Tunnel lining disease detection system and method based on deep learning

Abstract

The present invention discloses a tunnel lining disease detection system and method based on deep learning, which relates to the technical field of tunnel lining disease detection, including collecting environmental data and preprocessing it; extracting features from the preprocessed data through a convolutional neural network, integrating multimodal features using an attention mechanism, and constructing a comprehensive disease feature vector; using the comprehensive disease feature vector to train a deep learning model, and adjusting model parameters through a domain adaptation strategy; performing real-time disease detection and diagnosis through the adjusted model parameters; and formulating an intelligent maintenance strategy based on disease prediction and analysis. The invention realizes accurate detection, trend prediction, and dynamic maintenance of tunnel diseases by integrating multimodal acquisition, greatly improving management efficiency and safety, accurately capturing disease features through deep neural networks and data preprocessing, and combining domain adaptation technology to make the model more widely used, and intelligent strategies to prevent disease deterioration and optimize resource utilization.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel lining defect detection, and in particular to a tunnel lining defect detection system and method based on deep learning. Background Art

[0002] In recent years, with the rapid development of artificial intelligence technology, especially breakthroughs in the field of deep learning, tunnel lining disease detection methods based on deep learning have gradually become a research hotspot, providing new solutions for tunnel maintenance. Traditional tunnel disease detection mainly relies on manual inspections and simple mechanical auxiliary equipment. This method is not only time-consuming and labor-intensive, but also the detection accuracy is limited by personnel experience and environmental conditions. With the advancement of sensor technology and data analysis technology, the integration and analysis of multimodal data has become possible, which provides strong support for the early detection and precise positioning of tunnel diseases.

[0003] Although deep learning-based methods have shown great potential in tunnel defect detection, existing technologies still have some significant limitations. First, most methods focus on the analysis of single modal data and ignore the importance of other modal information, which limits the comprehensiveness and accuracy of defect detection. Second, the training of deep learning models usually relies on a large amount of labeled data, while in actual environments, the generalization ability of the model is limited under different geological conditions and operating environments. In addition, existing technologies lack effective adaptive mechanisms when dealing with dynamically changing environmental conditions, making it difficult to adjust model parameters in real time to meet the needs of defect detection in different scenarios. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a tunnel lining disease detection system and method based on deep learning to solve the problems of insufficient utilization of single modality data, weak model generalization ability and poor adaptability to dynamic environments in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, an embodiment of the present invention provides a tunnel lining disease detection method based on deep learning, which includes collecting environmental data and preprocessing it; extracting features from the preprocessed data through a convolutional neural network, integrating multimodal features using an attention mechanism, and constructing a comprehensive disease feature vector; using the comprehensive disease feature vector to train a deep learning model, and adjusting model parameters through a domain adaptation strategy; performing real-time disease detection and diagnosis through the adjusted model parameters; analyzing and predicting disease development trends based on disease diagnosis results; and formulating intelligent maintenance strategies based on disease prediction analysis.

[0008] As a preferred solution of the tunnel lining disease detection method based on deep learning of the present invention, the specific steps of collecting environmental data and preprocessing are as follows:

[0009] Through multiple types of sensors, image data, acoustic signals, vibration data and temperature distribution information in the tunnel environment are collected;

[0010] Median filtering is used to denoise image data;

[0011] Apply a bandpass filter to remove background noise;

[0012] Perform baseline correction on vibration data to remove environmental background vibration;

[0013] Implement temperature compensation to correct measurement deviations under environmental influences

[0014] Integrate the preprocessed environmental data into a multimodal dataset D;

[0015] As a preferred solution of the tunnel lining disease detection method based on deep learning described in the present invention, wherein: the pre-processed data is subjected to feature extraction through a convolutional neural network, multi-modal features are integrated using an attention mechanism, and a comprehensive disease feature vector is constructed. The specific steps are as follows:

[0016] Use the convolutional layer and pooling layer of the convolutional neural network to extract feature vectors from the environmental data in the tunnel, including image feature vector I, voiceprint signal feature vector A, vibration data feature vector V, and temperature distribution feature vector T;

[0017] Using the attention mechanism, the attention weight of the feature vector of each modality is calculated, and the expression is: ; Where B is a learnable vector for calculating attention weights, W i is the learnable weight matrix applied to the fully connected layer for each modality feature vector, b is the bias term, and α i is the attention weight, i is the specific modality for which the attention weight is being calculated, and j is one of the four modalities: image, acoustic signal, vibration data, and temperature distribution;

[0018] Through the attention weight, the feature vectors of each modality are weighted and summed to construct a comprehensive disease feature vector, which is expressed as: ; Where F is the comprehensive disease feature vector.

[0019] As a preferred solution of the tunnel lining disease detection method based on deep learning described in the present invention, wherein: the comprehensive disease feature vector is used to train the deep learning model, and the model parameters are adjusted through the domain adaptation strategy. The specific steps are as follows:

[0020] Divide the dataset D in the source tunnel environment into training set D t and validation set D V ;

[0021] Using cross entropy loss L CE Measures the difference between the model predictions and the actual disease;

[0022] Through the domain adaptation loss L DA , minimize the difference between the feature distribution of the source environment and the target environment, use MMD, calculate the distance between the kernel feature maps of the two distributions, and measure the similarity of the feature distribution of the source environment and the target environment. The expression is: Among them, F s Represents the comprehensive disease feature vector set of the source environment, F t Represents the comprehensive disease feature vector set of the target environment, L DA represents the domain adaptation loss, θ is the model parameter;

[0023] Combining the cross entropy loss of disease classification and the domain adaptation loss, a composite loss function L is defined to alternately minimize the disease classification loss and the domain adaptation loss to guide the learning and optimization of the model parameter θ. The expression is: Among them, L CE represents the cross entropy loss, λ is L CE and L DA Hyperparameters of relative importance, minL(θ) represents the minimum value of the minimized loss function L(θ) with respect to the model parameters θ;

[0024] At the end of each training, use the validation set D V Evaluate the model's loss and generalization ability on the validation set, and adjust the model parameter θ based on the model's loss and generalization ability. The expression is: ; where K is the disease state corresponding to F, f(F;θ) is the predicted output of the deep learning model for the input feature vector F, and θ * is the adjusted model parameter, J(θ * ;D V ) is the validation set D V The average loss of the above models.

[0025] As a preferred solution of the tunnel lining disease detection method based on deep learning described in the present invention, wherein: the real-time disease detection and diagnosis are performed through the adjusted model parameters, and the specific steps are as follows:

[0026] The adjusted model parameters θ * The data is transmitted back to the convolutional neural network via HTTPS for secondary feature extraction and weighted fusion to generate an updated comprehensive disease feature vector F. * ;

[0027] Based on the environmental data in the tunnel, a time-dependent and time-adaptive disease discrimination threshold η(t) is defined, combined with the updated comprehensive disease feature vector F * and the adjusted model parameters θ * , the probability of disease in the tunnel is diagnosed, and the expression is: ; Among them, P d represents the probability of disease existence, ψ is a nonlinear conversion function;

[0028] When Pd ≥ η (t), it means that the probability of the disease is beyond the safe range, and the system immediately generates an alarm and records the location, type and severity of the disease in detail;

[0029] When Pd<η(t), it means that the probability of disease existence is within a safe range, and the development of the disease should continue to be monitored.

[0030] As a preferred solution of the tunnel lining disease detection method based on deep learning of the present invention, the specific steps of analyzing and predicting the disease development trend based on the disease diagnosis results are as follows:

[0031] The probability of disease existence P based on real-time detection d Dynamic analysis of historical data, combined with periodic changes in the environment and the updated comprehensive disease feature vector F * , the development trend of the disease at the future time point t is predicted through comprehensive evaluation of nonlinear conversion functions, and the expression is: ; Among them, N d (t), is P at time t d Normalization factor, N l (t) is the lag P at time t d Normalization factor, N s (t) is the normalization factor for the periodic variation at time t, N F* (t) is the updated comprehensive disease feature vector F at time t * The normalization factor, N ε (t) is the normalization factor of the random disturbance term ε(t) at time t, P pd is the probability of disease development at a point in time.

[0032] As a preferred solution of the tunnel lining disease detection method based on deep learning described in the present invention, wherein: the intelligent maintenance strategy is formulated according to the disease prediction analysis, and the specific steps are as follows:

[0033] Prediction based on disease development trend pd (t), identify high-risk areas and time periods for disease development, and develop dynamic maintenance plans based on the high-risk areas and time periods for disease development;

[0034] Introduce the response function R(t), according to P pd The prediction results of (t) and the structural health status automatically adjust the intensity of maintenance activities, and the expression is: ; where H(t) represents the health index of the structure at time t, C(t) represents the cost constraint at time t, and γ1 is P pd (t) adjustment coefficient, γ2 is the adjustment coefficient of H(t), γ3 is the adjustment coefficient of C(t), ξ(t) represents the random error term at time t;

[0035] Define three key thresholds: the lower threshold R of the response function R(t) l , the middle threshold R of the response function R(t) m and the high threshold R of the response function R(t) h ;

[0036] When R(t)≤R l When the risk of disease development is at a controllable level, the structure is in good health, and regular maintenance plans are implemented. l <R(t)≤R h When R(t)>R h When the risk of disease development is extremely high and has threatened the overall safety of the structure, emergency maintenance actions should be implemented promptly.

[0037] In the second aspect, the present invention provides a tunnel lining disease detection system based on deep learning, including a data acquisition and preprocessing module, a feature fusion module, a disease detection and diagnosis module, a disease development trend prediction module and an intelligent maintenance strategy formulation module; the data acquisition and preprocessing module is used to collect tunnel environment data and preprocess the collected data; the feature fusion module is used to extract features from the preprocessed data and integrate multimodal features using an attention mechanism; the disease detection and diagnosis module is used for real-time disease detection and calculates the disease existence probability P d , and diagnose diseases based on the time-adaptive disease discrimination threshold η(t); the disease development trend prediction module is used to predict the disease development trend based on the disease existence probability P d , historical data, environmental periodic changes and comprehensive disease feature vector F * , predict the development trend of the disease at the future time point tP pd (t); Intelligent maintenance strategy formulation module is used to predict P based on disease development trend pd (t), formulate intelligent maintenance strategies.

[0038] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the deep learning-based tunnel lining disease detection method as described in the first aspect of the present invention is implemented.

[0039] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the tunnel lining disease detection method based on deep learning as described in the first aspect of the present invention is implemented.

[0040] The beneficial effects of the present invention are as follows: through multimodal data collection and preprocessing, convolutional neural networks and attention mechanisms are used to extract and integrate environmental features, and a deep learning model is constructed for disease detection and diagnosis. This method not only improves the accuracy of disease identification, but also enhances the generalization ability of the model through domain adaptation strategies. Through real-time monitoring and intelligent analysis, the system can predict the development trend of diseases, formulate dynamic maintenance strategies based on this, realize the optimal allocation of resources and the precise scheduling of maintenance activities, and ultimately achieve the purpose of extending the life of the tunnel, ensuring driving safety and reducing maintenance costs. This series of intelligent steps significantly improves the efficiency and intelligence level of tunnel structure health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0042] Figure 1 This is a flow chart of the tunnel lining disease detection method based on deep learning in Example 1.

[0043] Figure 2 This is a flow chart of the tunnel lining disease detection system based on deep learning in Example 1. DETAILED DESCRIPTION

[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0045] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0046] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0047] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a tunnel lining disease detection method based on deep learning, comprising the following steps:

[0048] S1. Collect environmental data and perform preprocessing.

[0049] Furthermore, multiple types of sensors (including high-definition cameras, sonar sensors, vibration sensors, and temperature sensors) are used to collect image data, acoustic signals, vibration data, and temperature distribution information in the tunnel environment;

[0050] Median filtering is used to denoise image data;

[0051] Apply a bandpass filter to remove background noise;

[0052] Perform baseline correction on vibration data to remove environmental background vibration;

[0053] Implement temperature compensation to correct measurement deviations under environmental influences

[0054] The preprocessed environmental data are integrated into a multimodal dataset D.

[0055] It should also be noted that the image data captured by the high-definition camera is denoised by median filtering to ensure image clarity and retention of defect details. The acoustic signal collected by the sonar sensor is filtered through a bandpass filter to filter irrelevant noise, which improves the purity of the signal and is conducive to detecting tiny abnormal sounds. The vibration data recorded by the vibration sensor is baseline corrected to eliminate the interference of environmental vibration and focus on the vibration mode directly related to the disease. The temperature distribution information measured by the temperature sensor can accurately reflect the state change of the lining material after temperature compensation, avoiding the influence of ambient temperature fluctuations on the detection results.

[0056] All these preprocessing steps work together to transform raw, potentially contaminated environmental data into a clean, clear, and targeted multimodal dataset D, creating favorable conditions for the subsequent training of deep learning models and the accurate extraction of disease features.

[0057] S2. Through the convolutional neural network, the preprocessed data is extracted, and the multimodal features are integrated using the attention mechanism to construct a comprehensive disease feature vector.

[0058] Furthermore, the convolutional layer and pooling layer of the convolutional neural network are used to extract feature vectors from the environmental data in the tunnel, including the image feature vector I, the voiceprint signal feature vector A, the vibration data feature vector V, and the temperature distribution feature vector T;

[0059] Using the attention mechanism, the attention weight of the feature vector of each modality is calculated, and the expression is: ; Where B is a learnable vector for calculating attention weights, W i is the learnable weight matrix applied to the fully connected layer for each modality feature vector, b is the bias term, and α i is the attention weight, i is the specific modality for which the attention weight is being calculated, and j is one of the four modalities: image, acoustic signal, vibration data, and temperature distribution;

[0060] Through the attention weight, the feature vectors of each modality are weighted and summed to construct a comprehensive disease feature vector, which is expressed as: ; Where F is the comprehensive disease feature vector.

[0061] It should also be noted that the learning parameters in the above formula will be optimized through the back-propagation algorithm during the training process to minimize the prediction error. This feature fusion method based on the attention mechanism allows the model to dynamically adjust the weights of different modalities according to the real-time characteristics of the data, which is impossible to achieve with traditional static fusion methods.

[0062] The comprehensive defect feature vector F can be regarded as a compact representation of all modal data, which contains key features from images, acoustic signals, vibration data and temperature distribution information. This vector is used as the input of subsequent classification or regression models for tunnel defect identification and severity assessment.

[0063] S3. Use the comprehensive disease feature vector to train the deep learning model and adjust the model parameters through the domain adaptation strategy.

[0064] Furthermore, the dataset D in the source tunnel environment is divided into the training set D t and validation set D V ;

[0065] Using cross entropy loss L CE Measuring the difference between the model prediction value and the actual disease, and promoting the model learning on the source environmental data;

[0066] Through the domain adaptation loss L DA , minimize the difference between the feature distribution of the source environment and the target environment, use MMD, calculate the distance between the kernel feature maps of the two distributions, and measure the similarity of the feature distribution of the source environment and the target environment. The expression is: Among them, F s Represents the comprehensive disease feature vector set of the source environment, F t Represents the comprehensive disease feature vector set of the target environment, L DA represents the domain adaptation loss, θ is the model parameter;

[0067] Cross entropy loss L CE The expression is: ; where N is the order of magnitude D t The number of samples in P i is the probability distribution predicted by the model for sample i, y i is the true label of sample i;

[0068] Combining the cross entropy loss of disease classification and the domain adaptation loss, a composite loss function L is defined to alternately minimize the disease classification loss and the domain adaptation loss to guide the learning and optimization of the model parameters θ to promote the generalization ability of the model in different environments. The expression is: Among them, L CE represents the cross entropy loss, λ is L CE and L DA Hyperparameters of relative importance, minL(θ) represents the minimum value of the minimized loss function L(θ) with respect to the model parameters θ;

[0069] At the end of each training, use the validation set D V Evaluate the model's loss and generalization ability on the validation set, and adjust the model parameter θ based on the model's loss and generalization ability. The expression is: ; where K is the disease state corresponding to F, f(F;θ) is the predicted output of the deep learning model for the input feature vector F, and θ * is the adjusted model parameter, J(θ * ;D V ) is the validation set D V The average loss of the above model;

[0070] The expression of f(F;θ) is: ; where b o is the bias of the output layer, b h is the bias of the hidden layer, W o is the output layer weight matrix, W h is the hidden layer weight matrix, σ is the activation function;

[0071] When evaluating the performance and generalization ability of the model, if J(θ * ; Dv) is high, which means that the model performs poorly on unseen data and may be overfitting. *; The value of Dv) is low, and the model performs well on the validation set, which indicates that the model has good generalization ability;

[0072] It should also be noted that MMD maximum mean difference is a method to measure the difference between two probability distributions. It is often used in fields such as domain adaptation, generative adversarial networks (GANs), and sample quality assessment in machine learning. MMD uses this technique to compare the difference between the expected values ​​of the feature maps of the two distributions, so it can capture the differences in distributions in high-dimensional space.

[0073] The specific process of data set division is as follows: First, the data set is cleaned and preprocessed to eliminate noise and inconsistency. Then, the inherent order of the data is broken through randomization, and then the data set is divided according to a preset ratio (such as the common 80%-20% or 70%-30%), with most of the data used as a training set for model learning, and a small part as a validation set for objective evaluation of model performance.

[0074] Cross entropy loss L CE The average cross entropy loss of all samples is calculated, which is a measure of the difference between the model's prediction and the actual standard. In machine learning, the smaller the loss value, the better, because it means that the model's prediction is closer to the actual value. During the training process, optimization algorithms (such as gradient descent) are used to adjust the model parameters θ to reduce this loss value.

[0075] The range of minL(θ) is the value of θ, which can theoretically range from close to zero (ideally, the model has perfect classification and the distribution of source and target environmental characteristics is consistent) to infinity (the model has many classification errors and the distribution of source and target environmental characteristics is completely mismatched). Smaller values ​​of θ indicate that the model performs well in disease detection and domain adaptation.

[0076] S4. Real-time disease detection and diagnosis are performed through the adjusted model parameters.

[0077] Furthermore, the adjusted model parameters θ * The data is transmitted back to the convolutional neural network via HTTPS for secondary feature extraction and weighted fusion to generate an updated comprehensive disease feature vector F. * ;

[0078] Based on the environmental data in the tunnel, a time-dependent and time-adaptive disease discrimination threshold η(t) is defined to distinguish between normal and abnormal states, and combined with the updated comprehensive disease feature vector F * and the adjusted model parameters θ * , the probability of disease in the tunnel is diagnosed, and the expression is: ; Among them, P drepresents the probability of disease existence, ψ is a nonlinear conversion function;

[0079] When Pd ≥ η (t), it means that the probability of the disease is beyond the safe range, and the system immediately generates an alarm and records the location, type and severity of the disease in detail;

[0080] When Pd<η(t), it means that the probability of disease existence is within a safe range, and the development of the disease should continue to be monitored.

[0081] It should also be noted that if the probability of disease occurrence is increasing over time, η(t) should be lowered to trigger the alarm earlier. Taking into account the impact of specific seasons (such as rainy season or winter) on disease incidence, η(t) needs to be adjusted during these periods. If certain environmental conditions (such as sudden changes in humidity or temperature) are observed to be significantly correlated with disease probability, then η(t) under these conditions should be more sensitive.

[0082] S5. Analyze and predict the development trend of the disease based on the disease diagnosis results.

[0083] Furthermore, the probability of disease existence P based on real-time detection d Dynamic analysis of historical data, combined with periodic changes in the environment and the updated comprehensive disease feature vector F * , the development trend of the disease at the future time point t is predicted through comprehensive evaluation of nonlinear conversion functions, and the expression is: ; Among them, N d (t), is P at time t d Normalization factor, N l (t) is the lag P at time t d Normalization factor, N s (t) is the normalization factor for the periodic variation at time t, N F* (t) is the updated comprehensive disease feature vector F at time t * The normalization factor, N ε (t) is the normalization factor of the random disturbance term ε(t) at time t, P pd is the probability of disease development at a point in time;

[0084] It should also be noted that when P pd When the value of (t) is close to 0, it means that the possibility of disease development is low and the system is in a low alert state. pd When the value of (t) is close to 1, it means that the possibility of disease development is high, and the system will trigger an alarm and immediate action is required;

[0085] The middle value indicates that the possibility of disease development is at a moderate level, and monitoring should be strengthened and emergency plans should be prepared.

[0086] This formula combines the current status of the disease, historical trends, cyclical changes, and real-time environmental characteristics to provide a comprehensive forecasting model that not only considers the immediate impact of the disease, but also incorporates long-term trends and seasonal factors. This comprehensive analysis helps to more accurately predict the development trend of the disease, thereby providing a scientific basis for decision makers to take timely prevention and control measures.

[0087] S6. Develop intelligent maintenance strategies based on disease prediction and analysis.

[0088] Furthermore, based on the prediction of disease development trend pd (t) Identify high-risk areas and time periods for disease development, and develop dynamic maintenance plans based on these high-risk areas and time periods. Maintenance plans are no longer static, but are adjusted as disease forecasts are updated to ensure that the schedule and resource allocation of maintenance activities accurately correspond to the predicted trends in disease development.

[0089] Introduce the response function R(t), according to P pd The prediction results of (t) and the structural health status automatically adjust the intensity of maintenance activities, and the expression is: ; where H(t) represents the health index of the structure at time t, C(t) represents the cost constraint at time t, and γ1 is P pd (t) adjustment coefficient, γ2 is the adjustment coefficient of H(t), γ3 is the adjustment coefficient of C(t), ξ(t) represents the random error term at time t;

[0090] The range of R(t) reflects the responsiveness of the maintenance strategy, ranging from low responsiveness (few maintenance activities) to high responsiveness (frequent and intensive maintenance activities);

[0091] Define three key thresholds: the lower threshold R of the response function R(t) l , the middle threshold R of the response function R(t) m and the high threshold R of the response function R(t) h ;

[0092] When R(t)≤R l When the risk of disease development is at a controllable level, the structure is in good health, and regular maintenance plans are implemented. l <R(t)≤R h When R(t)>R h When the risk of disease development is extremely high and has threatened the overall safety of the structure, emergency maintenance actions should be implemented promptly.

[0093] It should also be noted that the lower threshold R of the response function R(t) l, the middle threshold R of the response function R(t) m and the high threshold R of the response function R(t) h The setting needs to be based on historical data, structural characteristics and cost-benefit analysis to ensure the optimal allocation of resources while ensuring the safety of the structure. In this way, an accurate and responsive maintenance plan can be developed to effectively manage the risk of disease and extend the service life of the structure.

[0094] While giving priority to high-risk areas indicated by the forecast, ensure that maintenance work is carried out evenly throughout the structure. Through intelligent algorithms, the shortest path and the best maintenance sequence are calculated, allowing the maintenance team to cover as many critical areas as possible within a limited time while avoiding waste of resources;

[0095] This embodiment also provides a tunnel lining disease detection system based on deep learning, including: a data acquisition and preprocessing module, a feature fusion module, a disease detection and diagnosis module, a disease development trend prediction module and an intelligent maintenance strategy formulation module; the data acquisition and preprocessing module is used to collect tunnel environment data and preprocess the collected data; the feature fusion module is used to extract features from the preprocessed data and integrate multimodal features using an attention mechanism; the disease detection and diagnosis module is used for real-time disease detection and calculates the disease existence probability P d , and diagnose diseases based on the time-adaptive disease discrimination threshold η(t); the disease development trend prediction module is used to predict the disease development trend based on the disease existence probability P d , historical data, environmental periodic changes and comprehensive disease feature vector F * , predict the development trend of the disease at the future time point tP pd (t); Intelligent maintenance strategy formulation module is used to predict P based on disease development trend pd (t), formulate intelligent maintenance strategies.

[0096] The setting of these thresholds needs to be based on historical data, structural characteristics and cost-benefit analysis to ensure the optimal allocation of resources while ensuring the safety of the structure. In this way, accurate and responsive maintenance plans can be developed to effectively manage the risk of disease and extend the service life of the structure.

[0097] While giving priority to high-risk areas indicated by the forecast, ensure that maintenance work is carried out evenly throughout the structure. Through intelligent algorithms, the shortest path and the best maintenance sequence are calculated, allowing the maintenance team to cover as many critical areas as possible within a limited time while avoiding waste of resources;

[0098] This embodiment also provides a computer device, which is suitable for the case of a tunnel lining disease detection method based on deep learning, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the tunnel lining disease detection method based on deep learning proposed in the above embodiment.

[0099] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0100] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for detecting tunnel lining defects based on deep learning as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0101] In summary, the present invention significantly improves the intelligence and efficiency of tunnel disease management by integrating multimodal data acquisition, deep learning feature extraction and fusion, intelligent model training, real-time disease detection, disease trend prediction and intelligent maintenance strategy. Through sophisticated preprocessing and deep neural networks, the accurate capture of disease characteristics is ensured. Combined with domain adaptation and dynamic threshold adjustment, the wide applicability of the model and the immediate response to diseases are achieved. With the help of disease trend prediction and intelligent maintenance strategy, the deterioration of diseases is effectively prevented and maintenance resources are optimized. The overall solution greatly enhances the intelligence level of tunnel structure health management and ensures the safety and efficient operation of tunnels.

[0102] Example 2, referring to Table 1, is the second example of the present invention. In order to further verify the advancement of the present invention, experimental simulation data of a tunnel lining disease detection method based on deep learning is provided.

[0103] A 1000-meter-long tunnel was selected as the research object. The tunnel is located in an earthquake-prone area with a complex internal structure and various types of defects such as cracks, water seepage, and lining damage. The test subjects include a control group and three experimental groups, representing different implementations of the prior art and the present invention. The test process covers everything from data collection, preprocessing to feature extraction, model training, and even defect detection and intelligent maintenance strategies.

[0104] First, deploy a variety of sensors to continuously collect various types of data in the tunnel environment and pre-process the data.

[0105] Secondly, a convolutional neural network (CNN) is used to extract key features from the dataset D. Through the attention mechanism, the attention weight of each modal feature vector is calculated, and then the weighted sum is obtained to obtain the comprehensive disease feature vector F, which contains the key information of all modal data.

[0106] Next, the deep learning model is trained with the comprehensive disease feature vector F. The composite loss function L is defined by combining the two loss functions, and the model parameter θ is optimized by minimizing L.

[0107] Finally, the adjusted model parameters θ are used for secondary feature extraction and weighted fusion to generate an updated comprehensive disease feature vector F. Based on this feature vector and the disease discrimination threshold η(t), the diseases in the tunnel are diagnosed in real time, and a dynamic maintenance plan is formulated based on the prediction of disease development trends and intelligent adjustment of maintenance strategies.

[0108] Through the tunnel lining disease detection method of the present invention, a more intelligent and dynamic maintenance strategy can be formulated to automatically adjust the intensity of maintenance activities according to the development trend of the disease and the health status of the structure, which is difficult to achieve with traditional methods. This intelligent maintenance strategy can significantly improve maintenance efficiency, reduce unnecessary maintenance costs, and ensure the safe operation of the tunnel.

[0109] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A tunnel lining disease detection method based on deep learning, characterized by: include, Collect environmental data and pre-process it; The preprocessed data is subjected to feature extraction through convolutional neural networks, and the multimodal features are integrated using the attention mechanism to construct a comprehensive disease feature vector. Using comprehensive disease feature vectors, deep learning models are trained and model parameters are adjusted through domain adaptation strategies; Real-time disease detection and diagnosis are performed through adjusted model parameters; Analyze and predict the development trend of diseases based on disease diagnosis results; Develop intelligent maintenance strategies based on disease prediction analysis; The specific steps of collecting environmental data and preprocessing are as follows: Through multiple types of sensors, image data, acoustic signals, vibration data and temperature distribution information in the tunnel environment are collected; Median filtering is used to denoise image data; Apply a bandpass filter to remove background noise; Perform baseline correction on vibration data to remove environmental background vibration; Implement temperature compensation to correct measurement deviations under environmental influences; Integrate the preprocessed environmental data into a multimodal dataset D; The convolutional neural network is used to extract features from the preprocessed data, and the attention mechanism is used to integrate multimodal features to construct a comprehensive disease feature vector. The specific steps are as follows: Use the convolutional layer and pooling layer of the convolutional neural network to extract feature vectors from the environmental data in the tunnel, including image feature vector I, voiceprint signal feature vector A, vibration data feature vector V, and temperature distribution feature vector T; Using the attention mechanism, the attention weight of the feature vector of each modality is calculated, and the expression is: ； Among them, B is a learnable vector for calculating attention weights, W i is the learnable weight matrix applied to the fully connected layer for each modality feature vector, b is the bias term, and α i is the attention weight, i is the specific modality for which the attention weight is being calculated, j is one of the four modalities: image, acoustic signal, vibration data, and temperature distribution, and h i is the output of the feature vector of the i-th mode after passing through the activation function of the fully connected layer; Through the attention weight, the feature vectors of each modality are weighted and summed to construct a comprehensive disease feature vector, which is expressed as: ； Among them, F is the comprehensive disease feature vector; The comprehensive disease feature vector is used to train the deep learning model and the model parameters are adjusted through the domain adaptation strategy. The specific steps are as follows: Divide the dataset D in the source tunnel environment into training set D t and validation set D V ; Using cross entropy loss L CE Measures the difference between the model predictions and the actual disease; Through the domain adaptation loss L DA , minimize the difference between the feature distribution of the source environment and the target environment, use MMD, calculate the distance between the kernel feature maps of the two distributions, and measure the similarity of the feature distribution of the source environment and the target environment. The expression is: ； Among them, F s Represents the comprehensive disease feature vector set of the source environment, F t Represents the comprehensive disease feature vector set of the target environment, L DA represents the domain adaptation loss, θ is the model parameter; Combining the cross entropy loss of disease classification and the domain adaptation loss, a composite loss function L is defined to alternately minimize the disease classification loss and the domain adaptation loss to guide the learning and optimization of the model parameter θ. The expression is: ； Among them, L CE represents the cross entropy loss, λ is L CE and L DA Hyperparameters of relative importance, minL(θ) represents the minimum value of the minimized loss function L(θ) with respect to the model parameters θ; At the end of each training, use the validation set D V Evaluate the model's loss and generalization ability on the validation set, and adjust the model parameter θ based on the model's loss and generalization ability. The expression is: ； Where K is the disease state corresponding to F, f(F;θ) is the predicted output of the deep learning model for the input feature vector F, and θ * is the adjusted model parameter, J(θ * ;D V ) is the validation set D V The average loss of the above model; The real-time disease detection and diagnosis is performed by adjusting the model parameters, and the specific steps are as follows: The adjusted model parameters θ * The data is transmitted back to the convolutional neural network via HTTPS for secondary feature extraction and weighted fusion to generate an updated comprehensive disease feature vector F. * ; Based on the environmental data in the tunnel, a time-dependent and time-adaptive disease discrimination threshold η(t) is defined, combined with the updated comprehensive disease feature vector F * and the adjusted model parameters θ * , the probability of disease in the tunnel is diagnosed, and the expression is: ； Among them, P d represents the probability of disease existence, ψ is a nonlinear conversion function; When P d When ≥η(t), it means that the probability of the disease is beyond the safe range, and the system immediately generates an alarm and records the location, type and severity of the disease in detail; When P d <η(t), it means the probability of disease existence is within a safe range, and the disease development should continue to be monitored; The specific steps of analyzing and predicting the disease development trend based on the disease diagnosis results are as follows: The probability of disease existence P based on real-time detection d Dynamic analysis of historical data, combined with periodic changes in the environment and the updated comprehensive disease feature vector F * , the development trend of the disease at the future time point t is predicted through comprehensive evaluation of nonlinear conversion functions, and the expression is: ； Among them, N d (t), is P at time t d Normalization factor, N l (t) is the lag P at time t d Normalization factor, N s (t) is the normalization factor for the periodic variation at time t, N F* (t) is the updated comprehensive disease feature vector F at time t * The normalization factor, N ε (t) is the normalization factor of the random disturbance term ε(t) at time t, P pd is the probability of disease development at a point in time; According to the disease prediction analysis, the intelligent maintenance strategy is formulated, and the specific steps are as follows: Prediction based on disease development trend pd (t), identify high-risk areas and time periods for disease development, and develop dynamic maintenance plans based on the high-risk areas and time periods for disease development; Introduce the response function R(t), according to P pd The intensity of maintenance activities can be automatically adjusted based on the prediction results of (t) and the structural health status. The expression is: ； Where H(t) represents the health index of the structure at time t, C(t) represents the cost constraint at time t, and γ1 is P pd (t) adjustment coefficient, γ2 is the adjustment coefficient of H(t), γ3 is the adjustment coefficient of C(t), ξ(t) represents the random error term at time t; Define three key thresholds: the lower threshold R of the response function R(t) l , the middle threshold R of the response function R(t) m and the high threshold R of the response function R(t) h ; When R(t)≤R l When the risk of disease development is at a controllable level, the structure is in good health, and regular maintenance plans are implemented. l <R(t)≤R h When R(t)>R h When the risk of disease development is extremely high and has threatened the overall safety of the structure, emergency maintenance actions should be implemented promptly.

2. A tunnel lining disease detection system based on deep learning, based on the tunnel lining disease detection method based on deep learning according to claim 1, characterized in that: Including data collection and preprocessing module, feature fusion module, disease detection and diagnosis module, disease development trend prediction module and intelligent maintenance strategy formulation module; The data acquisition and preprocessing module is used to collect the environmental data in the tunnel and preprocess the collected data; The feature fusion module is used to extract features from the preprocessed data and integrate multimodal features using the attention mechanism; The disease detection and diagnosis module is used for real-time disease detection and calculation of the probability of disease existence P. d , and perform disease diagnosis based on the time-adaptive disease discrimination threshold η(t); The disease development trend prediction block is used to predict the disease development trend based on the disease existence probability P d , historical data, environmental periodic changes and comprehensive disease feature vector F * , predict the development trend of the disease at the future time point tP pd (t); Intelligent maintenance strategy formulation module is used to predict P based on disease development trend pd (t), formulate intelligent maintenance strategies.

3. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the tunnel lining disease detection method based on deep learning described in claim 1 are implemented.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the tunnel lining disease detection method based on deep learning described in claim 1 are implemented.

Images