Method for measuring water leakage in construction process of large underground cave depot

Abstract

The invention discloses a large underground cave depot construction process leakage water measurement method, which comprises the following steps: obtaining a cave depot image in a large underground cave depot construction process, carrying out image preprocessing on the cave depot image to obtain image data, extracting a plurality of visible light images and infrared images from the image data, and calculating the leakage water in the large underground cave depot construction process; fusing the visible light image and the infrared image to obtain an image fusion model, performing feature extraction and segmentation on the image fusion model to obtain a water leakage image, and inputting the water leakage image into a trained YOLOV8 network model for training to obtain a water leakage identification model; according to the method, an evaluation index system of a water leakage model is constructed, a water leakage measurement result is obtained through evaluation, non-contact thermal imaging is carried out on leakage points, cave depot water leakage characteristics lacked when a visible light image is not good under the illumination condition are supplemented, a basis is provided for cave depot water leakage detection, and the detection accuracy is improved. The saliency characteristic and the imaging quality of the leakage water in the fused image are effectively improved, and the detection efficiency is greatly improved.

Description

Technical Field

[0001] The invention belongs to the technical field of image processing, and in particular relates to a method for measuring water leakage during the construction process of a large underground cavern. Background Art

[0002] At present, with the development of new technologies of various sensors, information fusion technology integrates the acquired data in time and space, improves the accuracy and convenience of information processing, and effectively reduces the appearance of data redundancy and misleading information. Traditional methods of water leakage detection are no longer applicable. Manual inspections are difficult to find relatively hidden water leakages, and it is impossible to determine the location of water leakage. The timeliness and accuracy of detection can no longer meet current requirements. However, since water leakage is generally detected visually, visible light cameras, as sensors for visual detection of water leakage, can accurately capture environmental changes and information on water leakage areas. They have high clarity, but are greatly affected by light. Uneven lighting conditions or water leakage with varying intensity will affect the accuracy of water leakage detection, resulting in the loss of edge features in the water leakage area. Summary of the invention

[0003] In view of this, the present invention provides a method for measuring water leakage during the construction process of a large underground cavern, performs non-contact thermal imaging of the leakage point, supplements the cavern water leakage characteristics that are missing in the visible light image when the lighting conditions are poor, and provides a basis for cavern water leakage detection. It is specifically implemented by the following technical scheme.

[0004] The present invention provides a method for measuring water leakage during the construction process of a large underground cavern, comprising the following steps:

[0005] Obtaining cavern images during the construction process of a large underground cavern, and performing image preprocessing on the cavern images to obtain image data, wherein the image preprocessing includes image registration, image enhancement and image noise reduction;

[0006] Extract multiple visible light images and infrared images from the image data, and fuse the visible light images and infrared images to obtain an image fusion model, wherein the image fusion includes: firstly denoising the collected infrared image and visible light image to eliminate the interference of image noise on the target information, then scaling the high-resolution visible light image, then grayscale processing the visible light image and infrared image, and finally inputting the pre-processed infrared and visible light images into the image fusion network;

[0007] Perform feature extraction and segmentation on the image fusion model to obtain a water leakage image, and input the water leakage image into the trained YOLOV8 network model to obtain a water leakage recognition model;

[0008] An evaluation index system for the water leakage model is constructed and the water leakage measurement results are obtained by evaluation, wherein the evaluation index system for the water leakage model includes determining the index weights and constructing the evaluation model.

[0009] As a preferred embodiment of the above technical solution, a cavern image of a large underground cavern during construction is obtained, and image preprocessing is performed on the cavern image to obtain image data, including:

[0010] Image registration includes feature detection, feature matching, spatial transformation model estimation, resampling and interpolation. Feature detection extracts common features representing the source image; feature matching establishes the correspondence between the image to be registered and the reference image by comprehensively using a variety of feature description operators, similarity measurement standards and feature space correlation; spatial transformation estimation is used to calculate the mapping function based on feature matching; resampling and interpolation are used to perform spatial geometric transformation on the image to be registered. The mapping function will produce non-integer pixel coordinates, and an appropriate difference algorithm must be used to determine the grayscale value of the coordinate point;

[0011] Image enhancement includes infrared image normalization and histogram equalization to convert cave images into digital images. The expression of linear normalization is: Where D(m,n) represents the temperature value at the coordinate (m,n) of the converted digital image, T min Represents the minimum temperature value of the temperature matrix, T max represents the maximum temperature value of the temperature matrix; the expression of nonlinear normalization processing is Where m represents the scaling factor.

[0012] As a preferred embodiment of the above technical solution, the histogram equalization process includes:

[0013] The cave image is equally divided into m×n rectangular sub-blocks, each sub-block is of M×N size, and the image is divided in the order from top to bottom and from left to right without overlapping;

[0014] Perform histogram equalization transformation on all sub-blocks to obtain a change function;

[0015] Limiting the height of the local histogram requires calculating the cutoff value of each rectangular sub-block. Based on the preset limit threshold φ, the rectangular sub-block passes the threshold φ. If the number of gray levels of the nth rectangular sub-block is K n , while the rectangular sub-blocks are segmented, the average number of pixels assigned to each gray level is c represents the restriction coefficient, and the cutoff value φ=cn is obtained x n y / K n , n x Indicates the number of pixels in the rectangular sub-block in the x direction, n yIndicates the number of pixels in the rectangular sub-block in the y direction;

[0016] According to the cutoff value, the grayscale histograms of all rectangular sub-blocks are cut off in turn and the cutoff value is redistributed to the grayscale values ​​of each level of the hole library histogram. The number of cutoff pixels assigned to each pixel level is A. c =S n / K n , S n Represents the total number of truncated pixels, then the histogram pixel distribution after redistribution is At the same time, global histogram equalization is used to process the truncated rectangular sub-blocks to improve the image contrast and achieve uniform grayscale distribution;

[0017] Taking the grayscale value of the central pixel of the rectangular sub-block as the reference, bilinear interpolation is performed on the grayscale value of each pixel in the image.

[0018] As a preferred embodiment of the above technical solution, the infrared image and the visible light image are converted into grayscale images, the infrared temperature matrix is ​​normalized to map to the grayscale image, and the visible light image is converted into a grayscale image by calculating the weighted average of the RGB three channels. The expression for the normalization of the infrared image is: Among them, g(x,y) represents the infrared thermal image of the original two-dimensional matrix, f(x,y) represents the grayscale image of the mapped two-dimensional matrix, and k represents the scaling factor; the expression of visible light grayscale processing is Gray=0.299R×0.587G×0.114B, where R, G, and B represent the values ​​of the red, green, and blue channels respectively, and 0.299 / 0.587 / 0.114 represent the weight coefficients corresponding to the R, G, and B channels respectively;

[0019] The enhanced image data is subjected to denoising. The input image data p is used as the guide image l. The texture remains similar to the guide image l. The output image q is similar to the input image p to reduce the difference between the input image data p and the output image data q. The guided filter needs to determine the linear constant coefficient a. k and linear constant coefficient b k , determine the coefficient a by minimizing the cost function and the least square method k and coefficient b k , the corresponding expression is Among them, ω k represents the local window centered at pixel k, a k Represents the input image data p in window ω k The linear constant coefficients within b k Represents the guidance image l in window ω k Linear constant coefficient, w represents the penalty coefficient to prevent linear constant coefficient a k and linear constant coefficient b k tends to infinity; E(a k ,b k ) represents the linear constant coefficient a k and linear constant coefficient b k The guided filter is represented by |d|, which represents the local window ω. k The number of pixels in μ k and σ k They represent the mean and variance of the pixels of the guide image l in the local window respectively; q i represents the output image after filtering, with the constraint q i =p i -n i .

[0020] As a preferred embodiment of the above technical solution, the visible light image and the infrared image are fused to obtain an image fusion model, including:

[0021] The spatial domain image fusion algorithm is used to perform fusion in the grayscale pixel space of the original image according to specific rules, and the weighted average algorithm is used to calculate the average value of the image pixels. The expression for generating the fused image after weighting is F(x,y)=ω k A(x,y)+ω B *B(x,y), where ω k and ω B Both represent formaldehyde coefficient, ω k +ω B =0.5.

[0022] As a preferred embodiment of the above technical solution, the YOLOV8 network model includes a Backbone layer, a Neck layer, a Head layer and a Loss layer. The Backbone layer includes a CBS convolution module, a C2f module and an SPPF module. The CBS convolution module includes a convolution layer Conv, a batch normalization layer BN and a SiLU activation function. The Neck layer uses a path feature pyramid network to construct a feature pyramid.

[0023] As a preferred embodiment of the above technical solution, the Loss layer includes classification loss, CIoU loss, DF loss and mask loss. The classification loss adopts the binary cross entropy loss function, and its calculation expression is L cls =-w n (y n ·log(σ(x n ))+(1-y n )·log(1-σ(x n ))), where y n Represents the true value of the category, x nRepresents the predicted value of the category, σ() represents the Sigmond function;

[0024] The CIoU loss function introduces the height ratio of the predicted box to the true box in the loss calculation, and the corresponding expression is Among them, Dis_2 represents the length of the connecting line between the center point of the real frame and the center point of the predicted frame; Dis_c represents the diagonal distance of the minimum circumscribed matrix between the predicted frame and the actual frame; IoU represents the intersection over union ratio of the predicted frame to the actual frame, and v represents the influencing factor of the aspect ratio;

[0025] The expression of the influence factor v of the aspect ratio is: w′ represents the actual width of the border, h′ represents the actual height of the border, w P Indicates the width of the predicted border, h P Indicates the height of the predicted bounding box.

[0026] As a preferred embodiment of the above technical solution, the DF loss function models the target bounding box distribution as a generalized distribution, so that the neural network quickly focuses on the values ​​around the target bounding box until it is continuous. The corresponding expression is: Among them, y represents the real box coordinates of the manual annotation, y i ,y i+1 Represents the values ​​of two points close to y, S i , S i+1 Represents the Sigmoid function value of the corresponding point;

[0027] When i+1 Close to y i When DFL takes the minimum value, S i With S i+1 The value of

[0028] The mask loss function is the pixel-by-pixel binary cross entropy between the true mask and the predicted mask, and the corresponding calculation expression is LOSS m =BCE(M,M gt ), where M represents the prediction mask, M gt represents the true mask.

[0029] As a preferred embodiment of the above technical solution, an evaluation index system of a water leakage model is constructed and the water leakage measurement results are obtained by evaluation, including:

[0030] Construct the initial matrix X: Assume that there are m existing evaluation objects, each of which involves n evaluation indicators, then Among them, n represents the number of evaluation indicators, m represents the number of research objects, and X mn represents the nth value of the mth evaluation indicator;

[0031] Data normalization: Indicators of different types need to be normalized so that the data can be measured with a unified standard. For positive indicators: Among them, i∈m, j∈n; for negative indicators: x ij Represents the value of the jth index of the i-th image;

[0032] The variability of the calculated indicator is x j represents the sample mean of the jth indicator, S j represents the standard deviation of the jth indicator;

[0033] Calculate the conflict of indicators, build the indicator correlation matrix, and calculate the (1-r qj ) and: Among them, r qj represents the correlation coefficient between the qth index and the jth index, A j represents the number of conflicts of the jth indicator;

[0034] Calculate the objective weight of each evaluation indicator

[0035] As a preferred embodiment of the above technical solution, the water leakage measurement result includes at least one of the water leakage area, water leakage intensity, water leakage shape or water leakage position.

[0036] The present invention provides a method for measuring water leakage during the construction process of a large underground cavern, which comprises the following steps: obtaining cavern images during the construction process of the large underground cavern, performing image preprocessing on the cavern images to obtain image data, extracting multiple visible light images and infrared images from the image data, fusing the visible light images and the infrared images to obtain an image fusion model, performing feature extraction and segmentation on the image fusion model to obtain a water leakage image, inputting the water leakage image into a trained YOLOV8 network model to obtain a water leakage recognition model, constructing an evaluation index system for the water leakage model and evaluating to obtain a water leakage measurement result, performing non-contact thermal imaging on the leakage point, supplementing the missing cavern water leakage features when the visible light image is under poor lighting conditions, providing a basis for cavern water leakage detection, effectively improving the significant features and imaging quality of the water leakage in the fused image, and greatly improving the detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 A flow chart of a method for measuring water leakage during the construction process of a large underground cavern provided by the present invention;

[0039] Figure 2 A flow chart of the evaluation index system for constructing a water leakage model provided by the present invention. DETAILED DESCRIPTION

[0040] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0041] See also Figure 1 The present invention provides a method for measuring water leakage during the construction process of a large underground cavern, comprising the following steps:

[0042] S1: Obtaining cavern images during the construction process of a large underground cavern, and performing image preprocessing on the cavern images to obtain image data, wherein the image preprocessing includes image registration, image enhancement and image noise reduction;

[0043] S2: extract multiple visible light images and infrared images from the image data, and fuse the visible light images and infrared images to obtain an image fusion model, wherein the image fusion includes: firstly denoising the collected infrared images and visible light images to eliminate the interference of image noise on the target information, then scaling the high-resolution visible light images, then grayscale processing the visible light images and infrared images, and finally inputting the pre-processed infrared and visible light images into the image fusion network;

[0044] S3: Extract features and segment the image fusion model to obtain a water leakage image, and input the water leakage image into the trained YOLOV8 network model to obtain a water leakage recognition model;

[0045] S4: construct an evaluation index system for the water leakage model and evaluate the water leakage measurement results, wherein constructing the evaluation index system for the water leakage model includes determining the index weights and constructing the evaluation model.

[0046] In this embodiment, a cave image of a large underground cave construction process is obtained, and the cave image is preprocessed to obtain image data, including: image registration including feature detection, feature matching, spatial transformation model estimation, resampling and interpolation, feature detection extracts common features representing the source image; feature matching establishes the correspondence between the image to be registered and the reference image by comprehensively using a variety of feature description operators, similarity measurement standards and feature space correlation; spatial transformation estimation is used to calculate the mapping function based on feature matching; resampling and interpolation are used to perform spatial geometric transformation on the image to be registered, and the mapping function will produce non-integer pixel coordinates, and an appropriate difference algorithm is required to determine the grayscale value of the coordinate point; image enhancement includes infrared image normalization and histogram equalization to convert the cave image into a digital image, and the expression of linear normalization is: Where D(m,n) represents the temperature value at the coordinate (m,n) of the converted digital image, T min Represents the minimum temperature value of the temperature matrix, T max represents the maximum temperature value of the temperature matrix; the expression of nonlinear normalization processing is Where m represents the scaling factor.

[0047] It should be noted that the histogram equalization process includes: dividing the cave image into m×n rectangular sub-blocks in equal amounts, each sub-block has a size of M×N, and the image is divided in the order from top to bottom and from left to right without overlapping; performing histogram equalization transformation on all sub-blocks to obtain a change function; limiting the height of the local histogram requires calculating the cutoff value of each rectangular sub-block, and based on the preset limit threshold φ, the rectangular sub-block passes the threshold φ, if the number of gray levels of the nth rectangular sub-block is K n , while the rectangular sub-blocks are segmented, the average number of pixels assigned to each gray level is c represents the restriction coefficient, and the cutoff value φ=cn is obtained x n y / K n , n x Indicates the number of pixels in the rectangular sub-block in the x direction, n y Indicates the number of pixels in the rectangular sub-block in the y direction; the grayscale histograms of all rectangular sub-blocks are truncated in sequence according to the cutoff value and the cutoff value is redistributed to the grayscale values ​​of each level of the hole library histogram. The number of truncated pixels allocated to each pixel level is A c =S n / K n , S n Represents the total number of truncated pixels, then the histogram pixel distribution after redistribution is At the same time, global histogram equalization is used to process the truncated rectangular sub-blocks to improve the image contrast and achieve uniform grayscale distribution; the grayscale value of each pixel in the image is bilinearly interpolated with the grayscale value of the central pixel of the rectangular sub-block as a reference.

[0048] Specifically, due to the spatial differences between images captured by infrared and visible light sensors at different times and viewing angles, in order to ensure the quality of the fused image, before image fusion, it is necessary to select appropriate registration techniques for specific applications and image characteristics. The image data generated by the infrared imager contains temperature information, which needs to be converted into digital images to facilitate the subsequent image annotation and training, ensuring that the temperature data maintains its original information and characteristics during visualization. Linear normalization can convert the two-dimensional temperature matrix into a digital image, and nonlinear normalization can also achieve the conversion of digital images. In order to solve the problem of excessive enhancement and excessive noise amplification in local areas of leakage images, contrast-limited adaptive histogram equalization can enhance contrast and suppress noise by setting the upper limit of the local histogram peak. Due to the differences in the corresponding relationship between the pixels of the rectangular sub-block truncation equalization, the grayscale value of each pixel in the image is bilinearly interpolated with the central pixel grayscale value of the rectangular sub-block as the benchmark, which can improve the image contrast and achieve a more uniform grayscale distribution.

[0049] It should be understood that by acquiring cave images during the construction process of a large underground cave, and performing image preprocessing on the cave images to obtain image data, multiple visible light images and infrared images are extracted from the image data, and the visible light images and infrared images are fused to obtain an image fusion model, feature extraction and segmentation are performed on the image fusion model to obtain a water leakage image, the water leakage image is input into the trained YOLOV8 network model to train a water leakage recognition model, an evaluation index system for the water leakage model is constructed and evaluated to obtain water leakage measurement results, non-contact thermal imaging is performed on the leakage points, and the missing cave water leakage features of the visible light image under poor lighting conditions are supplemented, which provides a basis for cave water leakage detection, effectively improves the significant features and imaging quality of water leakage in the fused image, and greatly improves the efficiency of water leakage detection.

[0050] Optionally, the infrared image and the visible light image are converted into grayscale images. Normalization maps the infrared temperature matrix to the grayscale image. The visible light image is converted into a grayscale image by calculating the weighted average of the RGB three channels. The expression for the normalization of the infrared image is: Among them, g(x,y) represents the infrared thermal image of the original two-dimensional matrix, f(x,y) represents the grayscale image of the mapped two-dimensional matrix, and k represents the scaling factor; the expression of visible light grayscale processing is Gray=0.299R×0.587G×0.114B, where R, G, and B represent the values ​​of the red, green, and blue channels respectively, and 0.299 / 0.587 / 0.114 represent the weight coefficients corresponding to the R, G, and B channels respectively;

[0051] The enhanced image data is subjected to denoising. The input image data p is used as the guide image l. The texture remains similar to the guide image l. The output image q is similar to the input image p to reduce the difference between the input image data p and the output image data q. The guided filter needs to determine the linear constant coefficient a. k and linear constant coefficient b k , determine the coefficient a by minimizing the cost function and the least square method k and coefficient b k , the corresponding expression is Among them, ω k represents the local window centered at pixel k, a k Represents the input image data p in window ω k The linear constant coefficients within b k Represents the guidance image l in window ω k Linear constant coefficient, w represents the penalty coefficient to prevent linear constant coefficient a k and linear constant coefficient b k tends to infinity; E(a k ,b k ) represents the linear constant coefficient a k and linear constant coefficient b k The guided filter is represented by |d|, which represents the local window ω. k The number of pixels in μ k and σ k They represent the mean and variance of the pixels of the guide image l in the local window respectively; q i represents the output image after filtering, with the constraint q i =p i -n i .

[0052] In this embodiment, the visible light image and the infrared image are fused to obtain an image fusion model, including: using a spatial domain image fusion algorithm to perform fusion on the grayscale pixel space of the original image according to a specific rule, using a weighted average algorithm to calculate the average value of the image pixels, and the expression for generating the fused image after weighting is F(x, y) = ω k A(x,y)+ωB *B(x,y), where ω k and ω B Both represent formaldehyde coefficient, ω k +ω B =0.5. The YOLOV8 network model includes Backbone layer, Neck layer, Head layer and Loss layer. The Backbone layer includes CBS convolution module, C2f module and SPPF module. The CBS convolution module includes convolution layer Conv, batch normalization layer BN and SiLU activation function. The Neck layer uses a path feature pyramid network to construct a feature pyramid. The Loss layer includes classification loss, CIoU loss, DF loss and mask loss. The classification loss uses a binary cross entropy loss function, and its calculation expression is L cls =-w n (y n ·log(σ(x n ))+(1-y n )·log(1-σ(x n ))), where y n Represents the true value of the category, x n Represents the predicted value of the category, σ() represents the Sigmond function; the CIoU loss function introduces the height ratio of the predicted box to the true box in the loss calculation, and the corresponding expression is Among them, Dis_2 represents the length of the connecting line between the center point of the real frame and the center point of the predicted frame; Dis_c represents the diagonal distance of the minimum circumscribed matrix between the predicted frame and the actual frame; IoU represents the intersection over union ratio of the predicted frame and the actual frame, and v represents the influencing factor of the aspect ratio; the expression of the influencing factor v of the aspect ratio is w′ represents the actual width of the border, h′ represents the actual height of the border, w P Indicates the width of the predicted border, h P Indicates the height of the predicted bounding box.

[0053] It should be noted that the DF loss function models the target bounding box distribution as a generalized distribution, so that the neural network quickly focuses on the values ​​around the target bounding box until it is continuous. The corresponding expression is Among them, y represents the real box coordinates of the manual annotation, y i ,y i+1 Represents the values ​​of two points close to y, S i , S i+1 Represents the Sigmoid function value of the corresponding point;

[0054] When i+1 Close to y i When DFL takes the minimum value, Si With S i+1 The value of

[0055] The mask loss function is the pixel-by-pixel binary cross entropy between the true mask and the predicted mask, and the corresponding calculation expression is LOSS m =BCE(M,M gt ), where M represents the prediction mask, M gt represents the true mask.

[0056] Specifically, the convolution operation is performed by traversing the image from the upper left to the lower right with a specific step size through a preset convolution kernel of a fixed size. In this process, the weight value on the convolution kernel is multiplied by the pixel value covered and summed to obtain the convolution kernel calculation result. The matrix obtained after the traversal is the feature map. During the model training process, the update of parameters will change the output distribution of the later layer network, and this change will affect the subsequent iteration process, resulting in a slowdown in model training. The batch normalization layer calculates the mean and variance of the minimum batch samples to standardize the input data to ensure that the input data distribution of each layer of the network is stable, thereby accelerating the model gradient convergence speed and preventing the gradient from being small and exploding. When constructing a neural network model, if the feature information between layers is transmitted linearly, the model's ability to fit multidimensional data will be affected. Therefore, it is necessary to use activation functions to establish connections in different networks. By incorporating nonlinear properties into the neural network model, the model's description ability is enhanced. In the YOLOv8 model, the activation function is the SiLU function.

[0057] Optionally, see Figure 2 , build an evaluation index system for the water leakage model and evaluate the water leakage measurement results, including:

[0058] Construct the initial matrix X: Assume that there are m existing evaluation objects, each of which involves n evaluation indicators, then Among them, n represents the number of evaluation indicators, m represents the number of research objects, and X mn represents the nth value of the mth evaluation indicator;

[0059] Data normalization: Indicators of different types need to be normalized so that the data can be measured with a unified standard. For positive indicators: Among them, i∈m, j∈n; for negative indicators: x ij Represents the value of the jth index of the i-th image;

[0060] The variability of the calculated indicator is x j represents the sample mean of the jth indicator, S jrepresents the standard deviation of the jth indicator;

[0061] Calculate the conflict of indicators, build the indicator correlation matrix, and calculate the (1-r qj ) and: Among them, r qj represents the correlation coefficient between the qth index and the jth index, A j represents the number of conflicts of the jth indicator;

[0062] Calculate the objective weight of each evaluation indicator

[0063] In this embodiment, the water leakage measurement result includes at least one of the water leakage area, water leakage intensity, water leakage shape or water leakage position. In order to more efficiently utilize the water leakage identification result, i.e., the measurement result, and improve the effectiveness of the cavern water leakage severity assessment, a water leakage severity assessment model is constructed, which mainly includes two parts: establishing a severity assessment index system and constructing a water leakage severity assessment model. In establishing the index system, firstly, the influencing factors of the water leakage severity are analyzed according to the water leakage identification results, and then the adverse effects of various factors on the cavern are comprehensively analyzed to determine the classification criteria of different severity levels, thereby determining the assessment index and establishing the assessment index system.

[0064] It should be noted that the seepage area has a direct impact on the structural safety and stability of the cavern and is a key factor in assessing the severity of seepage. The size of the seepage area can reflect the scope and severity of the seepage problem. A larger seepage area means that more cavern areas are eroded, resulting in more extensive material damage and structural weakening, which not only increases the risk of local or overall structural failure of the cavern, but also may pose a threat to the long-term stability of the cavern. The seepage intensity is usually the water flow through a specific area per unit time, measured in square meters per hour. The temperature of the seepage area in the infrared image is used as the characterization value of the seepage intensity, and the seepage intensity is reflected based on the temperature change. The seepage shape plays a key role in assessing the severity of cavern seepage, which directly affects the mode and scope of the seepage on the cavern structure. Different seepage shapes mean different water flow distributions and erosion patterns, which have different degrees of impact on the structural stability and safety of the cavern.

[0065] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limiting, and thus other examples of the exemplary embodiments may have different values.

[0066] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0067] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for measuring water leakage during the construction of a large underground cavern, characterized in that: The following steps are involved: Obtaining cavern images during the construction process of a large underground cavern, and performing image preprocessing on the cavern images to obtain image data, wherein the image preprocessing includes image registration, image enhancement and image noise reduction; Extract multiple visible light images and infrared images from the image data, and fuse the visible light images and infrared images to obtain an image fusion model, wherein the image fusion includes: firstly denoising the collected infrared image and visible light image to eliminate the interference of image noise on the target information, then scaling the high-resolution visible light image, then grayscale processing the visible light image and infrared image, and finally inputting the pre-processed infrared and visible light images into the image fusion network; Perform feature extraction and segmentation on the image fusion model to obtain a water leakage image, and input the water leakage image into the trained YOLOV8 network model to obtain a water leakage recognition model; An evaluation index system for the water leakage model is constructed and the water leakage measurement results are obtained by evaluation, wherein the evaluation index system for the water leakage model includes determining the index weights and constructing the evaluation model.

2. The method for measuring water leakage during construction of a large underground cavern according to claim 1 is characterized in that: Obtain cavern images during the construction process of a large underground cavern, and perform image preprocessing on the cavern images to obtain image data, including: Image registration includes feature detection, feature matching, spatial transformation model estimation, resampling and interpolation. Feature detection extracts common features representing the source image; feature matching establishes the correspondence between the image to be registered and the reference image by comprehensively using a variety of feature description operators, similarity measurement standards and feature space correlation; spatial transformation estimation is used to calculate the mapping function based on feature matching; resampling and interpolation are used to perform spatial geometric transformation on the image to be registered. The mapping function will produce non-integer pixel coordinates, and an appropriate difference algorithm must be used to determine the grayscale value of the coordinate point; Image enhancement includes infrared image normalization and histogram equalization to convert cave images into digital images. The expression of linear normalization is: Where D(m,n) represents the temperature value at the coordinate (m,n) of the converted digital image, T min Represents the minimum temperature value of the temperature matrix, T max represents the maximum temperature value of the temperature matrix; the expression of nonlinear normalization processing is Where m represents the scaling factor.

3. The method for measuring water leakage during construction of a large underground cavern according to claim 2, characterized in that: The histogram equalization process includes: The cave image is equally divided into m×n rectangular sub-blocks, each sub-block is of M×N size, and the image is divided in the order from top to bottom and from left to right without overlapping; Perform histogram equalization transformation on all sub-blocks to obtain a change function; Limiting the height of the local histogram requires calculating the cutoff value of each rectangular sub-block. Based on the preset limit threshold φ, the rectangular sub-block passes the threshold φ. If the number of gray levels of the nth rectangular sub-block is K n , while the rectangular sub-blocks are segmented, the average number of pixels assigned to each gray level is c represents the restriction coefficient, and the cutoff value φ=cn is obtained x n y / K n , n x Indicates the number of pixels in the rectangular sub-block in the x direction, n y Indicates the number of pixels in the rectangular sub-block in the y direction; According to the cutoff value, the grayscale histograms of all rectangular sub-blocks are cut off in turn and the cutoff value is redistributed to the grayscale values ​​of each level of the hole library histogram. The number of cutoff pixels assigned to each pixel level is A. c =S n / K n , S n Represents the total number of truncated pixels, then the histogram pixel distribution after redistribution is At the same time, global histogram equalization is used to process the truncated rectangular sub-blocks to improve the image contrast and achieve uniform grayscale distribution; Taking the grayscale value of the central pixel of the rectangular sub-block as the reference, bilinear interpolation is performed on the grayscale value of each pixel in the image.

4. The method for measuring water leakage during construction of a large underground cavern according to claim 3 is characterized in that: Also includes: The infrared image and visible light image are converted into grayscale images. Normalization maps the infrared temperature matrix to the grayscale image. The visible light image calculates the weighted average of the RGB three channels and converts it into a grayscale image. The expression for the normalization of the infrared image is: Among them, g(x,y) represents the infrared thermal image of the original two-dimensional matrix, f(x,y) represents the grayscale image of the mapped two-dimensional matrix, and k represents the scaling factor; the expression of visible light grayscale processing is Gray=0.299R×0.587G×0.114B, where R, G, and B represent the values ​​of the red, green, and blue channels respectively, and 0.299 / 0.587 / 0.114 represent the weight coefficients corresponding to the R, G, and B channels respectively; The enhanced image data is subjected to denoising. The input image data p is used as the guide image l. The texture remains similar to the guide image l. The output image q is similar to the input image p to reduce the difference between the input image data p and the output image data q. The guided filter needs to determine the linear constant coefficient a. k and linear constant coefficient b k , determine the coefficient a by minimizing the cost function and the least square method k and coefficient b k , the corresponding expression is Among them, ω k represents the local window centered at pixel k, a k Represents the input image data p in window ω k The linear constant coefficients within b k Represents the guide image l in window ω k Linear constant coefficient, w represents the penalty coefficient to prevent linear constant coefficient a k and linear constant coefficient b k tends to infinity; E(a k ,b k ) represents the linear constant coefficient a k and linear constant coefficient b k The guided filter is represented by |d|, which represents the local window ω. k The number of pixels in μ k and σ k They represent the mean and variance of the pixels of the guide image l in the local window respectively; q i represents the output image after filtering, with the constraint q i =p i -n i .

5. The method for measuring water leakage during construction of a large underground cavern according to claim 1, characterized in that: The image fusion model is obtained by fusing the visible light image and the infrared image, including: The spatial domain image fusion algorithm is used to perform fusion in the grayscale pixel space of the original image according to specific rules, and the weighted average algorithm is used to calculate the average value of the image pixels. The expression for generating the fused image after weighting is F(x,y)=ω k A(x,y)+ω B *B(x,y), where ω k and ω B Both represent formaldehyde coefficient, ω k +ω B =0.

5.

6. The method for measuring water leakage during construction of a large underground cavern according to claim 1, characterized in that: The YOLOV8 network model includes Backbone layer, Neck layer, Head layer and Loss layer. The Backbone layer includes CBS convolution module, C2f module and SPPF module. The CBS convolution module includes convolution layer Conv, batch normalization layer BN and SiLU activation function. The Neck layer uses a path feature pyramid network to construct a feature pyramid.

7. The method for measuring water leakage during construction of a large underground cavern according to claim 6, characterized in that: The Loss layer includes classification loss, CIoU loss, DF loss and mask loss. The classification loss uses the binary cross entropy loss function, and its calculation expression is L cls =-w n (y n ·log(σ(x n ))+(1-y n )·log(1-σ(x n ))), where y n Represents the true value of the category, x n Represents the predicted value of the category, σ() represents the Sigmond function; The CIoU loss function introduces the height ratio of the predicted box to the true box in the loss calculation, and the corresponding expression is Among them, Dis_2 represents the length of the connecting line between the center point of the real frame and the center point of the predicted frame; Dis_c represents the diagonal distance of the minimum circumscribed matrix between the predicted frame and the actual frame; IoU represents the intersection over union ratio of the predicted frame to the actual frame, and v represents the influencing factor of the aspect ratio; The expression of the influence factor v of the aspect ratio is: w′ represents the actual width of the border, h′ represents the actual height of the border, w P Indicates the width of the predicted border, h P Indicates the height of the predicted bounding box.

8. The method for measuring water leakage during construction of a large underground cavern according to claim 7, characterized in that: The DF loss function models the target bounding box distribution as a generalized distribution, allowing the neural network to quickly focus on the values ​​around the target bounding box until it is continuous. The corresponding expression is Among them, y represents the real box coordinates of the manual annotation, y i ,y i+1 Represents the value of two points close to y, S i , S i+1 Represents the Sigmoid function value of the corresponding point; When i+1 Close to y i When DFL takes the minimum value, S i With S i+1 The value of The mask loss function is the pixel-by-pixel binary cross entropy between the true mask and the predicted mask, and the corresponding calculation expression is LOSS m =BCE(M,M gt ), where M represents the prediction mask, M gt represents the true mask.

9. The method for measuring water leakage during construction of a large underground cavern according to claim 1, characterized in that: Construct an evaluation index system for the water leakage model and evaluate the water leakage measurement results, including: Construct the initial matrix X: Assume that there are m existing evaluation objects, each of which involves n evaluation indicators, then Among them, n represents the number of evaluation indicators, m represents the number of research objects, and X mn represents the nth value of the mth evaluation indicator; Data normalization: Indicators of different types need to be normalized so that the data can be measured with a unified standard. For positive indicators: Among them, i∈m, j∈n; for negative indicators: x ij Represents the value of the jth index of the i-th image; The variability of the calculated indicator is x j represents the sample mean of the jth indicator, S j represents the standard deviation of the jth indicator; Calculate the conflict of indicators, build the indicator correlation matrix, and calculate the (1-r qj ) and: Among them, r qj represents the correlation coefficient between the qth index and the jth index, A j represents the number of conflicts of the jth indicator; Calculate the objective weight of each evaluation indicator 10. The method for measuring water leakage during construction of a large underground cavern according to claim 9, characterized in that: The water leakage measurement result includes at least one of the water leakage area, water leakage intensity, water leakage shape or water leakage position.

Images