Self-navigation tunnel crack detection method, system, equipment and medium

Abstract

The invention discloses a self-navigation tunnel crack detection method, system and device and a medium, and relates to the technical field of tunnel detection.The method comprises the steps that infrared thermal imaging data and laser scanning data on a self-navigation detection device are collected; fusing the preprocessed infrared thermal imaging data and laser scanning data through an improved convolutional neural network to generate tunnel state information; analyzing the tunnel state information through an improved support vector machine to obtain a crack state evaluation result; and according to a crack state evaluation result, a detection path of the self-navigation detection equipment is adjusted, and a detection task is completed. The tunnel detection efficiency, accuracy and safety are remarkably improved, and meanwhile cost and resource consumption are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel detection, and in particular to a self-navigating tunnel crack detection method, system, device and medium. Background Art

[0002] In the prior art, tunnel detection mainly relies on methods such as manual visual inspection, ground penetrating radar, and ultrasonic detection. Although these methods can evaluate the structural state of the tunnel to a certain extent, they also have the following common drawbacks: Low efficiency of manual detection: Manual visual inspection requires a large amount of manpower and time, and is easily affected by subjective factors, resulting in inaccurate detection results.

[0003] Complex self-navigating detection equipment: Ground penetrating radar and ultrasonic self-navigating detection equipment are usually large in size, complex in operation, and require professional technical personnel for operation and data analysis.

[0004] Inability to monitor in real time: The prior art cannot achieve real-time monitoring of the tunnel, and cannot timely detect cracks and water seepage problems in the tunnel.

[0005] Poor environmental adaptability: The prior art has poor effects in complex tunnel environments (such as humid, dark, etc.) and cannot provide accurate detection data.

[0006] Personnel safety risks: Tunnel detection usually requires personnel to enter the tunnel interior, posing certain safety risks, especially in tunnels with harsh environments or unstable structures.

[0007] Waste of human resources: According to regulations, two people must accompany when entering the tunnel for detection, which increases the investment in human resources. Summary of the Invention

[0008] In view of the above existing problems, the present invention is proposed.

[0009] Therefore, the technical problem solved by the present invention is that the existing tunnel detection methods are inefficient and inaccurate.

[0010] To solve the above technical problem, the present invention provides the following technical solution: A self-navigating tunnel crack detection method, which includes the following steps: Collect infrared thermal imaging data and laser scanning data on the self-navigating detection device; Fuse the preprocessed infrared thermal imaging data and laser scanning data through an improved convolutional neural network to generate tunnel state information; Analyze the tunnel state information through an improved support vector machine to obtain a crack state evaluation result, and the improved support vector machine includes an integrated graph convolution operation and a designed multi-task loss function; Adjust the detection path of the self-navigating detection device according to the crack state assessment result to complete the detection task.

[0011] As a preferred solution of a self-navigating tunnel crack detection method according to the present invention, wherein: The preprocessing includes denoising the infrared thermal imaging data and the laser scanning data by using an improved wavelet transform denoising algorithm; The improved wavelet transform denoising algorithm includes the following steps: Use the sliding window technique to estimate the noise standard deviation of the current signal segment in real time; Based on the noise standard deviation and combined with the environmental parameters in the tunnel, dynamically adjust the noise adjustment factor to obtain an adaptive threshold; Perform non-local means filtering on the infrared thermal imaging data and the laser scanning data to obtain the data after non-local means filtering; Perform wavelet transform on the data after non-local means filtering, and combine with the adaptive threshold to obtain the wavelet coefficients after threshold processing; Perform wavelet reconstruction on the wavelet coefficients after threshold processing to obtain the denoised infrared thermal imaging data and laser scanning data.

[0012] As a preferred solution of a self-navigating tunnel crack detection method according to the present invention, wherein: The fusion through the improved convolutional neural network includes the following steps: Extract the temperature gradient feature from the preprocessed infrared thermal imaging data, and extract the point cloud density and geometric shape features from the preprocessed laser scanning data; Based on the temperature gradient feature, point cloud density and geometric shape features, design a dual attention fusion mechanism to enhance the representation ability of the features, and obtain the feature map after being processed by the dual attention fusion module; Based on the feature map after being processed by the dual attention fusion module, introduce a feature pyramid structure to fuse the crack features of different scales and generate the tunnel state information; The dual attention fusion mechanism includes a channel attention mechanism and a spatial attention mechanism.

[0013] As a preferred solution of a self-navigating tunnel crack detection method according to the present invention, wherein: The adaptive threshold Is expressed as: , Wherein, Is the noise standard deviation, Is the signal length, Is the environmental noise adjustment factor.

[0014] As a preferred solution of a self-navigating tunnel crack detection method according to the present invention, wherein: The analysis of the tunnel state information by the improved support vector machine includes the following steps: Based on the tunnel state information, update the node features through graph convolution operations, capture the spatial neighborhood relationship between the nodes in the crack area, and generate high-quality node features; Based on the high-quality node features, design a multi-task loss function, optimize the crack category prediction and width prediction simultaneously, and obtain the crack state evaluation result.

[0015] As a preferred solution of a self-navigating tunnel crack detection method according to the present invention, wherein: The graph convolution operation is expressed as: , wherein, represents the output feature of the -th graph convolution layer, represents the neighborhood aggregation operation for normalizing the node features, represents the adjacency matrix, represents the degree matrix, is the -th layer feature, is the weight matrix, is the activation function.

[0016] As a preferred solution of a self-navigating tunnel crack detection method according to the present invention, wherein: The multi-task loss function is expressed as: , wherein, and are the balance coefficients, is the classification loss, is the regression loss; The regression loss is expressed as: , wherein M is the number of samples, represents the true crack width of the -th sample, represents the predicted value of the crack width of the -th sample by the model.

[0017] Another object of the present invention is to provide a self-navigating tunnel crack detection system.

[0018] To solve the above technical problems, the present invention provides the following technical solutions: A self-navigating tunnel crack detection system, comprising: A collection module, configured to collect infrared thermal imaging data and laser scanning data on the self-navigating detection device; A fusion module, configured to fuse the preprocessed infrared thermal imaging data and laser scanning data through an improved convolutional neural network to generate tunnel state information; A data analysis module, configured to analyze the tunnel state information through an improved support vector machine to obtain a crack state evaluation result, where the improved support vector machine includes an integrated graph convolution operation and a designed multi-task loss function; An operation adjustment module, configured to adjust the detection path of the self-navigating detection device according to the crack state evaluation result to complete the detection task.

[0019] The present invention provides an electronic device, comprising: A memory, configured to store a program; A processor, configured to execute the computer-executable instructions, and when the computer-executable instructions are executed by the processor, the steps of the self-navigating tunnel crack detection method are implemented.

[0020] The present invention provides a computer-readable storage medium, comprising: When the program is executed by the processor, the steps of the self-navigating tunnel crack detection method are implemented.

[0021] Advantages of the present invention: By integrating an infrared thermal imaging module and a laser scanning module on a self-navigating detection device (such as a robotic dog), the present invention can capture the temperature change and three-dimensional structure information on the tunnel surface in real time, and use an intelligent module for data fusion and analysis, so as to quickly and accurately judge the size, position and water seepage condition of the tunnel crack. Compared with the prior art, the present invention has stronger environmental adaptability, is less affected by the tunnel environment, and can provide reliable detection results under a wider range of tunnel conditions; at the same time, the feedback mechanism is more timely, can directly provide the detection results and relevant data for the operator, help the operator take measures in time, effectively improve the detection efficiency and quality, reduce resource waste, and ensure the safety and stability of the tunnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0023] Figure 1Schematic diagram of the basic process of a self-navigating tunnel crack detection method provided by an embodiment of the present invention; Figure 2 Schematic diagram of the complete process of a self-navigating tunnel crack detection method provided by an embodiment of the present invention. Detailed implementation manners

[0024] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following will describe the detailed implementation manners of the present invention with reference to the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0025] Embodiment 1, referring to Figure 1 and Figure 2 , an embodiment of the present invention provides a self-navigating tunnel crack detection method. As shown in Figure 1 , 2, it includes the following steps: S1: Collect infrared thermal imaging data and laser scanning data on the self-navigating detection device; In the embodiment of the present invention, a high-resolution infrared thermal imaging module is installed at the front end of the self-navigating detection device (such as a robotic dog). After startup, self-check and calibration are performed to ensure the normal operation of the device. This module captures the temperature changes on the tunnel surface in real time at a frame rate of 30fps and generates a temperature image. The temperature detection range of the infrared thermal imaging module is -20°C to 100°C, the sensitivity is ±0.1°C, the resolution is not less than 640×480 pixels, and it can work in an ambient temperature range of -10°C to 50°C. The captured temperature signal is converted into digital image data and transmitted to the intelligent module for further processing in real time.

[0026] In the embodiment of the present invention, a laser scanning module is also installed at the front end of the self-navigating detection device. After startup, self-check and calibration are performed. The laser scanning module scans the tunnel surface in real time at a frequency of 10Hz and generates high-precision three-dimensional point cloud data. The scanning range covers the entire area of the tunnel surface, the horizontal scanning angle is 360°, the vertical scanning angle is 90°, the point cloud density is not less than 100 points per square meter, and the measurement accuracy reaches the millimeter level (±1mm). The scanned point cloud data is converted into a digital format and stored in the data storage unit for subsequent analysis.

[0027] S2: Fuse the preprocessed infrared thermal imaging data and laser scanning data through an improved convolutional neural network to generate tunnel state information; In the embodiments of the present invention, the collected infrared thermal imaging data and laser scanning data are first preprocessed. An improved wavelet transform denoising algorithm is used to denoise the data, removing environmental noise and equipment noise to ensure the purity of the data. The data is calibrated through a calibration board to ensure the consistency and accuracy of the data, and the calibration accuracy reaches 0.1%. The preprocessed data provides high-quality input for subsequent fusion and analysis.

[0028] In the embodiments of the present invention, the improved wavelet transform denoising algorithm includes the following steps: (1) Adaptive noise estimation and threshold calculation: Based on the traditional wavelet transform denoising algorithm, considering the complexity of the tunnel environment, an adaptive noise estimation mechanism is introduced. The noise standard deviation is estimated in real time through a sliding window , and the threshold is dynamically adjusted according to environmental parameters such as the humidity and illumination of the tunnel. The formula is expressed as: , where, represents the threshold during wavelet coefficient threshold processing, which is used to remove wavelet coefficients smaller than this threshold to achieve the purpose of denoising, is the noise standard deviation (estimated in real time through a sliding window), represents the natural logarithm, N is the signal length, is the environmental noise adjustment factor, which is dynamically adjusted according to environmental parameters such as the humidity and illumination of the tunnel (for example, when the humidity > 80% = 1.2, enhancing the denoising intensity to better adapt to the noise characteristics in a humid environment; in areas with low illumination, the value of can be appropriately adjusted to optimize the denoising effect.). (2) Hybrid noise joint processing: For the possible hybrid noise (such as Gaussian noise and salt-and-pepper noise) in the tunnel environment, non-local means filtering ( ) is combined for processing. First, non-local means filtering is performed on the original data, and then wavelet transform denoising ( ) is performed to improve the robustness of denoising. The formula is: , where, represents the signal after non-local means filtering and wavelet transform denoising, represents the original signal, that is, the unprocessed original data obtained from the infrared thermal imaging data and laser scanning data.

[0029] Among them, the calculation formula of non-local means filtering is: , where, Represents the result after non - local means filtering on a pixel which is a certain pixel point in the infrared thermal imaging data or laser scanning data and is the value of the pixel value representing the pixel value at the position in the infrared thermal imaging data or laser scanning data where is a normalization constant used to ensure that the result after filtering is within a reasonable range and its role is to make the sum of weights equal to 1, that is ; represents the summation operation for all pixels within the search window is the similarity weight between pixel and calculated based on the temperature gradient of the infrared thermal imaging data and the geometric features of the laser point cloud. The similarity weight reflects the importance of pixel for calculating the filtered value of pixel . The more similar the temperature gradients and geometric features of two pixels are, the larger the weight . In this way, mixed noise can be more effectively removed and useful signals can be retained

[0030] (3) Wavelet reconstruction: Wavelet reconstruction is performed according to the processed wavelet coefficients to obtain the denoised signal, and the formula is the same as the traditional wavelet reconstruction formula: , where represents the denoised signal, that is, the processed infrared thermal imaging data and laser scanning data represents the summation operation for all and where represents the processed wavelet coefficients represents the wavelet basis function

[0031] S3: Analyze the tunnel state information through an improved support vector machine to obtain the crack state evaluation result. The improved support vector machine includes an integrated graph convolution operation and a designed multi - task loss function; In the embodiment of the present invention, the pre - processed data is fused through an improved convolutional neural network (CNN). Temperature gradient features are extracted from the infrared thermal imaging data, and point cloud density and geometric shape features are extracted from the laser scanning data. The fused data generates comprehensive tunnel state information, and the fusion accuracy reaches more than 95%. The fused data provides comprehensive reference information for subsequent crack and water seepage detection. The detailed process of data fusion using the improved convolutional neural network is as follows: (1) Dual - attention fusion module: To make more full use of the spatial correlation of multimodal data, a dual attention fusion module is designed to capture the channel and spatial importance of infrared and laser data respectively.

[0032] Channel Attention (CA): , where, represents the feature map processed by the channel attention module, represents the activation function, usually the Sigmoid function, which is used to map the output of the multi-layer perceptron (MLP) to the interval (0, 1) to generate the channel attention weights. The MLP is used to perform non-linear transformation on the features after global average pooling to learn the dependencies between channels. represents the global average pooling operation on the input original feature map F, compressing the feature map of each channel into a value to obtain the global information of the channel; represents element-wise multiplication, multiplying the generated channel attention weights with the corresponding elements of the original feature map F to enhance the feature responses of important channels.

[0033] The global information of the feature map is obtained through global average pooling, then processed by the multi-layer perceptron (MLP) to obtain the channel attention weights, which are multiplied element-wise with the original feature map to enhance the features of important channels.

[0034] Spatial Attention (SA): , where, represents the feature map processed by the spatial attention module. is also the activation function (such as the Sigmoid function), which is used to generate the spatial attention weights. represents the two-dimensional convolution operation, which performs convolution processing on the concatenated results of max-pooling and average-pooling to extract spatial features and generate the spatial attention weights. represents the input original feature map The results of max-pooling and average-pooling on are concatenated along the channel dimension to generate the spatial attention weights by integrating the maximum and average information. represents the max-pooling operation on the original feature map, highlighting the important feature positions in the feature map. represents the average-pooling operation on the original feature map, reflecting the overall information of the feature map. represents the average-pooling operation on the original feature map represents element-wise multiplication, multiplying the spatial attention weights with the original feature map Multiply the corresponding elements to highlight the characteristics of important spatial regions.

[0035] (2)Feature pyramid fusion: Introduce a feature pyramid structure to fuse crack features at different scales. The formula is: , where, represents the final feature map after fusion, which synthesizes crack features at different scales, that is the state information of the tunnel. represents the feature maps of different convolutional layers, corresponding to the microscopic, mesoscopic, and macroscopic features of the cracks respectively. represents the upsampling operation on the macroscopic feature map to raise the resolution to be the same or similar to that of the mesoscopic feature map for fusion. is usually a macroscopic feature map with high semantic information but low resolution. The upsampling operation enables to be aligned with other scale feature maps in the spatial dimension. represents the downsampling operation on the microscopic feature map to reduce the resolution to be the same or similar to that of the mesoscopic feature map , generally has high resolution but relatively weak semantic information. Downsampling can make play a better role when fusing with other feature maps. Through upsampling and downsampling operations, feature maps at different scales are fused to enrich feature expression and improve the crack detection ability.

[0036] S4: According to the crack state evaluation result, adjust the detection path of the self-navigating detection device to complete the detection task.

[0037] In the embodiment of the present invention, the fused data is analyzed by an improved support vector machine (SVM). The algorithm identifies the position and size of the cracks, and the crack detection accuracy reaches the millimeter level (±0.5 mm). At the same time, the seepage area is identified through infrared thermal imaging data, the temperature change threshold is set to ±2 °C, and the seepage detection accuracy reaches the centimeter level (±1 cm). The analysis results are transmitted to the intelligent module in real time to provide real-time feedback to the operator. Steps of the improved support vector machine algorithm: (1)Graph convolutional feature enhancement: The traditional support vector machine has low classification efficiency for high-dimensional features and does not consider the spatial continuity of cracks. Introduce a graph convolutional network (GCN) to model the spatial neighborhood relationship of cracks and enhance the learning ability of crack features. The formula is: , Among them, represents the output feature of the th layer of graph convolutional layer. represents the core operation part of graph convolution, represents the adjacency matrix, which is constructed based on the spatial distance of the laser point cloud and describes the connection relationship between the nodes in the crack area; represents the degree matrix, and its diagonal element is the sum of the elements in the corresponding row (or column) of used for the neighborhood aggregation operation to normalize the node features, is the feature of the th layer, which is the output of the previous layer of graph convolutional layer or the initial node feature, is the weight matrix, which is learned during the training process and is used to perform a linear transformation on the input features, adjust the weights of the features, and better fit the data, is the activation function. Through the graph convolution operation, it can better capture the spatial features of the cracks and improve the classification accuracy.

[0038] (2) Multi-task loss function: Design a multi-task loss function to optimize the crack category and width prediction simultaneously. The multi-task loss function formula is: , Among them, and are the balance coefficients, which are used to adjust the weights of the classification loss and the regression loss .

[0039] ① Classification loss : It is used to measure the accuracy of the model's prediction of the crack category (such as normal, cracked, water seepage, etc.).

[0040] , Among them, M is the number of samples, is the true label of the sample (usually in the form of one - hot encoding), is the probability distribution predicted by the model for the sample . In the application scenario of tunnel crack detection, each sample is actually represented by the features after the fusion of infrared thermal imaging data and laser scanning data.

[0041] ② Regression loss (crack width prediction) : It is used to measure the difference between the predicted value of the crack width by the model and the true width .

[0042] , In this way, during the training process, the classification of cracks and the width prediction task are considered simultaneously, improving the comprehensive performance of the model.

[0043] In the embodiment of the present invention, the intelligent module transmits the analysis result to the remote control device of the operator through a wireless communication unit (such as Wi-Fi or 4G). The remote control device should have a high-resolution display screen (not less than 1920×1080 pixels), which can clearly display the cracks and water seepage conditions in the tunnel. The data transmission delay should be less than 1 second to ensure that the operator can obtain the detection information in real time.

[0044] In the embodiment of the present invention, the operator adjusts the path of the self-navigating detection device through a friendly operation interface according to the tunnel state information displayed on the remote control device. The operator can mark the problematic areas on the remote control device and record detailed information, including the crack position, size, water seepage conditions, etc. The recorded data is stored in the cloud or local device for subsequent analysis and processing. The path adjustment accuracy should reach the centimeter level (±1 cm).

[0045] In the embodiment of the present invention, the detection path of the self-navigating detection device is adjusted, including the self-navigating detection device perceiving the entrance and internal environment of the tunnel through sensors (such as inertial measurement unit (IMU), global positioning system (GPS), lidar, etc.). A map-based path planning algorithm (such as A* algorithm) is adopted to plan the walking route of the self-navigating detection device according to the perceived environmental information. The path planning accuracy should reach the centimeter level (±1 cm), and the planning time should be less than 1 minute. The self-navigating detection device moves autonomously according to the planned path, enters the tunnel for detection, and the sensors continuously perceive the obstacles in the environment. The path is automatically adjusted through a lidar-based obstacle avoidance algorithm to avoid obstacles. The moving speed of the self-navigating detection device is controlled between 0.5 m / s and 1.5 m / s, and the obstacle avoidance distance is set between 0.5 m and 1 m to ensure the safe operation of the device.

[0046] In the embodiment of the present invention, the transformation of the self-navigating detection device (taking a quadruped robot as an example): 1. Structure design of the quadruped robot: (1) Infrared thermal imaging module: Installation location: Install an infrared thermal imaging camera at the front end of the quadruped robot to capture the temperature changes on the tunnel surface in real time. This module includes a high-resolution infrared detector and an optical lens, which can detect tiny temperature differences to identify water seepage areas.

[0047] Function: By detecting the temperature changes on the surface of the tunnel, identify the seepage areas. Seepage areas usually have temperature changes due to water evaporation, and the infrared thermal imaging module can capture these changes and convert them into temperature images.

[0048] Data processing: The infrared thermal imaging module converts the captured temperature data into digital signals and transmits them to the intelligent module for further processing.

[0049] (2) Laser scanning module: Installation location: Install a laser scanner at the front end of the robotic dog to generate a three-dimensional model of the tunnel surface in real time. This module includes a laser emitter, a receiver, and a scanning mechanism, which can quickly scan the tunnel surface and generate high-precision three-dimensional point cloud data.

[0050] Function: Generate a three-dimensional model of the tunnel surface through laser scanning technology for analyzing the size and location of cracks in the tunnel. The laser scanning module can accurately measure the geometric shape and dimensions of the tunnel surface, providing basic data for crack detection.

[0051] Data processing: The laser scanning module converts the scanned three-dimensional data into point cloud format and transmits it to the intelligent module for further processing.

[0052] 2. Intelligent module: Installation location: Install the intelligent module at the control center of the robotic dog, which is responsible for data fusion, analysis, and feedback. This module includes a central processing unit, a data storage unit, a data fusion unit, and a wireless communication unit.

[0053] Function: ① Data fusion: Integrate the data from the infrared thermal imaging module and the laser scanning module to generate comprehensive tunnel status information.

[0054] ② Data analysis: Analyze the fused data through preset algorithms and models to identify cracks and seepage areas and calculate the size and location of the cracks.

[0055] ③ Feedback and display: Transmit the analysis results to the operator's remote control device through the wireless communication unit to provide real-time tunnel status information.

[0056] ④ Self-navigation function: Integrate a self-navigation system, including sensors (such as IMU, GPS, lidar, etc.) and path planning algorithms, enabling the robotic dog to autonomously navigate into the tunnel and complete the detection task.

[0057] 3. Remote control device: Function: The operator can view the surface conditions of the tunnel in real time through a remote control device and record the problematic areas. The remote control device can be a tablet computer, a laptop, or other portable devices, equipped with a dedicated software interface for displaying and recording the detection data.

[0058] Data recording: The operator can mark the problematic areas on the remote control device and record detailed information such as the crack location, size, water seepage situation, etc., for subsequent analysis and processing.

[0059] In the embodiment of the present invention, the operation process of tunnel detection is as follows: 1. On-site preparation: Determine the entrance position of the tunnel and the detection path, and input the relevant information into the intelligent module of the robotic dog.

[0060] Check the battery power and device status of the robotic dog to ensure that the device can complete the entire detection task.

[0061] 2. Equipment startup and entry: The operator starts the robotic dog through the remote control device, and the device performs self-check and initialization.

[0062] The robotic dog autonomously navigates into the tunnel according to the preset path and starts the detection task.

[0063] During the process of entering the tunnel, the robotic dog perceives the environmental information in real time, automatically adjusts the path, and avoids obstacles.

[0064] 3. Real-time detection and data transmission: When the robotic dog moves in the tunnel, the infrared thermal imaging module and the laser scanning module work simultaneously to collect the temperature changes and three-dimensional structure information of the tunnel surface in real time.

[0065] The intelligent module preprocesses, fuses, and analyzes the collected data, identifies the crack and water seepage areas, and calculates the size and location of the cracks.

[0066] The analysis results are transmitted to the operator's remote control device in real time through the wireless communication unit, and the operator can view the crack and water seepage conditions of the tunnel in real time through the display screen of the remote control device.

[0067] 4. Operation adjustment and recording: The operator adjusts the detection path of the robotic dog according to the detection results displayed on the remote control device to ensure that the device can comprehensively detect all areas of the tunnel.

[0068] The operator can mark the problematic areas on the remote control device and record detailed information, including the crack location, size, water seepage situation, etc.

[0069] The recorded data is stored in the cloud or local devices for subsequent analysis and processing.

[0070] 5. Detection Completion and Exit: When the robotic dog completes the preset detection path, it automatically returns to the tunnel entrance.

[0071] The operator confirms the completion of the detection task through the remote control device and shuts down and retrieves the device.

[0072] Conduct a preliminary analysis of the detection data, evaluate the overall condition of the tunnel, and determine whether there are problem areas that require further processing.

[0073] It should be noted that the present invention uses infrared thermal imaging and laser scanning technologies to collect tunnel surface data in real time, and immediately analyzes and feedbacks the crack and water seepage conditions. Without waiting for the long detection process of traditional methods, it significantly shortens the detection cycle and speeds up the project progress. Traditional tunnel detection methods often require interrupting operations for sampling or detection, while this system can achieve continuous detection operations without interruption, further improving the detection efficiency. Traditional tunnel detection usually requires personnel to enter the tunnel interior, posing safety risks. The robotic dog of the present invention can independently enter the tunnel for detection, avoiding personnel entering dangerous areas and reducing safety risks. Traditional detection methods require a large amount of manpower and time. This system reduces the dependence on manpower through automated detection and reduces labor costs. In addition to the robotic dog, this method can also be applied to other detection devices such as drones and rail robots, and appropriate devices can be selected according to different tunnel environments and detection requirements. This system has good scalability and can add other detection functions according to needs, such as gas detection and structural stress detection, further enhancing the comprehensiveness and accuracy of tunnel detection.

[0074] Embodiment 2, in this embodiment, a self-navigating tunnel crack detection system is provided, including: An acquisition module for acquiring infrared thermal imaging data and laser scanning data on the self-navigating detection device; A fusion module for fusing the preprocessed infrared thermal imaging data and laser scanning data through an improved convolutional neural network to generate tunnel state information; A data analysis module for analyzing the tunnel state information through an improved support vector machine to obtain a crack state evaluation result, and the improved support vector machine includes an integrated graph convolution operation and a designed multi-task loss function; An operation adjustment module for adjusting the detection path of the self-navigating detection device according to the crack state evaluation result to complete the detection task.

[0075] This embodiment also provides an electronic device applicable to a self-navigating tunnel crack detection method, including: A memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement a self-navigating tunnel crack detection method as proposed in the above embodiments.

[0076] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements a self-navigating tunnel crack detection method as proposed in the above embodiments.

[0077] The storage medium proposed in this embodiment and the self-navigating tunnel crack detection method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be referred to the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0078] From the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software and necessary general-purpose hardware. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as a floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk, or optical disc of a computer, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of the various embodiments of the present invention.

Claims

1. A self-navigating tunnel crack detection method, characterized in that Including: Infrared thermal imaging data and laser scanning data collected from a navigation detection device; Fusing the preprocessed infrared thermal imaging data and laser scanning data through an improved convolutional neural network to generate tunnel state information; Analyzing the tunnel state information through an improved support vector machine to obtain a crack state evaluation result, where the improved support vector machine includes integrating graph convolution operations and designing a multi-task loss function; Adjusting the detection path of the self-navigation detection device according to the crack state evaluation result to complete the detection task.

2. The self-navigating tunnel crack detection method according to claim 1, characterized in that: The preprocessing includes denoising the infrared thermal imaging data and laser scanning data using an improved wavelet transform denoising algorithm; The improved wavelet transform denoising algorithm includes the following steps: Using a sliding window technique to estimate the noise standard deviation of the current signal segment in real time; Based on the noise standard deviation and combined with the environmental parameters in the tunnel, dynamically adjusting the noise adjustment factor to obtain an adaptive threshold; Performing non-local mean filtering on the infrared thermal imaging data and laser scanning data to obtain data after non-local mean filtering; Performing wavelet transform on the data after non-local mean filtering and combining with the adaptive threshold to obtain wavelet coefficients after threshold processing; Performing wavelet reconstruction on the wavelet coefficients after threshold processing to obtain denoised infrared thermal imaging data and laser scanning data.

3. The self-navigating tunnel crack detection method according to claim 2, characterized in that: The fusion through the improved convolutional neural network includes the following steps: Extracting temperature gradient features from the preprocessed infrared thermal imaging data and extracting point cloud density and geometric shape features from the preprocessed laser scanning data; Based on the temperature gradient features, point cloud density and geometric shape features, designing a dual attention fusion mechanism to enhance the representation ability of the features and obtaining a feature map processed by the dual attention fusion module; Based on the feature map processed by the dual attention fusion module, introducing a feature pyramid structure to fuse crack features at different scales to generate tunnel state information; The dual attention fusion mechanism includes a channel attention mechanism and a spatial attention mechanism.

4. The self-navigating tunnel crack detection method according to claim 3, characterized in that: The adaptive threshold is expressed as: ， wherein, is the noise standard deviation, is the signal length, is the environmental noise adjustment factor.

5. The self-navigating tunnel crack detection method according to claim 4, characterized in that: The analysis of the tunnel state information through the improved support vector machine includes the following steps: Based on the tunnel state information, updating node features through graph convolution operations, capturing the spatial neighborhood relationship between nodes in the crack area, and generating high-quality node features; Based on the high-quality node features, designing a multi-task loss function to simultaneously optimize crack category prediction and width prediction to obtain a crack state evaluation result.

6. The self-navigating tunnel crack detection method according to claim 5, wherein: The graph convolution operation is expressed as: ， Among them, represents the output feature of the -th layer of graph convolutional layer, represents the normalized adjacency matrix operation, represents the adjacency matrix, represents the degree matrix, is the feature of the -th layer, is the weight matrix, is the activation function.

7. The self-navigating tunnel crack detection method according to claim 6, characterized in that: The multi-task loss function is expressed as: ， Among them, and are balance coefficients, is the classification loss, is the regression loss; The regression loss is expressed as: ， where M is the number of samples, represents the true crack width of the th sample, and represents the predicted crack width of the th sample by the model.

8. A self-navigating tunnel crack detection system, which applies a self-navigating tunnel crack detection method as described in any one of claims 1 to 7, characterized in that, Including: A collection module for collecting infrared thermal imaging data and laser scanning data from a self-navigation detection device; A fusion module for fusing the preprocessed infrared thermal imaging data and laser scanning data through an improved convolutional neural network to generate tunnel state information; A data analysis module for analyzing the tunnel state information through an improved support vector machine to obtain a crack state evaluation result, where the improved support vector machine includes integrating graph convolution operations and designing a multi-task loss function; An operation adjustment module for adjusting the detection path of the self-navigation detection device according to the crack state evaluation result to complete the detection task.

9. An electronic device, characterized in that, Including: A memory for storing a program; A processor for loading the program to execute the steps of a self-navigating tunnel crack detection method according to any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by the processor, the steps of a self-navigating tunnel crack detection method according to any one of claims 1-7 are implemented.

Images