Ground penetrating radar tunnel lining disease identification method and device

Abstract

The invention discloses a ground penetrating radar tunnel lining disease identification method and device, and the method comprises the steps: carrying out the amplitude-frequency feature analysis of tunnel lining data collected by a ground penetrating radar, and obtaining the corresponding amplitude-frequency feature data; generating a radar reflection signal profile map according to the tunnel lining data, and performing disease marking on the radar reflection signal profile map to obtain a profile map after disease marking; inputting the amplitude-frequency characteristic data into a characteristic deconstruction analysis model, and outputting a first disease identification result; inputting the section map after the disease marking into a section map disease identification model, and outputting a second disease identification result; and carrying out weighted average on the first disease identification result and the second disease identification result to generate a final tunnel lining disease identification result. According to the invention, the identification precision and identification efficiency of tunnel lining diseases can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ground penetrating radar data analysis and tunnel internal defect detection, and in particular to a ground penetrating radar tunnel lining defect identification method and device. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention recited in the claims. No admission is made that the description herein is prior art by inclusion in this section.

[0003] With the rapid development of railway construction, tunnel engineering, as an important part of it, is crucial to ensure the safety and stability of railway operation. Tunnels are important channels for trains to cross mountains, rivers, cities and other terrains, and they bear key transportation functions. Lining, as the main component of the tunnel structure, has a direct impact on the smooth operation of railway transportation due to its integrity and safety.

[0004] However, damage and abnormal conditions of tunnel lining are relatively hidden. In the past few decades, more and more tunnel lining problems have appeared in railway projects, including cavities, voids, steel corrosion, concrete cracking, deviation deformation, corrosion and loosening. Since tunnel lining is located underground, it is difficult to detect. It not only requires a lot of manpower, material resources and time, but also easily leads to missed detection and false detection. Traditional manual detection methods can no longer meet the needs of accurate and effective detection. Summary of the invention

[0005] The embodiment of the present invention provides a method for identifying tunnel lining defects using ground penetrating radar, which is used to improve the identification accuracy and efficiency of tunnel lining defects. The method includes:

[0006] Perform amplitude-frequency characteristic analysis on the tunnel lining data collected by the ground penetrating radar to obtain corresponding amplitude-frequency characteristic data; generate a radar reflection signal profile according to the tunnel lining data, mark defects on the radar reflection signal profile, and obtain a profile with marked defects;

[0007] Inputting the amplitude-frequency feature data into a feature deconstruction analysis model, and outputting a first disease recognition result; the feature deconstruction analysis model is obtained by training a convolutional neural network based on the historical amplitude-frequency feature data and the corresponding historical first disease recognition results;

[0008] Inputting the disease-annotated cross-section image into a cross-section image disease recognition model, and outputting a second disease recognition result; the cross-section image disease recognition model is obtained by training a deep learning network based on the historical disease-annotated cross-section image and the corresponding historical second disease recognition results;

[0009] The first defect identification result and the second defect identification result are weighted averaged to generate the final tunnel lining defect identification result.

[0010] The embodiment of the present invention further provides a ground penetrating radar tunnel lining defect identification device, which is used to improve the identification accuracy and efficiency of tunnel lining defects. The device includes:

[0011] The amplitude-frequency characteristic data and profile acquisition module is used to perform amplitude-frequency characteristic analysis on the tunnel lining data collected by the ground penetrating radar to obtain the corresponding amplitude-frequency characteristic data; generate a radar reflection signal profile according to the tunnel lining data, mark defects on the radar reflection signal profile, and obtain a profile with marked defects;

[0012] A first disease identification result output module is used to input the amplitude-frequency feature data into a feature deconstruction analysis model and output a first disease identification result; the feature deconstruction analysis model is obtained by training a convolutional neural network based on the historical amplitude-frequency feature data and the corresponding historical first disease identification results;

[0013] A second disease identification result output module, used to input the disease-annotated cross-section image into a cross-section image disease identification model, and output a second disease identification result; the cross-section image disease identification model is obtained by training a deep learning network based on the historical disease-annotated cross-section image and the corresponding historical second disease identification results;

[0014] The final tunnel lining disease identification result generation module is used to perform weighted averaging on the first disease identification result and the second disease identification result to generate the final tunnel lining disease identification result.

[0015] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned ground penetrating radar tunnel lining defect identification method when executing the computer program.

[0016] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned ground penetrating radar tunnel lining defect identification method is implemented.

[0017] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned ground penetrating radar tunnel lining disease identification method is implemented.

[0018] In an embodiment of the present invention, the corresponding amplitude-frequency characteristic data is obtained by performing amplitude-frequency characteristic analysis on the tunnel lining data collected by the ground penetrating radar; a radar reflection signal profile is generated according to the tunnel lining data, and defects are marked on the radar reflection signal profile to obtain a profile with marked defects; the amplitude-frequency characteristic data is input into a feature deconstruction analysis model to output a first defect recognition result; the feature deconstruction analysis model is obtained by training a convolutional neural network based on historical amplitude-frequency characteristic data and corresponding historical first defect recognition results; the profile with marked defects is input into a profile defect recognition model to output a second defect recognition result; the profile defect recognition model is obtained by training a deep learning network based on historical defect-marked profiles and corresponding historical second defect recognition results; the first defect recognition result and the second defect recognition result are weighted averaged to generate a final tunnel lining defect recognition result. In the above process, the embodiment of the present invention analyzes and combines the two types of data (amplitude-frequency characteristic data and profile diagram), performs weighted averaging processing on the defect identification results of the feature deconstruction analysis model and the defect identification results in the profile diagram defect identification model, and realizes the identification of tunnel lining defects, thereby improving the identification accuracy and efficiency of tunnel lining defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0020] Figure 1 Flow chart of the method for identifying tunnel lining defects using ground penetrating radar in an embodiment of the present invention;

[0021] Figure 2 It is a structural diagram of a spatial attention network block in an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of the YOLOv5 network structure in an embodiment of the present invention;

[0023] Figure 4 is the detection result of the tunnel lining disease identification model in the embodiment of the present invention;

[0024] Figure 5 Schematic diagram of a ground penetrating radar tunnel lining defect identification device in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0026] Figure 1 This is a flow chart of a method for identifying tunnel lining defects using ground penetrating radar in an embodiment of the present invention. The method includes:

[0027] Step 101, performing amplitude-frequency characteristic analysis on the tunnel lining data acquired by the ground penetrating radar to obtain corresponding amplitude-frequency characteristic data; generating a radar reflection signal profile according to the tunnel lining data, marking defects on the radar reflection signal profile to obtain a profile with marked defects;

[0028] Step 102, inputting the amplitude-frequency feature data into a feature deconstruction analysis model, and outputting a first disease recognition result; the feature deconstruction analysis model is obtained by training a convolutional neural network based on the historical amplitude-frequency feature data and the corresponding historical first disease recognition results;

[0029] Step 103, inputting the disease-annotated cross-section image into a cross-section image disease recognition model, and outputting a second disease recognition result; the cross-section image disease recognition model is obtained by training a deep learning network based on the historical disease-annotated cross-section image and the corresponding historical second disease recognition results;

[0030] Step 104: Perform a weighted average of the first defect identification result and the second defect identification result to generate a final tunnel lining defect identification result.

[0031] Each step is described in detail below.

[0032] In step 101, the tunnel lining data collected by the ground penetrating radar is analyzed for amplitude-frequency characteristics to obtain corresponding amplitude-frequency characteristic data; a radar reflection signal profile is generated according to the tunnel lining data, and defects are marked on the radar reflection signal profile to obtain a profile with marked defects.

[0033] In one embodiment, before performing amplitude-frequency characteristic analysis on tunnel lining data acquired by ground penetrating radar to obtain corresponding amplitude-frequency characteristic data, the following steps are included:

[0034] Performing data preprocessing on the tunnel lining data to obtain preprocessed tunnel lining data;

[0035] The amplitude-frequency characteristics of the tunnel lining data collected by the ground penetrating radar are analyzed, including:

[0036] Conduct amplitude-frequency characteristic analysis on the pre-processed tunnel lining data;

[0037] Generate radar reflection signal profiles based on tunnel lining data, including:

[0038] Generate radar reflection signal profile based on preprocessed tunnel lining data.

[0039] In a specific embodiment, data preprocessing includes data processing steps such as zero bias removal, zero point correction, inter-channel equalization, bandpass filtering, gain processing, etc. The processed tunnel lining data is set with appropriate file conversion format and image parameters to generate corresponding profiles.

[0040] In step 102, the amplitude-frequency feature data is input into a feature deconstruction analysis model, and a first disease recognition result is output; the feature deconstruction analysis model is obtained by training a convolutional neural network based on historical amplitude-frequency feature data and corresponding historical first disease recognition results.

[0041] In one embodiment, the convolutional neural network is trained according to the historical amplitude-frequency feature data and the corresponding historical first disease recognition results, including:

[0042] The historical amplitude-frequency feature data and the corresponding historical first disease recognition results are input as training sets into a convolutional neural network based on the attention mechanism for training, and a trained feature deconstruction analysis model is obtained.

[0043] In a specific embodiment, an attention mechanism called Greedy Channel-Spatial Attention (GCS attention) is introduced to improve the accuracy of the model. Figure 2 is a structural diagram of a spatial attention network block in an embodiment of the present invention, such as Figure 2 As shown in the figure, it specifically includes: introducing a channel attention network block, first performing global maximum pooling and global average pooling on the amplitude-frequency feature data X. Then use two fully connected networks to compress and excite the channel feature map respectively, and then use the SoftMax function to score the feature map respectively. Use 2×C learnable parameters to multiply the attention score respectively, and then multiply it by the amplitude-frequency feature data X, and then add the two channels together. Here, an addition operation is performed to improve the fitting speed of the model. Then use the spatial attention network block, which weights the different positions of the input data so that the model can pay more attention to important spatial position information, and apply it to different levels in the convolutional neural network to improve the expression ability and performance of the model. It also adds the output and input to finally obtain the feature map of the output amplitude-frequency feature data X.

[0044] Among them, the amplitude-frequency feature data X is the global maximum pooling value F max (x i，j，c) and the global average pooling value F avg (x i，j，c ) is calculated as:

[0045]

[0046] F avg (x i，j，c )=max(x i，j，c )

[0047] Among them, H represents the height of the feature map, W represents the width of the feature map, C represents the number of channels of the feature map, and x i,j,c Represents the amplitude-frequency characteristic data at the i-th row and j-th column on the C-th channel.

[0048] In step 103, the cross-sectional view with disease annotation is input into a cross-sectional view disease recognition model, and a second disease recognition result is output; the cross-sectional view disease recognition model is obtained by training a deep learning network based on the cross-sectional view with historical disease annotation and the corresponding historical second disease recognition results.

[0049] In one embodiment, the deep learning network is trained according to the cross-sectional image with historical disease annotation and the corresponding historical second disease recognition result, including:

[0050] The cross-section images with historical disease annotations and the corresponding historical second disease recognition results are used as training sets and input into the YOLO-based v 5 algorithm is trained in the deep learning network to obtain a trained profile disease recognition model.

[0051] In a specific embodiment, YOLO is used v 5 algorithms, Figure 3 YOLO in the embodiment of the present invention v 5 Network structure diagram, such as Figure 3 As shown in the figure, a deep residual network is designed to better extract the characteristics of railway tunnel lining defects and diseases according to the characteristics of amplitude-frequency feature data, and a feature pyramid at four scales is constructed using the feature map of the deep residual main network. YOLO is used at a 512×512 input resolution. v The minimum feature map of 5 is reduced from 16×16 to 8×8, which makes the network have a larger receptive field, and features are fused with the deep residual network at four scales to form a real-time detection model for railway tunnel lining defects.

[0052] Firstly, the B-scan profile of railway tunnel lining defects obtained by ground penetrating radar is annotated with the defect location and defect category, and a labeling file is generated. The data set is divided into a training set and a test set. In order to avoid overfitting and improve model performance, a hybrid algorithm, random flipping and random cropping are used to expand the training set data.

[0053] Subsequently, the YOLOv5 network is used to predict the defect location coordinates, category and confidence. The specific method is as follows: a deep residual network is designed to extract the defect features of radar images and divide it into four branches. The features are fused with the deep residual network at four scales to form a multi-scale prediction network and output the predicted bounding box. The non-maximum suppression method is used to remove duplicate bounding boxes, and the railway tunnel lining defect disease detection model is obtained through iterative calculation of the gradient descent method.

[0054] In step 104, the first defect identification result and the second defect identification result are weighted averaged to generate a final tunnel lining defect identification result.

[0055] In one embodiment, it further includes:

[0056] The characteristic deconstruction analysis model and the profile defect recognition model are evaluated respectively through the model evaluation indexes to obtain the corresponding evaluation results;

[0057] According to the corresponding evaluation results, the weights of the feature deconstruction analysis model and the profile defect recognition model are adjusted respectively; the model evaluation indicators include accuracy, recall rate and number of frames transmitted per second.

[0058] In the specific embodiment, indicators such as mean accuracy, recall rate and frames per second (FPS) are used to evaluate the model, select the best model to detect tunnel lining defects and diseases, and evaluate the detection effect of the tunnel lining disease recognition model. The tunnel lining disease recognition model divides the disease recognition task into two branches. By analyzing the amplitude-frequency characteristics of the tunnel lining data and analyzing and combining the disease recognition of the profile, the linear combination coefficient set of each recognition task, that is, the prediction coefficient vector, is predicted, and a unique bounding box is obtained by selecting it through the non-maximum suppression algorithm. The results of the two branches are linearly combined to obtain the prediction result. The original B-Scan file of the ground penetrating radar (including tunnel lining data) and the profile image are input, and an image with the category, location and contour of the tunnel lining disease is output. Figure 4 is the detection result of the tunnel lining disease identification model in the embodiment of the present invention, such as Figure 4 As shown in the figure, the void type and cavity type can not only depict the outline of the disease but also detect the type and location of the disease. It is a tunnel lining disease identification model suitable for ground penetrating radar B-Scan data processing.

[0059] To verify the accuracy of the model, the model's disease recognition accuracy (precision) and recall (recall) were tested using an artificially constructed tunnel lining data set. The accuracy and recall formulas are as follows:

[0060]

[0061] Where TP is the number of samples predicted to be diseased targets but are actually diseased; FP is the number of samples predicted to be diseased targets but are actually clutter signals; and FN is the number of samples predicted to be noise signals but are actually diseased targets.

[0062] The present invention also provides a ground penetrating radar tunnel lining defect identification device, as described in the following embodiments. Since the principle of the device to solve the problem is similar to the ground penetrating radar tunnel lining defect identification method, the implementation of the device can refer to the implementation of the ground penetrating radar tunnel lining defect identification method, and the repeated parts will not be repeated.

[0063] Figure 5 This is a schematic diagram of a ground penetrating radar tunnel lining defect identification device in an embodiment of the present invention, the device comprising:

[0064] The amplitude-frequency characteristic data and profile acquisition module 501 is used to perform amplitude-frequency characteristic analysis on the tunnel lining data collected by the ground penetrating radar to obtain the corresponding amplitude-frequency characteristic data; generate a radar reflection signal profile according to the tunnel lining data, mark defects on the radar reflection signal profile, and obtain a profile with marked defects;

[0065] The first disease identification result output module 502 is used to input the amplitude-frequency feature data into the feature deconstruction analysis model and output the first disease identification result; the feature deconstruction analysis model is obtained by training the convolutional neural network based on the historical amplitude-frequency feature data and the corresponding historical first disease identification results;

[0066] The second disease identification result output module 503 is used to input the disease-annotated cross-section image into the cross-section image disease identification model and output the second disease identification result; the cross-section image disease identification model is obtained by training a deep learning network based on the historical disease-annotated cross-section image and the corresponding historical second disease identification results;

[0067] The final tunnel lining disease identification result generating module 504 is used to perform weighted averaging on the first disease identification result and the second disease identification result to generate the final tunnel lining disease identification result.

[0068] In one embodiment, a data preprocessing module is further included, which is specifically used to:

[0069] Performing data preprocessing on the tunnel lining data to obtain preprocessed tunnel lining data;

[0070] The amplitude-frequency characteristic data and profile acquisition module is specifically used for:

[0071] Conduct amplitude-frequency characteristic analysis on the pre-processed tunnel lining data;

[0072] The amplitude-frequency characteristic data and profile acquisition module is specifically used for:

[0073] Generate radar reflection signal profile based on preprocessed tunnel lining data.

[0074] In one embodiment, a feature deconstruction analysis model training module is also included, which is specifically used to:

[0075] The historical amplitude-frequency feature data and the corresponding historical first disease recognition results are input as training sets into a convolutional neural network based on the attention mechanism for training, and a trained feature deconstruction analysis model is obtained.

[0076] In one embodiment, a cross-section image disease recognition model training module is also included, which is specifically used to:

[0077] The cross-section images with historical disease annotations and the corresponding historical second disease recognition results are input as training sets into a deep learning network based on the YOLOv5 algorithm for training to obtain a trained cross-section disease recognition model.

[0078] In one embodiment, a model evaluation module is further included, which is specifically used to:

[0079] The characteristic deconstruction analysis model and the profile defect recognition model are evaluated respectively through the model evaluation indexes to obtain the corresponding evaluation results;

[0080] According to the corresponding evaluation results, the weights of the feature deconstruction analysis model and the profile defect recognition model are adjusted respectively; the model evaluation indicators include accuracy, recall rate and number of frames transmitted per second.

[0081] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned ground penetrating radar tunnel lining defect identification method when executing the computer program.

[0082] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned ground penetrating radar tunnel lining defect identification method is implemented.

[0083] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned ground penetrating radar tunnel lining disease identification method is implemented.

[0084] In an embodiment of the present invention, the corresponding amplitude-frequency characteristic data is obtained by performing amplitude-frequency characteristic analysis on the tunnel lining data collected by the ground penetrating radar; a radar reflection signal profile is generated according to the tunnel lining data, and defects are marked on the radar reflection signal profile to obtain a profile with marked defects; the amplitude-frequency characteristic data is input into a feature deconstruction analysis model to output a first defect recognition result; the feature deconstruction analysis model is obtained by training a convolutional neural network based on historical amplitude-frequency characteristic data and corresponding historical first defect recognition results; the profile with marked defects is input into a profile defect recognition model to output a second defect recognition result; the profile defect recognition model is obtained by training a deep learning network based on historical defect-marked profiles and corresponding historical second defect recognition results; the first defect recognition result and the second defect recognition result are weighted averaged to generate a final tunnel lining defect recognition result. In the above process, the embodiment of the present invention analyzes and combines the two types of data (amplitude-frequency characteristic data and profile diagram), performs weighted averaging processing on the defect identification results of the feature deconstruction analysis model and the defect identification results in the profile diagram defect identification model, and realizes the identification of tunnel lining defects, thereby improving the identification accuracy and efficiency of tunnel lining defects.

[0085] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may 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-ROM, optical storage, etc.) containing computer-usable program code.

[0086] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, 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 device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0087] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0089] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. 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 identifying tunnel lining defects using ground penetrating radar, characterized in that: include: Perform amplitude-frequency characteristic analysis on the tunnel lining data collected by the ground penetrating radar to obtain the corresponding amplitude-frequency characteristic data; Generate a radar reflection signal profile according to the tunnel lining data, mark defects on the radar reflection signal profile, and obtain a profile with marked defects; Inputting the amplitude-frequency feature data into a feature deconstruction analysis model, and outputting a first disease recognition result; the feature deconstruction analysis model is obtained by training a convolutional neural network based on the historical amplitude-frequency feature data and the corresponding historical first disease recognition results; Inputting the disease-annotated cross-section image into a cross-section image disease recognition model, and outputting a second disease recognition result; the cross-section image disease recognition model is obtained by training a deep learning network based on the historical disease-annotated cross-section image and the corresponding historical second disease recognition results; The first defect identification result and the second defect identification result are weighted averaged to generate the final tunnel lining defect identification result.

2. The method according to claim 1, characterized in that Before performing amplitude-frequency characteristic analysis on the tunnel lining data collected by the ground penetrating radar to obtain the corresponding amplitude-frequency characteristic data, the following steps are included: Performing data preprocessing on the tunnel lining data to obtain preprocessed tunnel lining data; The amplitude-frequency characteristics of the tunnel lining data collected by the ground penetrating radar are analyzed, including: Conduct amplitude-frequency characteristic analysis on the pre-processed tunnel lining data; Generate radar reflection signal profiles based on tunnel lining data, including: Generate radar reflection signal profile based on preprocessed tunnel lining data.

3. The method according to claim 1, characterized in that According to the historical amplitude-frequency characteristic data and the corresponding historical first disease recognition results, the convolutional neural network is trained, including: The historical amplitude-frequency feature data and the corresponding historical first disease recognition results are input as training sets into a convolutional neural network based on the attention mechanism for training, and a trained feature deconstruction analysis model is obtained.

4. The method according to claim 1, characterized in that According to the historical disease-labeled cross-section and the corresponding historical second disease recognition results, the deep learning network is trained, including: The cross-section images with historical disease annotations and the corresponding historical second disease recognition results are input as training sets into a deep learning network based on the YOLOv5 algorithm for training to obtain a trained cross-section disease recognition model.

5. The method according to claim 1, characterized in that Also includes: The characteristic deconstruction analysis model and the profile defect recognition model are evaluated respectively through the model evaluation indexes to obtain the corresponding evaluation results; According to the corresponding evaluation results, the weights of the feature deconstruction analysis model and the profile defect recognition model are adjusted respectively; the model evaluation indicators include accuracy, recall rate and number of frames transmitted per second.

6. A ground penetrating radar tunnel lining defect identification device, characterized in that: include: The amplitude-frequency characteristic data and profile acquisition module is used to perform amplitude-frequency characteristic analysis on the tunnel lining data collected by the ground penetrating radar to obtain the corresponding amplitude-frequency characteristic data; Generate a radar reflection signal profile according to the tunnel lining data, mark defects on the radar reflection signal profile, and obtain a profile with marked defects; A first disease identification result output module is used to input the amplitude-frequency feature data into a feature deconstruction analysis model and output a first disease identification result; the feature deconstruction analysis model is obtained by training a convolutional neural network based on the historical amplitude-frequency feature data and the corresponding historical first disease identification results; A second disease identification result output module, used to input the disease-annotated cross-section image into a cross-section image disease identification model, and output a second disease identification result; the cross-section image disease identification model is obtained by training a deep learning network based on the historical disease-annotated cross-section image and the corresponding historical second disease identification results; The final tunnel lining disease identification result generation module is used to perform weighted averaging on the first disease identification result and the second disease identification result to generate the final tunnel lining disease identification result.

7. The device according to claim 6, characterized in that It also includes a data preprocessing module, which is specifically used to: Performing data preprocessing on the tunnel lining data to obtain preprocessed tunnel lining data; The amplitude-frequency characteristic data and profile acquisition module is specifically used for: Conduct amplitude-frequency characteristic analysis on the pre-processed tunnel lining data; The amplitude-frequency characteristic data and profile acquisition module is specifically used for: Generate radar reflection signal profile based on preprocessed tunnel lining data.

8. The device according to claim 6, characterized in that It also includes a feature deconstruction analysis model training module, which is specifically used for: The historical amplitude-frequency feature data and the corresponding historical first disease recognition results are input as training sets into a convolutional neural network based on the attention mechanism for training, and a trained feature deconstruction analysis model is obtained.

9. The device according to claim 6, characterized in that It also includes a profile disease recognition model training module, which is specifically used for: The cross-section images with historical disease annotations and the corresponding historical second disease recognition results are input as training sets into a deep learning network based on the YOLOv5 algorithm for training to obtain a trained cross-section disease recognition model.

10. The device according to claim 6, characterized in that It also includes a model evaluation module, which is specifically used to: The characteristic deconstruction analysis model and the profile defect recognition model are evaluated respectively through the model evaluation indexes to obtain the corresponding evaluation results; According to the corresponding evaluation results, the weights of the feature deconstruction analysis model and the profile defect recognition model are adjusted respectively; the model evaluation indicators include accuracy, recall rate and number of frames transmitted per second.

11. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

13. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images