Dynamic detection method, device, equipment and storage medium for cracks in rail transit structures

Abstract

The embodiments of the present application disclose a method, device, equipment and storage medium for dynamic detection of structural cracks in rail transit. The method comprises: acquiring key areas in rail transit; presetting a collection interval, and collecting multiple crack images of the key areas under different illuminations according to the collection interval to form an image database; performing feature extraction and similarity matching training on multiple crack images in the image database to obtain an image matching neural network, and performing recognition training on multiple crack images to obtain an image recognition neural network; collecting target crack images, and using the image matching neural network to match the target crack images with original crack images in the image database; identifying structural cracks in a set of crack images obtained by matching and outputting a set of binary images; extracting basic parameters of the structural cracks in the binary images and comparing the basic parameters with historical data to monitor the changing trend of the structural cracks.

Description

Technical Field

[0001] The present application relates to the field of machine vision detection technology, and specifically to a method, device, equipment and storage medium for dynamic detection of cracks in rail transit structures. Background Art

[0002] With the vigorous development of my country's urban rail transit industry, during the construction process, accompanied by the increasingly complex excavation environment and the increasingly deep excavation depth, accidents caused by the instability of the retaining structures of the construction projects occur frequently. Structural cracks are one of the effective indicators for judging the instability of the retaining structures.

[0003] Therefore, in order to ensure the safe construction and normal operation of urban rail transit, long-term monitoring of cracks is very necessary. However, in urban rail transit projects under construction, it is difficult to fully cover the monitoring of structural cracks. At the same time, traditional monitoring methods are often based on manual observation, lack of accurate and effective monitoring means, and difficult to operate on the spot. Summary of the invention

[0004] The embodiments of the present application provide a method, device, equipment and storage medium for dynamic detection of cracks in rail transit structures, which are used to improve the problem of low accuracy of detection data when detecting structural cracks in transportation buildings using human visual detection methods.

[0005] A first aspect of an embodiment of the present application provides a method for dynamic detection of cracks in a rail transit structure, the method comprising:

[0006] Obtain key areas in rail transit, and name and define the key areas according to scenarios;

[0007] Preset a collection interval time, and collect multiple crack images of the key area under different illuminations according to the collection interval time to form an image database;

[0008] Extracting features from a plurality of crack images in the image database, performing similarity matching training based on the extracted features of the plurality of crack images to obtain an image matching neural network, and performing recognition training based on the extracted features of the plurality of crack images to obtain an image recognition neural network;

[0009] Acquire a target crack image, and use the image matching neural network to match the target crack image with an original crack image in the image database that is shot at a different time but is of the same scene;

[0010] Identify structural cracks in a group of crack images belonging to the same scene obtained by matching and output a group of binary images;

[0011] The basic parameters of the structural cracks in the binary image are extracted and compared with the historical data to monitor the change trend of the structural cracks.

[0012] In some embodiments of the present application, before the step of acquiring the target crack image and using the image matching neural network to match the target crack image with the original crack image in the image database that was shot at different times but in the same scene, the step also includes: testing the image matching effect and the image recognition effect of the image matching neural network and the image recognition neural network under different illumination conditions, and obtaining the optimal illumination range required for acquiring the image based on the image matching effect and the image recognition effect.

[0013] In some embodiments of the present application, the step of testing the image matching effect and the image recognition effect of the image matching neural network and the image recognition neural network under different illumination conditions, and obtaining the optimal illumination range required for capturing the image according to the image matching effect and the image recognition effect, comprises:

[0014] Preset a test light illumination interval threshold, take multiple light illuminations within the test light illumination interval threshold range to respectively perform image acquisition and input the acquired images into the image matching neural network and the image recognition neural network;

[0015] Matching and identifying the collected image with the image in the image database, recording the image matching data of the image matching neural network and the image recognition data of the image recognition neural network respectively, and performing index evaluation on the image matching data and the image recognition data respectively;

[0016] According to the evaluation result, the matching accuracy illumination interval of the image matching neural network and the recognition accuracy illumination interval of the image recognition neural network are obtained;

[0017] The intersection of the matching accuracy illumination interval and the recognition accuracy illumination interval is taken to obtain an optimal illumination interval that satisfies the image matching neural network and the image recognition neural network.

[0018] In some embodiments of the present application, the training step of the image matching neural network includes:

[0019] Select the architecture of the network model and determine the depth and width of the network based on image data and requirements;

[0020] Building an initial image matching neural network, randomly dividing the images in the image database based on a preset ratio, wherein the preset ratio is training set: validation set: test set = 6:2:2;

[0021] Inputting the training set into the initial image matching neural network, performing similarity matching training according to the scene category, and obtaining the image matching neural network;

[0022] Inputting the verification set into the image matching neural network, and adjusting the hyperparameters of the image matching neural network according to the verification set to improve the performance of the image matching neural network;

[0023] The test set is used to quantitatively evaluate the matching effect of the image matching neural network.

[0024] In some embodiments of the present application, the step of setting the training data of the image recognition neural network includes:

[0025] Selecting representative crack images from the image database, and manually annotating cracks in the crack images at a pixel level, so that each crack image corresponds to a label image;

[0026] defining the crack image and a label image corresponding to the crack image as a set of image data;

[0027] Each set of the image data is enlarged to obtain an enlarged data set.

[0028] In some embodiments of the present application, the step of expanding the data set includes:

[0029] The crack image and the label image corresponding to the crack image are synchronously subjected to operations of rotation, flipping, scaling, cropping, adding noise and changing brightness to increase the data volume.

[0030] In some embodiments of the present application, the basic parameters of the structural cracks include one or more of crack spacing, opening, extension length, strike, dip and inclination.

[0031] In a second aspect, an embodiment of the present application provides a rail transit structure crack dynamic detection device, the dynamic detection device comprising:

[0032] An acquisition definition module is used to acquire key areas in rail transit and name and define the key areas according to scenarios;

[0033] An image interval acquisition module, used for presetting an acquisition interval time, and acquiring a plurality of crack images of the key area under different illuminations according to the acquisition interval time, so as to form an image database;

[0034] A model training module, used for extracting features from multiple crack images in an image database, performing similarity matching training based on the extracted features of multiple crack images to obtain an image matching neural network, and performing recognition training based on the extracted features of multiple crack images to obtain an image recognition neural network;

[0035] An image matching module, used for collecting a target crack image, and using the image matching neural network to match the target crack image with an original crack image in the image database that was shot at a different time but is of the same scene;

[0036] An image recognition module is used to identify structural cracks in a group of matched images belonging to the same scene and output a group of binary images;

[0037] The extraction and analysis module is used to extract basic parameters of the structural cracks in the binary image and compare the basic parameters with historical data to monitor the change trend of the structural cracks.

[0038] An electronic device provided in the third aspect of an embodiment of the present application includes: a processor, a storage medium and a bus, the storage medium stores machine-readable instructions executable by the processor, when the electronic device is running, the processor and the storage medium communicate through the bus, and the processor executes the machine-readable instructions to perform the steps of the method provided in the first aspect.

[0039] A fourth aspect of the embodiments of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method provided in the first aspect are executed.

[0040] It can be seen that the method of the embodiment of the present application is mainly to realize the dynamic detection of structural cracks by using image matching neural network combined with image recognition technology. In detail, first, compared with the inefficiency of traditional manual monitoring methods, it is limited by the safety of manual operations and the difficulty of execution in complex environments. The method provided by the present application can automatically collect crack images in key areas and form a database, so that the monitoring work is not limited by manpower, continuity and comprehensiveness are guaranteed, and the operation is safe and reliable. Second, in the background technology, due to factors such as complex site conditions and variable light conditions, there are large errors in matching and identifying artificial cracks. By using image matching and recognition neural networks, accurate crack features can be extracted, and by self-learning to adapt to changes in different working environments, the development of cracks can be more accurately identified and tracked, effectively avoiding the problem of large errors in matching and identifying artificial cracks due to factors such as complex site conditions and variable light conditions. Third, the present application can analyze the changing trend of structural cracks in real time, provide early warning signals in time, and help to take countermeasures as soon as possible to avoid potential safety accidents. Fourth, compared with traditional methods, the present application uses visual technology to greatly improve the efficiency of crack detection, which is particularly critical for public infrastructure such as urban rail transit that needs to be continuously operated. Since crack monitoring can be performed automatically during non-operating hours, it does not affect daily operations, thus optimizing the coordination between maintenance and operations. Fifth, the detection method of this application can accumulate rich crack development data, providing data support for the long-term maintenance and improvement of urban rail transit. Through in-depth analysis of historical data, the crack development trend can be predicted based on scientific data, and maintenance plans and budgets can be arranged more reasonably. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 This is a schematic diagram of an application scenario of the rail transit structure crack dynamic detection device provided in an embodiment of the present application;

[0043] Figure 2 It is a schematic diagram of the flow of a method for dynamic detection of cracks in rail transit structures provided in an embodiment of the present application;

[0044] Figure 3 It is a schematic diagram of obtaining crack binary images belonging to the same scene after image matching and image recognition provided by an embodiment of the present application, and extracting basic parameters of structural cracks in this set of binary images through codes and performing comparative analysis;

[0045] Figure 4 This is a schematic diagram of the illumination interval of the image matching neural network and the image recognition neural network provided in the embodiment of the present application;

[0046] Figure 5 It is a structural schematic diagram of a rail transit structure crack dynamic detection device provided in an embodiment of the present application;

[0047] Figure 6 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0049] The embodiments of the present application provide a method, device, equipment and storage medium for dynamic detection of cracks in rail transit structures, wherein the device may be an electronic device and the storage medium may be a computer-readable storage medium. The dynamic detection device for cracks in rail transit structures may be integrated in an electronic device, which may be a server or a terminal.

[0050] Among them, the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, network acceleration services (Content Delivery Network, CDN), as well as big data and artificial intelligence platforms.

[0051] The terminal may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the server may be directly or indirectly connected via wired or wireless communication, and this application does not limit this.

[0052] For example, Figure 1As shown, the server can obtain key areas in rail transit, and name and define the key areas according to the scene; preset a collection interval, and collect multiple crack images of the key areas under different illumination according to the collection interval to form an image database; collect target crack images, use image matching neural networks to extract features and perform similarity matching training on the target crack images, so as to achieve matching between the latest captured images and the original images in the database with different shooting times but of the same scene; train image recognition neural networks to identify structural cracks in the target crack images, identify structural cracks in a group of images belonging to the same scene obtained by matching, and output a group of binary images; extract basic parameters of structural cracks in the binary images and compare and analyze them, so as to achieve dynamic early warning detection of structural cracks related to rail transit.

[0053] It should be noted that the above steps performed in the server may also be performed in the terminal, and this embodiment does not limit this.

[0054] In the embodiments of the present application, "plurality" refers to two or more than two. In the embodiments of the present application, "first" and "second" are used to distinguish descriptions and should not be understood as implying relative importance.

[0055] It should be noted that the description order of the following embodiments is not intended to limit the preferred order of the embodiments.

[0056] See also Figure 2 In this embodiment, a method for dynamic detection of cracks in rail transit structures is provided. Figure 2 As shown, the specific process of the rail transit structure crack dynamic detection method can be as follows:

[0057] S1. Obtain key areas in rail transit and name and define the key areas according to the scenarios.

[0058] First, for the acquisition of regions, detailed geographical and engineering analysis of key structures in rail transit is required to determine which areas are most likely to have structural cracks. For example, the station's foundation piles, crown beams, and concrete supports are usually monitored because these parts are under greater structural pressure and have a greater impact on safety.

[0059] Secondly, for the naming of regional scenes: "Critical areas" refer to those locations that are critical to the structural integrity of the rail transit system. These areas often play a key supporting role or are locations where the bearing capacity changes significantly. "Scene naming" for these areas means assigning a unique, descriptive name to each critical area for subsequent identification and indexing. For example, if there is a foundation pit retaining pile located near platform A, it can be named "A_Station_Wall_Pile_01". "A_Station" indicates that the location is at station A, "Wall_Pile" indicates that the structure is a retaining pile, and "01" is the number of the retaining pile. The naming of other areas can be deduced in the same way, and will not be repeated here. Of course, other naming methods can also be used, as long as they can facilitate subsequent identification and indexing.

[0060] For example, in the present invention, we can introduce a Multi-level Scene Naming System (MSNS). This system not only considers the geographical location, but also associates the area's usage history, structural characteristics, monitoring parameters, etc. MSNS can bring a higher level of data organization and analysis efficiency.

[0061] Multi-level scene naming:

[0062] First level: "Geography-Structure Code": Based on the geographical location and structure type of the region, such as "A_Station_PileStructure_01", where "A_Station" means the location is platform A, "PileStructure" means the type is retaining pile structure, and "01" is the structure number.

[0063] Level 2: “Function-Condition Coding”: Builds on the first level and adds a functional description of the area and a description of the site conditions. For example, “LoadBearing_HighMoisture” indicates that the area is the main load-bearing point and the ambient humidity is high.

[0064] Level 3: “Time-Event Code”: Codes that reflect the time correlation of the scene name and important historical events. “2020_Renovation” may mean that the scene underwent major repairs in 2020.

[0065] Creation and maintenance of scene library:

[0066] Build a comprehensive scenario library that includes not only scenario naming, but also detailed information for each area—such as construction data, historical crack events, geological reports, load statistics, etc.

[0067] The scene library is updated regularly to include new monitoring data or to correct existing data. Each database tuple contains the scene name, image set, and related monitoring reports and history records.

[0068] Data association of application scenario naming:

[0069] Integrating Geographic Information System (GIS) and Structural Health Monitoring (SHM) data with MSNS ensures that data can be cross-referenced to provide a more refined understanding of crack development and its impact on structural integrity.

[0070] Through the setting of step S1, an efficient information organization and retrieval method is provided for the detection of structural cracks. Scene naming allows for quick identification of specific areas and simplifies the management process of monitoring data. The scene naming definition allows easy association between data collected at different time points, which facilitates the tracking of crack characteristics. This method can adapt to new monitoring locations or change the monitoring status of existing locations without large-scale system changes. Through detailed scene naming, monitoring personnel can quickly index the complete historical record of specific cracks, including image data and associated reports. After adopting a multi-level coding system, the historical changes of cracks can be displayed through data visualization tools to help decision makers understand the patterns and trends of crack development. Different levels of scene naming help to integrate data sets from different research fields (such as structural engineering, geology, and operations management) and promote comprehensive analysis and prediction.

[0071] S2. Preset a collection interval time, and collect multiple crack images of the key area under different illuminations according to the collection interval time to form an image database.

[0072] Among them, the determination of the acquisition interval is based on the analysis of the crack development rate of the key area. For example, for areas where cracks develop faster, a shorter acquisition interval can be set, such as once a day, to capture any slight changes in the cracks. For relatively stable areas, the interval can be extended, such as once a week. The specific value of the acquisition interval can be selected according to the actual work experience of those skilled in the art, and the power consumption of the acquisition interval for the acquisition component or other components can be further considered. Under the combination of power consumption and experience, a more reasonable acquisition interval is taken. In this embodiment, the acquisition interval is preset to 1 hour, that is, the crack image is collected once every 1 hour. Of course, the acquisition interval here is set to be small, mainly to obtain as many and as detailed structural crack changes as possible, and to form a more accurate and rich image database. The acquisition interval for the target crack image can be longer. In this embodiment, the acquisition interval of the target crack image is 12 hours. Of course, the acquisition control of the target crack image can also be performed at any time according to actual needs. It should be noted that the target crack image refers to the image of the crack structure in the key area that currently needs to be detected, that is, by acquiring the image of the current crack structure and then using a neural network to match and identify the image of the current crack structure with the crack image in the image database, the changing trend of the structural cracks in the key area can be monitored.

[0073] The selection and arrangement of acquisition equipment required for image acquisition are diverse, and there is no fixed equipment requirement and acquisition form. For example, depending on the scene, a variety of image acquisition equipment can be used, among which the commonly used ones include: digital cameras, drones, industrial cameras, etc.

[0074] In addition, considering the need to explore the impact of illumination on the training results of the neural network in the future, the camera height and shooting angle must be consistent during the image acquisition process. For the multiple crack images collected, they are stored and managed according to the key areas defined by the acquisition time and naming. Each image has a clear time point and scene identification to facilitate subsequent retrieval and analysis, forming a complete image database.

[0075] Through the settings of steps S1 and S2, timely data acquisition can be achieved so that any crack changes can be recorded and processed in a timely manner, significantly improving the sensitivity and accuracy of the system in responding to crack development. Through standardized acquisition procedures and periodic image updates, the data at different time points are guaranteed to be comparable, providing an accurate basis for subsequent image analysis and crack tracking. The established image database not only provides real-time crack conditions, but also contains historical data on the evolution of cracks over time. These data can help analyze the development trend of cracks and provide data support for long-term monitoring.

[0076] It should also be noted that, for the aforementioned method of presetting the acquisition interval, in some embodiments, the preset acquisition interval can be set more accurately. Specifically, an algorithm is set to automatically analyze the crack development speed and adjust the acquisition frequency according to real-time data to achieve more refined monitoring. For the acquisition image, the same location can be repeatedly acquired in combination with different acquisition angles and multiple lighting conditions to obtain more comprehensive crack information. A simple data processing algorithm is deployed on the acquisition device to achieve preliminary image analysis and processing, reducing the transmission and storage of invalid data.

[0077] S3. Extract features from multiple crack images in the image database, perform similarity matching training based on the extracted features of multiple crack images to obtain an image matching neural network, and perform recognition training based on the extracted features of multiple crack images to obtain an image recognition neural network.

[0078] S4. Collect the target crack image, and use the image matching neural network to match the target crack image with the original crack image in the image database that is shot at different times but is of the same scene.

[0079] Among them, the traditional method generally adopts manual inspection, crack marking, measurement and recording of changes in related parameters. It is necessary to arrange personnel to monitor regularly and at fixed points, which is troublesome. The error of naked eye observation may also be large. The changes of cracks that may appear in a short time may not be distinguished by the naked eye. When the cracks change greatly, the risk of problems in key areas is already high. In addition, the use of artificial naked eye for monitoring is not applicable to all scenarios. For example, in tunnels with poor lighting conditions, human eye observation will be greatly limited, and it may not be possible to manually mark or mark damage near the cracks, which will also hinder the monitoring of structural cracks. The feature extraction through image matching neural network obtains the significant features of cracks in the image. Compared with traditional methods, the features are more robust and have higher discrimination. Therefore, the accuracy of matching the latest image with the original image of the same scene in the database is greatly improved. During the training process, through a large number of sample inputs, the neural network learns to identify which features are key indicators pointing to the same scene. Such learning reduces the mismatch between pictures taken in different scenes or at different times and improves the reliability of matching. The neural network is trained to recognize the same scene even when photographed at different times and under different environmental conditions, so that the monitoring work is no longer restricted by strict environmental control and is suitable for dynamic and long-term structural monitoring.

[0080] The specific principle process of image matching neural network feature extraction and similarity matching training:

[0081] Construct training and test sets: Select representative crack images as training data. Representative crack images refer to images with obvious structural cracks in key areas. At the same time, prepare a test set to verify the model performance.

[0082] Network architecture selection: Choose an appropriate convolutional neural network (CNN) architecture, such as VGG, ResNet, or Inception, which have been shown to be very effective in image recognition and feature extraction.

[0083] Network training:

[0084] Feature extraction layer (convolutional layer) training: Through automatic feature learning, the convolutional layer can extract key features from the crack image, such as the shape, size, texture, etc. of the crack.

[0085] Similarity learning: Different network structures and training methods may lead to different principles and implementation details. The common implementation principle is: compare the distance between the embedding vector of the representative crack image and the embedding vectors of all other crack images in the image database, so as to find the image most similar to the representative crack image. The design of the output layer enables the network to perform similarity calculations on the features of the input layer. Based on an effective distance metric (such as Euclidean distance or cosine similarity), compare the distance between the embedding vector of the representative crack image and the embedding vector of all other crack images in the image database, so as to find the image most similar to the representative crack image.

[0086] Choice of loss function and optimizer: Use cross entropy or ternary loss as the loss function to train the network, ensuring that not only images of the same scene are matched to the same category, but also images of different scenes are distinguished into different categories. Optimizers such as Adam or SGD are responsible for updating the network weights to improve matching accuracy.

[0087] Verification and adjustment: Evaluate the network on the test set and adjust the model based on the verification results, such as modifying the number of network layers, changing the activation function, adjusting the learning rate, etc., to achieve the best matching performance.

[0088] Among them, the training steps of the image matching neural network are as follows:

[0089] Select the architecture of the network model and determine the depth and width of the network based on image data and requirements;

[0090] Building an initial image matching neural network, randomly dividing the images in the image database based on a preset ratio, wherein the preset ratio is training set: validation set: test set = 6:2:2;

[0091] Inputting the training set into the initial image matching neural network, performing similarity matching training according to the scene category, and obtaining the image matching neural network;

[0092] Inputting the verification set into the image matching neural network, and adjusting the hyperparameters of the image matching neural network according to the verification set to improve the performance of the image matching neural network;

[0093] The test set is used to quantitatively evaluate the matching effect of the image matching neural network.

[0094] Among them, in order to ensure effective feature matching between images, first select a suitable neural network architecture, such as ResNet or InceptionNet, according to the image data and monitoring requirements. The depth and width of the network will be determined according to the complexity of the crack features. The depth determines the feature level that the model can learn, and the width is related to the capacity of the model. The images in the image database are randomly divided into training sets, validation sets and test sets according to a preset ratio, with a ratio of 6:2:2. The training set is used for model learning and feature extraction, the validation set is used for model performance optimization and hyperparameter adjustment, and the test set is used to evaluate the final matching effect of the model. The training set is input into the image matching neural network for feature extraction and matching learning. Through multiple iterative training, the model gradually learns how to identify crack features and match between images. During the training process, the performance of the model is continuously verified through the validation set, and the model parameters are adjusted to improve the matching accuracy. After the training is completed, the data of the test set is used to quantitatively evaluate the matching effect of the image matching neural network. This evaluation is based on a series of indicators, such as accuracy, recall and F1 score, to ensure that the constructed model maintains high matching accuracy in practical applications.

[0095] S5. Identify structural cracks in a group of crack images belonging to the same scene obtained by matching and output a group of binary images.

[0096] Among them, after extracting features of multiple crack images and performing recognition training to obtain an image recognition neural network, for each scene, a group of images with good matching effects are selected, and the structural cracks therein are clearly marked, so as to provide a labeled data set for the training of the image recognition neural network. For the image recognition task, an efficient deep learning model is selected, such as a convolutional neural network (CNN). More specifically, U-Net, Mask R-CNN, etc. can be considered. These models involve pixel-level classification and are very suitable for the identification of structural cracks. Using the pre-prepared data set, the model is allowed to learn how to identify crack features through multiple rounds of iterations, and the network parameters are optimized to achieve the best recognition effect. In this process, the network hyperparameters, such as learning rate, batch size, optimizer type, etc., are adjusted to improve the recognition accuracy and reduce the possibility of overfitting. The trained image recognition neural network can identify the location of cracks in pixels. According to the output of the network, each pixel point is demarcated as a crack or non-crack, and a binary image is generated by threshold processing, in which the crack is usually represented by white and the background is black.

[0097] In some embodiments, considering that the limited lighting conditions in the tunnel limit the effect of neural network image matching and image recognition to some extent, it is necessary to quantify the range of illumination required to meet the accuracy requirements. Therefore, before the step of acquiring the target crack image and using the image matching neural network to match the target crack image with the original crack image of the same scene in the image database at different shooting times, it also includes: testing the image matching effect and image recognition effect of the image matching neural network and the image recognition neural network under different illumination conditions, and obtaining the optimal illumination interval required for acquiring the image based on the image matching effect and the image recognition effect. The process of acquiring structural crack images in the area under different illuminations includes: determining the overall range of illumination to be explored based on the actual engineering environment and lighting equipment, and setting appropriate illumination intervals to divide it into groups for research. By inputting images collected under different illumination conditions into the image matching neural network and the image recognition neural network, it is possible to quantify an illumination range that is beneficial for both the image matching neural network and the image recognition neural network to perform image matching. In the subsequent acquisition process of the target crack image, external devices such as flashlights, lighting lamps, reflectors, etc. can be used to change the illumination so that the illumination during acquisition meets the illumination conditions of the matching neural network and the recognition neural network, thereby improving the image matching accuracy and recognition accuracy.

[0098] Furthermore, for testing the image matching effect and image recognition effect of the image matching neural network and the image recognition neural network under different illumination conditions, the step of obtaining the optimal illumination range required for capturing the image according to the image matching effect and the image recognition effect includes:

[0099] Preset a test light illumination interval threshold, take multiple light illuminations within the test light illumination interval threshold range to respectively perform image acquisition and input the acquired images into the image matching neural network and the image recognition neural network;

[0100] Matching and identifying the collected image with the image in the image database, recording the image matching data of the image matching neural network and the image recognition data of the image recognition neural network respectively, and performing index evaluation on the image matching data and the image recognition data respectively;

[0101] According to the evaluation result, the matching accuracy illumination interval of the image matching neural network and the recognition accuracy illumination interval of the image recognition neural network are obtained;

[0102] The intersection of the matching accuracy illumination interval and the recognition accuracy illumination interval is taken to obtain an optimal illumination interval that satisfies the image matching neural network and the image recognition neural network.

[0103] It should be noted that the range of the test illumination interval threshold needs to be larger than the range of the matching accuracy illumination interval and the range of the recognition accuracy illumination interval, and the range of the matching accuracy illumination interval and the range of the recognition accuracy illumination interval are both within the range of the test illumination interval threshold, thereby ensuring that the numerical range of the illumination test is large enough and the data has sufficient universality. For indicator evaluation, the best illumination interval can be obtained intuitively through visual observation or quantitative indicator comparison, such as using binarization, comparing similarity coefficients and intersection ratios.

[0104] The above embodiments provide a quantitative method to define the illumination range required to achieve the optimal performance of both the image matching neural network and the image recognition neural network models, thereby laying a data foundation for subsequent image acquisition work.

[0105] See also Figure 4 , a simple example is as follows:

[0106] Overexposure or underexposure of images will greatly reduce the accuracy of neural network matching and recognition. Therefore, in the early stage, structural crack images under various illuminations are intentionally collected and processed and input into two neural networks for training. Then, images with different illuminations are selected to test the two neural networks at the same time, and the corresponding evaluation indicators are qualitatively observed and quantitatively calculated. The calculation results are plotted in the form of data graphs (such as Figure 4 As shown), the illumination range that meets the image matching accuracy requirement is [a, b], and the illumination range that meets the image recognition accuracy requirement is [c, d]. The optimal illumination range for the two neural networks is [c, b].

[0107] In some embodiments, the step of setting training data for the image recognition neural network includes:

[0108] Selecting representative crack images from the image database, and manually annotating cracks in the crack images at a pixel level, so that each crack image corresponds to a label image;

[0109] defining the crack image and a label image corresponding to the crack image as a set of image data;

[0110] Each set of the image data is enlarged to obtain an enlarged data set.

[0111] Among them, pictures reflecting typical characteristics of cracks are screened out from the image database. These characteristics may include different crack widths, lengths, shapes, and crack images taken under different lighting and angles. The purpose of selecting diverse image samples is to improve the learning and generalization capabilities of the neural network and ensure that the network can identify cracks with high accuracy in practical applications. The selected crack images are annotated pixel by pixel using image editing tools, such as dedicated image annotation software (such as LabelMe). The purpose of annotation is to accurately mark the crack boundaries in the image so that the network can learn how to distinguish cracks from non-crack areas in the subsequent training stage. The annotated crack images are converted into binary images, in which the cracks are white (value 1) and the non-crack background is black (value 0). Such binary images can intuitively reflect the accurate shape and distribution of cracks and are ideal inputs for training neural networks. Various forms of data enhancement operations, such as rotation, scaling, cropping, color transformation, etc., are performed on the original image and its corresponding annotated image to expand the data set and increase sample diversity. Data enhancement can simulate the changes of cracks under different environmental conditions and help the neural network learn more robust crack characteristics.

[0112] Through the above training steps, refined pixel-level annotations are obtained to ensure that the network can learn the accurate representation of cracks in the image, so as to accurately identify cracks in practical applications. And because the data set includes a variety of crack images after changes, the trained neural network can adapt to larger scenes and condition changes. The expanded data set contains a wider range of crack types and environmental conditions, reducing the risk of the model overfitting the original training samples and improving the generalization ability of the model. Using the binary image output by the network, various parameters of the crack, such as length, width, etc., can be automatically extracted, which is conducive to subsequent crack analysis and structural health assessment.

[0113] S6. Extract basic parameters of the structural cracks in the binary image and compare the basic parameters with historical data to monitor the changing trend of the structural cracks.

[0114] Among them, after completing the training and binarization output of the image recognition neural network, the basic parameters of the crack are extracted. The binary image simplifies the morphology of the crack so that it only contains black and white colors, white represents the crack itself, and black is the part outside the crack. Image processing techniques such as edge detection and contour tracking are used to analyze the white crack area and systematically extract the basic characteristics of the crack. These basic parameters usually include the width, length, area, direction and expansion mode of the crack. The morphology of the crack is quantified into a numerical value using a related algorithm, such as estimating the area of ​​the crack by calculating the number of white pixels in the binary image, or obtaining the length and width of the crack by analyzing the geometric properties of the crack contour. The extracted crack parameters are compared with historical data to monitor the change trend of the crack. For example, the growth rate of the crack length or the degree of change of the crack width in two consecutive monitoring results is compared. An early warning is issued based on whether the change in the crack parameters exceeds the preset safety threshold. For example, if the crack width increases by more than 10% within a certain period of time, the system will automatically issue an early warning signal to prompt the relevant maintenance department to conduct inspection or repair.

[0115] In this embodiment, the accuracy and consistency of crack monitoring are ensured by parameter extraction and quantitative processing. The system can accurately measure the specific changes of cracks, thereby providing accurate crack status information. Automated parameter extraction and comparative analysis significantly improve the efficiency and response speed of monitoring, which is especially critical for comprehensive crack monitoring of large-scale rail networks. Threshold monitoring is achieved by automated comparative analysis, and the introduction of dynamic early warning helps to detect problems in a timely manner and carry out preventive maintenance in advance, thereby avoiding potential cracks from causing serious structural damage. Continuous monitoring and analysis of crack parameter changes helps to understand the evolution of cracks over time and provide a scientific basis and decision-making support for the long-term stability of rail transit structures. The introduction of dynamic early warning detection can formulate and implement more effective maintenance plans in advance, avoiding higher maintenance costs caused by rushing to deal with crack problems.

[0116] It should be noted that the key areas include: the station pit retaining pile body, the crown beam pit side, the concrete support, the side wall of the construction shaft, the station in the operating line and the top of the section between adjacent stations.

[0117] The step of collecting a plurality of crack images of the key area to form an image database includes: increasing or decreasing the brightness of the shooting area, controlling the camera to shift or rotate at a certain angle in the horizontal or vertical direction, controlling the camera's shooting field of view to become narrower or wider, and controlling the camera to adopt a motion blur mode. One or more of the following methods;

[0118] The crack images of all different areas captured within the acquisition interval and the crack images of the same area captured at different times are combined into an image database.

[0119] The steps to expand the dataset include:

[0120] The crack image and the label image corresponding to the crack image are synchronously subjected to operations of rotation, flipping, scaling, cropping, adding noise and changing brightness to increase the data volume.

[0121] The basic parameters of structural cracks include one or more of crack spacing, opening, extension length, strike, dip and inclination.

[0122] See also Figure 3 In order to more easily understand the dynamic early warning detection of structural cracks implemented by the present application, the following will be described by way of example in combination with the technical solutions in the aforementioned embodiments:

[0123] In rail transit structure monitoring, crack identification is defined as a binary classification problem involving two output categories: background and crack. The goal is to determine whether each pixel in the image belongs to the background or the crack. In order to evaluate the effect of the trained neural network model on the crack identification task, two indicators were selected: Dice Similarity coefficient and intersection ratio ( IoU ), both of which are pixel-level evaluation indicators. Specifically, Dice The similarity coefficient is a pixel-level evaluation index with a value between 0 and 1. The closer its value is to 1, the better the segmentation effect of the neural network model. IoU ) is the ratio of the intersection and union of the predicted mask result of the image recognition neural network for the crack and the true mask.

[0124] The confusion matrix is ​​used to analyze the classification results in more detail. The confusion matrix includes four parts: the true class ( TP ), false positive class ( FP ), false negative class ( FN ) and true negatives ( TN ).in, TP Indicates the number of pixels that are actually cracks and are correctly predicted to be cracks, FP Indicates the number of pixels that are actually background but are mistakenly predicted as cracks, FNIndicates the number of pixels that are actually cracks but are mistakenly predicted as background. TN It is the number of pixels that are actually background and are correctly predicted to be background.

[0125] Dice The similarity coefficient calculation formula is:

[0126] ,

[0127] Intersection-over-union ratio ( IoU ) is calculated as:

[0128] ,

[0129] The binary image of the cracks is processed using a dedicated code to extract the basic parameters of the cracks, such as crack spacing, opening, extension length, strike, dip and inclination. This process includes finding the outline of the cracks in the image and applying the minimum circumscribed oblique rectangle algorithm to identify and measure the cracks. As for the dedicated code, it refers to the programming script or software module used to implement a specific image processing task. In the crack identification task, the dedicated code can include the following steps:

[0130] Image preprocessing: such as noise filtering and contrast enhancement.

[0131] Feature extraction: e.g. measurement of crack width and length.

[0132] Crack location: For example, using edge detection algorithms to find the boundaries of cracks.

[0133] Crack Classification: Use the trained neural network model to identify and label cracks.

[0134] Parameter quantification: Convert the image data of the crack into the corresponding actual size data.

[0135] These dedicated codes are written by software developers or data scientists using programming languages ​​​​(such as Python, MATLAB, etc.), and implemented using image processing libraries (such as OpenCV), machine learning frameworks (such as TensorFlow, PyTorch), etc., which will not be described in detail here.

[0136] In order to overcome the possible errors in image acquisition, such as camera offset, a same marker is placed in the camera field of view each time an image is acquired. This ensures a reasonable comparison between images, and the conversion relationship between the pixel data in the image and the actual crack parameters is established through the algorithm. Regular crack parameter comparison and analysis is carried out. By comparing the changes in the crack parameters continuously monitored, if the change in one or more parameters of the crack exceeds the established threshold, an early warning signal will be automatically issued to prompt maintenance personnel to conduct corresponding inspections or take appropriate maintenance measures.

[0137] It should be noted that, in addition to Dice Similarity coefficient and intersection ratio ( IoU ), there are also multiple evaluation indicators that can be used to evaluate the performance of crack identification tasks, including but not limited to: precision ( Precision ): Also known as the positive prediction value, it is the ratio of correctly classified positive samples (cracks) to the total number of samples classified as positive (cracks). Recall rate ( Recall ): Also known as the true class rate, it is the ratio of correctly classified positive samples to all actual positive samples. F1 F1 Score: It is the harmonic mean of precision and recall, giving a single metric that takes both precision and recall into account. Accuracy Specificity ( ): is the proportion of all correctly classified samples (cracks correctly identified and non-cracks correctly rejected) to all samples. Specificity ): Also known as the true negative rate, it is the ratio of correctly classified negative samples (non-cracks) to all actual negative samples. Matthews correlation coefficient ( MCC ): is a comprehensive consideration TP , FP , FN , TN The evaluation index has a range of [-1, +1], where +1 indicates perfect prediction and -1 indicates complete inconsistency. Area under the receiver operating characteristic curve (ROC Curve) (AUC-ROC): AUC represents the area formed by the curve and the coordinate axis, which can reflect the model's ability to distinguish between positive and negative classes. Average accuracy ( Mean Accuracy ): The average accuracy of multiple test sets. Area under the Precision-Recall Curve (AUC-PR): The Precision-Recall curve shows the relationship between precision and recall at different thresholds. AUC-PR is especially useful in unbalanced datasets.

[0138] The method described in the above embodiment will be further described in detail below.

[0139] like Figure 5 FIG. 1 is a schematic diagram of a structure of a dynamic detection device for cracks in a rail transit structure provided by an embodiment of the present application, and the device comprises:

[0140] An acquisition definition module 10 is used to acquire key areas in rail transit and name and define the key areas according to scenarios; optionally, the acquisition definition module includes an acquisition module and a naming definition module, the acquisition module is used to acquire key areas, and the naming definition module is used to name and define the key areas according to scenarios;

[0141] An image interval acquisition module 20, used for presetting an acquisition interval time, and acquiring a plurality of crack images of the key area under different illuminations according to the acquisition interval time, so as to form an image database; optionally, the image interval acquisition module includes a preset time module for presetting the acquisition interval time and an interval acquisition module for acquiring crack images;

[0142] A model training module 30 is used to extract features from multiple crack images in an image database, perform similarity matching training based on the extracted features of the multiple crack images to obtain an image matching neural network, and perform recognition training based on the extracted features of the multiple crack images to obtain an image recognition neural network;

[0143] An image matching module 40 is used to collect a target crack image and use the image matching neural network to match the target crack image with an original crack image in the image database that was shot at a different time but is of the same scene;

[0144] An image recognition module 50 is used to identify structural cracks in a group of images belonging to the same scene obtained by matching and output a group of binary images;

[0145] The extraction and analysis module 60 is used to extract basic parameters of the structural cracks in the binary image and compare the basic parameters with historical data to monitor the change trend of the structural cracks. A dedicated code can be implanted in the extraction and analysis module.

[0146] It can be seen that the method of the embodiment of the present application can realize dynamic detection of structural cracks by using image matching neural network combined with image recognition technology. In detail, first, compared with the inefficiency of traditional manual monitoring methods, it is limited by the safety of manual operations and the difficulty of execution in complex environments. The method provided by the present application can automatically collect crack images in key areas and form a database, so that the monitoring work is not limited by manpower, continuity and comprehensiveness are guaranteed, and the operation is safe and reliable. Second, in the background technology, due to factors such as complex site conditions and variable light conditions, there are large errors in matching and identifying artificial cracks. By using image matching and recognition neural networks, accurate crack features can be extracted, and by self-learning to adapt to changes in different working environments, the development of cracks can be more accurately identified and tracked, effectively avoiding the problem of large errors in matching and identifying artificial cracks due to factors such as complex site conditions and variable light conditions. Third, the present application can analyze the changing trend of structural cracks in real time, provide early warning signals in a timely manner, and is conducive to taking countermeasures as soon as possible to avoid potential safety accidents. Fourth, compared with traditional methods, the present application uses visual technology to greatly improve the efficiency of crack detection, which is particularly critical for public infrastructure such as urban rail transit that needs to be continuously operated. Since crack monitoring can be performed automatically during non-operating hours, it does not affect daily operations, thus optimizing the coordination between maintenance and operations. Fifth, the detection method of this application can accumulate rich crack development data, providing data support for the long-term maintenance and improvement of urban rail transit. Through in-depth analysis of historical data, the crack development trend can be predicted based on scientific data, and maintenance plans and budgets can be arranged more reasonably.

[0147] Corresponding to the above method embodiment, the embodiment of the present invention further provides a computer device. The computer device described below and the method for dynamic detection of cracks in a rail transit structure described above can refer to each other.

[0148] The computer equipment includes:

[0149] Memory for storing computer programs;

[0150] The processor is used to implement the steps of the method for dynamic detection of cracks in rail transit structures of the above method embodiment when executing the computer program:

[0151] Obtain key areas in rail transit, and name and define the key areas according to scenarios;

[0152] Preset a collection interval time, and collect multiple crack images of the key area under different illuminations according to the collection interval time to form an image database;

[0153] Extracting features from a plurality of crack images in the image database, performing similarity matching training based on the extracted features of the plurality of crack images to obtain an image matching neural network, and performing recognition training based on the extracted features of the plurality of crack images to obtain an image recognition neural network;

[0154] Acquire a target crack image, and use the image matching neural network to match the target crack image with an original crack image in the image database that is shot at a different time but is of the same scene;

[0155] Identify structural cracks in a group of crack images belonging to the same scene obtained by matching and output a group of binary images;

[0156] The basic parameters of the structural cracks in the binary image are extracted and compared with the historical data to monitor the change trend of the structural cracks.

[0157] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application, which may be a computing device with data processing capabilities.

[0158] The device may include: a processor 501 and a memory 502 .

[0159] The memory 502 is used to store programs, and the processor 501 calls the programs stored in the memory 502 to execute the above method embodiment. The specific implementation method and technical effect are similar and will not be repeated here.

[0160] The memory 502 stores program codes, and when the program codes are executed by the processor 501, the processor 501 executes various steps of the method according to various exemplary embodiments of the present application described in the above “Exemplary Method” section of this specification.

[0161] Processor 501 can be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, and can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware processor for execution, or can be executed by a combination of hardware and software modules in the processor.

[0162] The memory 502 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The memory may include at least one type of storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory, a random access memory (Random Access Memory, RAM), a static random access memory (Static Random Access Memory, SRAM), a programmable read-only memory (Programmable Read Only Memory, PROM), a read-only memory (Read Only Memory, ROM), an electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), a magnetic memory, a disk, an optical disk, etc. The memory is any other medium that can be used to carry or store a desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 502 in the embodiment of the present application can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0163] Optionally, the present application also provides a program product, such as a computer-readable storage medium, including a program, which is used to execute the above method embodiment when executed by a processor.

[0164] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0165] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0166] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0167] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (English: processor) to perform some steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (English: Read-Only Memory, abbreviated: ROM), random access memory (English: Random Access Memory, abbreviated: RAM), disk or optical disk and other media that can store program codes.

Claims

1. A dynamic detection method for cracks in rail transit structures, It is characterized in that The method comprises: Obtain key areas in rail transit, and name and define the key areas according to scenarios; Preset a collection interval time, and collect multiple crack images of the key area under different illuminations according to the collection interval time to form an image database; Extracting features from a plurality of crack images in the image database, performing similarity matching training based on the features of the plurality of crack images to obtain an image matching neural network, and performing recognition training based on the features of the plurality of crack images to obtain an image recognition neural network; Acquire a target crack image, and use the image matching neural network to match the target crack image with an original crack image in the image database that is shot at a different time but is of the same scene; Identify structural cracks in a group of crack images belonging to the same scene obtained by matching and output a group of binary images; The basic parameters of the structural cracks in the binary image are extracted and compared with the historical data to monitor the change trend of the structural cracks.

2. The dynamic detection method for cracks in rail transit structures according to claim 1, It is characterized in that Before the step of acquiring the target crack image and using the image matching neural network to match the target crack image with the original crack image in the image database that was shot at different times but in the same scene, the step also includes: testing the image matching effect and image recognition effect of the image matching neural network and the image recognition neural network under different illumination conditions, and obtaining the optimal illumination range required for acquiring the image based on the image matching effect and the image recognition effect.

3. The method for dynamic detection of cracks in rail transit structures according to claim 2, It is characterized in that The step of testing the image matching effect and the image recognition effect of the image matching neural network and the image recognition neural network under different illumination conditions, and obtaining the optimal illumination range required for capturing an image according to the image matching effect and the image recognition effect, comprises: Preset a test light illumination interval threshold, take multiple light illuminations within the test light illumination interval threshold range to respectively perform image acquisition and input the acquired images into the image matching neural network and the image recognition neural network; Matching and identifying the collected image with the image in the image database, recording the image matching data of the image matching neural network and the image recognition data of the image recognition neural network respectively, and performing index evaluation on the image matching data and the image recognition data respectively; According to the evaluation result, the matching accuracy illumination interval of the image matching neural network and the recognition accuracy illumination interval of the image recognition neural network are obtained; The intersection of the matching accuracy illumination interval and the recognition accuracy illumination interval is taken to obtain an optimal illumination interval that satisfies the image matching neural network and the image recognition neural network.

4. According to claim 1, the dynamic detection method for cracks in rail transit structures, It is characterized in that The training steps of the image matching neural network include: Select the architecture of the network model and determine the depth and width of the network based on image data and requirements; Building an initial image matching neural network, randomly dividing the images in the image database based on a preset ratio, wherein the preset ratio is training set: validation set: test set = 6:2:2; Inputting the training set into the initial image matching neural network, performing similarity matching training according to the scene category, and obtaining the image matching neural network; Inputting the verification set into the image matching neural network, so that the image matching neural network adjusts hyper parameters according to the verification set to improve the performance of the image matching neural network; The test set is used to quantitatively evaluate the matching effect of the image matching neural network.

5. The method for dynamic detection of cracks in rail transit structures according to claim 1, It is characterized in that The step of setting the training data of the image recognition neural network comprises: Selecting representative crack images from the image database, and manually annotating cracks in the crack images at a pixel level, so that each crack image corresponds to a label image; defining the crack image and a label image corresponding to the crack image as a set of image data; Each set of the image data is enlarged to obtain an enlarged data set.

6. The method for dynamic detection of cracks in rail transit structures according to claim 5, It is characterized in that The step of obtaining the expanded data set comprises: The crack image and the label image corresponding to the crack image are synchronously subjected to operations of rotation, flipping, scaling, cropping, adding noise and changing brightness to increase the data volume.

7. The method for dynamic detection of cracks in rail transit structures according to any one of claims 1 to 6, It is characterized in that The basic parameters of the structural cracks include one or more of crack spacing, opening, extension length, strike, dip and inclination.

8. A dynamic detection device for cracks in rail transit structures, It is characterized in that The dynamic detection device comprises: An acquisition definition module is used to acquire key areas in rail transit and name and define the key areas according to scenarios; An image interval acquisition module, used for presetting an acquisition interval time, and acquiring a plurality of crack images of the key area under different illuminations according to the acquisition interval time, so as to form an image database; A model training module, used for extracting features from multiple crack images in an image database, performing similarity matching training based on the extracted features of multiple crack images to obtain an image matching neural network, and performing recognition training based on the extracted features of multiple crack images to obtain an image recognition neural network; An image matching module, used for collecting a target crack image, and using the image matching neural network to match the target crack image with an original crack image in the image database that was shot at a different time but is of the same scene; An image recognition module is used to identify structural cracks in a group of matched images belonging to the same scene and output a group of binary images; The extraction and analysis module is used to extract basic parameters of the structural cracks in the binary image and compare the basic parameters with historical data to monitor the change trend of the structural cracks.

9. An electronic device, It is characterized in that include: A processor, a storage medium and a bus, wherein the storage medium stores program instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the program instructions to perform the steps of any method described in claims 1 to 7.

10. A computer-readable storage medium, It is characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are executed.

Images