Tunnel face underground water intelligent identification method and system based on infrared thermal imaging

Abstract

The application discloses a tunnel face groundwater intelligent identification method and system based on infrared thermal imaging, which comprises the following steps: collecting tunnel face infrared thermal imaging video; acquiring tunnel face infrared image dataset by using frame extraction and preprocessing workflow; constructing tunnel face infrared image sample set after rapid labeling and data enhancement; adopting a transfer learning method to partially unfreeze three pre-training models based on a YOLOv5 target detection framework trained on three large image datasets to obtain three identification base models; integrating the three base models to obtain a final identification model; and using the model to identify the water outflow state of the infrared image in time sequence and output the identification result. The application can overcome the dark light condition in the tunnel, eliminate the influence of the subjective factors of personnel in the groundwater identification, improve the identification efficiency, reduce the workload of personnel, overcome the shortcomings of insufficient sampling, and improve the identification safety.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel face groundwater identification, and in particular to a tunnel face groundwater intelligent identification method and system based on infrared thermal imaging. Background Art

[0002] The primary effect of groundwater in tunnels is the dissolution of soluble cements in rock masses and structural surfaces, as well as the erosion of fine particles in fillings. This groundwater softens and loosens rock, argillaceifies fillings, reduces rock mass strength, and increases the dynamic and static water pressures in the tunnel's surrounding rock. This often leads to engineering disasters such as sudden water and mud inrush, face instability and collapse, surface subsidence, and seepage, significantly impacting the safety of personnel within the tunnel and the proper operation of equipment and machinery. Therefore, groundwater measurement is of great engineering significance.

[0003] During tunnel construction, traditional groundwater measurement methods primarily rely on visual observation of water seepage at the tunnel face and tunnel walls, categorizing groundwater discharge states as dry, moist or dripping, drenching or linear, and gushing. Due to the dim lighting inside tunnels, traditional groundwater identification results are subject to subjective influences from surveyors, resulting in low efficiency, high workload, and insufficient sampling, leading to insufficient supporting data for groundwater analysis at the tunnel face. Furthermore, visual observation requires close proximity to the tunnel face, which compromises the safety of surveyors. Therefore, a method for identifying groundwater in the surrounding rock of a tunnel face that improves measurement efficiency and safety is urgently needed. Summary of the Invention

[0004] In order to solve the problems existing in the prior art, the present invention provides a method and system for intelligent identification of groundwater in the tunnel face based on infrared thermal imaging, which eliminates the influence of subjective factors in the traditional tunnel face surrounding rock groundwater measurement process, avoids the shortcomings of insufficient sampling in traditional measurement, improves measurement efficiency and safety, and solves the problems mentioned in the above background technology.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solution: a method for intelligently identifying groundwater at a tunnel face based on infrared thermal imaging, comprising the following steps:

[0006] S1. Standardized acquisition of infrared thermal imaging video of the tunnel face;

[0007] S2. Based on the acquired infrared thermal imaging video of the tunnel face, a frame extraction and pre-processing workflow is used to obtain a tunnel face infrared image dataset.

[0008] S3. Rapidly annotate and enhance the acquired tunnel face infrared image dataset to construct a tunnel face infrared image sample set;

[0009] S4. Using the transfer learning method, three pre-trained models based on the YOLOv5 object detection framework trained on three large image datasets were partially unfrozen to obtain three intelligent groundwater identification models for surrounding rock of tunnel faces.

[0010] S5. Integrate the three base models to obtain an intelligent identification model for groundwater in the surrounding rock of the tunnel face;

[0011] S6. Use the tunnel face surrounding rock groundwater intelligent identification model to identify the water outflow status of the infrared image in time sequence. Based on the identification results, automatically extract the anchor frame area in the infrared image, segment the water outflow area, and calculate the area of ​​the water outflow area;

[0012] S7. Register the high-definition photos and infrared images and output the identification results of groundwater in the surrounding rock of the tunnel face.

[0013] Preferably, in step S2, it specifically includes:

[0014] S21. Use the VideoCapture() function of the OpenCV library in Python to read the infrared thermal imaging video of the tunnel face;

[0015] S22. Use the get function to read the frame rate of the infrared thermal imaging video;

[0016] S23, using the read() function of OpenCV to extract 5 frames of images per second from the tunnel face infrared video, saving them in ".jpg" format, and obtaining the original tunnel face infrared image dataset;

[0017] S24, using OpenCV's GaussianBlur() function to perform noise reduction and smoothing on the tunnel face infrared image;

[0018] S25. Then, the Laplacian function Laplacian() of OpenCV is used to sharpen the infrared image to make the infrared image clear, and finally the tunnel face infrared image dataset is obtained.

[0019] Preferably, in step S3, it specifically includes:

[0020] S31. First, use the Labelimg annotation tool to manually annotate a small number of tunnel face infrared images. Then, use the manually annotated infrared images to train the Faster R-CNN target detection model. The model uses the RPN Loss plus Fast R-CNN Loss loss function.

[0021] S32, then using the trained Faster R-CNN model to predict the unlabeled tunnel face infrared image, and storing the prediction result information as a YOLO annotation format result file;

[0022] S33. Use the Labelimg annotation tool to open the YOLO annotation format result file predicted by the model, correct and fine-tune the prediction results of the model, and complete the rapid annotation of the tunnel face infrared image dataset;

[0023] S34. Use the Random Grid Shuffle method of the Albumentations image enhancement library to simultaneously perform data enhancement on the rapidly annotated tunnel face infrared image and the annotation result file, and finally obtain a tunnel face infrared image sample set.

[0024] Preferably, in step S34, the number of photos in the tunnel face infrared image data set after data enhancement becomes 5 to 10 times that of the unenhanced infrared image data set.

[0025] Preferably, in step S4, the three large image datasets are respectively the "Pascal VOC", "ImageNet" and "COCO" large image datasets; the partial unfreezing specifically refers to: freezing the training parameters before the image feature output layer of the pre-trained model to make it untrainable; unfreezing the training parameters after the image feature output layers of the three pre-trained models to make them trainable.

[0026] Preferably, step S5 specifically includes: integrating the three base models obtained in step S4 using a five-fold cross validation and stacking method to obtain an intelligent groundwater identification model for the surrounding rock of the tunnel face.

[0027] Preferably, in step S6, segmenting the water outlet area specifically comprises: using an integrated SAM semantic segmentation model to interactively semi-automatically segment the water outlet area of ​​the local image, thereby obtaining a segmented image;

[0028] The calculation of the area of ​​the water region is specifically as follows: using the threshold() function of OpenCV to perform binarization processing on the segmented image, and then using the countNonZero() function to calculate the area of ​​the water region in the binarized image.

[0029] Preferably, in step S7, the identification result data of the groundwater in the surrounding rock of the tunnel face includes: a water discharge area time history curve for indicating the change of the water discharge area of ​​the tunnel face over time, a water discharge state of the tunnel face for measuring the amount of water discharge from the tunnel face, an average temperature and a temperature standard deviation for measuring whether there is an abnormal change in the water discharge from the tunnel face, and a tunnel face temperature range and a maximum and minimum temperature for warning of the occurrence of high-temperature hot water or sudden water inrush at the tunnel face;

[0030] The time history curve of the water outlet area is expressed as follows:

[0031] A = f(t);

[0032] Where A represents the area of ​​the water-out region of the segmented tunnel face, and t represents the time of water-out;

[0033] The water discharge state of the face is divided into gushing water discharge, drenching water discharge or linear water discharge, moist water discharge or dripping water discharge; the water seepage of the gushing water discharge is greater than 125L / (min.10m), the water seepage of the drenching water discharge or linear water discharge is 25-125L / (min.10m), and the water seepage of the moist or dripping water discharge is ≤25L / (min.10m);

[0034] The average temperature T m and temperature standard deviation T σ , the formula is as follows:

[0035]

[0036]

[0037] Where, T m represents the average temperature of the tunnel face, T j represents the temperature value of the jth pixel, P u represents the total number of pixels of the tunnel face infrared image, T σ represents the standard deviation of the infrared temperature of the tunnel face;

[0038] The tunnel face temperature range is expressed as follows:

[0039] T ave ∈[T max ,T min ];

[0040] The maximum and minimum temperatures of the tunnel face are expressed as follows:

[0041] T max =MAX(T j ,j=1,2,3,4…P u );

[0042] T min =MIN(T j ,j=1,2,3,4…P u );

[0043] Where, T ave represents the temperature range of the tunnel face infrared image, T max Indicates the maximum temperature of the tunnel face, T min Indicates the minimum temperature of the tunnel face. MAX() is the maximum value function, and MIN() is the minimum value function.

[0044] On the other hand, to achieve the above-mentioned purpose, the present invention further provides the following technical solution: an intelligent groundwater identification system for a tunnel face based on infrared thermal imaging, the system specifically comprising:

[0045] Infrared thermal imaging video acquisition module: used for standardized acquisition of tunnel face infrared thermal imaging video;

[0046] Tunnel face infrared image dataset acquisition module: This module is used to acquire the tunnel face infrared image dataset using the acquired tunnel face infrared thermal imaging video as the data basis, and uses the frame extraction and pre-processing workflow to obtain the tunnel face infrared image dataset;

[0047] A tunnel face infrared image sample set construction module is used to quickly annotate and enhance the acquired tunnel face infrared image dataset to construct a tunnel face infrared image sample set;

[0048] Base model acquisition module: This module uses transfer learning to partially unfreeze three pre-trained models based on the YOLOv5 object detection framework trained on three large image datasets, thereby obtaining three base models for intelligent identification of groundwater in surrounding rock of tunnel faces.

[0049] Groundwater intelligent identification model acquisition module: used to integrate the three base models to obtain the tunnel face surrounding rock groundwater intelligent identification model;

[0050] Water outflow status identification and identification result output module: It is used to use the intelligent identification model of groundwater in the surrounding rock of the tunnel face to identify the water outflow status of infrared images in a time sequence. Based on the identification results, it automatically extracts the anchor frame area in the infrared image, segments the water outflow area, and calculates the area of ​​the water outflow area; aligns high-definition photos and infrared images, and outputs the identification results of groundwater in the surrounding rock of the tunnel face.

[0051] The beneficial effects of the present invention are: the present invention can overcome the dark light situation in the tunnel through the intelligent identification method, eliminate the influence of subjective factors of personnel in groundwater identification, improve identification efficiency, reduce personnel workload, overcome the shortcomings of insufficient sampling, and improve identification safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a schematic diagram of the steps of the intelligent identification method for groundwater at the tunnel face of the present invention;

[0053] Figure 2 Schematic diagram of high-definition photos and infrared video screenshots collected by the standard in the embodiment; Figure 2 (a) is a high-definition photo collected by the standard. Figure 2 (b) is a screenshot of the infrared video collected by the standard;

[0054] Figure 3Schematic diagram of a photo image dataset obtained by extracting frames in the embodiment, and an infrared image after Gaussian blur and Laplace sharpening, Figure 3 (a) is the tunnel face infrared image dataset obtained by extracting frames. Figure 3 (b) is the infrared image of the tunnel face after Gaussian blur. Figure 3 (a) is the infrared image of the tunnel face after Laplace sharpening;

[0055] Figure 4 Schematic diagram of manually annotated photos, model prediction results, fine-tuning, and data enhancement in the embodiment. Figure 4 (a) is a schematic diagram of manual labeling of Labelimg, Figure 4 (b) is a schematic diagram of the prediction results of the Faster R-CNN model. Figure 4 (c) is a schematic diagram of fine-tuning the prediction results. Figure 4 (d) is a schematic diagram of data enhancement;

[0056] Figure 5 Schematic diagram of freezing and unfreezing YOLOv5 model training parameters for transfer learning in an embodiment;

[0057] Figure 6 Schematic diagram of the principle of five-fold cross validation and Stacking algorithm in the embodiment;

[0058] Figure 7 Schematic diagram of the identification photo, anchor frame area photo, IAST segmentation photo, and binarization photo in the embodiment; Figure 7 (a) is a schematic diagram of the model prediction result identification photo, Figure 7 (b) is a schematic diagram of automatically capturing the anchor frame area. Figure 7 (c) is a schematic diagram of ISAT automatic image segmentation. Figure 7 (d) is a schematic diagram of the image after binarization processing;

[0059] Figure 8 Schematic diagram of the time course curve of the water outlet area in the embodiment;

[0060] Figure 9 Schematic diagram of the module of the intelligent identification system for groundwater at the tunnel face in the embodiment;

[0061] In the figure, 110 is an infrared thermal imaging video acquisition module; 120 is a tunnel face infrared image data set acquisition module; 130 is a tunnel face infrared image sample set construction module; 140 is a base model acquisition module; 150 is a groundwater intelligent identification model acquisition module; 160 is a water discharge status identification and identification result output module. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0063] See also Figures 1-8 The present invention provides a technical solution: a method for intelligently identifying groundwater in tunnel faces based on infrared thermal imaging, which eliminates the influence of subjective factors in the traditional tunnel face surrounding rock groundwater measurement process, avoids the shortcomings of insufficient sampling in traditional measurement, and improves measurement efficiency and safety. Figure 1 As shown, the specific steps include:

[0064] S1. Standardized acquisition of infrared thermal imaging video of the tunnel face. After arranging the absolute coordinate target positions, use an infrared thermal imager with a high-definition camera and specific parameter configuration to shoot infrared thermal imaging video of the tunnel face.

[0065] Place three targets with absolute coordinates on the left and right sides of the lower tunnel face and in the center of the tunnel face. Set up a tripod 25 meters away from the tunnel face and place an infrared thermal imager below the tunnel axis. Turn on the infrared thermal imager, set the infrared color mode to iron red mode, adjust the infrared lens focus, enter the current ambient humidity, enter the emissivity value of the current tunnel face rock type, and set the video format to ".mp4". Figure 2 As shown. When the entire infrared lens includes the tunnel face, adjust to the electric autofocus mode and shoot a 20-second tunnel face infrared thermal imaging video. In order to map the results of infrared image target recognition to high-definition photos to guide on-site construction, it is also necessary to shoot high-definition photos, adjust the focus of the high-definition camera lens, and save the high-definition photos in ".jpg" format. When the entire high-definition camera lens includes the tunnel face, adjust to the autofocus mode and shoot a high-definition photo of the tunnel face, as shown. Figure 2 As shown in (a).

[0066] Through standardized tunnel face infrared thermal imaging acquisition, the influence of subjective factors in the traditional tunnel face surrounding rock groundwater measurement process is eliminated. Infrared data also eliminates the influence of dark tunnel light, improving acquisition efficiency and safety, thereby obtaining high-quality infrared thermal imaging videos, such as Figure 2 (b) shows a schematic diagram of a standard captured infrared video screenshot.

[0067] S2. Based on the infrared thermal imaging video of the tunnel face obtained in S1, a frame extraction and pre-processing workflow is used to obtain a tunnel face infrared image dataset.

[0068] The VideoCapture() function of the OpenCV library in Python reads the infrared thermal imaging video of the tunnel face obtained in step S1 and saved locally. The get(cv2.CAP_PROP_FPS) function reads the frame rate of the infrared thermal imaging video. The read() function of OpenCV extracts 5 frames per second of the 50 Hz tunnel face infrared video and saves the tunnel face infrared images in .jpg format to a local folder. This video extraction process generates a raw tunnel face infrared image dataset. The GaussianBlur() function of OpenCV is used to reduce noise and smooth the tunnel face infrared images, with the Gaussian kernel size set to 5. The Laplacian() function of OpenCV is then used to sharpen the infrared image, enhancing the high-frequency portion of the image and making it clearer. The Laplacian operator size is set to 3, the scaling factor is set to 1, the offset value is set to 1, and the boundary type is set to cv2.BORDER_DEFAULT. This results in a pre-processed tunnel face infrared image dataset.

[0069] like Figure 3 middle Figure 3 As shown in (a), the frame extraction technology greatly expands the number of data sets, avoids the labor of manual photography or screenshots, and uses the Gaussian blur algorithm to reduce the infrared image noise caused by dust, vibration, and humidity in the tunnel environment. Figure 3 (b) As shown. The Laplace algorithm can restore the clarity of the image and sharpen the image after Gaussian blur, as shown in Figure 3 (c) As shown in FIG. 1 , step S2 relies on image processing technology to expand the amount of infrared image data with high quality, avoids the disadvantage of insufficient sampling in traditional measurement, and improves measurement efficiency.

[0070] S3. Use the Labelimg annotation tool and the semi-automatic annotation method based on the Faster R-CNN target detection framework to quickly annotate the tunnel face infrared image dataset obtained in step S2. Then use the RandomGridShuffle (random grid shuffle) method of the Albumentations image enhancement library to perform data enhancement, thereby constructing a tunnel face infrared image sample set for YOLOv5 target detection model training, as shown in the following figure: Figure 4 shown.

[0071] First, a small number of tunnel face infrared images were manually labeled using the Labelimg labeling tool, such as Figure 4 (a); the manually annotated infrared images are then used to train the Faster R-CNN target detection model. The model uses the loss function of RPNLoss plus Fast R-CNN Loss, as shown in Equations (1) to (7):

[0072]

[0073] Fast R-CNN Loss=L(p,u,t u ,v)=L cls (p,u)+λ[u≥1]L loc (t u ,v ) (2)

[0074]

[0075]

[0076] L cls (p,u)=-logp u (5)

[0077]

[0078]

[0079] In formulas (1) to (7), p i Indicates the probability that the i-th anchor box is predicted to be the true label; when the prediction is a positive sample, p i * 1 otherwise 0; t i Represents the boundary regression parameter for predicting the i-th anchor box; t i * N represents the regression parameter of the bounding box corresponding to the label of the i-th anchor box; cls Indicates the number of all samples; N reg Indicates the number of anchor box positions; p is the probability distribution predicted by the classifier; u is the true label category of the corresponding target; t u is the regression parameter of the corresponding category u predicted by the bounding box regressor; v corresponds to the boundary regression parameter of the true target; λ is 10.

[0080] Then, the trained Faster R-CNN model is used to predict the unlabeled tunnel face infrared image, such as Figure 4 As shown in (b), the prediction result information is stored as a result file in the YOLO annotation format; finally, the model prediction result file is opened using the Labelimg annotation tool to correct and fine-tune the model prediction results, as shown in Figure 4 As shown in (c), the annotation of the tunnel face infrared image dataset is quickly completed. The Random Grid Shuffle (random grid disordered arrangement) method of the Albumentations image enhancement library is used to enhance the annotated original tunnel face infrared image and the annotation result file at the same time, as shown in Figure 4As shown in (d), the number of photos in the enhanced tunnel face infrared image dataset can be increased to 5 to 10 times that of the original tunnel face infrared images, thereby constructing the final tunnel face infrared image sample set for YOLOv5 model training.

[0081] Using a small number of manually annotated samples, we implement semi-automatic annotation and fine-tuning for sample-assisted annotation, improving both efficiency and accuracy. The Random Grid Shuffle method in the Albumentations image enhancement library can expand the number and diversity of tunnel face infrared image datasets, helping to improve the model's generalization, accuracy, and reliability.

[0082] S4. Use the transfer learning method to partially unfreeze three pre-trained models based on the YOLOv5 object detection framework that were trained on the large-scale image datasets "Pascal VOC", "ImageNet" and "COCO", thereby obtaining three basic models for intelligent identification of groundwater in the surrounding rock of the tunnel face.

[0083] Download pre-trained models for the YOLOv5 framework based on three large image datasets: Pascal_VOC, ImageNet, and COCO. Load the pre-trained models into the Python environment using the PyTorch framework. Apply transfer learning to obtain the training weights, model structure, and image feature output layer of each pre-trained model. The image feature output layer represents the unique features of an image. Freeze the training parameters before the image feature output layer of the pre-trained model, making it "untrainable." Unfreeze the training parameters after the image feature output layer of the three pre-trained models, making them "trainable." Figure 5 As shown in FIG, three intelligent identification base models of surrounding rock water discharge status of tunnel faces are obtained. After step S4, the intelligent identification base model of surrounding rock water discharge status of tunnel faces has the universal ability to extract photo features.

[0084] The use of transfer learning methods saves computing resources and improves computing efficiency. At the same time, the intelligent identification model of groundwater in the surrounding rock of the tunnel face has the universal ability to extract photo features, which can improve the generalization and accuracy of the model. The introduction of transfer learning reduces the workload of manual labeling and the workload of on-site photography, and can achieve a model with better training effects on a smaller data set.

[0085] S5. Use the five-fold cross-validation and stacking method to integrate the three base models obtained in step S4 to obtain the final tunnel face surrounding rock groundwater intelligent identification model based on the YOLOv5 target detection framework.

[0086] The infrared image sample set of the tunnel face obtained in step S3 is divided into a training set and a test set. The training set is divided into five parts, and the five-fold cross-validation method is used to train each individual base model, that is, for each base model, four training sets are used to train the model, and one training set is used for verification, and the verification obtains A. At the same time, the base model trained with the four training sets is used for prediction of the test set to obtain the prediction result B. The five-fold cross-validation of a base model produces five A's and five B's. By stacking the five A's and the weighted averages of the five B's into columns, the above steps are repeated for the three base models using the Stacking method. The principles of the five-fold cross-validation and Stacking algorithm are as follows: Figure 6 As shown in the figure, a new dataset is obtained in the form of an (m, p)-dimensional matrix, where row A is the training set and row B is the test set, m is the number of samples, and p is 3. Linear regression is performed on the new dataset (m, p), resulting in an intelligent groundwater identification model for the surrounding rock of the tunnel face.

[0087] The trained intelligent groundwater identification model for tunnel faces is an integrated intelligent identification model, reducing the workload of on-site surveyors and improving the accuracy of groundwater identification in the surrounding rock of the tunnel face. Five-fold cross-validation reduces model overfitting and improves model generalization. The stacking algorithm leverages the strengths of each base model to improve the accuracy of the model for identifying water discharge in the surrounding rock of the tunnel face.

[0088] S6. Using the model obtained in step S5, the infrared images extracted from the tunnel face infrared video are analyzed for water-out status in a time-series manner. Based on the analysis results, anchor regions are automatically extracted from the infrared images. The integrated SAM semantic segmentation model is used to automatically segment the water-out region within the anchor region. Finally, the area of ​​the segmented water-out region is calculated.

[0089] The water outflow state identification model of the tunnel face surrounding rock trained in step S5 is used to identify the water outflow state of the infrared images obtained by extracting frames from the infrared video of the tunnel face in time sequence, such as Figure 7 middle Figure 7 As shown in (a), based on the anchor frame coordinate information output by the model recognition result, the original image is directly cut according to the position and size of the anchor frame to obtain the local image of the water outlet area, as shown in Figure 7 (b) As shown. The integrated SAM semantic segmentation model is used to interactively and semi-automatically segment the water area of ​​the local image to obtain the segmented image, as shown in Figure 7 (c) Then use OpenCV's threshold() function to binarize the segmented image, as shown in Figure 7As shown in (d), the countNonZero() function is then used to calculate the area of ​​the water-discharged region in the binary processed image. At the same time, the area of ​​the water-discharged region is statistically analyzed in time series, and finally the trend of the water-discharged state of the tunnel face over time is analyzed.

[0090] Based on the prediction results of the tunnel face surrounding rock groundwater intelligent identification model, infrared images of the tunnel face can automatically extract the tunnel face water-discharge areas and classify the tunnel face water-discharge status. The integrated SAM semantic segmentation model is used to quickly segment the water-discharge areas in the local image, ensuring segmentation accuracy while reducing the time required to construct the semantic segmentation dataset and train the semantic segmentation model. The countNonZero() function in the OpenCV library can be used to quickly and accurately calculate the area of ​​the segmented water-discharge areas.

[0091] S7. Register the high-definition photos and infrared images and output the identification results of groundwater in the surrounding rock of the tunnel face.

[0092] The absolute coordinates of the three targets are used to register the high-definition photo. The scale-invariant feature transform (SIFT) algorithm is then used to extract and match the feature points of the infrared thermal imaging photo and the high-definition photo. The random sampling consensus algorithm (RANSAC) is used to eliminate mismatches and obtain the optimal transformation matrix between the two images, achieving accurate registration. Groundwater identification results are output on the high-definition photo to guide on-site construction. The output data types include: Time-history curves of the water-discharge area, as shown in Equation (8):

[0093] A=f(t)(8)

[0094] In formula (8), A represents the area of ​​the water-outlet area of ​​the segmented tunnel face, t represents the time of water-outlet, and formula (8) represents the function of the area of ​​the water-outlet area changing with the time of water-outlet. The time history curve of the water-outlet area is the change of the water-outlet area of ​​the tunnel face with time, as shown in the figure below: Figure 8 shown.

[0095] The water discharge state at the tunnel face is divided into "Surge Water", "Rain-like effluent", "Linear effluent", "Wet effluent", and "Driplike effluent". The relationship between each water discharge state and the amount of water seepage is shown in Table 1. The water discharge state is used to measure the amount of water discharge at the tunnel face.

[0096]

[0097] The average temperature and temperature standard deviation are as follows:

[0098] The average temperature Tm and temperature standard deviation T σ , the formula is as follows:

[0099]

[0100]

[0101] Where, T m represents the average temperature of the tunnel face, T j represents the temperature value of the jth pixel, P u represents the total number of pixels of the tunnel face infrared image, T σ Indicates the standard deviation of the infrared temperature at the tunnel face; the average temperature and temperature standard deviation are used to measure whether there are abnormal changes in the water discharge at the tunnel face.

[0102] The maximum and minimum temperatures of the tunnel face are expressed as follows, respectively, using equations (11) and (12):

[0103] T max =MAX(T j ,j=1,2,3,4…P u ) (11);

[0104] T min =MIN(T j ,j=1,2,3,4…P u ) (12);

[0105] The tunnel face temperature range is expressed as follows:

[0106] T ave ∈[T max ,T min ] (13);

[0107] Where, T ave represents the temperature range of the tunnel face infrared image, T max Indicates the maximum temperature of the tunnel face, T min Indicates the minimum temperature of the tunnel face. MAX() is a maximum value function, and MIN() is a minimum value function. The temperature range and the maximum and minimum temperatures are used to warn of the occurrence of high-temperature hot water or sudden water inrush at the tunnel face.

[0108] Displaying groundwater identification results in high-definition photographs is beneficial for guiding on-site construction. By analyzing the data output by the intelligent groundwater identification model for the tunnel face surrounding rock, extensive and in-depth analysis of the impact of groundwater on tunnel construction can be achieved. This overcomes the subjective influence, insufficient sampling, and low efficiency of manual groundwater identification.

[0109] Based on the same inventive concept as the above method embodiment, the embodiment of the present application also provides a tunnel face groundwater intelligent identification system based on infrared thermal imaging, which can realize the functions provided by the above method embodiment, such as Figure 9 As shown, the system includes:

[0110] Infrared thermal imaging video acquisition module 110: used for collecting infrared thermal imaging videos of tunnel faces in a standardized manner;

[0111] The tunnel face infrared image data set acquisition module 120 is used to acquire a tunnel face infrared image data set using a frame extraction and pre-processing workflow based on the acquired tunnel face infrared thermal imaging video.

[0112] A tunnel face infrared image sample set construction module 130 is used to quickly annotate and enhance the acquired tunnel face infrared image dataset to construct a tunnel face infrared image sample set;

[0113] Base model acquisition module 140: uses a transfer learning method to partially unfreeze three pre-trained models based on the YOLOv5 object detection framework trained on three large image datasets to obtain three tunnel face surrounding rock and groundwater intelligent identification base models;

[0114] Groundwater intelligent identification model acquisition module 150: used to integrate the three base models to obtain the tunnel face surrounding rock groundwater intelligent identification model;

[0115] Water outflow status identification and identification result output module 160: is used to use the intelligent identification model of groundwater in the surrounding rock of the tunnel face to identify the water outflow status of the infrared image in a time sequence, automatically extract the anchor frame area in the infrared image, segment the water outflow area, and calculate the area of ​​the water outflow area based on the identification result; align the high-definition photos and infrared images, and output the identification result of the groundwater in the surrounding rock of the tunnel face.

[0116] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for intelligent identification of groundwater in a tunnel face based on infrared thermal imaging, characterized in that: The steps include: S1. Standardized acquisition of infrared thermal imaging video of the tunnel face; S2. Based on the acquired infrared thermal imaging video of the tunnel face, a frame extraction and pre-processing workflow is used to obtain a tunnel face infrared image dataset. S3. Rapidly annotate and enhance the acquired tunnel face infrared image dataset to construct a tunnel face infrared image sample set; S4. Using the transfer learning method, three pre-trained models based on the YOLOv5 object detection framework trained on three large image datasets were partially unfrozen to obtain three intelligent groundwater identification models for surrounding rock of tunnel faces. S5. Integrate the three base models to obtain an intelligent identification model for groundwater in the surrounding rock of the tunnel face; S6. Use the tunnel face surrounding rock groundwater intelligent identification model to identify the water outflow status of the infrared image in time sequence. Based on the identification results, automatically extract the anchor frame area in the infrared image, segment the water outflow area, and calculate the area of ​​the water outflow area; S7. Register the high-definition photos and infrared images and output the identification results of groundwater in the surrounding rock of the tunnel face.

2. The method for intelligently identifying groundwater at a tunnel face based on infrared thermal imaging according to claim 1 is characterized in that: In step S2, it specifically includes: S21. Use the VideoCapture() function of the OpenCV library in Python to read the infrared thermal imaging video of the tunnel face; S22. Use the get function to read the frame rate of the infrared thermal imaging video; S23, using the read() function of OpenCV to extract 5 frames of images per second from the tunnel face infrared video, saving them in ".jpg" format, and obtaining the original tunnel face infrared image dataset; S24, using OpenCV's GaussianBlur() function to perform noise reduction and smoothing on the tunnel face infrared image; S25. Then, the Laplacian function Laplacian() of OpenCV is used to sharpen the infrared image to make the infrared image clear, and finally the tunnel face infrared image dataset is obtained.

3. The method for intelligently identifying groundwater at a tunnel face based on infrared thermal imaging according to claim 1 is characterized in that: In step S3, it specifically includes: S31. First, use the Labelimg annotation tool to manually annotate a small number of tunnel face infrared images. Then, use the manually annotated infrared images to train the Faster R-CNN target detection model. The model uses the RPN Loss plus Fast R-CNN Loss loss function. S32, then using the trained Faster R-CNN model to predict the unlabeled tunnel face infrared image, and storing the prediction result information as a YOLO annotation format result file; S33. Use the Labelimg annotation tool to open the YOLO annotation format result file predicted by the model, correct and fine-tune the prediction results of the model, and complete the rapid annotation of the tunnel face infrared image dataset; S34. Use the Random Grid Shuffle method of the Albumentations image enhancement library to simultaneously perform data enhancement on the rapidly annotated tunnel face infrared image and the annotation result file, and finally obtain a tunnel face infrared image sample set.

4. The method for intelligently identifying groundwater at a tunnel face based on infrared thermal imaging according to claim 3 is characterized by: In step S34, the number of photos in the tunnel face infrared image data set after data enhancement becomes 5 to 10 times that of the unenhanced infrared image data set.

5. The method for intelligently identifying groundwater at a tunnel face based on infrared thermal imaging according to claim 1 is characterized in that: In step S4, the three large image datasets are "PascalVOC", "ImageNet" and "COCO" large image datasets; the partial unfreezing specifically refers to: freezing the training parameters before the image feature output layer of the pre-trained model to make it untrainable; unfreezing the training parameters after the image feature output layers of the three pre-trained models to make them trainable.

6. The method for intelligently identifying groundwater at a tunnel face based on infrared thermal imaging according to claim 1 is characterized in that: In step S5, it specifically includes: using five-fold cross validation and stacking method to integrate the three base models obtained in step S4 to obtain an intelligent identification model of groundwater in the surrounding rock of the tunnel face.

7. The method for intelligently identifying groundwater at a tunnel face based on infrared thermal imaging according to claim 1 is characterized in that: In step S6, the segmentation of the water outlet area specifically comprises: using an integrated SAM semantic segmentation model to interactively semi-automatically segment the water outlet area of ​​the local image, thereby obtaining a segmented image; The calculation of the area of ​​the water region is specifically as follows: using the threshold() function of OpenCV to perform binarization processing on the segmented image, and then using the countNonZero() function to calculate the area of ​​the water region in the binarized image.

8. The method for intelligently identifying groundwater at a tunnel face based on infrared thermal imaging according to claim 1 is characterized in that: In step S7, the identification result data of the groundwater in the surrounding rock of the tunnel face includes: a water discharge area time history curve for indicating the change of the water discharge area of ​​the tunnel face over time, a water discharge status of the tunnel face for measuring the amount of water discharge from the tunnel face, an average temperature and a temperature standard deviation for measuring whether there is an abnormal change in the water discharge from the tunnel face, and a tunnel face temperature range and a maximum and minimum temperature for warning of the occurrence of high-temperature hot water or sudden water inrush at the tunnel face; The time history curve of the water outlet area is expressed as follows: A = f(t); Where A represents the area of ​​the water-out region of the segmented tunnel face, and t represents the time of water-out; The water discharge state of the face is divided into gushing water discharge, drenching water discharge or linear water discharge, moist water discharge or dripping water discharge; the water seepage of the gushing water discharge is greater than 125L / (min.10m), the water seepage of the drenching water discharge or linear water discharge is 25-125L / (min.10m), and the water seepage of the moist or dripping water discharge is ≤25L / (min.10m); The average temperature T m and temperature standard deviation T σ , the formula is as follows: Where, T m represents the average temperature of the tunnel face, T j represents the temperature value of the jth pixel, P u represents the total number of pixels of the tunnel face infrared image, T σ represents the standard deviation of the infrared temperature of the tunnel face; The tunnel face temperature range is expressed as follows: T ave ∈[T max ,T min ]; The maximum and minimum temperatures of the tunnel face are expressed as follows: T max =MAX(T j ,j=1,2,3,4…P u ); T min =MIN(T j ,j=1,2,3,4…P u ); Where, T ave represents the temperature range of the tunnel face infrared image, T max Indicates the maximum temperature of the tunnel face, T min Indicates the minimum temperature of the tunnel face. MAX() is the maximum value function, and MIN() is the minimum value function.

9. An intelligent groundwater identification system for tunnel faces based on infrared thermal imaging, characterized by: The system specifically includes: Infrared thermal imaging video acquisition module (110): used for standardizing the acquisition of infrared thermal imaging videos of tunnel faces; A tunnel face infrared image data set acquisition module (120) is used to acquire a tunnel face infrared image data set using a frame extraction and pre-processing workflow based on the acquired tunnel face infrared thermal imaging video as a data basis; A tunnel face infrared image sample set construction module (130) is used to quickly annotate and enhance the acquired tunnel face infrared image data set to construct a tunnel face infrared image sample set; A base model acquisition module (140) is configured to partially unfreeze three pre-trained models based on the YOLOv5 target detection framework trained on three large image datasets using a transfer learning method, thereby obtaining three tunnel face surrounding rock and groundwater intelligent identification base models; Groundwater intelligent identification model acquisition module (150): used for integrating the three base models to obtain the tunnel face surrounding rock groundwater intelligent identification model; Water outflow status identification and identification result output module (160): used to use the tunnel face surrounding rock groundwater intelligent identification model to identify the water outflow status of infrared images in time sequence, automatically extract the anchor frame area in the infrared image, segment the water outflow area, and calculate the area of ​​the water outflow area based on the identification result; align the high-definition photo and the infrared image, and output the identification result of the tunnel face surrounding rock groundwater.

Images