existing research, some progress has been made in the identification of geological catalog information of tunnel faces. For example, patent CN115761038A proposed a method for geological sketching of tunnel faces based on image spectroscopy technology, which identifies the spatial distribution of minerals and water discharge in surrounding rocks through spectral information of tunnel faces; patent CN117011759A proposed a method for analyzing multivariate geological information of tunnel face surrounding rocks, which obtains structural surface parameters of tunnel face surrounding rocks by analyzing high-density point clouds, and conducts a comprehensive analysis of the stability of tunnel face surrounding rocks based on the analysis results of multivariate geological information of tunnel face high-definition photos, infrared thermal imaging videos and high-density point clouds. However, the above existing methods do not fully analyze and identify the factors of tunnel face geological information analysis, and the processing steps are cumbersome and inefficient. They cannot classify and process key information in a timely manner, and the economic benefits are low. Summary of the invention

[0006] In order to solve the shortcomings of the above-mentioned prior art, the present invention provides a method and system for identifying multivariate geological information of the tunnel face based on a convolutional neural network. Taking into account the multivariate geological information of the tunnel face, such as groundwater, rock structure and degree of rock weathering, the optimal classification model for efficient and accurate identification of each type of geological information is established respectively. The method can perform parallel rapid and intelligent identification of the multivariate geological information of the tunnel face, effectively solving the problems of incomplete analysis and identification factors and low efficiency in the existing methods, and meeting the time requirements for geological cataloging work in the tunnel construction process.

[0007] In a first aspect, the present invention provides a method for identifying multivariate geological information of a tunnel face based on a convolutional neural network.

[0008] A method for identifying multivariate geological information of a tunnel face based on a convolutional neural network, comprising:

[0009] Acquire tunnel face image data;

[0010] The acquired image data is input into the tunnel face geological information recognition model, and the recognition results of three geological information of the tunnel face are output, namely, the groundwater type, rock structure type and rock weathering degree of the tunnel face are output; wherein, the construction of the recognition model includes:

[0011] Collect tunnel face image data on site, build a data set, and divide it into training set and test set;

[0012] Based on the training set, multiple classification models based on different convolutional neural networks were trained for each type of tunnel face geological information; based on the test set, each classification model was comprehensively evaluated according to the model evaluation indicators, and the optimal classification model was selected and constructed; three optimal classification models were constructed in parallel to obtain the tunnel face geological information recognition model.

[0013] In a second aspect, the present invention provides a multi-dimensional geological information recognition system for a tunnel face based on a convolutional neural network.

[0014] A multi-dimensional geological information recognition system for a tunnel face based on a convolutional neural network, comprising:

[0015] An image acquisition module is used to acquire tunnel face image data;

[0016] The multi-dimensional geological information recognition module of the tunnel face is used to input the acquired image data into the geological information recognition model of the tunnel face, and output the recognition results of three geological information of the tunnel face, that is, output the groundwater type, rock structure type and rock weathering degree of the tunnel face; wherein, the construction of the recognition model includes:

[0017] Collect tunnel face image data on site, build a data set, and divide it into training set and test set;

[0018] Based on the training set, multiple classification models based on different convolutional neural networks were trained for each type of tunnel face geological information; based on the test set, each classification model was comprehensively evaluated according to the model evaluation indicators, and the optimal classification model was selected and constructed; three optimal classification models were constructed in parallel to obtain the tunnel face geological information recognition model.

[0019] In a third aspect, the present invention further provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps of the method described in the first aspect are completed.

[0020] In a fourth aspect, the present invention further provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the method described in the first aspect.

[0021] One or more of the above technical solutions have the following beneficial effects:

[0022] 1. The present invention provides a method and system for identifying multivariate geological information of the tunnel face based on a convolutional neural network. Taking into account multivariate geological information of the tunnel face, such as groundwater, rock mass structure and rock mass weathering degree, the optimal classification model for efficient and accurate identification of each type of geological information is established. The features in the image are extracted through deep learning to achieve parallel, rapid, intelligent and accurate identification and classification of multivariate geological information of the tunnel face, effectively solving the problems of incomplete analysis and identification factors and low efficiency in existing methods, and meeting the time requirements for geological cataloging work during tunnel construction. Compared with traditional methods, it can process more complex image data, greatly reduce interference from human factors, and improve accuracy.

[0023] 2. Based on the parallel processing and feature extraction capabilities of convolutional neural networks, the present invention helps tunnel designers and construction personnel to understand the geological conditions of the tunnel more accurately, thereby formulating more reasonable construction plans, optimizing construction strategies, and reducing risks and uncertainties in the construction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0025] Figure 1 A flow chart of constructing a tunnel face geological information identification model in an embodiment of the present invention;

[0026] Figure 2 Schematic diagram of image data enhancement processing in an embodiment of the present invention;

[0027] Figure 3A schematic diagram of a classification model training process in an embodiment of the present invention; DETAILED DESCRIPTION

[0028] It should be noted that the following detailed descriptions are exemplary only, are intended to describe specific embodiments, are intended to provide further explanation of the present invention, and are not intended to limit exemplary embodiments according to the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those of ordinary skill in the art to which the present invention belongs. In addition, it should also be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0029] Embodiment 1

[0030] This embodiment provides a method for identifying multivariate geological information of a tunnel face based on a convolutional neural network. Aiming at the full coverage identification of multivariate geological information factors of a tunnel face, a tunnel face geological information image dataset including three types of tunnel face groundwater, rock mass structure, and rock mass weathering degree is constructed. A tunnel face groundwater development classification model, a rock mass structure type classification model, and a rock mass weathering degree classification model based on Inception_v3, ResNet101, Densenet121, Efficientnet_B2, and MobileNet_v2 are established. The constructed datasets are used to train the above models. The model training is accelerated through transfer learning and data enhancement methods. Finally, combined with the model experimental results, a number of model evaluation indicators such as model loss value, accuracy, recognition speed and memory usage are comprehensively considered, and the optimal classification models for groundwater at the face, rock structure type and rock weathering degree are determined to be Efficientnet_B2, MobileNet_v2 and Densenet121 models, respectively. Their recognition accuracy rates on the test set reached 93.88%, 96.36% and 86% respectively, realizing efficient, intelligent and accurate recognition of three types of geological information: groundwater at the face, rock structure type and rock weathering degree.

[0031] The method for identifying multivariate geological information of a tunnel face based on a convolutional neural network proposed in this embodiment specifically includes:

[0032] Acquire tunnel face image data;

[0033] The acquired image data is input into the tunnel face geological information recognition model, and the recognition results of three geological information of the tunnel face are output, namely, the groundwater type, rock structure type and rock weathering degree of the tunnel face are output; wherein, the construction of the recognition model includes:

[0034] Collect tunnel face image data on site, build a data set, and divide it into training set and test set;

[0035] Based on the training set, multiple classification models based on different convolutional neural networks were trained for each type of tunnel face geological information; based on the test set, each classification model was comprehensively evaluated according to the model evaluation indicators, and the optimal classification model was selected and constructed; three optimal classification models were constructed in parallel to obtain the tunnel face geological information recognition model.

[0036] The following content introduces the method for identifying multivariate geological information of a tunnel face based on a convolutional neural network proposed in this embodiment in more detail.

[0037] like Figure 1 As shown, considering that there is no public tunnel face image dataset at present, tunnel face image data needs to be collected on site first. That is, step S1, collect tunnel face image data on site, construct a dataset, and divide it into a training set and a test set.

[0038] Step S1.1, collect tunnel face image data on site, and the geological engineer at the construction site classifies and summarizes the collected images, annotates and constructs three original image data sets of face geological information, namely, constructs the original image data set of groundwater in the face, the original image data set of rock mass structure and the original image data set of rock mass weathering degree.

[0039] Among them, during image acquisition, in order not to affect the construction progress and at the same time ensure the acquisition of clear and high-quality tunnel face images, the acquisition time is determined to be the time period between slag discharge, risk elimination and arch erection; at the same time, the fill light is set at a set distance in front of the tunnel face, such as about 6m, with one on each side to ensure uniform lighting on the tunnel face and clear rock mass characterization information, so as to ensure the accuracy of subsequent image processing and recognition; for the same tunnel face, multiple images at different positions, angles and heights are taken, the images obtained are used as original images, and the original images are classified and summarized to construct corresponding original image data sets.

[0040] Step S1.2: By image enhancement processing, the original image data set is expanded, an image data set of each type of tunnel face geological information is constructed, and a training set and a test set are divided.

[0041] Among them, based on the original image dataset, such as Figure 2As shown in the figure, three image enhancement methods, namely brightness processing, geometric transformation and random noise, are used to transform the original image to form the final face geological information image dataset, and the original dataset is randomly divided into training set and test set in a ratio of 8:2, laying the foundation for subsequent deep learning. Among them, brightness processing includes overall brightening, darkening and blurring of the original image; geometric transformation processing includes horizontal flipping, vertical flipping, rotating 30°, rotating 60°, rotating 90° and magnifying 1.5 times of the original image; random noise processing refers to adding random salt and pepper noise to the original image.

[0042] Step S2: Based on the training set, multiple classification models based on different convolutional neural networks are trained for each type of tunnel face geological information; based on the test set, each classification model is comprehensively evaluated according to the model evaluation index, and the optimal classification model is selected and constructed; three optimal classification models are constructed in parallel to obtain a tunnel face geological information recognition model.

[0043] Specifically, according to existing literature research and experimental comparison, this embodiment selects five different convolutional neural network models, namely Inception_v3, ResNet101, Densenet121, Efficientnet_B2 and MobileNet_v2, to respectively establish a classification model for groundwater development at the face, a classification model for rock structure type and a classification model for rock weathering degree, and uses the data set constructed in the above step S1 to train and verify each classification model, and evaluates each classification model through a series of preset model evaluation indicators to understand the performance of each network model in processing specific tasks and the advantages and limitations of the model, and then determine a more advantageous model to improve the recognition efficiency and accuracy when applied to specific objects.

[0044] like Figure 3 As shown, step S2.1, three types of tunnel face geological information include tunnel face groundwater, rock mass structure, and rock mass weathering degree. For any tunnel face geological information of tunnel face groundwater, rock mass structure, and rock mass weathering degree, classification models based on Inception_v3, ResNet101, Densenet121, Efficientnet_B2, and MobileNet_v2 are first established respectively.

[0045] Step S2.2: Based on the same training set, select the forward propagation algorithm, the back propagation algorithm, the loss function and the network optimizer to train each classification model; wherein, during the model training process, the model parameters are updated through the back propagation algorithm and the optimizer so that the loss value of the model training converges and the model training is completed.

[0046] Step S2.3: Based on the test set, evaluate the accuracy of each classification model, adjust the model hyperparameters to improve performance and prevent overfitting; and comprehensively evaluate each classification model according to the model evaluation index to determine the optimal classification model. Among them, the optimal classification model for groundwater classification at the tunnel face is determined to be Efficientnet_B2, the optimal classification model for rock mass structure classification is determined to be MobileNet_v2, and the optimal classification model for rock mass weathering degree classification is determined to be Densenet121.

[0047] Furthermore, in this embodiment, each classification model is comprehensively evaluated according to the model evaluation index to determine the optimal classification model. The model evaluation index includes the model loss value (the loss function in this embodiment selects the cross entropy loss function), accuracy, precision, recall, frame rate per second (FPS), confusion matrix size and model size (i.e., the size of the memory occupied), specifically:

[0048] (1) Accuracy: The proportion of samples whose prediction results are consistent with the actual results. The formula is:

[0049]

[0050] In the above formula, TP represents the number of samples that are predicted by the model to be positive and are actually positive, TN represents the number of samples that are predicted by the model to be negative and are actually negative, FP represents the number of samples that are predicted by the model to be positive and are actually negative, and FN represents the number of samples that are predicted by the model to be negative and are actually positive.

[0051] (2) Precision: The proportion of samples predicted to be positive and actually positive among the samples predicted to be positive by the model. The formula is:

[0052]

[0053] (3) Recall rate: The ratio of samples predicted by the model to be positive and actually positive to all positive samples. The formula is:

[0054]

[0055] (4) Confusion matrix: A table used to evaluate the performance of a classification model. It is a two-dimensional matrix used to compare the difference between the prediction results of the classification model and the actual results. The vertical axis of the confusion matrix represents the actual category of the data, the horizontal axis represents the category predicted by the model, and the numbers on the diagonal represent the probability that the model prediction is correct. The closer the number is to 1, the better the model classification effect.

[0056] (5) Frame rate per second: a measure of the model recognition speed. FPS is the number of frames that the target network can process per second, that is, 1÷inference time. The larger the FPS index, the more images the model can recognize per second, the faster the recognition speed, and the higher the efficiency in actual engineering applications.

[0057] (6) Based on the above model evaluation indicators, model size is also introduced as an evaluation indicator. The size of the model is determined according to the properties of the network model itself. The smaller the model, the lower the training cost and the shorter the training time, which is also more conducive to the deployment of the model on mobile devices in the later stage.

[0058] For each trained classification model, a test set is used for testing, and based on the test results, a comprehensive evaluation is performed according to the model evaluation index. This embodiment adopts a weighted scoring method for comprehensive evaluation based on multiple indicators, that is, each indicator is assigned a value according to the importance of each indicator; and each classification model is scored according to the ranking of each classification model under the current indicator and the assignment of the current indicator; for each classification model, the score of the classification model under each indicator is comprehensively considered to obtain the comprehensive evaluation result of the classification model; according to the comprehensive evaluation result of each classification model, the final optimal classification model is determined.

[0059] Specifically, for a network model, the most important are confusion matrix, model loss value, accuracy, precision and recall. Each of these five indicators is assigned 10 points, the top three are given 10, 9 and 8 points respectively, and the last two are given 6 and 4 points; the second is the frame rate per second FPS and model size. Each of these two indicators is assigned 5 points, and the value decreases from 5 points according to the ranking; the model with the highest total score calculated based on the above assignments is the optimal classification model.

[0060] In this embodiment, the performance of different deep learning models on the same data set is compared, and it is determined that the optimal classification model for groundwater at the face is Efficientnet_B2, the optimal classification model for rock structure classification is MobileNet_v2, and the optimal classification model for rock weathering degree classification is Densenet121.

[0061] (1) Regarding the geological information of the tunnel face, namely, groundwater at the tunnel face:

[0062] The five classification models of Inception_v3, ResNet101, Densenet121, Efficientnet_B2 and MobileNet_v2 were trained using the tunnel face groundwater image dataset (including training set and test set). During the training process, the model was trained for 50 rounds, the batch size was set to 16, the input image size was 224×224, the loss function selected the cross entropy loss, the optimizer selected the commonly used adaptive optimizer Adam, the initial learning rate was set to 0.001, and the Softmax output channel was set to the five types of tunnel face groundwater, namely, tunnel face dryness, seepage, linear water, strand water and gushing water.

[0063] Through training, it can be determined that the model loss value based on Efficientnet_B2 is the lowest, which is stable at around 0.06, the model loss value of ResNet101 is stable at around 0.09, the model loss value of Densenet121 is stable at around 0.1, the model loss value of MobileNet_v2 is stable at around 0.15, and the model loss value of Inception_v3 is stable at around 0.29; ResNet101, Efficientnet_B2 and MobileNet_v2 models can process 10 images per second, the Inception_v3 model can process 7 images per second, and the Densenet121 model can process 6 images per second; Accuracy is used to compare the recognition accuracy of the five models. The Densenet121 and Efficientnet_B2 models have the highest accuracy, reaching 97.33% and the curves are almost overlapping. The model accuracy of ResNet101 and MobileNet_v2 is second, which is 95.41% and 96.91% respectively. The model accuracy of Inception_v3 is the lowest at 87.88%, and the model accuracy of Efficientnet_B2 is the highest at 97.94%. The recall rate of Efficientnet_B2 is 97.82%, the model accuracy of Densenet121 is 97.32%, and the model accuracy of ResNet101 is 96.71%. The model accuracy of MobileNet_v2 fluctuates within a small range, with an accuracy value of 94.61%. The model accuracy of ion_v3 fluctuates around 89.95%; the recall curves and precision curves of the five models on the training set are basically consistent, the recall curves of Densenet121, ResNet101 and Efficientnet_B2 models are approximately overlapped, and the model recall values ​​are 96.71%, 97.48% and 97.82% respectively, the model recall of MobileNet_v2 is 93.72%, and the model recall value of Inception_v3 fluctuates around 86.59%. Based on the above analysis, the comprehensive evaluation results of each classification model are calculated and determined, and the final optimal classification model is determined to be the Efficientnet_B2 model, which has high model recognition accuracy, fast recognition speed, and small memory usage in the task of groundwater image classification at the tunnel face, which is convenient for the model to be deployed in the actual system later.

[0064] (2) Regarding the geological information of the tunnel face, namely the rock mass structure of the tunnel face:

[0065] The five classification models of Inception_v3, ResNet101, Densenet121, Efficientnet_B2 and MobileNet_v2 were trained under the TensorFlow framework using the tunnel face rock structure image dataset (including training set and test set). The cross entropy was selected as the loss function for model optimization. The experiments were carried out under the same computer hardware. The model training was set to 50 rounds, the batch size was 16, the input image size was 224×224, the commonly used adaptive optimizer Adam was selected as the optimizer, the initial learning rate was set to 0.001, and the Softmax output channel was set to four types of rock structure types, namely layered structure, block structure, fragmented structure and scattered structure.

[0066] Through training, it can be determined that the loss value of Densenet121 fluctuates the most, indicating that the model may be overfitting, followed by ResNet101 and Efficientnet_B2, and Inception_v3 and MobileNet_v2 have the smallest volatility; MobileNet_v2 and ResNet101 models can process 9 images per second, Densenet121 models can process 4 images per second, and Inception_v3 and Efficientnet_B2 models can process 5 images per second; The test accuracy of Inception_v3 and MobileNet_v2 models fluctuates the least, stabilizing at around 98.42%, and the test accuracy of ResNet101 and Efficientnet_B2 models is The test accuracy of the Densenet121 model fluctuates the most, fluctuating around 93.33%; the precision curves of the three models Densenet121, ResNet101 and Efficientnet_B2 almost overlap, with the highest reaching 99.11%, the precision of the MobileNet_v2 model is second at 98.68%, and the precision of the Inception_v3 model is the lowest at 95.53%; the recall curves of the three models Densenet121, ResNet101 and Efficientnet_B2 almost overlap, with the highest reaching 98.99%, the recall rate of the MobileNet_v2 model is second at 98.47%, and the recall rate of the Inception_v3 model is the lowest at 94.91%. Based on the above analysis, the comprehensive evaluation results of each classification model are calculated and determined, and the optimal classification model for the final rock structure image classification task is determined to be the MobileNet_v2 model. This model has high recognition accuracy, fast recognition speed, and small memory usage, which is convenient for subsequent model deployment.

[0067] (3) Regarding the geological information of the face, i.e. the degree of weathering of the rock mass:

[0068] The five classification models of Inception_v3, ResNet101, Densenet121, Efficientnet_B2 and MobileNet_v2 were trained under the TensorFlow framework using the image dataset of rock weathering degree of the tunnel face (including training set and test set). The cross entropy was selected as the loss function for model optimization. The experiments were carried out under the same computer hardware. The model training was set to 50 rounds, the batch size was 16, the input image size was 224×224, the commonly used adaptive optimizer Adam was selected as the optimizer, the initial learning rate was set to 0.001, and the Softmax output channel was set to the four types of rock weathering degree, namely slightly weathered, moderately weathered, strongly weathered and fully weathered.

[0069] Similar to the above method, this step will not be described in detail. Through training, it can be determined that: the loss value of Densenet121 is stable at around 0.02 on the training set, and the convergence effect is good; the test accuracy of the Densenet121 model is relatively stable, fluctuating around 86%; Densenet121 has the highest recall rate, fluctuating around 90.8%; the rock weathering degree classification model based on Densenet121 has a certain probability of identifying other types for the three categories of moderate weathering, strong weathering and full weathering, but the recognition accuracy rate reaches 88%. The Densenet121 model has a high model recognition accuracy and small memory usage in the rock weathering degree image classification task, which is convenient for later model deployment.

[0070] Based on the three optimal classification models determined above, a face geological information recognition model is constructed through parallel setting. On this basis, the tunnel face image data is obtained, and the acquired image data is input into the face geological information recognition model, and the recognition results of the three geological information of the tunnel face are output, that is, the groundwater type, rock structure type and rock weathering degree of the tunnel face are output, so as to realize the rapid and accurate classification and recognition of the multi-dimensional geological information of the face.

[0071] Embodiment 2

[0072] This embodiment provides a multivariate geological information recognition system for a tunnel face based on a convolutional neural network, comprising:

[0073] An image acquisition module is used to acquire tunnel face image data;

[0074] The multi-dimensional geological information recognition module of the tunnel face is used to input the acquired image data into the geological information recognition model of the tunnel face, and output the recognition results of three geological information of the tunnel face, that is, output the groundwater type, rock structure type and rock weathering degree of the tunnel face; wherein, the construction of the recognition model includes:

[0075] Collect tunnel face image data on site, build a data set, and divide it into training set and test set;

[0076] Based on the training set, multiple classification models based on different convolutional neural networks were trained for each type of tunnel face geological information; based on the test set, each classification model was comprehensively evaluated according to the model evaluation indicators, and the optimal classification model was selected and constructed; three optimal classification models were constructed in parallel to obtain the tunnel face geological information recognition model.

[0077] Embodiment 3

[0078] This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the method for identifying multivariate geological information of a tunnel face based on a convolutional neural network as described above are completed.

[0079] Embodiment 4

[0080] This embodiment also provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in the method for identifying multivariate geological information of a tunnel face based on a convolutional neural network as described above are completed.

[0081] The steps involved in the above embodiments 2 to 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0082] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0083] The above description is only a preferred embodiment of the present invention. Although the specific implementation mode of the present invention is described in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present invention.

Claims

1. A method for identifying multivariate geological information of a tunnel face based on a convolutional neural network, characterized in that: include: Acquire tunnel face image data; The acquired image data is input into the tunnel face geological information recognition model, and the recognition results of three geological information of the tunnel face are output, namely, the groundwater type, rock structure type and rock weathering degree of the tunnel face are output; wherein, the construction of the recognition model includes: Collect tunnel face image data on site, build a data set, and divide it into training set and test set; Based on the training set, multiple classification models based on different convolutional neural networks were trained for each type of tunnel face geological information; based on the test set, each classification model was comprehensively evaluated according to the model evaluation indicators, and the optimal classification model was selected and constructed; three optimal classification models were constructed in parallel to obtain the tunnel face geological information recognition model.

2. The method for identifying multivariate geological information of a tunnel face based on a convolutional neural network according to claim 1, characterized in that: The model evaluation indicators include model loss value, accuracy, precision, recall rate, frame rate per second, confusion matrix size and occupied memory size.

3. The method for identifying multivariate geological information of a tunnel face based on a convolutional neural network according to claim 1, characterized in that: The construction of the data set includes: Collect tunnel face image data on site, annotate and construct three types of original image data sets of face geological information; The original image data set is expanded through image enhancement processing to construct an image data set of each type of tunnel face geological information; the image enhancement includes brightness processing, geometric transformation and random noise.

4. The method for identifying multivariate geological information of a tunnel face based on a convolutional neural network according to claim 1, characterized in that: When collecting tunnel face image data on site, the collection time is the period between slag discharge, risk elimination and arch erection.

5. The method for identifying multivariate geological information of a tunnel face based on a convolutional neural network according to claim 1, characterized in that: The three types of face geological information include face groundwater, rock mass structure, and rock mass weathering degree; For any of the tunnel face geological information, including groundwater, rock mass structure, and rock mass weathering degree, classification models based on Inception_v3, ResNet101, Densenet121, Efficientnet_B2, and MobileNet_v2 are established respectively; Based on the same training set, select the forward propagation algorithm, back propagation algorithm, loss function and network optimizer to train each classification model; Based on the test set, each classification model was comprehensively evaluated according to the model evaluation indicators, and the optimal classification model for groundwater classification at the tunnel face was determined to be Efficientnet_B2, the optimal classification model for rock structure classification was determined to be MobileNet_v2, and the optimal classification model for rock weathering degree classification was determined to be Densenet121.

6. The method for identifying multivariate geological information of a tunnel face based on a convolutional neural network according to claim 5, characterized in that: The types of groundwater at the tunnel face include five types, namely, dry tunnel face, seepage water, linear water, strand water and gushing water; The rock mass structure types include four types, namely, layered structure, massive structure, fragmented structure and dispersed structure; The weathering degree of the rock mass includes four types, namely, slightly weathered, moderately weathered, strongly weathered and fully weathered.

7. A multivariate geological information recognition system for a tunnel face based on a convolutional neural network, characterized in that: include: An image acquisition module is used to acquire tunnel face image data; The multi-dimensional geological information recognition module of the tunnel face is used to input the acquired image data into the geological information recognition model of the tunnel face, and output the recognition results of three geological information of the tunnel face, that is, output the groundwater type, rock structure type and rock weathering degree of the tunnel face; wherein, the construction of the recognition model includes: Collect tunnel face image data on site, build a data set, and divide it into training set and test set; Based on the training set, multiple classification models based on different convolutional neural networks were trained for each type of tunnel face geological information; based on the test set, each classification model was comprehensively evaluated according to the model evaluation indicators, and the optimal classification model was selected and constructed; three optimal classification models were constructed in parallel to obtain the tunnel face geological information recognition model.

8. The multivariate geological information recognition system for a tunnel face based on a convolutional neural network according to claim 7, characterized in that: The model evaluation indicators include model loss value, accuracy, precision, recall rate, frame rate per second, confusion matrix size and occupied memory size.

9. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of a method for identifying multivariate geological information of a tunnel face based on a convolutional neural network are completed as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of a method for identifying multivariate geological information of a tunnel face based on a convolutional neural network as described in any one of claims 1 to 6.

Images