Tunnel wall detection method, device, equipment and medium

Abstract

The invention relates to a tunnel wall detection method and device, equipment and a medium, and the method comprises the steps: obtaining a tunnel image, and extracting the image features of the tunnel image through a feature extraction layer of a detection model; the detection model comprises the feature extraction layer, a water leakage detection branch network, a crack detection branch network and a deformation detection branch network; respectively processing the image features according to the water leakage detection branch network, the crack detection branch network and the deformation detection branch network to obtain a water leakage detection result, a crack detection result and a deformation detection result; and determining the health state of the tunnel according to the water leakage detection result, the crack detection result and the deformation detection result. According to the technical scheme, the tunnel health state can be accurately evaluated, and the tunnel disease comprehensive detection capability is improved.

Description

Technical Field

[0001] This disclosure relates to the field of image processing technology, and in particular to a method, apparatus, equipment and medium for tunnel wall detection. Background Technology

[0002] In practical engineering, tunnel walls may suffer from erosion and damage, affecting the stability of the tunnel structure. Multiple defects can occur simultaneously and interact with each other, increasing the probability of safety accidents within the tunnel. Currently, shield tunnel wall inspection mainly relies on manual inspection and some relatively traditional detection techniques. Manual inspection typically involves workers using simple tools, such as flashlights and measuring tapes, to closely examine the tunnel walls and determine if there are defects such as water leakage, cracks, or deformation. While this method can detect obvious defects to some extent, it has many limitations. Traditional detection techniques, such as ultrasonic testing and infrared detection based on physical principles, can provide some detection data in certain aspects, but they also face many challenges.

[0003] Existing tunnel wall inspection methods have the following problems:

[0004] (1) Low inspection efficiency: Manual inspection requires staff to check the tunnel walls section by section and surface by surface, which is slow. Shield tunnels are long, and manual inspection is time-consuming and labor-intensive. In some long shield tunnels, completing a full inspection may take several days or even weeks, which is difficult to meet the needs of rapid inspection and affects the normal operation and maintenance schedule of the tunnel.

[0005] (2) Limited detection accuracy: Manual inspection mainly relies on visual observation and simple tool measurement, which can easily miss some small cracks, initial signs of water leakage, and areas that are not easy to observe directly. The experience and judgment standards of different staff members vary, making it difficult to guarantee the accuracy and consistency of the inspection results. Traditional inspection techniques are also unable to accurately detect the specific location, extent, and severity of defects when faced with complex tunnel environments and diverse types of defects.

[0006] (3) Difficulty in achieving comprehensive assessment: Tunnel wall defects such as water leakage, cracks, and deformation are often interconnected. However, most existing detection methods only target a single defect and cannot comprehensively assess the health status of the tunnel wall. Manual inspections are difficult to quantify multiple defects simultaneously, and traditional detection technologies lack the ability to fuse and process multi-source data, making it impossible to judge the overall health status of the tunnel wall. Summary of the Invention

[0007] To address the aforementioned technical problems, this disclosure provides a method, apparatus, equipment, and medium for tunnel wall detection.

[0008] In a first aspect, embodiments of this disclosure provide a tunnel wall detection method, including:

[0009] A tunnel image is acquired, and image features of the tunnel image are extracted through a feature extraction layer of a detection model; the detection model includes the feature extraction layer, a water seepage detection branch network, a crack detection branch network, and a deformation detection branch network.

[0010] The image features are processed according to the leakage detection branch network, crack detection branch network and deformation detection branch network respectively to obtain leakage detection results, crack detection results and deformation detection results;

[0011] The health status of the tunnel is determined based on the results of the water leakage detection, crack detection, and deformation detection.

[0012] The training process of the detection model includes:

[0013] Acquire tunnel sample images; the tunnel sample images include positive sample images and negative sample images, and the negative sample images include leakage water annotation information;

[0014] Based on the leakage detection branch network, leakage prediction information corresponding to the tunnel sample image is determined, so as to determine the first loss through the leakage prediction information and the leakage annotation information;

[0015] The step of determining the leakage prediction information corresponding to the tunnel sample image based on the leakage detection branch network includes:

[0016] The sample image features of the tunnel sample image are obtained, and a spatial attention map is generated through a convolutional layer;

[0017] The spatial attention map and the sample image features are weighted to obtain a weighted feature map;

[0018] Based on the weighted feature map, the leakage prediction information corresponding to the tunnel sample image is determined.

[0019] Secondly, embodiments of this disclosure provide a tunnel wall detection device, comprising:

[0020] An acquisition module is used to acquire tunnel images and extract image features of the tunnel images through a feature extraction layer of a detection model; the detection model includes the feature extraction layer, a water seepage detection branch network, a crack detection branch network, and a deformation detection branch network.

[0021] The detection module is used to process the image features according to the leakage detection branch network, crack detection branch network and deformation detection branch network respectively to obtain leakage detection results, crack detection results and deformation detection results;

[0022] The determination module is used to determine the health status of the tunnel based on the leakage detection results, crack detection results, and deformation detection results;

[0023] The training process of the detection model includes:

[0024] Acquire tunnel sample images; the tunnel sample images include positive sample images and negative sample images, and the negative sample images include leakage water annotation information;

[0025] Based on the leakage detection branch network, leakage prediction information corresponding to the tunnel sample image is determined, so as to determine the first loss through the leakage prediction information and the leakage annotation information;

[0026] The step of determining the leakage prediction information corresponding to the tunnel sample image based on the leakage detection branch network includes:

[0027] The sample image features of the tunnel sample image are obtained, and a spatial attention map is generated through a convolutional layer;

[0028] The spatial attention map and the sample image features are weighted to obtain a weighted feature map;

[0029] Based on the weighted feature map, the leakage prediction information corresponding to the tunnel sample image is determined.

[0030] Thirdly, embodiments of this disclosure provide an electronic device, including: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the tunnel wall detection method described in the first aspect.

[0031] Fourthly, embodiments of this disclosure provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the tunnel wall detection method described in the first aspect.

[0032] Compared with the prior art, the technical solution provided in this disclosure has the following advantages: the image features of the tunnel image are extracted by the feature extraction layer of the detection model, and the image features are processed according to the leakage detection branch network, crack detection branch network and deformation detection branch network respectively to obtain the leakage detection results, crack detection results and deformation detection results. Then, the health status of the tunnel is determined according to the leakage detection results, crack detection results and deformation detection results. Thus, the health status of the tunnel can be accurately assessed, and the risks of leakage, cracks and deformation of the tunnel wall can be accurately judged, meeting the needs of large-scale tunnel maintenance, improving the comprehensive detection capability of tunnel defects, and introducing an attention mechanism so that the leakage detection network focuses on the corners and joint areas, improving the feature extraction capability of the model in complex environments. Attached Figure Description

[0033] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0034] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a schematic flowchart of a tunnel wall detection method provided in an embodiment of the present disclosure;

[0036] Figure 2 This is a schematic flowchart of a model training method provided in an embodiment of the present disclosure;

[0037] Figure 3 This is a schematic diagram of the structure of a tunnel wall detection device provided in an embodiment of the present disclosure. Detailed Implementation

[0038] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0039] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0040] Figure 1This is a flowchart illustrating a tunnel wall detection method provided in an embodiment of the present disclosure. The method provided in this embodiment can be executed by a tunnel wall detection device, which can be implemented using software and / or hardware and can be integrated into any electronic device with computing capabilities.

[0041] like Figure 1 As shown, the tunnel wall detection method provided in this disclosure embodiment may include:

[0042] Step 101: Obtain the tunnel image and extract the image features of the tunnel image through the feature extraction layer of the detection model.

[0043] In this embodiment, the detection model includes a feature extraction layer, a water leakage detection branch network, a crack detection branch network, and a deformation detection branch network. The detection model uses an encoder-decoder architecture as its basic framework. The encoder extracts low-level features from the image, while the decoder performs task-specific processing for different detection tasks.

[0044] As an example, the encoder (with a shared underlying feature extraction layer) uses CNN (Convolutional Neural Networks) as the backbone network, including but not limited to ResNet. The feature extraction layer performs convolution, pooling, and other operations on the input tunnel image to gradually extract the abstract features of the image. The decoder includes a leakage detection branch network, a crack detection branch network, and a deformation detection branch network. The leakage detection branch network processes shared features through a series of convolutional and fully connected layers, outputting a classification result of leakage risk. This classification result includes, for example, no risk, risky but handled, and risky but not handled, as well as the location information of the leakage area. Optionally, the location information of the leakage area can be represented by a heatmap or a segmentation mask. The crack detection branch network is constructed using convolutional and fully connected layers to process shared features, outputting information such as the presence of cracks, crack type, and crack location and length. The crack detection branch network can use object detection algorithms (such as Faster R-CNN) to locate and identify cracks, or it can use semantic segmentation methods to accurately segment crack areas. The deformation detection branch network is used to detect the deformation of the tunnel wall. By analyzing shared features, it outputs information such as the degree of tunnel wall deformation and the location of deformed areas. The degree of tunnel wall deformation is represented, for example, by calculating the deformation amount or deformation ratio. The deformation detection branch network can be implemented using regression algorithms or pixel-level deformation analysis methods.

[0045] Step 102: Process the image features according to the leakage detection branch network, crack detection branch network and deformation detection branch network respectively to obtain the leakage detection result, crack detection result and deformation detection result.

[0046] In this embodiment, image features are processed according to the leakage detection branch network to obtain leakage detection results, image features are processed according to the crack detection branch network to obtain crack detection results, and image features are processed according to the deformation detection branch network to obtain deformation detection results.

[0047] The training process of the detection model includes: acquiring tunnel sample images; the tunnel sample images include positive sample images and negative sample images, and the negative sample images include at least one of leakage water annotation information, crack annotation information, and deformation annotation information. Based on the leakage water detection branch network, leakage water prediction information corresponding to the tunnel sample images is determined, and a first loss is determined using the leakage water prediction information and leakage water annotation information; based on the crack detection branch network, crack prediction information corresponding to the tunnel sample images is determined, and a second loss is determined using the crack prediction information and crack annotation information; based on the deformation detection branch network, deformation prediction information corresponding to the tunnel sample images is determined, and a third loss is determined using the deformation prediction information and deformation annotation information; based on the sum of the first loss, second loss, and third loss, a target loss is determined, and the detection model is trained using the target loss.

[0048] As an example, a multi-task loss function is constructed, where the target loss equals the sum of the first, second, and third losses. The first loss is the loss function for the water leakage risk detection task, which can be a weighted sum of classification and localization losses; for example, cross-entropy loss is used for classification, and mean squared error loss is used for localization. The second loss is the loss function for the crack detection task, and an appropriate loss function can be selected based on the specific detection method, such as the loss function for object detection or semantic segmentation. The third loss is the loss function for the deformation detection task; for example, mean squared error loss is used to regress the deformation amount. In this example, dataset construction includes: collecting a dataset of tunnel images containing different scenarios such as water leakage, cracks, and deformation; labeling each sample with information such as water leakage risk category, crack location and type, deformation area and degree, etc., ensuring the accuracy and consistency of the labeling. Data preprocessing: preprocessing operations are performed on the images, such as cropping, scaling, and normalization, to make the image data meet the model input requirements. Simultaneously, data augmentation techniques (such as random flipping, rotation, and adding noise) can be used to expand the dataset and improve the model's generalization ability. During training, stochastic gradient descent or its variants are used to minimize the multi-task loss function. The weights of each task's loss in the loss function are set according to the importance and difficulty of each task. The model's performance on each task is evaluated periodically, including metrics such as accuracy, recall, F1 score (for classification tasks), mean squared error (for regression tasks), and intersection-over-union ratio (for segmentation tasks). Training parameters, such as learning rate and weight decay, are adjusted based on the evaluation results to optimize model performance.

[0049] In one embodiment of this disclosure, obtaining tunnel sample images includes: generating tunnel wall images under different environmental conditions using image synthesis technology based on existing negative sample images, and adding them to a set of negative sample images; undersampling positive sample images and oversampling negative sample images to obtain tunnel sample images. In this embodiment, image synthesis technology is used to simulate tunnel wall images under different environmental conditions. Simultaneously, data balancing techniques such as oversampling and undersampling are employed. By synthesizing new samples to increase the amount of data in the risky category, oversampling is performed on the risky category with less data, and undersampling is performed on the risk-free category with more data, thus achieving a relatively balanced amount of data across categories.

[0050] Step 103: Determine the health status of the tunnel based on the results of water leakage detection, crack detection, and deformation detection.

[0051] In this embodiment, when the leakage detection result indicates a leakage area, the crack detection result indicates a crack area, and the deformation detection result indicates a deformation area, a high-precision color-coded label map can be generated using image segmentation technology to intuitively display the location of the abnormal area, providing intuitive positioning and warning.

[0052] As an example, a geographic information system (GIS) can be used to convert pixel coordinates into actual geographic coordinates, enabling precise positioning and helping maintenance personnel quickly locate areas requiring treatment. When a risky, untreated area is detected, an automatic warning signal can be issued, alerting maintenance personnel through various means such as SMS, email, and mobile app notifications. This real-time warning mechanism ensures that water leakage problems are addressed promptly, preventing further deterioration and guaranteeing the tunnel's safety and normal operation.

[0053] According to the technical solution of this disclosure, image features of the tunnel image are extracted through the feature extraction layer of the detection model. The image features are then processed according to the leakage detection branch network, crack detection branch network, and deformation detection branch network to obtain leakage detection results, crack detection results, and deformation detection results. Based on the leakage detection results, crack detection results, and deformation detection results, the health status of the tunnel is determined. Thus, the health status of the tunnel can be accurately assessed, and the risks of leakage, cracks, and deformation in the tunnel wall can be accurately judged, meeting the needs of large-scale tunnel maintenance and improving the comprehensive detection capability of tunnel defects.

[0054] Furthermore, in actual tunnel wall health assessment scenarios, complex environmental factors such as dust, dirt, uneven lighting, and partial obstruction within the tunnel can lead to misjudgments or omissions. For example, when the tunnel wall is covered in dust and the lighting is dim, it may obscure signs of water leakage, causing the model to misjudge a risky situation as risk-free. Alternatively, when temporary construction facilities partially obscure the tunnel wall, the model may fail to accurately identify areas of water leakage. The following section explains the water leakage detection branch network.

[0055] Based on the above embodiments, Figure 2 This is a flowchart illustrating a model training method provided in an embodiment of the present disclosure, as shown below. Figure 2 As shown, the method includes:

[0056] Step 201: Obtain the first tunnel sample image and the sensor sample data corresponding to the first tunnel sample image.

[0057] The first tunnel sample image includes images of tunnels with water leakage. Sensors include a humidity sensor and / or a water level sensor. The humidity sensor detects humidity changes in the tunnel wall or surrounding environment, using localized humidity increases to help identify leakage areas. The water level sensor monitors the water level height and rate of change inside or around the tunnel, indicating leakage or water accumulation risks through abnormal water levels. Multiple sensors can be installed at different locations within the tunnel. Camera calibration allows mapping based on sensor installation locations to determine the positional information of each sensor's data within the tunnel image. Sample image features of the first tunnel sample image are extracted. Then, through calibration steps such as time synchronization, spatial calibration, and data normalization, the sensor data and sample image features are fused to generate sample fusion features, which are used as network input for model training. The time synchronization includes ensuring that sensor data and image acquisition time are strictly aligned to avoid misjudgment due to time difference; spatial calibration includes mapping the sensor installation location to the image acquisition area (e.g., through coordinate mapping) to ensure that the data are spatially correlated; and data normalization includes standardizing humidity and water level data, for example, scaling them to the 0-1 range to facilitate fusion with image features.

[0058] During the inference process, the image features are processed according to the leakage detection branch network, including: acquiring sensor data collected by sensors; the sensors include humidity sensors and / or water level sensors; mapping according to the installation location of the sensors to determine the location information of each sensor data in the tunnel image; fusing each sensor data with image features according to the location information to generate fused features, and processing the fused features according to the leakage detection branch network.

[0059] Step 202: Generate a spatial attention map, and perform weighted prediction based on the spatial attention map and sample image features to obtain leakage prediction information.

[0060] In this embodiment, sample image features of tunnel sample images are obtained, and a spatial attention map is generated through a convolutional layer; the spatial attention map and sample image features are weighted to obtain a weighted feature map; and the leakage prediction information corresponding to the tunnel sample image is determined based on the weighted feature map.

[0061] The network architecture for the spatial attention mechanism is designed as follows: An encoder-decoder structure is adopted, with a spatial attention module introduced in the encoder stage to focus on corners and seams. The input to the spatial attention module is the feature map F∈R of a certain layer of the encoder. H×W×C (Height × Width × Number of Channels), the output is a weighted feature map. Where A∈R H×W×1 This is a spatial attention map. This represents element-wise multiplication. Feature compression: Global average pooling and max pooling are performed on the input feature map F along the channel dimension to obtain two spatial descriptors F1 and F2 respectively. avg =GlobalAveragePool(F)∈R H×W×1 and F max =GlobalMaxPool(F)∈R H×W×1 Both capture global and local salient features respectively. Feature concatenation and convolution: F avg and F max F is obtained by splicing along the channel dimension concat ∈R H×W×2 A spatial attention map A is generated through a convolutional layer (containing sigmoid activation), where A = σ(Conv 1×1 (F concat ))∈[0,1] H×W×1 , where Conv 1×1 A 1×1 convolution kernel is used to fuse channel information, and σ is the Sigmoid function used to normalize the weights to 0-1. Feature weighting: The attention map A is multiplied element-wise with the original feature map F to obtain the weighted feature map F′. To enhance the characteristic response of leakage-related areas (high weight) and suppress irrelevant areas (low weight).

[0062] Specifically, a spatial attention module is inserted after each residual block or downsampling layer of the encoder.

[0063] Step 203: Calculate the first loss to train the model using the first loss.

[0064] In this embodiment, determining the first loss using leakage prediction information and leakage annotation information includes: calculating cross-entropy loss using leakage prediction information and leakage annotation information; calculating attention loss using spatial attention maps and target regions in tunnel sample images; target regions include corner regions and joint regions; and determining the first loss based on cross-entropy loss and attention loss.

[0065] As an example, the loss function is as follows, and in the segmentation task, the attention mechanism is used to optimize the objective: L total =L CE (Y pred Y gt )+λ·L attn , where L CE Cross-entropy loss is used for segmenting leaky water areas, L attn For the attention-supervised loss, the mean squared error between the attention map and the labeled mask can be calculated, where λ is a hyperparameter used to balance the two losses. In this example, after training, the spatial attention map A is extracted and superimposed on the input image to verify whether the attention is focused on areas such as corners and seams.

[0066] In this embodiment, an attention mechanism is introduced, enabling the leakage detection network to focus on corners and joint areas, improving the model's feature extraction capabilities in complex environments. By fusing image features with other sensor data through multimodal data fusion, the characteristics of the leakage area can be captured more comprehensively. Image data provides visual information, while sensor data provides physical quantity information. The multimodal fusion network can complement these different modal features, allowing the model to more comprehensively understand the characteristics of the leakage area and reduce false positives and false negatives. Furthermore, the multimodal fusion network can effectively cope with interference factors in complex environments, enhancing robustness. For example, in situations with uneven lighting, stains, or reflections inside the tunnel, sensor data can provide reliable supplementary information, thereby enhancing the model's robustness. In other words, multimodal data provides data redundancy; even if data from one modality is abnormal or lost, data from other modalities can still provide sufficient information to support detection, ensuring stable system operation. Multimodal fusion networks can automate the detection of water leakage risks in shield tunnel walls, enabling automated and batch processing. In practical engineering, the amount of detection data for tunnel walls is enormous. Multimodal fusion networks can efficiently process this data, promptly detect and warn of water leakage risks, and meet the needs of large-scale tunnel maintenance. The fusion of multimodal data increases data diversity, allowing the model to learn richer features and improving its generalization ability.

[0067] Figure 3 This is a schematic diagram of the structure of a tunnel wall detection device provided in an embodiment of the present disclosure, as shown below. Figure 3As shown, the tunnel wall detection device includes: an acquisition module 31, a detection module 32, and a determination module 33.

[0068] The acquisition module 31 is used to acquire tunnel images and extract image features of the tunnel images through the feature extraction layer of the detection model; the detection model includes a feature extraction layer, a water seepage detection branch network, a crack detection branch network, and a deformation detection branch network.

[0069] The detection module 32 is used to process the image features according to the leakage detection branch network, the crack detection branch network and the deformation detection branch network respectively, so as to obtain the leakage detection result, the crack detection result and the deformation detection result.

[0070] The determination module 33 is used to determine the health status of the tunnel based on the results of water leakage detection, crack detection, and deformation detection.

[0071] The training process of the detection model includes:

[0072] Acquire tunnel sample images; tunnel sample images include positive sample images and negative sample images, and the negative sample images include leakage water annotation information;

[0073] Based on the leakage detection branch network, the leakage prediction information corresponding to the tunnel sample image is determined, so as to determine the first loss through the leakage prediction information and leakage annotation information;

[0074] Based on the leakage detection branch network, leakage prediction information corresponding to the tunnel sample images is determined, including:

[0075] The sample image features of the tunnel sample image are obtained, and a spatial attention map is generated through a convolutional layer;

[0076] The spatial attention map and sample image features are weighted to obtain a weighted feature map;

[0077] The leakage prediction information corresponding to the tunnel sample image is determined based on the weighted feature map.

[0078] In one embodiment of this disclosure, the negative sample image further includes crack annotation information and deformation annotation information, and the training process of the detection model further includes:

[0079] Based on the crack detection branch network, crack prediction information corresponding to the tunnel sample image is determined, so as to determine the second loss through crack prediction information and crack annotation information;

[0080] Based on the deformation detection branch network, the deformation prediction information corresponding to the tunnel sample image is determined, so as to determine the third loss through the deformation prediction information and deformation annotation information;

[0081] The target loss is determined based on the sum of the first loss, the second loss, and the third loss, and the detection model is trained using the target loss.

[0082] In one embodiment of this disclosure, the training process of the detection model is implemented by a training module, which is specifically used for:

[0083] Based on existing negative sample images, tunnel wall images under different environmental conditions are generated using image synthesis technology and added to the set of negative sample images;

[0084] The tunnel sample image is obtained by undersampling the positive sample image and oversampling the negative sample image.

[0085] In one embodiment of this disclosure, the training module is specifically used for:

[0086] The cross-entropy loss is calculated using leakage prediction information and leakage labeling information.

[0087] Attention loss is calculated using target regions in spatial attention maps and tunnel sample images; target regions include corner areas and seam areas.

[0088] The first loss is determined based on the cross-entropy loss and the attention loss.

[0089] In one embodiment of this disclosure, the detection module 32 is specifically used for:

[0090] Acquire sensor data collected by sensors; sensors include humidity sensors and / or water level sensors;

[0091] Mapping is performed based on the sensor's installation location to determine the location information of each sensor's data in the tunnel image;

[0092] The sensor data and image features are fused based on the location information to generate fused features, and the fused features are processed according to the leakage detection branch network.

[0093] In one embodiment of this disclosure, the device further includes:

[0094] The early warning module generates an early warning message if the health status indicates that there are untreated risk areas.

[0095] The tunnel wall detection device provided in this disclosure can execute any tunnel wall detection method provided in this disclosure, and has the corresponding functional modules and beneficial effects for executing the method. Content not described in detail in the device embodiments of this disclosure can be referred to the description in any method embodiment of this disclosure.

[0096] This disclosure also provides an electronic device including one or more processors and a memory. The processor may be a central processing unit (CPU) or other processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may execute the program instructions to implement the methods of the embodiments of this disclosure above and / or other desired functions. Various contents such as input signals, signal components, and noise components may also be stored in the computer-readable storage medium.

[0097] In one example, the electronic device may also include input and output devices, which are interconnected via a bus system and / or other forms of connection. Furthermore, the input device may include, for example, a keyboard, a mouse, etc. The output device can output various information to the outside, including determined distance information, direction information, etc. The output device may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc. In addition, depending on the specific application, the electronic device may include any other suitable components such as a bus, input / output interfaces, etc.

[0098] In addition to the methods and apparatus described above, embodiments of this disclosure may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform any of the methods provided in the embodiments of this disclosure.

[0099] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of embodiments of this disclosure. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0100] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer program instructions that, when executed by a processor, cause the processor to perform any of the methods provided in the embodiments of this disclosure.

[0101] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0102] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0103] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting tunnel walls, characterized in that, The method includes: A tunnel image is acquired, and image features of the tunnel image are extracted through a feature extraction layer of a detection model; the detection model includes the feature extraction layer, a water seepage detection branch network, a crack detection branch network, and a deformation detection branch network. The image features are processed according to the leakage detection branch network, crack detection branch network and deformation detection branch network respectively to obtain leakage detection results, crack detection results and deformation detection results; The health status of the tunnel is determined based on the results of the water leakage detection, crack detection, and deformation detection. The training process of the detection model includes: Acquire tunnel sample images; the tunnel sample images include positive sample images and negative sample images, and the negative sample images include leakage water annotation information; Based on the leakage detection branch network, leakage prediction information corresponding to the tunnel sample image is determined, so as to determine the first loss through the leakage prediction information and the leakage annotation information; The step of determining the leakage prediction information corresponding to the tunnel sample image based on the leakage detection branch network includes: The sample image features of the tunnel sample image are obtained, and a spatial attention map is generated through a convolutional layer; The spatial attention map and the sample image features are weighted to obtain a weighted feature map; Based on the weighted feature map, the leakage prediction information corresponding to the tunnel sample image is determined.

2. The method as described in claim 1, characterized in that, The negative sample image also includes crack annotation information and deformation annotation information, and the training process of the detection model also includes: Based on the crack detection branch network, crack prediction information corresponding to the tunnel sample image is determined, so as to determine the second loss through the crack prediction information and the crack annotation information; Based on the deformation detection branch network, the deformation prediction information corresponding to the tunnel sample image is determined, so as to determine the third loss through the deformation prediction information and the deformation annotation information; The target loss is determined based on the sum of the first loss, the second loss, and the third loss, and the detection model is trained using the target loss.

3. The method as described in claim 2, characterized in that, The acquisition of tunnel sample images includes: Based on existing negative sample images, tunnel wall images under different environmental conditions are generated using image synthesis technology and added to the set of negative sample images; The tunnel sample image is obtained by undersampling the positive sample image and oversampling the negative sample image.

4. The method as described in claim 2, characterized in that, The determination of the first loss through the leakage prediction information and the leakage labeling information includes: The cross-entropy loss is calculated using the leakage prediction information and the leakage labeling information. Attention loss is calculated using the spatial attention map and the target regions in the tunnel sample image; the target regions include corner areas and seam areas. The first loss is determined based on the cross-entropy loss and the attention loss.

5. The method as described in claim 1, characterized in that, The image features are processed according to the leakage detection branch network, including: Acquire sensor data collected by the sensors; the sensors include a humidity sensor and / or a water level sensor; The location information of each sensor data under the tunnel image is determined by mapping the sensor's installation location. The sensor data and image features are fused together based on the location information to generate fused features, and the fused features are processed according to the leakage detection branch network.

6. The method as described in claim 1, characterized in that, After determining the health status of the tunnel, the following is also included: If the health status indicator shows an unresolved risk area, a warning message will be generated.

7. A tunnel wall detection device, characterized in that, include: An acquisition module is used to acquire tunnel images and extract image features of the tunnel images through a feature extraction layer of a detection model; the detection model includes the feature extraction layer, a water seepage detection branch network, a crack detection branch network, and a deformation detection branch network. The detection module is used to process the image features according to the leakage detection branch network, crack detection branch network and deformation detection branch network respectively to obtain leakage detection results, crack detection results and deformation detection results; The determination module is used to determine the health status of the tunnel based on the leakage detection results, crack detection results, and deformation detection results; The training process of the detection model includes: Acquire tunnel sample images; the tunnel sample images include positive sample images and negative sample images, and the negative sample images include leakage water annotation information; Based on the leakage detection branch network, leakage prediction information corresponding to the tunnel sample image is determined, so as to determine the first loss through the leakage prediction information and the leakage annotation information; The step of determining the leakage prediction information corresponding to the tunnel sample image based on the leakage detection branch network includes: The sample image features of the tunnel sample image are obtained, and a spatial attention map is generated through a convolutional layer; The spatial attention map and the sample image features are weighted to obtain a weighted feature map; Based on the weighted feature map, the leakage prediction information corresponding to the tunnel sample image is determined.

8. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the tunnel wall detection method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the tunnel wall detection method according to any one of claims 1-6.

Images