Tunnel safety monitoring system and method based on edge calculation

Abstract

The invention provides a tunnel safety monitoring system and method based on edge computing, the tunnel safety monitoring system is used for tunnel safety monitoring, and the tunnel safety monitoring system comprises a sensor, an edge computing node and a cloud platform. The sensor comprises a displacement sensor based on the Beidou technology and an image acquisition device based on the Leiysight technology, and tunnel inner wall images of the target tunnel in the safety monitoring process are continuously acquired through the image acquisition device based on the Leiysight technology; extracting structure edge features of the target tunnel inner wall area from the tunnel inner wall image according to the edge convolution kernel in each angle direction; fusing the tunnel inner wall image according to the structure edge features and the gray projection features to obtain a recognition feature map; and analyzing the potential safety hazard of the target tunnel according to the identification feature map. According to the scheme, the detail features of the tunnel inner wall image can be identified under the influence of complex distribution of tunnel inner wall surface defects and uneven illumination distribution, so that tunnel safety alarm can be carried out in time.

Description

Technical Field

[0001] The present application relates to the technical field of tunnel safety monitoring, and more specifically, to a tunnel safety monitoring system and method based on edge computing. Background Art

[0002] With the rapid development of the Internet of Things and the Industrial Internet, the amount of data has exploded. Traditional cloud computing architectures can no longer meet the needs of real-time performance, bandwidth, and privacy protection. Edge computing has emerged as the times require. It deploys computing, storage, and network resources close to data sources or user terminals. By processing local data at edge nodes, it reduces dependence on the cloud and significantly reduces data transmission latency and bandwidth usage. At the same time, edge computing can enhance data privacy protection and avoid the risks of centralized storage and processing of sensitive data in the cloud. In addition, edge computing also supports distributed intelligent decision-making, improves system reliability and autonomy, and provides key support for achieving efficient, secure, and intelligent digital transformation.

[0003] In existing edge computing, when data is generated, the edge node first pre-processes the data, such as data cleaning, compression and feature extraction, to reduce the transmission volume and improve efficiency. Subsequently, the data is analyzed and decided in real time through locally deployed algorithms (such as machine learning models or rule engines), and key results or a small amount of data are uploaded to the cloud. However, in tunnel safety monitoring, the edge computing pre-processes the tunnel inner wall image. Due to the complex distribution of surface defects of the tunnel inner wall and the uneven distribution of light, traditional image processing methods are difficult to accurately extract the detailed features of the tunnel inner wall image (such as texture, edge), resulting in a decrease in the accuracy of the target tunnel safety assessment. Therefore, how to identify the detailed features of the tunnel inner wall image under the influence of the complex distribution of surface defects of the tunnel inner wall and the uneven distribution of light in order to issue a timely tunnel safety alarm has become a difficult problem faced by the industry. Summary of the Invention

[0004] The present application provides a tunnel safety monitoring system and method based on edge computing, which can identify the detailed features of the tunnel inner wall image under the influence of the complex distribution of surface defects on the tunnel inner wall and uneven light distribution, so as to issue tunnel safety alarms in a timely manner.

[0005] In a first aspect, the present application provides a tunnel safety monitoring method based on edge computing, which is used for a tunnel safety monitoring system to perform tunnel safety monitoring. The tunnel safety monitoring system includes sensors, edge computing nodes, and a cloud platform. The sensors are deployed at various monitoring points in a target tunnel. The method includes the following steps:

[0006] Sensors installed at various monitoring points in the target tunnel collect images of the inner wall of the target tunnel, and transmit the images to the edge computing nodes in real time;

[0007] Determining edge convolution kernels in different angular directions when the edge computing node performs edge extraction on the tunnel inner wall image, and performing multi-directional inner wall edge recognition on the tunnel inner wall image based on the edge convolution kernels in each angular direction, to obtain structural edge features of the target tunnel inner wall area under the current monitoring state;

[0008] Determining a grayscale compression coefficient of the tunnel inner wall image in a luminance channel, and performing projection compression on the tunnel inner wall image based on the grayscale compression coefficient combined with grayscale statistical parameters of the luminance channel to obtain grayscale projection features of a target tunnel inner wall area under current lighting conditions;

[0009] Performing heterogeneous feature fusion on the tunnel inner wall image according to the structural edge features and the grayscale projection features to obtain an identification feature map of the inner wall area of the target tunnel for safety identification;

[0010] The safety hazards of the inner wall of the target tunnel are analyzed based on the identification feature map, and the safety hazard information of the inner wall of the target tunnel is output by the edge computing node to the cloud platform.

[0011] In some embodiments, determining edge convolution kernels at different angles when the edge computing node performs edge extraction on the tunnel inner wall image specifically includes:

[0012] Obtain historical crack monitoring data of the target tunnel;

[0013] performing distribution prediction of crack directions of the target tunnel based on the historical crack monitoring data to obtain predicted distribution probabilities of different crack directions of the target tunnel;

[0014] Calculate the predicted distribution entropy of the target tunnel cracks through the predicted distribution probability of each different crack direction;

[0015] The edge computing node calibrates the angle direction when performing edge extraction on the tunnel inner wall image based on the predicted distribution entropy to obtain multiple different angle directions;

[0016] Based on the edge detection operator, edge convolution kernels in different angular directions are set, thereby obtaining edge convolution kernels in different angular directions when performing edge extraction on the tunnel inner wall image.

[0017] In some embodiments, the tunnel inner wall image is subjected to multi-directional inner wall edge recognition based on edge convolution kernels in various angular directions to obtain structural edge features of the target tunnel inner wall area in the current monitoring state, specifically including:

[0018] Extracting the convolution block corresponding to each pixel point in the tunnel inner wall image;

[0019] Select a pixel as the selected pixel, convolve the convolution block of the selected pixel with the edge convolution kernel in each angular direction, and then obtain the convolution pixel of the selected pixel;

[0020] Continue to determine the convolution pixels of the remaining pixels;

[0021] Constructing an edge detail map corresponding to the tunnel inner wall image based on all convolution pixels;

[0022] Structural edge features of the target tunnel inner wall area under the current monitoring state are identified from the edge detail image.

[0023] In some embodiments, determining the grayscale compression coefficient of the tunnel inner wall image in the brightness channel specifically includes:

[0024] Converting the tunnel inner wall image into a grayscale inner wall image;

[0025] Calculating a brightness change rate sequence corresponding to the grayscale inner wall image;

[0026] The grayscale compression coefficient of the tunnel inner wall image in the brightness channel is determined by the brightness change rate sequence.

[0027] In some embodiments, projecting and compressing the tunnel inner wall image based on the grayscale compression coefficient in combination with the grayscale statistical parameters of the brightness channel to obtain the grayscale projection features of the target tunnel inner wall area under the current lighting conditions specifically includes:

[0028] Acquiring grayscale data of the tunnel inner wall image in a brightness channel;

[0029] Determine the grayscale statistical parameters of the brightness channel according to the grayscale data;

[0030] Setting an adaptive adjustment factor when projecting and compressing the tunnel inner wall image based on the grayscale statistical parameter;

[0031] Projecting and compressing the grayscale of each pixel of the tunnel inner wall image on the brightness channel using the adaptive adjustment factor and the grayscale compression coefficient to obtain a brightness projection compression image corresponding to the tunnel inner wall image;

[0032] The grayscale projection features of the target tunnel inner wall area under the current lighting conditions are extracted from the brightness projection compression map.

[0033] In some embodiments, the sensor includes: a displacement sensor based on Beidou technology and an image acquisition device based on radar technology.

[0034] In some embodiments, the displacement sensors based on Beidou technology are deployed in key parts of the tunnel to obtain high-precision displacement data, wherein the key parts of the tunnel include: the arch, the arch waist, and the side walls. The image acquisition device based on radar technology is installed at various monitoring points on the inner wall of the tunnel. The image acquisition device based on radar technology continuously collects images of the inner wall of the target tunnel during the safety monitoring process.

[0035] In a second aspect, the present application provides a tunnel safety monitoring system based on edge computing, which includes sensors, edge computing nodes, a cloud platform, and a monitoring unit. The sensors are deployed at various monitoring points in the target tunnel, and the monitoring unit includes:

[0036] An acquisition module is configured to collect images of the inner wall of the target tunnel through sensors deployed at various monitoring points in the target tunnel, and transmit the images to the edge computing node in real time;

[0037] a processing module, configured to determine edge convolution kernels in different angular directions when the edge computing node performs edge extraction on the tunnel inner wall image, and perform multi-directional inner wall edge recognition on the tunnel inner wall image based on the edge convolution kernels in each angular direction, to obtain structural edge features of the target tunnel inner wall area under the current monitoring state;

[0038] The processing module is further configured to determine a grayscale compression coefficient of the tunnel inner wall image in a luminance channel, and perform projection compression on the tunnel inner wall image based on the grayscale compression coefficient in combination with grayscale statistical parameters of the luminance channel to obtain grayscale projection features of a target tunnel inner wall area under current lighting conditions;

[0039] The processing module is further configured to perform heterogeneous feature fusion on the tunnel inner wall image according to the structural edge features and the grayscale projection features to obtain an identification feature map of the inner wall area of the target tunnel for safety identification;

[0040] An execution module is used to analyze the safety hazards of the inner wall of the target tunnel based on the identification feature map, and the edge computing node outputs the safety hazard information of the inner wall of the target tunnel to the cloud platform.

[0041] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a code, and the processor is configured to obtain the code and execute the above-mentioned tunnel safety monitoring method based on edge computing.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned tunnel safety monitoring method based on edge computing.

[0043] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0044] In the tunnel safety monitoring system and method based on edge computing provided by the present application, first, the tunnel inner wall image of the target tunnel is collected by sensors arranged at various monitoring points of the target tunnel, and the tunnel inner wall image is transmitted to the edge computing node in real time; secondly, the edge convolution kernels in different angular directions when the edge computing node performs edge extraction on the tunnel inner wall image are determined, and the inner wall edge of the tunnel inner wall image is multi-directionally recognized based on the edge convolution kernels in various angular directions to obtain the structural edge features of the target tunnel inner wall area under the current monitoring state; further, the tunnel inner wall image is determined. The grayscale compression coefficient of the image on the brightness channel is obtained, and the tunnel inner wall image is projected and compressed based on the grayscale compression coefficient combined with the grayscale statistical parameters of the brightness channel to obtain the grayscale projection features of the target tunnel inner wall area under the current lighting conditions; then, heterogeneous feature fusion is performed on the tunnel inner wall image according to the structural edge features and the grayscale projection features to obtain an identification feature map of the inner wall area when the target tunnel is safely identified; finally, the safety hazards of the target tunnel inner wall are analyzed based on the identification feature map, and the safety hazard information of the target tunnel inner wall is output by the edge computing node to the cloud platform.

[0045] It can be seen that the present application identifies the detailed features of the tunnel inner wall image under the influence of the complex distribution of tunnel inner wall surface defects and uneven illumination distribution; first, the edge computing node sets the edge convolution kernel in different angular directions when performing edge extraction on the tunnel inner wall image based on the predicted distribution probability of the target tunnel crack direction, thereby providing conditions for highlighting the grayscale gradient changes in the tunnel inner wall image, thereby improving the recognition accuracy of the tunnel inner wall image; secondly, the tunnel inner wall image is subjected to multi-directional inner wall edge recognition based on the edge convolution kernel in each angular direction, and the structural edge features of the target tunnel inner wall area under the current monitoring state are obtained to provide the edge information of the tunnel structure, thereby improving the accuracy and stability of automated tunnel safety monitoring, thereby avoiding the loss of some recognition information caused by the complex directional distribution of tunnel inner wall surface defects; further, the tunnel inner wall image is projected and compressed based on the grayscale compression coefficient combined with the grayscale statistical parameters of the brightness channel to obtain the target tunnel inner wall image. The grayscale projection features of the inner wall area under the current lighting conditions are used to effectively identify abnormal structures that may exist in the tunnel inner wall image, avoid recognition interference caused by uneven lighting inside the tunnel, and effectively reduce the impact of local lighting changes; then, the tunnel inner wall image is subjected to heterogeneous feature fusion based on the structural edge features and grayscale projection features to obtain an identification feature map of the inner wall area when the target tunnel is safely identified, so as to simultaneously retain the edge information and brightness feature information of the tunnel structure, so that the identification feature map can adapt to tunnel environments with different lighting, materials and damage types, thereby effectively enhancing the ability to identify tunnel defects; finally, the safety hazards of the target tunnel are analyzed based on the identification feature map, and the safety hazard information of the target tunnel is output to the cloud platform by the edge computing node; in summary, the technical solution provided by this application can identify the detailed features of the tunnel inner wall image under the influence of the complex distribution of surface defects on the tunnel inner wall and uneven lighting distribution, so as to make timely tunnel safety alarms. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is an exemplary flow chart of a tunnel safety monitoring method based on edge computing according to some embodiments of the present application;

[0047] Figure 2 is an exemplary flow chart of determining structural edge features according to some embodiments of the present application;

[0048] Figure 3 is an exemplary flow chart of determining a grayscale compression coefficient according to some embodiments of the present application;

[0049] Figure 4 is a schematic structural diagram of a monitoring unit according to some embodiments of the present application;

[0050] Figure 5This is a structural diagram of a computer device for implementing a tunnel safety monitoring method based on edge computing according to some embodiments of the present application. DETAILED DESCRIPTION

[0051] In order to better understand the technical solution of the present application, the technical solution of the present application will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0052] refer to Figure 1 , which is an exemplary flow chart of a tunnel safety monitoring method based on edge computing according to some embodiments of the present application. The tunnel safety monitoring method based on edge computing 100 mainly includes the following steps:

[0053] In step 101, an image of the inner wall of the target tunnel is collected by sensors installed at various monitoring points of the target tunnel, and the image of the inner wall is transmitted to an edge computing node in real time.

[0054] In specific implementation, the image of the inner wall of the target tunnel is collected by sensors arranged at various monitoring points of the target tunnel. The sensors include: a displacement sensor based on Beidou technology and an image acquisition device based on radar technology. The displacement sensor based on Beidou technology is arranged at key parts of the tunnel (such as the vault, arch waist, side wall, etc.) to obtain high-precision displacement data. The image acquisition device based on radar technology is installed at various monitoring points on the inner wall of the tunnel, wherein each monitoring point is a pre-set point. The image acquisition device based on radar technology can continuously collect images of the inner wall of the target tunnel during the safety monitoring process. In addition, in this application, 5G communication can be used to transmit the image of the inner wall of the tunnel to the edge computing node in real time. It should be noted that the image of the inner wall of the tunnel described in this application is a digitized image of the inner wall structure of the tunnel, which can be used to analyze the surface condition of the inner wall of the tunnel and detect defects that may affect the safety of the tunnel.

[0055] It should also be noted that the edge computing node in this application refers to a computing device deployed close to the data source (such as tunnel monitoring equipment), which is used to locally process data, reduce cloud computing pressure, and improve real-time performance. In the tunnel safety monitoring system, the edge computing node is responsible for receiving, processing, and analyzing tunnel wall images and outputting safety hazard information without transmitting all original data to a remote server or cloud.

[0056] In step 102, the edge convolution kernels in different angular directions when the edge computing node performs edge extraction on the tunnel inner wall image are determined, and the inner wall edges of the tunnel inner wall image are identified in multiple directions based on the edge convolution kernels in each angular direction to obtain the structural edge features of the target tunnel inner wall area under the current monitoring state.

[0057] In some embodiments, determining edge convolution kernels at different angles when the edge computing node performs edge extraction on the tunnel inner wall image can be implemented by the following steps, namely:

[0058] Obtain historical crack monitoring data of the target tunnel;

[0059] performing distribution prediction of crack directions of the target tunnel based on the historical crack monitoring data to obtain predicted distribution probabilities of different crack directions of the target tunnel;

[0060] Calculate the predicted distribution entropy of the target tunnel cracks through the predicted distribution probability of each different crack direction;

[0061] The edge computing node calibrates the angle direction when performing edge extraction on the tunnel inner wall image based on the predicted distribution entropy to obtain multiple different angle directions;

[0062] Based on the edge detection operator, edge convolution kernels in different angular directions are set, thereby obtaining edge convolution kernels in different angular directions when performing edge extraction on the tunnel inner wall image.

[0063] In the specific implementation, first, the historical crack monitoring data of the target tunnel is obtained. The historical crack monitoring data can be obtained by collecting a large number of crack monitoring images through the camera in the tunnel safety monitoring system, which will not be repeated here. The historical crack monitoring data includes crack monitoring images of multiple different time nodes; secondly, the crack direction of the target tunnel is predicted based on the historical crack monitoring data to obtain the predicted distribution probability of different crack directions of the target tunnel, that is, the crack direction distribution of each crack monitoring image in the historical crack data is extracted by Hough transform, and the probability distribution of different crack directions is calculated by using the histogram statistics method to obtain the predicted distribution probability of different crack directions of the target tunnel. The Hough transform (Hough Transform (HT) is a feature extraction method for detecting straight lines, arcs or specific shapes in images, and is particularly suitable for applications such as crack detection, road boundary detection, and medical image analysis. Furthermore, the predicted distribution entropy of the target tunnel crack is calculated by the predicted distribution probability of each different crack direction, that is, the predicted probability distribution of each different crack direction is input as an input parameter into a preset entropy model, and the entropy model outputs the predicted distribution entropy of the target tunnel crack, wherein the preset entropy model can adopt the Shannon entropy model. In addition, in other implementation examples, other entropy models can also be adopted, which are not limited here. Then, the edge computing node calibrates the angular direction of the tunnel inner wall image when performing edge extraction based on the predicted distribution entropy to obtain multiple different angular directions, that is, the edge computing node compares the predicted distribution entropy with the distribution entropy threshold, and when the predicted distribution entropy is less than or equal to the distribution entropy threshold, the predicted distribution entropy is calibrated. The angular directions of the tunnel inner wall image during edge extraction are calibrated as horizontal and vertical directions. When the predicted distribution entropy is greater than the distribution entropy threshold, the angular directions of the tunnel inner wall image during edge extraction are calibrated as horizontal, vertical and skew symmetric directions. The skew symmetric directions are 45° and 135°. The distribution entropy threshold is a pre-set standard predicted distribution entropy, which can be set according to actual requirements and is not limited here. Finally, edge convolution kernels in different angular directions are set based on the edge detection operator to obtain edge convolution kernels in different angular directions when the tunnel inner wall image is subjected to edge extraction, namely, edge convolution kernels in the horizontal and vertical directions are generated by the Sobel directional gradient operator, and the convolution kernels in the main diagonal direction and the sub-diagonal direction generated by the Roberts cross operator are used as edge convolution kernels in the skew symmetric directions of 45° and 135°, respectively. No further details will be given here.

[0064] It should be noted that, in this embodiment, the historical crack monitoring data represents a collection of multiple historical crack monitoring images, and the historical crack monitoring images represent images containing cracks on the inner wall of the tunnel; the predicted distribution probability in this application represents the predicted probability distribution of different crack directions of the target tunnel, and the predicted distribution probability is used to describe the possibility of cracks appearing in different directions; the predicted distribution entropy in this embodiment represents a quantitative indicator for measuring the uncertainty of the crack direction, which is used to describe the degree of discreteness of the crack direction distribution of the target tunnel, that is, whether the cracks are evenly distributed in different directions, or whether they are concentrated in a specific direction; the angular direction in this embodiment represents the direction of the convolution kernel in the image; the edge convolution kernel in this application represents a convolution operator for detecting edge information in the image, which is used to highlight the gradient change area of the image. In the edge recognition of the tunnel inner wall image, edge detection in a single direction may cause some crack information to be missing, affecting subsequent crack recognition. Therefore, by determining the edge convolution kernels in different directions, conditions can be provided for highlighting the grayscale gradient changes in the tunnel inner wall image, thereby improving the recognition accuracy of the tunnel inner wall image.

[0065] In some embodiments, reference Figure 2 As shown in FIG, this figure is an exemplary flow chart of determining structural edge features according to some embodiments of the present application. In this embodiment, the tunnel inner wall image is subjected to multi-directional inner wall edge recognition based on the edge convolution kernel in each angular direction. The structural edge features of the target tunnel inner wall area under the current monitoring state can be obtained by the following steps:

[0066] First, in step 1021, a convolution block corresponding to each pixel point in the tunnel inner wall image is extracted;

[0067] Next, in step 1022, a pixel is selected as a selected pixel, and the convolution block of the selected pixel is convolved with the edge convolution kernel in each angular direction to obtain a convolution pixel of the selected pixel.

[0068] Further, in step 1023, the convolution pixels of the remaining pixels are continuously determined;

[0069] Then, in step 1024, an edge detail map corresponding to the tunnel inner wall image is constructed based on all the convolution pixels;

[0070] Finally, in step 1025 , the structural edge features of the target tunnel inner wall area under the current monitoring state are identified from the edge detail image.

[0071] In the specific implementation, first, the convolution block corresponding to each pixel point in the tunnel inner wall image is extracted by the image processing tool OpenCV, that is, the convolution block corresponding to each pixel point in the tunnel inner wall image is obtained. The size of the convolution block is usually determined by the size of the edge convolution kernel in each angular direction, usually 3×3 in size; secondly, a pixel point is selected as the selected pixel point, and the convolution block of the selected pixel point is convolved with the edge convolution kernel in each angular direction to obtain the convolution pixel of the selected pixel point, that is: a pixel point is selected as the selected pixel point Pixel point, multiply the convolution block of the selected pixel point by the edge convolution kernel in each angular direction element by element and sum them, obtain the convolution sub-pixel of the selected pixel point in each angular direction, calculate the square sum of all the convolution sub-pixels, take the square root of the quotient of the square sum calculation result and the value 2, and obtain the convolution pixel of the selected pixel point, wherein the convolution sub-pixel represents the convolution pixel in different angular directions; further, by "convolving the convolution block of the selected pixel point with the edge convolution kernel in each angular direction, thereby obtaining the convolution pixel of the selected pixel point" Continue to determine the convolution pixels of the remaining pixels; then, construct an edge detail map corresponding to the tunnel inner wall image based on all the convolution pixels, that is, combine the convolution pixels corresponding to each pixel to obtain the edge detail map corresponding to the tunnel inner wall image; finally, identify the structural edge features of the target tunnel inner wall area under the current monitoring state from the edge detail map, that is, set an edge judgment value for the edge detail map, compare the edge judgment value with the pixel value of each pixel in the edge detail map, and use pixels corresponding to values greater than the edge judgment value as edge pixels, and then extract an edge pixel statistic, wherein the edge pixel statistic represents the total number of all edge pixels, and the quotient of the edge pixel statistic and the total number of pixels in the edge detail image is used as the structural edge feature of the target tunnel inner wall area under the current monitoring state. The edge judgment value can be set according to actual needs, for example, the average pixel value of the edge pixels in the historical tunnel inner wall images can be used as the edge judgment value. In addition, in other embodiments, other calculation methods can be used to calculate the structural edge features of the target tunnel, which are not limited here.

[0072] It should be noted that, in this embodiment, the convolution block represents a matrix area formed by a certain pixel point selected in the image and its surrounding neighborhood; in this embodiment, the convolution pixel represents the new pixel value obtained by the pixel point after the convolution operation; in this embodiment, the edge detail map represents the tunnel inner wall image with highlighted edge details, and the edge detail map contains rich edge information, which makes the edge features of the tunnel inner wall image clearer and more complete, thereby providing effective analysis data for tunnel safety monitoring; in this application, the structural edge feature represents the structural edge information in the tunnel inner wall image. The determination of the structural edge feature is a key step in accurately identifying structural problems such as cracks, peeling, and lining joints in tunnel safety monitoring. The structural edge feature effectively provides the edge information of the tunnel structure, which can improve the accuracy and stability of automated tunnel safety monitoring, thereby avoiding the loss of some identification information caused by the complex directional distribution of surface defects on the tunnel inner wall.

[0073] It should also be noted that the multi-directional inner wall edge recognition in the present application refers to the process of multi-directional identification of the inner wall edges in the inner wall image, wherein the inner wall edge recognition of the tunnel inner wall image is performed in multiple directions based on the edge convolution kernels in various angular directions, namely: extracting the convolution blocks corresponding to each pixel point in the tunnel inner wall image; selecting a pixel point as the selected pixel point, convolving the convolution block of the selected pixel point with the edge convolution kernels in various angular directions, and then obtaining the convolution pixel of the selected pixel point; continuing to determine the convolution pixels of the remaining pixels; constructing the edge detail map corresponding to the tunnel inner wall image based on all the convolution pixels; identifying the structural edge features of the target tunnel inner wall area under the current monitoring state from the edge detail map, and thus completing the multi-directional inner wall edge recognition of the tunnel inner wall image.

[0074] In step 103, the grayscale compression coefficient of the tunnel inner wall image in the brightness channel is determined, and the tunnel inner wall image is projected and compressed based on the grayscale compression coefficient combined with the grayscale statistical parameters of the brightness channel to obtain the grayscale projection characteristics of the target tunnel inner wall area under the current lighting conditions.

[0075] In some embodiments, reference Figure 3 As shown in FIG. 1 , this figure is an exemplary flow chart of determining a grayscale compression coefficient according to some embodiments of the present application. In this embodiment, determining the grayscale compression coefficient of the tunnel inner wall image in the brightness channel can be achieved by using the following steps:

[0076] First, in step 1031, the tunnel inner wall image is converted into a grayscale inner wall image;

[0077] Then, in step 1032, a brightness change rate sequence corresponding to the grayscale inner wall image is calculated;

[0078] Finally, in step 1033, the grayscale compression coefficient of the tunnel inner wall image in the brightness channel is determined through the brightness change rate sequence.

[0079] In the specific implementation, first, the tunnel inner wall image is converted into a grayscale image by RGB weighted averaging. In addition, in other embodiments, other methods can be used to convert the tunnel inner wall image into a grayscale inner wall image, which is not limited here; secondly, the brightness change rate sequence corresponding to the grayscale inner wall image is calculated, that is: the brightness change rate of each pixel in the grayscale inner wall image is calculated by the Sobel operator, and all the brightness change rates are combined in sequence to obtain the brightness change rate sequence corresponding to the grayscale inner wall image, which will not be repeated here; finally, the tunnel inner wall is determined by the brightness change rate sequence. The grayscale compression coefficient of the image in the brightness channel is as follows: the maximum brightness change rate and the minimum brightness change rate in the brightness change rate sequence are extracted to obtain the contrast scale of the tunnel inner wall image, the difference between the maximum brightness change rate and the minimum brightness change rate is calculated, and the contrast scale and the difference calculation result are used as the grayscale compression coefficient of the tunnel inner wall image in the brightness channel. The contrast scale of the tunnel inner wall image is set to 255. In addition, in other embodiments, other calculation methods can also be used to calculate the grayscale compression coefficient of the tunnel inner wall image in the brightness channel, which is not limited here.

[0080] It should be noted that the grayscale inner wall image in this embodiment represents a grayscale image converted from the original color image, and the grayscale inner wall image only contains brightness information but not color information. In tunnel monitoring, the grayscale inner wall image can show the structural features and possible defects of the inner surface of the tunnel; the brightness change rate sequence in this embodiment represents a combination of multiple brightness change rates, and the brightness change rate represents the degree of change of the brightness value of a pixel point relative to its neighboring pixel points; the grayscale compression coefficient in this application represents a parameter used to adjust the distribution of brightness values in an image. Specifically, the grayscale compression coefficient is used to control the degree of compression of the image brightness (grayscale) value, so that the brightness range of the image is more suitable for subsequent processing or analysis. By determining the grayscale compression coefficient, the details in the tunnel inner wall image can be effectively enhanced, and areas with large brightness changes, such as cracks and defects, can be highlighted.

[0081] In some embodiments, projecting and compressing the tunnel inner wall image based on the grayscale compression coefficient combined with the grayscale statistical parameters of the brightness channel to obtain the grayscale projection features of the target tunnel inner wall area under the current lighting conditions can be achieved by the following steps, namely:

[0082] Acquiring grayscale data of the tunnel inner wall image in a brightness channel;

[0083] Determine the grayscale statistical parameters of the brightness channel according to the grayscale data;

[0084] Setting an adaptive adjustment factor when projecting and compressing the tunnel inner wall image based on the grayscale statistical parameter;

[0085] Projecting and compressing the grayscale of each pixel of the tunnel inner wall image on the brightness channel using the adaptive adjustment factor and the grayscale compression coefficient to obtain a brightness projection compression image corresponding to the tunnel inner wall image;

[0086] The grayscale projection features of the target tunnel inner wall area under the current lighting conditions are extracted from the brightness projection compression map.

[0087] In the specific implementation, first, the grayscale data of the tunnel inner wall image on the brightness channel is obtained through the image processing tool OpenCV, and the grayscale data includes multiple grayscale values, that is, the grayscale value of each pixel in the tunnel inner wall image in the brightness channel; secondly, the grayscale statistical parameters of the brightness channel are determined according to the grayscale data, that is: the grayscale mean and grayscale standard deviation of the grayscale data are calculated, and the grayscale mean and the grayscale standard deviation are combined as the grayscale statistical parameters of the brightness channel; further, based on the grayscale statistical parameters, the adaptive adjustment factor of the tunnel inner wall image during projection compression is set, that is: the grayscale mean in the grayscale statistical parameters is set. The mean is summed with 0.2 times the grayscale standard deviation, and the quotient of the summation result and the grayscale average value is used as the independent variable value of the exponential function with e as the base to obtain the adaptive adjustment factor of the tunnel inner wall image during projection compression. The grayscale average value is always not equal to 0. In addition, in other embodiments, other calculation methods can be used to calculate the adaptive adjustment factor of the tunnel inner wall image during projection compression, which is not limited here; then, the grayscale of each pixel of the tunnel inner wall image on the brightness channel is projected and compressed by using the adaptive adjustment factor and the grayscale compression coefficient to obtain the brightness projection compression map corresponding to the tunnel inner wall image, that is: the power law transformation function is used. The number is used as the projection compression model, the grayscale compression coefficient is used as the product coefficient of the power-law transformation function, the adaptive adjustment factor is used as the power exponent of the power-law transformation function, the grayscale of each pixel of the tunnel inner wall image on the brightness channel is used as the input parameter of the projection compression model, and the projection compression model outputs the grayscale compression value corresponding to each pixel point, and each grayscale compression value replaces the original grayscale value to obtain the brightness projection compression map corresponding to the tunnel inner wall image. The power-law transformation function is a nonlinear transformation function widely used in the field of image processing; finally, the target tunnel inner wall area under the current lighting conditions is extracted from the brightness projection compression map. Grayscale projection features, namely: setting a high brightness judgment value of the brightness projection compression map, comparing the high brightness judgment value with the pixel value of each pixel in the brightness projection compression map, taking the pixel points corresponding to the values greater than the high brightness judgment value as high-frequency pixel points, and then extracting high-frequency pixel point statistics, the high-frequency pixel point statistics representing the total number of all high-frequency pixel points, and taking the quotient of the high-frequency pixel point statistics and the total number of pixels in the brightness projection compression map as the grayscale projection features of the target tunnel inner wall area under the current lighting conditions. In addition, other methods can also be used to calculate the grayscale projection features of the target tunnel inner wall area under the current lighting conditions, which are not limited here.

[0088] It should be noted that, in this embodiment, the grayscale statistical parameters represent a set of numerical indicators that describe the grayscale value distribution characteristics of the tunnel inner wall image in the brightness channel. The grayscale statistical parameters include the grayscale mean and the grayscale standard deviation. The grayscale statistical parameters can effectively reflect the overall brightness, contrast, grayscale change trend and other information of the image; the adaptive adjustment factor in this embodiment represents a parameter for adjusting the degree of grayscale compression; the brightness projection compression map in this embodiment represents a new image obtained by compressing and projecting the grayscale information of the tunnel inner wall image in the brightness channel. The brightness projection compression map retains the main brightness information of the tunnel inner wall structure and compresses the grayscale to enhance the structural features, providing effective support for the subsequent feature extraction and analysis of the tunnel inner wall. Data support: In this application, the grayscale projection feature represents the feature information extracted after the tunnel inner wall image is projected and compressed on the brightness channel. The grayscale projection feature characterizes the brightness distribution pattern and structural characteristics of the tunnel inner wall. In tunnel safety monitoring, the grayscale distribution of defects such as cracks and peeling is often different from the background. The grayscale projection feature can help identify these abnormal structures. In addition, the lighting conditions inside the tunnel are complex, and directly using the original brightness value may be affected by the intensity of the light source and shadows. The grayscale projection feature extracts the brightness distribution pattern in a statistical way, which can effectively identify abnormal structures that may exist in the tunnel inner wall image, avoid recognition interference caused by uneven lighting inside the tunnel, and effectively reduce the impact of local lighting changes.

[0089] It should also be noted that, in the present application, projection compression refers to the process of compressing the grayscale of an image, wherein the tunnel inner wall image is projected and compressed based on the grayscale compression coefficient combined with the grayscale statistical parameters of the brightness channel, namely: obtaining the grayscale data of the tunnel inner wall image on the brightness channel; determining the grayscale statistical parameters of the brightness channel based on the grayscale data; setting an adaptive adjustment factor for projection compression of the tunnel inner wall image based on the grayscale statistical parameters; projecting and compressing the grayscale of each pixel of the tunnel inner wall image on the brightness channel through the adaptive adjustment factor and the grayscale compression coefficient to obtain a brightness projection compression map corresponding to the tunnel inner wall image; extracting the grayscale projection features of the target tunnel inner wall area under the current lighting conditions from the brightness projection compression map, thereby completing the projection compression of the tunnel inner wall image.

[0090] In step 104, heterogeneous feature fusion is performed on the tunnel inner wall image according to the structural edge features and the grayscale projection features to obtain an identification feature map of the inner wall area of the target tunnel for safety identification.

[0091] In some embodiments, heterogeneous feature fusion is performed on the tunnel inner wall image based on the structural edge features and the grayscale projection features to obtain an identification feature map of the inner wall area of the target tunnel for safety identification. The following steps can be used, namely:

[0092] Determining the edge fusion coefficient and the brightness fusion coefficient of the inner wall of the target tunnel for safety monitoring according to the structural edge features and the grayscale projection features;

[0093] Extracting an edge detail image and a brightness projection compression image corresponding to the tunnel inner wall image;

[0094] Decomposing the edge detail image and the brightness projection compression image at multiple scales to obtain edge detail sub-images and brightness projection compression sub-images at different scale levels;

[0095] Selecting a scale level as a selected scale level, and fusing the edge detail sub-image and the brightness projection compression sub-image at the selected scale level based on the edge fusion coefficient and the brightness fusion coefficient to obtain a fused sub-image corresponding to the selected scale level;

[0096] Continue to determine the fusion subgraphs corresponding to the remaining scale levels;

[0097] Based on all the fused sub-graphs, an identification feature map of the inner wall area of the target tunnel is generated for safety identification.

[0098] In a specific implementation, first, the edge fusion coefficient and the brightness fusion coefficient of the inner wall of the target tunnel when performing safety monitoring are determined according to the structural edge features and the grayscale projection features, that is, the structural edge features and the grayscale projection features are normalized by using minimum-maximum normalization, and the normalized structural edge features and grayscale projection features are used as the edge fusion coefficient and the brightness fusion coefficient of the inner wall of the target tunnel when performing safety monitoring, wherein the normalized structural edge features and grayscale projection features have a unified dimension. In addition, in other embodiments, other calculation methods can also be used to calculate the target The edge fusion coefficient and brightness fusion coefficient of the inner wall of the tunnel are not limited here when the tunnel is monitored for safety. Secondly, the edge detail map and brightness projection compression map corresponding to the tunnel inner wall image are obtained from the tunnel safety monitoring database. Further, the edge detail map and the brightness projection compression map are multi-scale decomposed to obtain edge detail sub-maps and brightness projection compression sub-maps at different scale levels, that is: the edge detail map and the brightness projection compression map are multi-scale decomposed using a Laplacian pyramid to obtain edge detail sub-maps and brightness projection compression sub-maps at different scale levels, for example, by using a Gaussian pyramid to decompose the edge. The detail image and the brightness projection compression image are downsampled to obtain edge detail sub-images and brightness projection compression sub-images at different scale levels. In addition, other multi-scale decomposition algorithms can be used for decomposition, which is not limited here, wherein the edge detail sub-image and the brightness projection compression sub-image are images of the same scale size; further, a scale level is selected as the selected scale level, and the edge detail sub-image and the brightness projection compression sub-image at the selected scale level are fused based on the edge fusion coefficient and the brightness fusion coefficient to obtain a fused sub-image corresponding to the selected scale level, that is, the edge fusion coefficient and the brightness fusion coefficient are combined. The fusion coefficients are respectively used as weight values of the edge detail sub-image and the brightness projection compressed sub-image at the selected scale level, and weighted fusion is performed based on the Laplacian pyramid to obtain a fused sub-image corresponding to the selected scale level. This will not be described in detail here. In addition, in other embodiments, other fusion methods may be used for fusion, which are not limited here. Then, the fused sub-images corresponding to the remaining scale levels are further determined by the determination method of "fusing the edge detail sub-image and the brightness projection compressed sub-image at the selected scale level based on the edge fusion coefficient and the brightness fusion coefficient to obtain a fused sub-image corresponding to the selected scale level".Finally, pyramid reconstruction is used to inversely transform all fused sub-images to generate a recognition feature map of the inner wall area for safe identification of the target tunnel. First, starting with the fused sub-image at the highest scale level, the current fused sub-image is upsampled layer by layer (usually restored to size through bilinear interpolation or Gaussian interpolation). The upsampled image is then pixel-by-pixel superimposed with the fused sub-image at the next level to form a new intermediate image. This step is repeated until all fused sub-images at all scale levels are processed, ultimately restoring a recognition feature map with the same size as the original image.

[0099] It should be noted that, in this embodiment, the edge fusion coefficient represents a value for adjusting the fusion effect of the edge detail image, and the brightness fusion coefficient represents a value for adjusting the fusion effect of the brightness projection compression image; in this embodiment, the edge detail sub-image represents an edge information sub-image at different scale levels decomposed from the edge detail image, and the brightness projection compression sub-image represents a brightness information sub-image at different scale levels decomposed from the edge brightness projection compression image; in this embodiment, the fused sub-image represents a sub-image generated at different scale levels, specifically, the fused sub-image refers to a sub-image generated at different scale levels after weighted fusion of the edge detail sub-image and the brightness projection compression sub-image in the multi-scale decomposition process; the identification in this application The feature map represents the characteristic map of the tunnel inner wall obtained by fusing multiple features. Specifically, the recognition feature map is a feature image that can identify the safety status of the target tunnel, generated by performing heterogeneous feature fusion of the edge detail map and the brightness projection compression map on the basis of multi-scale decomposition. The recognition feature map is used for safety monitoring, crack detection or defect identification of the tunnel inner wall. This image fuses the structural edge features (provided by the edge detail map) and the grayscale projection features (provided by the brightness projection compression map), and can simultaneously retain the edge information and brightness feature information of the tunnel structure. This enables the recognition feature map to adapt to tunnel environments with different lighting, materials and damage types, improves the generalization ability of the algorithm, and effectively enhances the ability to identify tunnel defects.

[0100] It should also be noted that in the present application, heterogeneous feature fusion refers to the process of fusing multiple image features, wherein the heterogeneous feature fusion of the tunnel inner wall image is performed according to the structural edge feature and the grayscale projection feature, that is: the edge fusion coefficient and the brightness fusion coefficient of the tunnel inner wall when the target tunnel is safety monitored are determined according to the structural edge feature and the grayscale projection feature respectively; the edge detail map and the brightness projection compression map corresponding to the tunnel inner wall image are extracted; the edge detail map and the brightness projection compression map are multi-scale decomposed to obtain edge detail sub-maps and brightness projection compression sub-maps at different scale levels; a scale level is selected as the selected scale level, and the edge detail sub-map and the brightness projection compression sub-map at the selected scale level are fused based on the edge fusion coefficient and the brightness fusion coefficient to obtain the fused sub-map corresponding to the selected scale level; the fused sub-maps corresponding to the remaining scale levels are continued to be determined; the identification feature map of the inner wall area of the target tunnel is generated according to all the fused sub-maps when safety identification is performed, that is, the heterogeneous feature fusion of the tunnel inner wall image is completed.

[0101] In step 105, the safety hazard of the inner wall of the target tunnel is analyzed based on the identification feature map, and the edge computing node outputs the safety hazard information of the inner wall of the target tunnel to the cloud platform.

[0102] In some embodiments, analyzing the safety hazard of the inner wall of the target tunnel based on the identification feature map and outputting the safety hazard information of the inner wall of the target tunnel by the edge computing node to the cloud platform can be achieved by the following steps, namely:

[0103] Performing crack identification on the identification feature map to obtain crack identification features of the target tunnel;

[0104] The edge computing node compares the crack identification feature with the preset crack comparison feature and outputs the safety hazard information of the target tunnel inner wall to the cloud platform.

[0105] In specific implementation, first, Canny edge detection is used to identify cracks in the identification feature map to obtain the crack identification feature of the inner wall of the target tunnel, where the crack identification feature is the crack length. Then, the edge computing node uses Euclidean distance to compare the crack identification feature with the preset crack comparison feature. When the crack identification feature is greater than or equal to the preset crack comparison feature, it is determined that there is a safety hazard on the inner wall of the target tunnel, and the edge computing node outputs the corresponding safety hazard information (i.e., information about the current safety hazard on the inner wall of the tunnel) to the cloud platform. When the crack identification feature is less than the preset crack comparison feature, it is determined that the inner wall of the target tunnel is within a safe range and no processing is performed.

[0106] It should be noted that the preset crack comparison feature in this embodiment represents a pre-set standard crack feature, which is used to identify and evaluate the standard for tunnel cracks. It is usually established based on historical crack data, engineering specifications, experimental analysis and other information, and will not be repeated here.

[0107] In addition, another aspect of the present application, in some embodiments, the present application provides a tunnel safety monitoring system based on edge computing, the system including sensors, edge computing nodes, a cloud platform and a monitoring unit, with reference to Figure 4 , which is a schematic diagram of the structure of a monitoring unit according to some embodiments of the present application. The monitoring unit 200 includes: a collection module 201, a processing module 202 and an execution module 203, which are described as follows:

[0108] Acquisition module 201, in this application, acquisition module 201 is mainly used to collect tunnel inner wall images of the target tunnel through sensors deployed at various monitoring points of the target tunnel, and transmit the tunnel inner wall images to the edge computing node in real time;

[0109] Processing module 202, in this application, is mainly used to determine the edge convolution kernels in different angular directions when the edge computing node performs edge extraction on the tunnel inner wall image, and perform multi-directional inner wall edge recognition on the tunnel inner wall image based on the edge convolution kernels in each angular direction, to obtain the structural edge features of the target tunnel inner wall area under the current monitoring state;

[0110] The processing module 202 is further configured to determine a grayscale compression coefficient of the tunnel inner wall image in a luminance channel, and perform projection compression on the tunnel inner wall image based on the grayscale compression coefficient in combination with grayscale statistical parameters of the luminance channel to obtain grayscale projection features of a target tunnel inner wall region under current lighting conditions;

[0111] In addition, the processing module 202 is further configured to perform heterogeneous feature fusion on the tunnel inner wall image according to the structural edge features and the grayscale projection features to obtain an identification feature map of the inner wall area of the target tunnel for safety identification;

[0112] Execution module 203. In this application, execution module 203 is mainly used to analyze the safety hazards of the inner wall of the target tunnel based on the identification feature map, and the edge computing node outputs the safety hazard information of the inner wall of the target tunnel to the cloud platform.

[0113] In addition, the present application also provides a computer device, which includes a memory and a processor, the memory stores code, and the processor is configured to obtain the code and execute the above-mentioned edge computing-based tunnel safety monitoring method.

[0114] In some embodiments, reference Figure 5 , which is a schematic diagram of the structure of a computer device for implementing a tunnel safety monitoring method based on edge computing according to some embodiments of the present application. The tunnel safety monitoring method based on edge computing in the above embodiment can be achieved by Figure 5 The computer device 300 shown in FIG. 1 is implemented as shown in FIG. 1 , and the computer device 300 includes at least one processor 301 , a communication bus 302 , a memory 303 , and at least one communication interface 304 .

[0115] The processor 301 can be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices for controlling the execution of the tunnel safety monitoring method based on edge computing in this application.

[0116] The communication bus 302 may be used to transmit information between the aforementioned components.

[0117] The memory 303 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including a compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 303 may exist independently and be connected to the processor 301 via the communication bus 302. The memory 303 may also be integrated with the processor 301.

[0118] Memory 303 is used to store program code for executing the solution of the present application, and is controlled by processor 301 for execution. Processor 301 is used to execute the program code stored in memory 303. The program code may include one or more software modules. The determination of the tunnel safety monitoring method based on edge computing in the above embodiment can be implemented by processor 301 and one or more software modules in the program code in memory 303.

[0119] The communication interface 304 uses any device such as a transceiver to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.

[0120] In a specific implementation, as an embodiment, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. The processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0121] The aforementioned computer device may be a general-purpose computer device or a dedicated computer device. In a specific implementation, the computer device may be a desktop computer, a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of the present application do not limit the type of computer device.

[0122] In addition, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned tunnel safety monitoring method based on edge computing.

[0123] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0124] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A tunnel safety monitoring method based on edge computing, used for tunnel safety monitoring by a tunnel safety monitoring system, wherein the tunnel safety monitoring system comprises sensors, edge computing nodes, and a cloud platform, wherein the sensors are deployed at various monitoring points in the target tunnel, and wherein: The method comprises the following steps: Sensors installed at various monitoring points in the target tunnel collect images of the inner wall of the target tunnel, and transmit the images to the edge computing nodes in real time; Determining edge convolution kernels in different angular directions when the edge computing node performs edge extraction on the tunnel inner wall image, and performing multi-directional inner wall edge recognition on the tunnel inner wall image based on the edge convolution kernels in each angular direction, to obtain structural edge features of the target tunnel inner wall area under the current monitoring state; Determining a grayscale compression coefficient of the tunnel inner wall image in a luminance channel, and performing projection compression on the tunnel inner wall image based on the grayscale compression coefficient combined with grayscale statistical parameters of the luminance channel to obtain grayscale projection features of a target tunnel inner wall area under current lighting conditions; Performing heterogeneous feature fusion on the tunnel inner wall image according to the structural edge features and the grayscale projection features to obtain an identification feature map of the inner wall area of the target tunnel for safety identification; The safety hazards of the inner wall of the target tunnel are analyzed based on the identification feature map, and the safety hazard information of the inner wall of the target tunnel is output by the edge computing node to the cloud platform.

2. The method according to claim 1, wherein Determining edge convolution kernels at different angles when the edge computing node performs edge extraction on the tunnel inner wall image specifically includes: Obtain historical crack monitoring data of the target tunnel; performing distribution prediction of crack directions of the target tunnel based on the historical crack monitoring data to obtain predicted distribution probabilities of different crack directions of the target tunnel; Calculate the predicted distribution entropy of the target tunnel cracks through the predicted distribution probability of each different crack direction; The edge computing node calibrates the angle direction when performing edge extraction on the tunnel inner wall image based on the predicted distribution entropy to obtain multiple different angle directions; Based on the edge detection operator, edge convolution kernels in different angular directions are set, thereby obtaining edge convolution kernels in different angular directions when performing edge extraction on the tunnel inner wall image.

3. The method according to claim 1, wherein The tunnel inner wall image is subjected to multi-directional inner wall edge recognition based on the edge convolution kernel in each angular direction, and the structural edge features of the target tunnel inner wall area under the current monitoring state are obtained, specifically including: Extracting the convolution block corresponding to each pixel point in the tunnel inner wall image; Select a pixel as the selected pixel, convolve the convolution block of the selected pixel with the edge convolution kernel in each angular direction, and then obtain the convolution pixel of the selected pixel; Continue to determine the convolution pixels of the remaining pixels; Constructing an edge detail map corresponding to the tunnel inner wall image based on all convolution pixels; Structural edge features of the target tunnel inner wall area under the current monitoring state are identified from the edge detail image.

4. The method according to claim 1, wherein Determining the grayscale compression coefficient of the tunnel inner wall image in the brightness channel specifically includes: Converting the tunnel inner wall image into a grayscale inner wall image; Calculating a brightness change rate sequence corresponding to the grayscale inner wall image; The grayscale compression coefficient of the tunnel inner wall image in the brightness channel is determined by the brightness change rate sequence.

5. The method according to claim 1, wherein The tunnel inner wall image is projected and compressed based on the grayscale compression coefficient in combination with the grayscale statistical parameters of the brightness channel to obtain the grayscale projection features of the target tunnel inner wall area under the current lighting conditions, specifically including: Acquiring grayscale data of the tunnel inner wall image in a brightness channel; Determine the grayscale statistical parameters of the brightness channel according to the grayscale data; Setting an adaptive adjustment factor when projecting and compressing the tunnel inner wall image based on the grayscale statistical parameter; Projecting and compressing the grayscale of each pixel of the tunnel inner wall image on the brightness channel using the adaptive adjustment factor and the grayscale compression coefficient to obtain a brightness projection compression image corresponding to the tunnel inner wall image; The grayscale projection features of the target tunnel inner wall area under the current lighting conditions are extracted from the brightness projection compression map.

6. The method according to claim 1, wherein The sensors include: a displacement sensor based on Beidou technology and an image acquisition device based on radar technology.

7. The method according to claim 6, wherein The displacement sensors based on Beidou technology are deployed in key locations of the tunnel to obtain high-precision displacement data, including the vault, haunch, and side walls. The image acquisition devices based on radar technology are installed at various monitoring points on the inner wall of the tunnel. The image acquisition devices based on radar technology continuously capture images of the inner wall of the target tunnel during safety monitoring.

8. A tunnel safety monitoring system based on edge computing, comprising sensors, edge computing nodes, a cloud platform and a monitoring unit, wherein the sensors are deployed at various monitoring points in the target tunnel, characterized in that: The monitoring unit includes: An acquisition module is configured to collect images of the inner wall of the target tunnel through sensors deployed at various monitoring points in the target tunnel, and transmit the images to the edge computing node in real time; a processing module, configured to determine edge convolution kernels in different angular directions when the edge computing node performs edge extraction on the tunnel inner wall image, and perform multi-directional inner wall edge recognition on the tunnel inner wall image based on the edge convolution kernels in each angular direction, to obtain structural edge features of the target tunnel inner wall area under the current monitoring state; The processing module is further configured to determine a grayscale compression coefficient of the tunnel inner wall image in a luminance channel, and perform projection compression on the tunnel inner wall image based on the grayscale compression coefficient in combination with grayscale statistical parameters of the luminance channel to obtain grayscale projection features of a target tunnel inner wall area under current lighting conditions; The processing module is further configured to perform heterogeneous feature fusion on the tunnel inner wall image according to the structural edge features and the grayscale projection features to obtain an identification feature map of the inner wall area of the target tunnel for safety identification; An execution module is used to analyze the safety hazards of the inner wall of the target tunnel based on the identification feature map, and the edge computing node outputs the safety hazard information of the inner wall of the target tunnel to the cloud platform.

9. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a code, and the processor is configured to obtain the code and execute the tunnel safety monitoring method based on edge computing as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the tunnel safety monitoring method based on edge computing as described in any one of claims 1 to 7 is implemented.

Images