Tunnel crack tiny defect detection method based on generative adversarial network

Abstract

The invention discloses a tunnel crack tiny defect detection method based on a generative adversarial network. The method comprises the following steps: S1, acquiring a space-time crack data set containing tunnel surface tiny crack information; s2, generating a preprocessed space-time fracture data set; s3, generating a crack expansion image data set; s4, designing a generative adversarial network model which comprises a generator and a discriminator; s5, inputting the preprocessed space-time crack data set into the generative adversarial network model; s6, deploying the trained generative adversarial network model into a tunnel monitoring system; s7, performing classification according to the form, size and expansion trend of the crack, and dividing the crack into different risk grades; and S8, according to the crack detection result and the expansion trend thereof. The tunnel crack detection accuracy and the crack propagation prediction capability are improved, the automation level of the system is remarkably improved, and efficient and accurate crack detection and monitoring can be realized in a complex environment.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel crack detection, and in particular to a method for detecting tiny defects in tunnel cracks based on a generative adversarial network. Background Art

[0002] In the existing technology, tunnel crack detection mainly relies on manual inspection and traditional technical means based on image processing. Manual inspection usually requires professionals to use high-resolution cameras, laser scanners or ultrasonic equipment to conduct regular inspections of the tunnel surface and identify cracks with the naked eye. Manual inspection is not only time-consuming and labor-intensive, but also easily affected by human subjective factors, resulting in poor accuracy in early detection of cracks. It is difficult for manual detection methods to discover and quantify tiny cracks. In addition, the tunnel environment is usually complex, the lighting conditions are unstable, and external noise interference is serious, which further increases the difficulty of manual inspection.

[0003] Tunnel crack detection technology based on image processing has made certain progress in recent years. The presence of cracks can be detected by processing the collected tunnel surface images, including denoising, contrast enhancement, edge detection and image segmentation operations. However, traditional image processing technology has many limitations when facing complex tunnel environments. For example, noise interference and illumination changes may make it difficult for image processing algorithms to accurately separate crack areas, especially for early tiny cracks. The detection accuracy of traditional algorithms is low. In addition, existing technologies mainly focus on static detection of cracks, which makes it difficult to accurately predict the expansion trend of cracks and cannot provide information and early warnings on the dynamic development of cracks.

[0004] Another shortcoming of the existing technology is the lack of automation and real-time performance. Tunnel crack detection usually requires a large amount of image data processing and analysis. Traditional methods mostly rely on semi-automatic or manual intervention, resulting in low detection efficiency and inability to achieve real-time monitoring and early warning of large-scale infrastructure. As a result, potential safety hazards during crack expansion may not be discovered and handled in time, thereby increasing the risk of tunnel structure.

[0005] In summary, the existing technologies have the following main defects: first, manual detection and traditional image processing methods are difficult to accurately detect tiny cracks, especially in complex environments, and the detection accuracy is limited; second, the existing technologies are difficult to effectively predict the expansion trend of cracks and lack the ability to monitor the dynamic changes of cracks; finally, the existing crack detection technology is insufficient in automation and real-time performance, and cannot meet the needs of modern tunnel monitoring systems for real-time warning and risk assessment. Summary of the invention

[0006] One purpose of the present invention is to propose a method for detecting tiny defects in tunnel cracks based on a generative adversarial network. The present invention not only improves the accuracy of tunnel crack detection and the ability to predict crack expansion, but also significantly improves the automation level of the system, and can achieve efficient and accurate crack detection and monitoring in complex environments.

[0007] A method for detecting micro-defects of tunnel cracks based on a generative adversarial network according to an embodiment of the present invention comprises the following steps:

[0008] S1. Perform multi-time series scanning of the tunnel surface by a camera or laser scanner to obtain a spatiotemporal crack dataset containing information on micro-cracks on the tunnel surface;

[0009] S2. performing denoising, contrast enhancement, edge detection and image segmentation operations on the collected spatiotemporal crack dataset to generate a preprocessed spatiotemporal crack dataset;

[0010] S3. Based on the physical expansion characteristics of actual cracks in the tunnel, a crack evolution model is constructed using the preprocessed spatiotemporal crack dataset. The crack evolution model simulates the crack expansion process under different time and stress environments and generates a crack expansion image dataset;

[0011] S4. Design a generative adversarial network model, including a generator and a discriminator;

[0012] S5. Input the preprocessed spatiotemporal crack dataset into the generative adversarial network model. The generator generates synthetic images of tiny cracks and simulated images of their expansion trends. The discriminator compares the generated crack images with the real images, and optimizes the performance of the generative adversarial network model in crack detection and crack expansion trend prediction through multiple rounds of training iterations.

[0013] S6. Deploy the trained generative adversarial network model to the tunnel monitoring system;

[0014] S7. Based on the crack extension simulation image dataset output by the generative adversarial network, the system locates tiny cracks and classifies them according to their morphology, size and extension trend, and divides the cracks into different risk levels;

[0015] S8. Based on the crack detection results and their expansion trends, the system generates an early warning report including the crack location, size, severity and expansion trend.

[0016] Optionally, the S1 specifically includes the following contents:

[0017] S11. Use the camera to scan the tunnel surface multiple times in time series, obtain tunnel surface images at different time points, and generate a multi-time series tunnel surface spatiotemporal crack data set:

[0018]

[0019] in, Indicates that at time point t i (i=1,2,…,n) acquired tunnel surface image;

[0020] S12. For each time point t i , by performing multi-angle scanning at different locations of the tunnel, a multi-view tunnel surface spatiotemporal crack dataset is generated:

[0021]

[0022] in, Indicates that at time point t i The tunnel surface image obtained from the j-th viewing angle, m is the number of viewing angles;

[0023] S13. Based on the tunnel surface images acquired at different time points and viewing angles, a spatiotemporal crack dataset of tunnel crack micro-defects containing temporal and spatial information is constructed:

[0024]

[0025] in, Indicates that at time point t i The tunnel surface image containing tiny cracks obtained from the j-th viewing angle;

[0026] S14. For each tunnel surface image Extract crack features and mark crack locations (x crack ,y crack ), where x crack and crack The crack images and their coordinate positions under the same viewing angle are combined to construct a spatiotemporal crack dataset D crackspatial-temporal :

[0027]

[0028] in, Indicates that at time point t i and viewing angle v j The image matrix obtained below is For images The crack feature set extracted from represents the kth feature of the crack, including the length, width, depth and shape of the crack, P crack (t i ,v j )=(x crack ,y crack ) is the position coordinate of the crack in the image.

[0029] Optionally, S2 specifically includes the following contents:

[0030] S21. Use the non-local mean filtering algorithm to collect the spatiotemporal crack data set Perform denoising on each tunnel surface image The noise in the image is removed to generate a denoised tunnel surface image.

[0031] S22. The denoised tunnel surface image ( Perform contrast enhancement processing to enhance the contrast between the crack area and the background area, and generate a contrast-enhanced tunnel surface image

[0032] S23. Using Canny edge detection algorithm to detect the contrast of tunnel surface image Perform edge detection, extract edge information of cracks, and generate tunnel surface images including crack edges

[0033] S24. Tunnel surface image after edge detection Perform tunnel surface image segmentation processing, use the tunnel surface image segmentation method based on region growing algorithm to separate the crack area from the background area, and generate a tunnel surface image including the crack area

[0034] S25. The final spatiotemporal crack dataset after denoising, contrast enhancement, edge detection and tunnel surface image segmentation is obtained:

[0035]

[0036] Optionally, S3 specifically includes the following contents:

[0037] S31. Based on the preprocessed spatiotemporal crack dataset D seg Combined with the physical expansion characteristics of actual tunnel cracks, a set of crack expansion feature vectors is constructed Each feature vector Contains physical parameters of crack length, width, depth and shape;

[0038] S32. Establish a crack evolution model under different time points and stress environments based on the crack extension feature vector set:

[0039]

[0040] Among them, M crack (t,σ) is the crack evolution model, which represents the crack expansion behavior under time t and stress environment σ; Fcrack (t) is the crack characteristic vector at time t; α i is the weighting factor of the specific physical characteristics of the crack on the propagation process; is the influencing factor of the i-th crack characteristics under time i and stress environment σ; β i is the attenuation factor of stress on crack expansion, which indicates the changing trend of the influence of crack expansion under stress environment over time; S(t,σ) is the function of stress changing with time and environment, from t0 to t n The time frame for crack expansion;

[0041] S33. According to the crack evolution model M crack (t,σ), simulate the crack expansion process under different stress conditions, and adjust the crack expansion trend according to the actual stress environment parameters to generate a crack expansion prediction image at a future time point:

[0042]

[0043] Among them, I forecast (t,σ) is the predicted image of crack expansion at the future time point t and stress environment σ; γ is the adjustment parameter of the crack expansion trend, which controls the influence of the expansion speed; δ i is the adjustment factor of the characteristic expansion rate of the i-th crack; It represents the rate of change of the crack evolution model over time, reflecting the crack expansion speed over time; represents the rate at which the i-th crack characteristic changes with time; Indicates the effect of stress changes on crack propagation;

[0044] S34. Apply the fracture evolution model to each fracture image in the spatiotemporal fracture dataset The crack evolution model is used to simulate and generate a crack extension image dataset at future time points:

[0045]

[0046] Optionally, the S4 specifically includes the following contents:

[0047] S41. Design a generative adversarial network model, including a generator G and a discriminator D. The generator G is based on the crack evolution model M. crack (t,σ) generates a synthetic crack image. The discriminator D is used to distinguish between the real crack image and the generated image and verify the rationality of the generated crack extension simulation;

[0048] S42. Training input of generator G: crack extension image dataset D forecast , the generator uses the crack evolution model M crack(t,σ) Generate the crack extension simulation image I of the current crack gen (t,σ) and crack extension simulation image I sim (t,σ):

[0049] I gen (t,σ)=G(M crack (t,σ),z);

[0050] Among them, G is the generator, z is the random noise vector used to generate the random characteristics of the crack, and I gen (t,σ) is the generated crack extension simulation image;

[0051] S43. The input of the discriminator D is the crack extension simulation image I generated by the generator gen (t,σ) and the true crack image I real (t,σ), the discriminator determines the authenticity of the image by comparing the generated image and the real image, and optimizes the performance of the generator and the discriminator through the loss function:

[0052]

[0053] Among them, L D is the loss function of the discriminator, is the time change rate of each crack characteristic, reflecting the dynamic change of crack extension; is the time rate of change of the generated image, reflecting the adjustment of the crack propagation speed in the generated image;

[0054] S44. Through the feedback of the discriminator D, the generator G optimizes the generation ability, generates a crack extension simulation image that is highly similar to the real crack image, and generates an optimized crack extension simulation image I sim (t,σ), the discriminator verifies the rationality of the generated crack extension simulation image by comparing the optimized crack extension simulation image I sim The consistency between (t,σ) and the actual crack extension trend confirms that the generator simulates an image that is consistent with the actual crack extension, and generates an optimized crack extension simulation image dataset D sim .

[0055] Optionally, the S7 specifically includes the following contents:

[0056] S71. Crack extension simulation image dataset D based on the output of generative adversarial network sim , the system locates the starting point and end point of each crack, extracts the geometric features of the crack, and defines the crack morphology feature vector as:

[0057] F geo ={L crack ,W crack ,Dcrack ,θ crack};

[0058] Among them, L crack is the crack length, W crack is the crack width, D crack is the crack depth, θ crack is the crack expansion angle;

[0059] S72. By analyzing the expansion trend of cracks in real-time spatiotemporal crack data, the system extracts crack expansion characteristics and defines the crack expansion feature vector as:

[0060] F ext = {v crack ,a crack ,t ext};

[0061] Among them, v crack is the crack expansion speed, a crack is the crack extension acceleration, t ext The duration of crack expansion;

[0062] S73. Combined with crack morphology feature vector F geo and the extended eigenvector F ext , the cracks are classified according to the risk level through the threshold-based classification algorithm, and the risk level is defined as R crack , the classification rule is expressed as:

[0063]

[0064] Among them, v th1 ,v th2 is the threshold of crack extension rate, a th1 ,a th2 Threshold of crack growth acceleration, L th1 ,L th2 is the threshold of crack length;

[0065] S74. Based on the above classification results, the system classifies the cracks into different risk levels R crack To mark:

[0066] When R crack =1, the crack expansion speed, acceleration and length are all within the low threshold range, indicating that the crack expansion trend is relatively slow and the structural risk is low;

[0067] When R crack =2, the crack expansion speed, acceleration or length exceeds the lower threshold but does not reach the higher threshold, indicating that the crack has a moderate risk and the expansion trend may intensify, requiring timely monitoring;

[0068] When R crack =3, the crack expansion speed, acceleration or length reaches or exceeds the high threshold, indicating that the crack expansion trend is significant and may cause serious damage to the structure, and immediate measures need to be taken;

[0069] A risk assessment report is generated based on the shape, size and expansion trend of the cracks. The report includes the crack location information, size, shape, expansion speed, acceleration and risk level.

[0070] The beneficial effects of the present invention are:

[0071] (1) The present invention adopts a generative adversarial network model, and realizes high-precision detection of micro cracks in tunnels through the collaborative work of a generator and a discriminator. Compared with traditional image processing algorithms, the generative adversarial network uses a crack extension image data set to generate simulated images of cracks. In the early stage of cracks, the generative adversarial network model can capture the details of micro cracks that are difficult to identify by traditional methods, thereby improving the ability to detect micro defects. Experiments show that the method of the present invention can improve the accuracy and robustness of micro crack detection in complex environments, effectively reduce errors caused by environmental noise and illumination changes, and the accuracy of crack identification is improved by more than 15% compared with traditional methods.

[0072] (2) The present invention constructs a crack evolution model, simulates the crack expansion process under different stress environments based on the physical properties of tunnel cracks and real-time detection data, and combines the crack expansion simulation images generated by the generative adversarial network. The present invention can effectively predict the crack expansion trend at future time points. Different from the traditional static detection method, the present invention introduces dynamic stress parameters and time series analysis to simulate and predict the crack expansion speed, direction and acceleration, and provides a dynamic crack expansion path, which significantly improves the accuracy and timeliness of crack expansion prediction and reduces the error by about 10%.

[0073] (3) The present invention designs an automated crack risk assessment system. Combined with the spatiotemporal crack data set detected in real time, the system automatically extracts the geometric features and expansion characteristics of the cracks, and classifies the cracks using a threshold-based risk level classification algorithm. The system can locate the crack position in real time under different environments, identify the risk level of crack expansion, and generate detailed crack expansion trends and risk assessment reports, significantly reducing the need for manual intervention, achieving real-time early warning and risk assessment, and improving the level of automation of monitoring. The detection efficiency is increased by more than 30%, greatly reducing labor costs and error rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0075] Figure 1 A flow chart of a method for detecting small defects in tunnel cracks based on a generative adversarial network proposed by the present invention;

[0076] Figure 2 This is a flow chart for constructing a crack evolution model in a tunnel crack micro-defect detection method based on a generative adversarial network proposed in the present invention. DETAILED DESCRIPTION

[0077] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0078] refer to Figure 1-2 , a method for detecting small defects of tunnel cracks based on a generative adversarial network, comprising the following steps:

[0079] S1. Perform multi-time series scanning of the tunnel surface by a camera or laser scanner to obtain a spatiotemporal crack dataset containing information on micro-cracks on the tunnel surface;

[0080] S2. performing denoising, contrast enhancement, edge detection and image segmentation operations on the collected spatiotemporal crack dataset to generate a preprocessed spatiotemporal crack dataset;

[0081] S3. Based on the physical expansion characteristics of actual cracks in the tunnel, a crack evolution model is constructed using the preprocessed spatiotemporal crack dataset. The crack evolution model simulates the crack expansion process under different time and stress environments and generates a crack expansion image dataset;

[0082] S4. Design a generative adversarial network model, including a generator and a discriminator;

[0083] S5. Input the preprocessed spatiotemporal crack dataset into the generative adversarial network model. The generator generates synthetic images of tiny cracks and simulated images of their expansion trends. The discriminator compares the generated crack images with the real images, and optimizes the performance of the generative adversarial network model in crack detection and crack expansion trend prediction through multiple rounds of training iterations.

[0084] S6. Deploy the trained generative adversarial network model to the tunnel monitoring system;

[0085] S7. Based on the crack extension simulation image dataset output by the generative adversarial network, the system locates tiny cracks and classifies them according to their morphology, size and extension trend, and divides the cracks into different risk levels;

[0086] S8. Based on the crack detection results and their expansion trends, the system generates an early warning report including the crack location, size, severity and expansion trend.

[0087] In this implementation, S1 specifically includes the following contents:

[0088] S11. Use the camera to scan the tunnel surface multiple times in time series, obtain tunnel surface images at different time points, and generate a multi-time series tunnel surface spatiotemporal crack data set:

[0089]

[0090] in, Indicates that at time point t i (i=1,2,…,n) acquired tunnel surface image;

[0091] S12. For each time point t i , by performing multi-angle scanning at different locations of the tunnel, a multi-view tunnel surface spatiotemporal crack dataset is generated:

[0092]

[0093] in, Indicates that at time point t i The tunnel surface image obtained from the j-th viewing angle, m is the number of viewing angles;

[0094] S13. Based on the tunnel surface images acquired at different time points and viewing angles, a spatiotemporal crack dataset of tunnel crack micro-defects containing temporal and spatial information is constructed:

[0095]

[0096] in, Indicates that at time point t i The tunnel surface image containing tiny cracks obtained from the j-th viewing angle;

[0097] S14. For each tunnel surface image Extract crack features and mark crack locations (x crack ,y crack ), where x crack and crack The crack images and their coordinate positions under the same viewing angle are combined to construct a spatiotemporal crack dataset

[0098]

[0099] in, Indicates that at time point t i and viewing angle v jThe image matrix obtained below is For images The crack feature set extracted from represents the kth feature of the crack, including the length, width, depth and shape of the crack, P crack (t i ,v j )=(x crack ,y crack ) is the position coordinate of the crack in the image.

[0100] In this implementation, S2 specifically includes the following contents:

[0101] S21. Use the non-local mean filtering algorithm to collect the spatiotemporal crack data set Perform denoising on each tunnel surface image The noise in the image is removed to generate a denoised tunnel surface image.

[0102] S22. The denoised tunnel surface image ( Perform contrast enhancement processing to enhance the contrast between the crack area and the background area, and generate a contrast-enhanced tunnel surface image

[0103] S23. Using Canny edge detection algorithm to detect the contrast of tunnel surface image Perform edge detection, extract edge information of cracks, and generate tunnel surface images including crack edges

[0104] S24. Tunnel surface image after edge detection Perform tunnel surface image segmentation processing, use the tunnel surface image segmentation method based on region growing algorithm to separate the crack area from the background area, and generate a tunnel surface image including the crack area

[0105] S25. The final spatiotemporal crack dataset after denoising, contrast enhancement, edge detection and tunnel surface image segmentation is obtained:

[0106]

[0107] In this implementation, S3 specifically includes the following contents:

[0108] S31. Based on the preprocessed spatiotemporal crack dataset D seg Combined with the physical expansion characteristics of actual cracks in the tunnel, a set of crack expansion feature vectors is constructed Each feature vector Contains physical parameters of crack length, width, depth and shape;

[0109] S32. Establish a crack evolution model under different time points and stress environments based on the crack extension feature vector set:

[0110]

[0111] Among them, M crack (t,σ) is the crack evolution model, which represents the crack expansion behavior under time t and stress environment σ; F crack (t) is the crack characteristic vector at time t; α i is the weighting factor of the specific physical characteristics of the crack on the propagation process; is the influencing factor of the i-th crack characteristics under time i and stress environment σ; β i is the attenuation factor of stress on crack expansion, which indicates the changing trend of the influence of crack expansion under stress environment over time; S(t,σ) is the function of stress changing with time and environment, from t0 to t n The time frame for crack expansion;

[0112] S33. According to the crack evolution model M crack (t,σ), simulate the crack expansion process under different stress conditions, and adjust the crack expansion trend according to the actual stress environment parameters to generate a crack expansion prediction image at a future time point:

[0113]

[0114] Among them, I forecast (t,σ) is the predicted image of crack expansion at the future time point t and stress environment σ; γ is the adjustment parameter of the crack expansion trend, which controls the influence of the expansion speed; δ i is the adjustment factor of the characteristic expansion rate of the i-th crack; It represents the rate of change of the crack evolution model over time, reflecting the crack expansion speed over time; represents the rate at which the i-th crack characteristic changes with time; Indicates the effect of stress changes on crack propagation;

[0115] S34. Apply the fracture evolution model to each fracture image in the spatiotemporal fracture dataset The crack evolution model is used to simulate and generate a crack extension image dataset at future time points:

[0116]

[0117] In this implementation, S4 specifically includes the following contents:

[0118] S41. Design a generative adversarial network model, including a generator G and a discriminator D. The generator G is based on the crack evolution model M. crack (t,σ) generates a synthetic crack image. The discriminator D is used to distinguish between the real crack image and the generated image and verify the rationality of the generated crack extension simulation;

[0119] S42. Training input of generator G: crack extension image dataset D forecast , the generator uses the crack evolution model M crack (t,σ) Generate the crack extension simulation image I of the current crack gen (t,σ) and crack extension simulation image I sim (t,σ):

[0120] I gen (t,σ)=G(M crack (t,σ),z);

[0121] Among them, G is the generator, z is the random noise vector used to generate the random characteristics of the crack, and I gen (t,σ) is the generated crack extension simulation image;

[0122] S43. The input of the discriminator D is the crack extension simulation image I generated by the generator gen (t,σ) and the true crack image I real (t,σ), the discriminator determines the authenticity of the image by comparing the generated image and the real image, and optimizes the performance of the generator and the discriminator through the loss function:

[0123]

[0124] Among them, L D is the loss function of the discriminator, is the time change rate of each crack characteristic, reflecting the dynamic change of crack extension; is the time rate of change of the generated image, reflecting the adjustment of the crack propagation speed in the generated image;

[0125] S44. Through the feedback of the discriminator D, the generator G optimizes the generation ability, generates a crack extension simulation image that is highly similar to the real crack image, and generates an optimized crack extension simulation image I sim (t,σ), the discriminator verifies the rationality of the generated crack extension simulation image by comparing the optimized crack extension simulation image I sim The consistency between (t,σ) and the actual crack extension trend confirms that the generator simulates an image that is consistent with the actual crack extension, and generates an optimized crack extension simulation image dataset D sim .

[0126] In this implementation, S7 specifically includes the following contents:

[0127] S71. Crack extension simulation image dataset D based on the output of generative adversarial network sim , the system locates the starting point and end point of each crack, extracts the geometric features of the crack, and defines the crack morphology feature vector as:

[0128] F geo ={L crack ,W crack ,D crack ,θ crack};

[0129] Among them, L crack is the crack length, W crack is the crack width, D crack is the crack depth, θ crack is the crack expansion angle;

[0130] S72. By analyzing the expansion trend of cracks in real-time spatiotemporal crack data, the system extracts crack expansion characteristics and defines the crack expansion feature vector as:

[0131] F ext = {v crack ,a crack ,t ext};

[0132] Among them, v crack is the crack expansion speed, a crack is the crack extension acceleration, t ext The duration of crack expansion;

[0133] S73. Combined with crack morphology feature vector F geo and the extended eigenvector F ext , the cracks are classified according to the risk level through the threshold-based classification algorithm, and the risk level is defined as R crack , the classification rule is expressed as:

[0134]

[0135] Among them, v th1 ,v th2 is the threshold of crack extension rate, a th1 ,a th2 Threshold of crack growth acceleration, L th1 ,L th2 is the threshold of crack length;

[0136] S74. Based on the above classification results, the system classifies the cracks into different risk levels R crack To mark:

[0137] When R crack =1, the crack expansion speed, acceleration and length are all within the low threshold range, indicating that the crack expansion trend is relatively slow and the structural risk is low;

[0138] When R crack =2, the crack expansion speed, acceleration or length exceeds the lower threshold but does not reach the higher threshold, indicating that the crack has a moderate risk and the expansion trend may intensify, requiring timely monitoring;

[0139] When R crack =3, the crack expansion speed, acceleration or length reaches or exceeds the high threshold, indicating that the crack expansion trend is significant and may cause serious damage to the structure, and immediate measures need to be taken;

[0140] A risk assessment report is generated based on the shape, size and expansion trend of the cracks. The report includes the crack location information, size, shape, expansion speed, acceleration and risk level.

[0141] Embodiment 1:

[0142] In this embodiment 1, the implementer applied a tunnel crack micro-defect detection system based on a generative adversarial network in a highway tunnel that has been in operation for more than 15 years, and conducted continuous monitoring of the development of cracks in the tunnel for 6 months. The tunnel is 2.2 kilometers long and has been affected by unfavorable conditions such as heavy traffic, changeable climate, and frequent vibrations for a long time. Cracks of varying degrees have appeared in many places on the inner wall of the tunnel. In some hidden and difficult-to-access areas, manual inspection is difficult to detect tiny cracks in time, posing a safety hazard.

[0143] In this test, the system first scanned the tunnel surface multiple times using high-resolution laser scanners and cameras, and collected tunnel surface image data sets acquired from different time periods and angles. The tunnel surface images covered different areas of the tunnel, including the tunnel top, walls, and stress concentration areas at tunnel exits and turns. The implementers input the data sets preprocessed by the system into the generative adversarial network model for further analysis.

[0144] On July 15, 2023, the system detected for the first time a tiny crack at the northern entrance of the tunnel, with an initial width of only 0.12 mm and a length of about 2 cm. Traditional manual detection failed to detect the crack in time because it was in a visual blind spot and the size of the crack was extremely small. However, the generative adversarial network model accurately identified the existence of the crack through high-precision image processing and crack expansion prediction functions, and generated a simulated image of the crack's expansion trend.

[0145] According to the system's real-time prediction model, simulation data shows that the cracks may expand to 0.25 mm wide and 5 cm long in the next three months, especially in the summer under high temperature and high humidity conditions, the expansion rate will be accelerated. The system also generates an early warning report, which records in detail the location information of the cracks (northern entrance, 150 meters from the tunnel entrance, 2 meters from the top of the tunnel), the initial width and length of the cracks, and the predicted expansion rate. Based on the prediction results, the system automatically classifies the cracks as medium-risk cracks (risk level 2) and sends them to tunnel maintenance managers through real-time early warning signals.

[0146] On August 20, 2023, a new crack appeared in the middle area of ​​the tunnel. The system detected that the crack was 0.18 mm wide and 1.8 cm long at the beginning, located in the area of ​​the tunnel wall close to the ground. The system predicted through historical data and crack expansion models that the crack may expand in a short period of time due to the large traffic volume and stress concentration in the area. The system predicts that the width of this crack may increase to 0.3 mm in the next month and expand to 6 cm in length after 3 months. The system generated a risk assessment report based on the crack expansion trend and historical data, marking the crack as a high-risk crack (risk level 3). Based on the report generated by the system, the management immediately took reinforcement measures in the area to prevent further expansion of the crack.

[0147] In addition, during the entire 6-month monitoring period, the system automatically detected more than 50 cracks of varying degrees. Through the generative adversarial network model, the system can not only detect tiny cracks, but also predict the expansion trend of cracks in a specific time period in the future. Specifically, in a data analysis, the system predicted the cracks at the turn on the west side of the tunnel. The initial width of the cracks was 0.22 mm. The system predicted through the generative adversarial network model that the width of the cracks may increase to 0.6 mm in the next 5 months. The subsequent test results in January 2024 verified the prediction. The actual expansion of the cracks was almost exactly the same as the system's prediction, with a final width of 0.58 mm.

[0148] In order to verify the difference between the present invention and the traditional method, the implementers compared the method of the present invention with the traditional manual detection and image processing methods. The traditional method can only detect larger cracks in the test tunnel, with an average crack detection rate of 80% for cracks with a width exceeding 0.3 mm, and a detection rate of only 35% for tiny cracks (less than 0.2 mm). In contrast, the detection system of the present invention can detect cracks with a width as low as 0.1 mm, and the detection accuracy of tiny cracks reaches 94%.

[0149] In addition, in terms of expansion trend prediction, the traditional method has a large error in predicting crack expansion, especially in the stress concentration area. The error of the traditional method is about ±0.1 mm, while the error of the method of the present invention is controlled within ±0.03 mm. Specifically, when the system predicts the expansion of a crack in the middle of the tunnel, the traditional method predicts that the crack will expand from 0.3 mm to 0.8 mm in 3 months, but the actual expansion value is 0.9 mm, with an error of 0.1 mm; the system of the present invention predicts that the crack will expand to 0.88 mm under the same conditions, with an error of only 0.02 mm.

[0150] In terms of automation and real-time performance, the system of the present invention demonstrates extremely high efficiency. The crack detection and prediction tasks for the entire tunnel section can be completed in only 2 hours a day, and a complete risk assessment report can be generated within 30 minutes after the detection is completed. The traditional method takes about 10 hours for each detection, and data analysis and report generation require 2-3 days.

[0151] In summary, the present invention not only improves the accuracy of tunnel crack detection through the introduction of generative adversarial networks, especially shows extremely high sensitivity in the early detection of tiny cracks, but also has extremely high accuracy in predicting crack expansion trends. The system can automatically generate crack expansion simulation and risk assessment reports in real time, providing tunnel managers with timely maintenance recommendations, significantly improving monitoring efficiency and safety. The application of this technology greatly reduces the time and cost of manual inspection, while avoiding safety accidents caused by crack expansion.

[0152] The present invention adopts a generative adversarial network model, and realizes high-precision detection of tiny cracks in tunnels through the collaborative work of a generator and a discriminator. Compared with traditional image processing algorithms, the generative adversarial network uses a crack extension image data set to generate simulated images of cracks. In the early stage of cracks, the generative adversarial network model can capture the details of tiny cracks that are difficult to identify by traditional methods, thereby improving the ability to detect tiny defects. Experiments show that the method of the present invention can improve the accuracy and robustness of tiny crack detection in complex environments, effectively reduce errors caused by environmental noise and illumination changes, and the accuracy of crack identification is improved by more than 15% compared with traditional methods.

[0153] The present invention constructs a crack evolution model, simulates the crack expansion process under different stress environments based on the physical properties of tunnel cracks and real-time detection data, and combines the crack expansion simulation images generated by the generative adversarial network. The present invention can effectively predict the crack expansion trend at future time points. Different from the traditional static detection method, the present invention introduces dynamic stress parameters and time series analysis to simulate and predict the crack expansion speed, direction and acceleration, provides a dynamic crack expansion path, significantly improves the accuracy and timeliness of crack expansion prediction, and reduces the error by about 10%.

[0154] The present invention designs an automated crack risk assessment system, combines the spatiotemporal crack data set detected in real time, and the system automatically extracts the geometric features and expansion characteristics of the cracks, and classifies the cracks using a threshold-based risk level classification algorithm. The system can locate the crack position in real time under different environments, identify the risk level of crack expansion, and generate detailed crack expansion trends and risk assessment reports, which significantly reduces the need for manual intervention, realizes real-time early warning and risk assessment, improves the automation level of monitoring, increases detection efficiency by more than 30%, and greatly reduces labor costs and error rates.

[0155] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for detecting small defects in tunnel cracks based on a generative adversarial network, characterized in that: The steps include: S1. Perform multi-time series scanning of the tunnel surface by a camera or laser scanner to obtain a spatiotemporal crack dataset containing information on micro-cracks on the tunnel surface; S2. performing denoising, contrast enhancement, edge detection and image segmentation operations on the collected spatiotemporal crack dataset to generate a preprocessed spatiotemporal crack dataset; S3. Based on the physical expansion characteristics of actual cracks in the tunnel, a crack evolution model is constructed using the preprocessed spatiotemporal crack dataset. The crack evolution model simulates the crack expansion process under different time and stress environments and generates a crack expansion image dataset; S4. Design a generative adversarial network model, including a generator and a discriminator; S5. Input the preprocessed spatiotemporal crack dataset into the generative adversarial network model. The generator generates synthetic images of tiny cracks and simulated images of their expansion trends. The discriminator compares the generated crack images with the real images, and optimizes the performance of the generative adversarial network model in crack detection and crack expansion trend prediction through multiple rounds of training iterations. S6. Deploy the trained generative adversarial network model to the tunnel monitoring system; S7. Based on the crack extension simulation image dataset output by the generative adversarial network, the system locates tiny cracks and classifies them according to their morphology, size and extension trend, and divides the cracks into different risk levels; S8. Based on the crack detection results and their expansion trends, the system generates an early warning report including the crack location, size, severity and expansion trend.

2. According to claim 1, a method for detecting micro defects of tunnel cracks based on a generative adversarial network is characterized in that: The S1 specifically includes the following contents: S11. Use the camera to scan the tunnel surface multiple times in time series, obtain tunnel surface images at different time points, and generate a multi-time series tunnel surface spatiotemporal crack data set: in, Indicates that at time point t i (i=1,2,…,n) acquired tunnel surface image; S12. For each time point t i , by performing multi-angle scanning at different locations of the tunnel, a multi-view tunnel surface spatiotemporal crack dataset is generated: in, Indicates that at time point t i The tunnel surface image obtained from the j-th viewing angle, m is the number of viewing angles; S13. Based on the tunnel surface images acquired at different time points and viewing angles, a spatiotemporal crack dataset of tunnel crack micro-defects containing temporal and spatial information is constructed: in, Indicates that at time point t i The tunnel surface image containing tiny cracks obtained from the j-th viewing angle; S14. For each tunnel surface image Extract crack features and mark crack locations (x crack ,y crack ), where x crack and crack The crack images and their coordinate positions under the same viewing angle are combined to construct a spatiotemporal crack dataset in, Indicates that at time point t i and viewing angle v j The image matrix obtained below is For images The crack feature set extracted from represents the kth feature of the crack, including the length, width, depth and shape of the crack, P crack (t i ,v j )=(x crack ,y crack ) is the position coordinate of the crack in the image.

3. The method for detecting micro defects of tunnel cracks based on a generative adversarial network according to claim 1, characterized in that: The S2 specifically includes the following contents: S21. Use the non-local mean filtering algorithm to collect the spatiotemporal crack data set Perform denoising on each tunnel surface image The noise in the image is removed to generate a denoised tunnel surface image. S22. The denoised tunnel surface image ( Perform contrast enhancement processing to enhance the contrast between the crack area and the background area, and generate a contrast-enhanced tunnel surface image S23. Using Canny edge detection algorithm to detect the contrast of tunnel surface image Perform edge detection, extract edge information of cracks, and generate tunnel surface images including crack edges S24. Tunnel surface image after edge detection Perform tunnel surface image segmentation processing, use the tunnel surface image segmentation method based on region growing algorithm to separate the crack area from the background area, and generate a tunnel surface image including the crack area S25. The final spatiotemporal crack dataset after denoising, contrast enhancement, edge detection and tunnel surface image segmentation is obtained:

4. The method for detecting micro defects of tunnel cracks based on a generative adversarial network according to claim 1, characterized in that: The S3 specifically includes the following contents: S31. Based on the preprocessed spatiotemporal crack dataset D seg Combined with the physical expansion characteristics of actual tunnel cracks, a set of crack expansion feature vectors is constructed Each feature vector Contains physical parameters of crack length, width, depth and shape; S32. Establish a crack evolution model under different time points and stress environments based on the crack extension feature vector set: Among them, M crack (t,σ) is the crack evolution model, which represents the crack expansion behavior under time t and stress environment σ; F crack (t) is the crack characteristic vector at time t; α i is the weighting factor of the specific physical characteristics of the crack on the propagation process; is the influencing factor of the i-th crack characteristics under time i and stress environment σ; β i is the attenuation factor of stress on crack expansion, which indicates the changing trend of the influence of crack expansion under stress environment over time; S(t,σ) is the function of stress changing with time and environment, from t0 to t n The time frame for crack expansion; S33. According to the crack evolution model M crack (t,σ), simulate the crack expansion process under different stress conditions, and adjust the crack expansion trend according to the actual stress environment parameters to generate a crack expansion prediction image at a future time point: Among them, I forecast (t,σ) is the predicted image of crack expansion at the future time point t and stress environment σ; γ is the adjustment parameter of the crack expansion trend, which controls the influence of the expansion speed; δ i is the adjustment factor of the characteristic expansion rate of the i-th crack; It represents the rate of change of the crack evolution model over time, reflecting the crack expansion speed over time; represents the rate at which the i-th crack characteristic changes with time; Indicates the effect of stress changes on crack growth; S34. Apply the fracture evolution model to each fracture image in the spatiotemporal fracture dataset The crack evolution model is used to simulate and generate a crack extension image dataset at future time points:

5. The method for detecting micro defects of tunnel cracks based on a generative adversarial network according to claim 1, characterized in that: The S4 specifically includes the following contents: S41. Design a generative adversarial network model, including a generator G and a discriminator D. The generator G is based on the crack evolution model M. crack (t,σ) generates a synthetic crack image. The discriminator D is used to distinguish between the real crack image and the generated image and verify the rationality of the generated crack extension simulation; S42. Training input of generator G: crack extension image dataset D forecast , the generator uses the crack evolution model M crack (t,σ) Generate the crack extension simulation image I of the current crack gen (t,σ) and crack extension simulation image I sim (t,σ): I gen (t,σ)=G(M crack (t,σ),z); Among them, G is the generator, z is the random noise vector used to generate the random characteristics of the crack, and I gen (t,σ) is the generated crack extension simulation image; S43. The input of the discriminator D is the crack extension simulation image I generated by the generator gen (t,σ) and the true crack image I real (t,σ), the discriminator determines the authenticity of the image by comparing the generated image and the real image, and optimizes the performance of the generator and the discriminator through the loss function: Among them, L D is the loss function of the discriminator, is the time change rate of each crack characteristic, reflecting the dynamic change of crack extension; is the time rate of change of the generated image, reflecting the adjustment of the crack propagation speed in the generated image; S44. Through the feedback of the discriminator D, the generator G optimizes the generation ability, generates a crack extension simulation image that is highly similar to the real crack image, and generates an optimized crack extension simulation image I sim (t,σ), the discriminator verifies the rationality of the generated crack extension simulation image by comparing the optimized crack extension simulation image I sim The consistency between (t,σ) and the actual crack extension trend confirms that the generator simulates an image that is consistent with the actual crack extension, and generates an optimized crack extension simulation image dataset D sim .

6. The method for detecting micro defects of tunnel cracks based on a generative adversarial network according to claim 1, characterized in that: The S7 specifically includes the following contents: S71. Crack extension simulation image dataset D based on the output of generative adversarial network sim , the system locates the starting point and end point of each crack, extracts the geometric features of the crack, and defines the crack morphology feature vector as: F geo ={L crack ,W crack ,D crack ,θ crack }; Among them, L crack is the crack length, W crack is the crack width, D crack is the crack depth, θ crack is the crack expansion angle; S72. By analyzing the expansion trend of cracks in real-time spatiotemporal crack data, the system extracts crack expansion characteristics and defines the crack expansion feature vector as: F ext ={v crack ,a crack ,t ext }; Among them, v crack is the crack expansion speed, a crack is the crack extension acceleration, t ext The duration of crack expansion; S73. Combined with crack morphology feature vector F geo and the extended eigenvector F ext , the cracks are classified according to the risk level through the threshold-based classification algorithm, and the risk level is defined as R crack , the classification rule is expressed as: Among them, v th1 ,v th2 is the threshold of crack extension rate, a th1 ,a th2 Threshold of crack growth acceleration, L th1 ,L th2 is the threshold of crack length; S74. Based on the above classification results, the system classifies the cracks into different risk levels R crack To mark: When R crack =1, the crack expansion speed, acceleration and length are all within the low threshold range, indicating that the crack expansion trend is relatively slow and the structural risk is low; When R crack =2, the crack expansion speed, acceleration or length exceeds the lower threshold but does not reach the higher threshold, indicating that the crack has a moderate risk and the expansion trend may intensify, requiring timely monitoring; When R crack =3, the crack expansion speed, acceleration or length reaches or exceeds the high threshold, indicating that the crack expansion trend is significant and may cause serious damage to the structure, and immediate measures need to be taken; A risk assessment report is generated based on the shape, size and expansion trend of the cracks. The report includes the crack location information, size, shape, expansion speed, acceleration and risk level.

Images