Tunnel face image descriptive model construction method, system, equipment and medium

Abstract

The invention discloses a tunnel face image descriptive model construction method, system and device and a medium, and relates to the technical field of tunnel engineering and computer vision, and the method comprises the steps: obtaining a tunnel face geological sketch of each mileage and each tunnel face image of the corresponding mileage; geological description is extracted from the geological sketch of each tunnel face, and standardization processing is carried out; extracting image high-dimensional features from the preprocessed tunnel face image; constructing and forming a training sample through the high-dimensional features of each image and the geological description which corresponds to the mileage and is subjected to standardized processing; training a preset initial model by using the training sample, and determining the initial model meeting a preset training ending condition as a Bert tunnel face image description model; according to the method, a complete technical path for directly generating the geological sketch text from the image is provided, error accumulation and complex deployment caused by series connection of multiple models are avoided, and the accuracy and efficiency of geological logging of tunnel construction are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel engineering and computer vision, more particularly, it relates to a tunnel face image descriptive model construction method, system, device and medium. BACKGROUND

[0002] In the process of tunnel construction, the geological information of the tunnel face is an important basis for evaluating the stability of surrounding rock, formulating support parameters and optimizing construction technology. The traditional geological recording of the tunnel face mainly relies on manual field observation and recording, or is completed by multiple independent models such as lithology identification model, structure surface identification model and underground water identification model. However, these methods have problems such as low efficiency, non-uniform standards, complex model links, high misjudgment rate, etc. Especially in the case of insufficient light, broken surface of the tunnel face and unobvious geological features, the accuracy and stability of manual identification and multi-model joint identification are difficult to guarantee. In recent years, with the development of deep learning and computer vision, it is possible to use tunnel face images to realize automatic extraction of geological information, but existing researches are mostly limited to single tasks such as crack detection and rock identification, and lack of end-to-end technical route for directly generating complete standardized geological description from tunnel face images. SUMMARY

[0003] The purpose of the present application is to provide a tunnel face image descriptive model construction method, system, device and medium to realize automatic extraction, standardized processing and high-precision description generation of tunnel face geological information, thereby improving the efficiency and accuracy of tunnel construction geological information recording, and solving the problems of existing tunnel face geological information acquisition methods, such as relying on manual recording, low information standardization, image quality being seriously affected by factors such as on-site light and dust, and lack of model construction method for automatically generating standardized geological description.

[0004] The above technical purpose of the present application is realized by the following technical scheme: In a first aspect, the present application provides a tunnel face image descriptive model construction method, comprising the following specific steps: Obtaining tunnel face geological sketches at each mileage in the construction process, and each tunnel face image at the corresponding mileage; Identifying and extracting geological descriptions from each tunnel face geological sketch by a deep learning algorithm, and performing standardized processing on the geological descriptions; Preprocessing each tunnel face image, and extracting image high-dimensional features from the preprocessed tunnel face image by a deep learning algorithm; Constructing training samples from each image high-dimensional feature and the corresponding mileage and standardized processed geological description; The initial model is trained by using the training sample until the initial model reaches the preset training end condition, and the initial model reaching the preset training end condition is determined as the Bert tunnel face image description model.

[0005] Based on the above technical solution, the application can be further improved as follows.

[0006] Further, the geological description is obtained by the following way: The text description area of the tunnel face geological sketch is detected by the trained yolov5 target detection model to obtain a detection area. The Convolutional Recurrent Neural Network algorithm is used to extract the text of the detection area to obtain the geological description.

[0007] Further, the above-mentioned pre-processing of each tunnel face image includes image quality enhancement processing and brightness color repair processing.

[0008] Further, the training sample includes a double-modal data set of each mileage, and the double-modal data set is formed by pairing the image high-dimensional features of each mileage and the corresponding mileage geological description after standardization processing.

[0009] Further, the image high-dimensional features are extracted by the Compact Convolutional Transformer model combined with convolution and Transformer.

[0010] In a second aspect, the application provides a tunnel face image description model construction system applied to the tunnel face image description model construction method of any one of the first aspect, comprising: The construction data acquisition module is used to acquire the tunnel face geological sketch of each mileage in the construction process and the tunnel face image of each mileage corresponding to the mileage; The geological sketch processing module is used to identify and extract the geological description from each tunnel face geological sketch by using a deep learning algorithm, and to standardize the geological description; The high-dimensional feature extraction module is used to pre-process each tunnel face image, and to extract the image high-dimensional features from the pre-processed tunnel face image by using a deep learning algorithm; The training sample construction module is used to construct the training sample by using each image high-dimensional feature and the corresponding mileage geological description after standardization processing; The description model determining module is configured to train a preset initial model by using training samples until the initial model reaches a preset training end condition, and determine the initial model reaching the preset training end condition as the Bert tunnel face image description model.

[0011] Further, in the construction data obtaining module, the geological description is obtained by the following method: The text description area of the tunnel face geological sketch is detected by using the trained yolov5 target detection model, and a detection area is obtained. The Convolutional Recurrent Neural Network algorithm is used to extract the text in the detection area, and the geological description is obtained.

[0012] Further, in the high-dimensional feature extraction module, each tunnel face image is preprocessed, and the preprocessing includes image quality enhancement processing and brightness color repair processing. The image quality enhancement processing is completed by using a Real-ESRGAN model. The brightness color repair processing is completed by using a Retinex-Net algorithm.

[0013] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the tunnel face image description model construction method of any one of the first aspect.

[0014] In a fourth aspect, the present application provides a non-transitory computer readable storage medium, which stores computer instructions, and the computer instructions make the computer execute the tunnel face image description model construction method of any one of the first aspect.

[0015] Compared with the prior art, the present application has at least the following beneficial effects: 1. The present application realizes the technical route of directly generating geological sketch text from tunnel face images, avoiding the cumbersome process of constructing multiple independent models such as lithology identification, structure surface identification, and underground water identification in traditional methods, and forming an integrated processing path of image input-text output.

[0016] 2. The training and integration of multiple models are omitted, the overall system structure is more simple, the deployment cost and operation and maintenance cost are significantly reduced, and it is beneficial to rapid deployment and long-term stable operation on the construction site.

[0017] 3. The standardized text description is generated by a unified model, avoiding differences between different recorders or different models, ensuring the uniformity and standardization of geological information recording, and improving the usability of data in subsequent analysis and archiving.

[0018] 4. The one-step conversion mode reduces intermediate data processing links, making the geological information generation more real-time. At the same time, the method can maintain high robustness under different light, dust and other complex construction conditions, and has stronger adaptability. BRIEF DESCRIPTION OF DRAWINGS

[0019] The accompanying drawings, which are included to provide a further understanding of the embodiments of the present application and are incorporated in and constitute a part of this application, illustrate embodiments of the present application and, together with the description, serve to explain the principles of the present application. In the drawings: Figure 1 The method flowchart of the construction method in the embodiments of the present application; Figure 2 The schematic diagram of the standardization processing of the geological description in the embodiments of the present application; Figure 3 The schematic diagram of the enhanced resolution of the tunnel face image in the embodiments of the present application, in which Figure 3 (a) is a low-quality image under one kind of scene, (c) is a low-quality image under another kind of scene, (b) and (d) are respectively corresponding to (a) and (c) and are super-resolution enhanced images; Figure 4 The schematic diagram of the enhanced illumination of the tunnel face image in the embodiments of the present application, in which Figure 4 (a) is a low-illumination image under one kind of scene, (c) is a low-illumination image under another kind of scene, (b) and (d) are respectively corresponding to (a) and (c) and are illumination enhanced images; Figure 5 The training flowchart of the Bert tunnel face image description model in the embodiments of the present application. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of embodiments of the application provided in the accompanying drawings is not intended to limit the scope of the application claimed, but merely represents selected embodiments of the application. All other embodiments obtained based on the embodiments in the application by those of ordinary skill in the art without creative labor fall within the scope of the protection of the application.

[0022] It should be noted that similar reference numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0023] In the description of embodiments of the application, "a plurality of" represents at least 2.

[0024] Example 1: Since there is currently a lack of a technical method capable of directly generating standardized geological sketch text from tunnel face image, this embodiment provides a tunnel face image descriptive model construction method, which should have the characteristics of data acquisition automation, geological description extraction intelligence, image enhancement and feature extraction refinement, and text generation standardization, in order to improve the efficiency and accuracy of geological logging, reduce the influence of human subjective factors, and promote the development of tunnel construction informatization and intelligence, such as Figure 1 As shown in the figure, the method comprises the following specific steps: S1, obtaining the tunnel face geological sketch at each mileage during construction, and the tunnel face image corresponding to each mileage.

[0025] Among them, during the tunneling process, the construction personnel will draw the face geological sketch after the completion of each construction mileage, this face geological sketch includes: drawing the face geological sketch according to the specification by the geological personnel or the supervision unit during the tunneling process, and making detailed annotations on the rock layer distribution, joint fissure, fault, water content, etc.; when the tunnel construction is completed, collect and organize the tunnel face geological sketch of each corresponding mileage section of the whole line, ensure that the sketch is clear and identifiable, and the annotations are complete, so as to provide a reliable data basis for subsequent extraction and analysis of the text description of the face, and the face geological sketch is shown in Table 1.

[0026] Table 1

[0027] Among them, the geological condition description is: the lithology of the face is sandstone, purple red, weak weathering (W), thick bedded, the bedding strikes a large angle with the tunnel axis, the tendency is on the left side of the large mileage end line, the inclination is about 39 degrees, there is a right side bedding bias on the face, the joint fissure is relatively developed, the combination degree is general, the rock mass is relatively broken to complete, the surrounding rock is in mosaic and fragmented structure, the face is damp, the excavation free arch part is easy to fall, and the surrounding rock grade is recommended to be grade III.

[0028] Further, to realize the data construction of the correspondence between the text and the image, the real scene images of the tunnel face corresponding to the geological sketch at the mileage are synchronously collected during the tunnel excavation process. The image collection can use a high-definition industrial camera, a single-lens reflex camera or an existing monitoring device on the construction site to ensure that the features of the tunnel face are completely presented in the image and completely correspond to the above-mentioned sketch data in terms of mileage number.

[0029] S2, identifying and extracting the geological description from each tunnel face geological sketch by a deep learning algorithm, and standardizing the geological description.

[0030] Optionally, the above-mentioned geological description can be automatically extracted from the collected tunnel face geological sketch image based on a deep learning target detection and character recognition algorithm. Specifically, the geological description information can be obtained by the following way: S21, detecting the text description area of the tunnel face geological sketch by the yolov5 target detection model trained to obtain a detection area.

[0031] The detection and positioning algorithm includes the following steps: The yolov5 obtains the four coordinate positions of the detection frame by convolution operation, pooling operation, SiLU activation as formulas (1)-(3), and finally by Sigmoid function.

[0032] (1) In the formula, represents the pixel value of the depth k at the position (i, j) in the input feature map of the convolution layer, represents the pixel value of the depth p at the position (i+m, j+n) in the input data, represents the weight value of the depth p at the position (m, n) in the convolution kernel, and the depth k in the output feature map, represents the bias term of the depth k in the output feature map, and f represents the size of the convolution kernel.

[0033] Further, (2) In the formula, represents the pooling function (such as maximum pooling and average pooling), represents the input tensor all values within the pooling window.

[0034] Further, in the above, (3) In the formula, represents the output of the activation function, represents the input, e is a mathematical constant, which is approximately equal to 2.71828.

[0035] Further, in the above: (4) wherein, is the input value, is the pixel value of the model output, whose value is between 0-1.

[0036] S22, using Convolutional Recurrent Neural Network algorithm to extract the text in the detection area, and obtaining the geological description; wherein, Convolutional Recurrent Neural Network (CRNN) is a hybrid architecture combining convolutional neural network (CNN) and recurrent neural network (RNN), which can be used to process sequence data and image data, and can capture the spatial structure features (such as local patterns in images) and sequence dependencies (such as time sequence dependencies in text) of data at the same time.

[0037] wherein, the detection output contains the bounding box coordinates of each text region; then, the detected text region image is subjected to optical character recognition by using Convolutional Recurrent Neural Network (CRNN) algorithm, so as to accurately extract the text information therein, and obtain the original geological description text; through convolution operation, pooling operation and ReLU activation as formula (5)-(7), the extracted text is outputted through the text extraction formula, as formula (8).

[0038] (5) wherein, represents the pixel value of depth k at position (i, j) in the convolution layer input feature map, 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, represents the bias term of depth k in the output feature map, and f represents the size of the convolution kernel.

[0039] Further, in the above: (6) wherein, represents the pooling function (such as maximum pooling and average pooling, etc.), represents the input tensor all values within the pooling window.

[0040] Further, in the above: (7) wherein, represents the output of the activation function, represents input.

[0041] Further, in the above: , (8) wherein, represents a convolution feature sequence extracted by formula (5) - (7), represents the hidden state of the RNN, represents the output text sequence, and the RNN represents a neural network (continuous function mapping), is a probability sequence decoding algorithm.

[0042] In the above steps S21 - S22, first, a labeled training yolov5 target detection model is used to accurately detect and locate the text description area in the tunnel face geological sketch, ensuring that the text box area can be accurately identified in a complex line background; then, a Convolutional Recurrent Neural Network (CRNN) algorithm is used to recognize characters and model sequences in the detected text area, realizing automatic extraction of handwritten or printed text, and thus automatically obtaining the original geological description text information.

[0043] Optionally, when the geological description is standardized, due to differences in terminology, expression, format, etc. used by different construction personnel in the geological sketch, the original geological description text style is inconsistent; this embodiment uses a large language model (such as DeepSeek, etc.) to standardize the original text extracted as shown in Figure 2 , unifies the geological description into a standardized format, including the standardized expression of lithology, color, weathering degree, structure surface occurrence, groundwater condition, surrounding rock structure characteristics, stability, etc. elements, so as to ensure the consistency and comparability of the data in subsequent analysis and modeling.

[0044] S3, pre-process each tunnel face image, and extract image high-dimensional features from the pre-processed tunnel face image through a deep learning algorithm.

[0045] Among them, due to insufficient lighting, dust interference, camera shaking and other factors on the construction site, the collected tunnel face images may have low resolution, insufficient brightness, color distortion and other problems, so pre-processing is needed. The schematic diagram of each image before and after processing is shown in Figures 3-4 , in Figure 3 , (a) and (c) are low-quality images, (b) and (d) are super-resolution enhanced images corresponding to (a) and (c) respectively, and Figure 4In the embodiment, (a) and (c) are low-light images, and (b) and (d) are light-enhanced images corresponding to (a) and (c) respectively.

[0046] Optionally, the tunnel face image is preprocessed, and the preprocessing includes image quality enhancement processing and brightness and color restoration processing; the image quality enhancement processing is completed by a Real-ESRGAN model; the brightness and color restoration processing is completed by a Retinex-Net algorithm; the Real-ESRGAN is an open source image / video super resolution algorithm, and the image restoration and resolution improvement can be realized by downloading a pre-compiled program or GitHub source code; the Retinex-Net algorithm combines the ideas of deep learning and Retinex theory, and can effectively improve the visual quality of the image; the Retinex-Net algorithm can enhance and defog the image by simulating the processing mode of the human visual system.

[0047] In the embodiment, the Real-ESRGAN model can be used to perform super-resolution reconstruction on the image resolution, improve the image detail performance, and make the micro cracks and rock layer textures clearer; meanwhile, the Retinex-Net is used to restore the brightness and color of the image, improve the uneven illumination and color deviation problems caused by construction light, dust, water mist and the like, and restore the real color and texture characteristics of the tunnel face under the natural light condition as much as possible; the two models still perform convolution operation, pooling operation, full connection operation and ReLu activation, as shown in formulas (9)-(12), and finally obtain the distribution of each pixel point through a Sigmoid function, as shown in formula (13).

[0048] (9) wherein, represents a pixel value of a depth k at a position (i, j) in a convolution layer input feature map, represents a pixel value of a depth p at a position (i+m, j+n) in input data, represents a weight value of a depth p at a position (m, n) in a convolution kernel, and a depth k in an output feature map, represents a bias term of a depth k in an output feature map, and f represents a convolution kernel size.

[0049] (10) wherein, represents a pooling function (such as maximum pooling and average pooling), represents an input tensor all values within a pooling window.

[0050] (11) wherein, denotes the output of the fully connected layer, W denotes the weight matrix, and b denotes the bias term.

[0051] (12) wherein, denotes the output of the activation function, denotes the input.

[0052] (13) wherein, is the input value, is the pixel value of the model output, which is between 0 and 1.

[0053] Optionally, the image high-dimensional features are extracted by a Compact Convolutional Transformer model combined with convolution and Transformer; wherein the Compact Convolutional Transformer (CCT) is a hybrid architecture model combining the advantages of convolutional neural network (CNN) and Transformer, which can improve the efficiency of computer vision tasks and reduce the demand for data size; the Transformer model architecture uses a Self-Attention structure to replace the commonly used RNN network structure in NLP tasks, which can perform parallel computing compared to the RNN network structure.

[0054] The Transformer is essentially an Encoder-Decoder architecture; therefore, the Transformer in the middle part can be divided into two parts: an encoding component and a decoding component.

[0055] In this embodiment, the Compact Convolutional Transformer (CCT) network can be used to extract features from the enhanced tunnel face image, generating a high-dimensional feature vector that can represent key information such as lithology, structural plane features, color texture, etc.; this feature vector is input as a deep feature of the image end, used for subsequent image-text joint modeling, which is still extracted through convolution operation, pooling operation, full connection operation, ReLu activation, and multi-head attention, as shown in formulas (14)-(18).

[0056] (14) wherein, denotes the pixel value of depth k at position (i, j) in the input feature map of the convolution layer, denotes the pixel value of depth p at position (i+m, j+n) in the input data, represents the weight value of the position (m, n) in the depth p in the convolution kernel, and the depth k in the output feature map, represents the bias item of the depth k in the output feature map, and f represents the size of the convolution kernel.

[0057] (15) wherein, represents a pooling function (such as maximum pooling and average pooling), represents an input tensor all values within the pooling window.

[0058] (16) wherein, represents the output of the fully connected layer, W represents the weight matrix, and b represents the bias item.

[0059] (17) wherein, represents the output of the activation function, represents the input.

[0060] , ; (18) wherein, Z is a feature vector extracted by a fully connected layer, is a model learnable vector, head is a self-attention calculated between Q, K and V, and Softmax is a function operation, is a multi-head mechanism, Concat is a function operation, and h represents the number of self-attention heads, is a linear transformation.

[0061] S4, a training sample is formed by each image high-dimensional feature and the corresponding geological description processed by standardization.

[0062] wherein, the training sample includes a double-modal data set of each mileage, and the double-modal data set is formed by pairing each image high-dimensional feature and the corresponding geological description processed by standardization; specifically, the standardized geological description of the working face is paired with the corresponding enhanced high-definition working face image to form a double-modal data set corresponding to the image and the text, and the double-modal data set can be divided into a training set, a verification set and a test set according to a certain proportion, and is respectively used for training, verification and testing of the model.

[0063] S5, training the preset initial model by using the training samples until the initial model reaches a preset training end condition, and determining the initial model reaching the preset training end condition as the Bert tunnel face image description model.

[0064] In the embodiment, a Bert model is used to construct a tunnel face image description model. In the training process, as shown in Figure 5 illustrated, the Bert model receives image feature vectors and corresponding text descriptions from the CCT network, realizes fusion learning of image and text features, and enables the model to directly generate standardized geological description texts according to the face image. The model realizes generation of the geological description texts through fusion of the image and text features, forward calculation, and prediction of text encoding as shown in formulas (19) and (20).

[0065] (19) wherein, is the extracted image feature (i.e., the image high-dimensional feature mentioned above), is a word vector embedding of the geological description, and Concat is a function operation, represents the image feature and the word vector embedding of the geological description concatenated.

[0066] ; (20) wherein, TransformerLayer represents a neural network layer calculation, is a hidden state matrix of an (l-1)-th layer Transformer encoder, wherein is an encoding operation for text encoding of a model extraction result, is a hidden state matrix of an l-th layer Transformer encoder, represents a text output generated after decoding by the Decoder.

[0067] Specifically, after the model is trained, it can be applied in real time at the construction site to realize automatic conversion from the face image to the standardized geological description, and improve the efficiency and accuracy of geological information recording in the construction process.

[0068] The tunnel face image descriptive model construction method provided in the embodiment is based on tunnel face construction site collected images and corresponding geological sketches, and a model capable of automatically generating standardized geological description is constructed through image quality enhancement, character recognition, text standardization processing, and image-text feature fusion steps. Meanwhile, the method realizes a technical path of directly generating geological sketch text from the tunnel face image, significantly improves the efficiency and accuracy of geological information recording in the construction process, overcomes the problems of low efficiency and subjective bias of traditional manual recording, avoids the complex process of constructing multiple geological identification models, and has the characteristics of simple deployment and strong adaptability.

[0069] Embodiment 2: The tunnel face image descriptive model construction system provided in the embodiment is applied to the tunnel face image descriptive model construction method of embodiment 1, and can include: A construction data acquisition module is configured to acquire tunnel face geological sketches at each mileage in the construction process and each tunnel face image at the corresponding mileage. In the construction data acquisition module, the geological description is obtained by the following method: A yolov5 target detection model is used to detect the text description area of the tunnel face geological sketch, and a detection area is obtained. A Convolutional Recurrent Neural Network algorithm is used to extract text from the detection area to obtain the geological description.

[0070] A geological sketch processing module is configured to identify and extract geological descriptions from each tunnel face geological sketch by a deep learning algorithm, and to standardize the geological descriptions. A high-dimensional feature extraction module is configured to preprocess each tunnel face image and extract image high-dimensional features from the preprocessed tunnel face image by a deep learning algorithm. In the high-dimensional feature extraction module, the preprocessing of each tunnel face image includes image quality enhancement processing and brightness color repair processing, wherein: The image quality enhancement processing is completed by a Real-ESRGAN model. The brightness color repair processing is completed by a Retinex-Net algorithm.

[0071] A training sample construction module is configured to construct training samples from each image high-dimensional feature and the corresponding mileage and standardized geological description. A descriptive model determination module is configured to train a preset initial model using the training samples until the initial model meets a preset training end condition, and to determine the initial model that meets the preset training end condition as a Bert tunnel face image descriptive model.

[0072] Embodiment 3: An electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the tunneling face image descriptive model construction method of embodiment 1 when executing the computer program.

[0073] Embodiment 4: A non-transitory computer readable storage medium is provided, and the non-transitory computer readable storage medium stores computer instructions, and the computer instructions cause a computer to execute the tunneling face image descriptive model construction method of embodiment 1.

[0074] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer usable program code.

[0075] The present application is described with reference to flowcharts and / or block diagrams according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks. Figure 1 The functions specified in a flow or multiple flows and / or blocks.

[0076] These computer program instructions can also be stored in a computer readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks. Figure 1 The functions specified in a flow or multiple flows and / or blocks.

[0077] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a process for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1one or more processes and / or blocks Figure 1 the steps of the functions specified in one or more blocks.

[0078] Those skilled in the art can understand that all or part of the steps of the above-mentioned facts and methods can be completed by instructing the relevant hardware through programs, and the programs involved or the programs can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: at this time, the corresponding method steps are derived, and the storage medium can be ROM / RAM, magnetic disc, optical disc, etc.

[0079] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for constructing a tunnel face image descriptive model, characterized in that, The method comprises the following specific steps: obtaining tunnel face geological sketches at various mileages in the construction process and various tunnel face images at the corresponding mileages; extracting geological descriptions from the various tunnel face geological sketches through a deep learning algorithm and performing standardization processing on the geological descriptions; preprocessing the various tunnel face images and extracting image high-dimensional features from the preprocessed tunnel face images through a deep learning algorithm; constructing training samples from the various image high-dimensional features and the geological descriptions at the corresponding mileages and after standardization processing; training a preset initial model using the training samples until the initial model reaches a preset training end condition, and determining the initial model that reaches the preset training end condition as a Bert tunnel face image description model.

2. The method according to claim 1, wherein, The geological descriptions are obtained in the following manner: detecting the text description area of the tunnel face geological sketch through a trained yolov5 target detection model to obtain a detection area; extracting text from the detection area using a Convolutional Recurrent Neural Network algorithm to obtain the geological descriptions.

3. The method of claim 1, wherein, The preprocessing of the various tunnel face images comprises image quality enhancement processing and brightness and color restoration processing; the image quality enhancement processing is completed through a Real-ESRGAN model; and the brightness and color restoration processing is completed through a Retinex-Net algorithm.

4. The method of claim 1, wherein, The training samples comprise a bimodal data set at each mileage, which is formed by pairing the image high-dimensional features at each mileage and the geological descriptions at the corresponding mileage after standardization processing.

5. The method of claim 1, wherein, The image high-dimensional features are extracted through a Compact Convolutional Transformer model in combination with convolution and Transformer.

6. A system for constructing a descriptive model of a tunnel face image, characterized in that The method comprises: a construction data acquisition module for obtaining tunnel face geological sketches at various mileages in the construction process and various tunnel face images at the corresponding mileages; a geological sketch processing module for extracting geological descriptions from the various tunnel face geological sketches through a deep learning algorithm and performing standardization processing on the geological descriptions; a high-dimensional feature extraction module for preprocessing the various tunnel face images and extracting image high-dimensional features from the preprocessed tunnel face images through a deep learning algorithm; a training sample construction module for constructing training samples from the various image high-dimensional features and the geological descriptions at the corresponding mileages and after standardization processing; a description model determination module for training a preset initial model using the training samples until the initial model reaches a preset training end condition, and determining the initial model that reaches the preset training end condition as a Bert tunnel face image description model.

7. The tunnel face image descriptive model building system of claim 6, wherein, In the construction data acquisition module, the geological descriptions are obtained in the following manner: The trained yolov5 target detection model is used for detecting the text description area of the tunnel face geological sketch, and a detection area is obtained; A Convolutional Recurrent Neural Network algorithm is used for text extraction on the detection area, and the geological description is obtained.

8. The tunnel face image descriptive model construction system of claim 6, wherein, The high-dimensional feature extraction module pre-processes each tunnel face image, and the pre-processing includes image quality enhancement processing and brightness color repair processing. The image quality enhancement processing is completed by a Real-ESRGAN model. The brightness color repair processing is completed by a Retinex-Net algorithm.

9. An electronic device, comprising: The computer program is stored in the memory and can be run on the processor, and the processor executes the computer program to implement the tunnel face image description model construction method in any one of claims 1-5.

10. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the tunnel face image description model construction method in any one of claims 1-5.

Images