A method for measuring leakage water in a large underground vault construction process

Abstract

The application discloses a large underground warehouse construction process leakage water measuring method, which obtains the warehouse image of the large underground warehouse construction process, carries out image preprocessing on the warehouse image to obtain image data, extracts a plurality of visible light images and infrared images from the image data, fuses the visible light images and the infrared images to obtain an image fusion model, extracts features and segments the image fusion model to obtain a leakage water image, inputs the leakage water image into a trained YOLOV8 network model to train a leakage water recognition model, constructs an evaluation index system of the leakage water model and evaluates to obtain a leakage water measuring result, carries out non-contact thermal imaging on the leakage point, supplements the missing warehouse leakage water features of the visible light image under poor illumination conditions, provides a basis for warehouse leakage water detection, effectively improves the saliency features and imaging quality of the leakage water in the fused image, and greatly improves the detection efficiency.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to a method for measuring water leakage during the construction of large underground caverns. Background Technology

[0002] Currently, the development of various new sensor technologies and information fusion technologies enable the spatiotemporal integration of acquired data, improving the accuracy and convenience of information processing and effectively reducing data redundancy and misleading information. Traditional methods of leak detection are no longer applicable. Manual inspection is insufficient to detect relatively hidden leaks and cannot accurately pinpoint the location of leaks; the timeliness and accuracy of detection no longer meet current requirements. However, since leaks are generally detected visually, visible light cameras, as sensors for visual leak detection, can accurately capture environmental changes and leak area information with high clarity. However, they are significantly affected by lighting conditions; uneven or varying lighting conditions can affect the accuracy of leak detection, leading to the loss of edge features of the leak area. Summary of the Invention

[0003] In view of this, the present invention provides a method for measuring water leakage during the construction of large underground caverns. The method uses non-contact thermal imaging to supplement the missing cavern water leakage characteristics in visible light images when the lighting conditions are poor, thus providing a basis for cavern water leakage detection. The specific technical solution adopted is as follows.

[0004] This invention provides a method for measuring water leakage during the construction of large underground caverns, comprising the following steps: Images of a large underground cavern during its construction process are acquired, and these images are preprocessed to obtain image data. The image preprocessing includes image registration, image enhancement, and image noise reduction. Multiple visible light and infrared images are extracted from image data, and the visible light and infrared images are fused to obtain an image fusion model. The image fusion includes: first, denoising the acquired infrared and visible light images to eliminate the interference of image noise on the target information; second, scaling the high-resolution visible light image; third, grayscale processing of the visible light and infrared images; and finally, inputting the preprocessed infrared and visible light images into the image fusion network. The image fusion model is used to extract and segment features to obtain the leakage image. The leakage image is then input into the trained YOLOV8 network model to train the leakage recognition model. An evaluation index system for a leakage model is constructed, and the leakage measurement results are evaluated. The construction of the evaluation index system for the leakage model includes determining the index weights and constructing the evaluation model.

[0005] As a preferred embodiment of the above technical solution, images of the underground cavern during its construction process are acquired, and image preprocessing is performed 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 images. Feature matching establishes the correspondence between the image to be registered and the reference image by comprehensively using various feature description operators, similarity measurement standards, and feature spatial correlation. Spatial transformation estimation is used to calculate the mapping function parameters based on feature matching. Resampling and interpolation are used when performing spatial geometric transformations on the image to be registered. The mapping function will generate non-integer pixel coordinates, and interpolation algorithms are needed to determine the gray values ​​of the coordinate points. Image enhancement includes infrared image normalization and histogram equalization, converting cave images into digital images. The expression for linear normalization is: ,in, Represents the coordinates of the converted digital image. Temperature value at that location, This represents the minimum temperature value in the temperature matrix. This represents the maximum temperature value in the temperature matrix; the expression for nonlinear normalization is... , where m represents the scaling factor.

[0006] As a preferred embodiment of the above technical solution, the histogram equalization process includes: The cave image is divided into equal parts. Rectangular sub-blocks, each sub-block having a size of [size missing]. The image is segmented in the order from top to bottom and from left to right, with no overlap between segments; Perform histogram equalization on all sub-blocks to obtain the transformation function; Limiting the height of a local histogram requires calculating the cutoff value for each rectangular sub-block, based on a pre-set limit threshold. The rectangular sub-blocks are passed through a threshold. If the number of gray levels of the nth rectangular sub-block is During the rectangular sub-block segmentation, the average number of pixels allocated to each gray level is: 'c' represents the restriction coefficient, and the cutoff value is obtained. , This indicates the number of pixels in the rectangular sub-block in the x-direction. This indicates the number of pixels in the rectangular sub-block along the y-direction; Based on the truncation value, the grayscale histograms of all rectangular sub-blocks are truncated sequentially, and the truncation value is redistributed to the grayscale values ​​of each level of the hole-bank histogram. The number of truncated pixels assigned to each pixel level is [number missing]. , Let represent the total number of truncated pixels. Then, the pixel distribution of the histogram after pixel redistribution is: Meanwhile, global histogram equalization is used to process the truncated rectangular sub-blocks to improve image contrast and achieve uniform grayscale distribution. Using the gray value of the center pixel of the rectangular sub-block as a reference, bilinear interpolation is performed on the gray value of each pixel in the image.

[0007] As a preferred embodiment of the above technical solution, infrared and visible light images are converted into grayscale images. Normalization maps the temperature matrix of the infrared image to the grayscale image. The weighted average of the RGB channels of the visible light image is calculated and converted into a grayscale image. The expression for the normalization processing of the infrared image is as follows: ,in, This represents the infrared thermal image of the original two-dimensional matrix. This represents a grayscale image of a mapped two-dimensional matrix, where k represents the scaling factor; the expression for visible light grayscale processing is... 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 then subjected to noise reduction processing. (Input image data) p As a guide image l Texture preservation and guiding image l resemblance ， Output image q Then, similar to the input image p To reduce the size of the input image data p and output image data q Differences between ， Guided filtering requires determining linear constant coefficients. and linear constant coefficients The coefficients are determined by minimizing the cost function and using the least squares method. Sum of coefficients The corresponding expression is ,in, This represents a local window centered at pixel k. Indicates input image data p In the window Linear constant coefficients within, ; Indicates a guide image l In the window Linear constant coefficients w represents the penalty coefficient, preventing linear constant coefficients. and linear constant coefficients It tends towards infinity; Represents linear constant coefficients and linear constant coefficients The guided filter representation; Represents a local window The number of pixels within; and These represent the guide images. l Mean and variance of pixels within a local window; This represents the output image obtained after filtering, with the following constraints: .

[0008] As a preferred embodiment of the above technical solution, an image fusion model is obtained by fusing visible light images and infrared images, including: A spatial domain image fusion algorithm is used to perform fusion in the grayscale pixel space of the original image according to specific rules. A weighted average algorithm is then used to calculate the average pixel value of the image. The expression for the weighted fused image is as follows: ,in, and All of these represent weighting coefficients. .

[0009] 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 convolutional module, a C2f module, and an SPPF module. The CBS convolutional module includes a Convolutional 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.

[0010] 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, the calculation expression of which is: ,in, This represents the true value of the category. This represents the predicted value for that category. Represents the Sigmoid function; The CIoU loss function incorporates the height ratio of the predicted bounding box to the ground truth bounding box in the loss calculation; the corresponding expression is: ,in, This represents the length of the line connecting the center point of the ground truth bounding box and the center point of the predicted bounding box. The diagonal distance between the minimum outermost matrix of the predicted bounding box and the actual bounding box is represented; IoU represents the intersection-union ratio of the predicted bounding box and the actual bounding box; and v represents the aspect ratio influence factor. The expression for the aspect ratio influence factor v is: , This indicates the actual width of the border. This indicates the actual height of the border. This indicates the predicted border width. This indicates the height of the predicted border.

[0011] As a preferred embodiment of the above technical solution, the DF loss function models the target object bounding box distribution as a generalized distribution, enabling the neural network to quickly focus on the values ​​around continuous positions of the target bounding box. The corresponding expression is: Where y represents the coordinates of the manually labeled true bounding box. , This represents the values ​​of two points adjacent to y. , This represents the Sigmoid function value at the corresponding point; when near When the DFL loss value decreases, the DFL loss value decreases; when the DFL reaches its minimum value, and The value is ; The mask loss function is the pixel-wise binary cross-entropy between the ground truth mask and the predicted mask, and the corresponding calculation expression is: ,in, Indicates the prediction mask, This represents the actual mask.

[0012] As a preferred embodiment of the above technical solution, an evaluation index system for the leakage model is constructed and the leakage measurement results are evaluated, including: Constructing the initial matrix X: Given m existing evaluation objects, each involving n evaluation indicators, then... Where n represents the number of evaluation indicators, and m represents the number of research subjects. This 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 using a unified standard. For positive indicators: ,in, , For negative indicators: , This represents the value of the j-th index in the i-th image; The variability of the calculated index is , , This represents the sample mean of the j-th indicator. This represents the standard deviation of the j-th indicator; Calculate the conflict of indicators, construct an indicator correlation matrix, and calculate the correlation coefficients in each column of indicators. The sum of: , ,in, This represents the correlation coefficient between the q-th indicator and the j-th indicator. This represents the number of conflicts for the j-th indicator; Calculate the objective weight of each evaluation indicator .

[0013] As a preferred embodiment of the above technical solution, the leakage measurement results include at least one of the following: leakage area, leakage intensity, leakage shape, or leakage location.

[0014] This invention provides a method for measuring water leakage during the construction of a large underground cavern. The method involves acquiring images of the cavern during construction, preprocessing these images to obtain image data, extracting multiple visible light and infrared images from the image data, fusing these images to obtain an image fusion model, extracting and segmenting features from the image fusion model to obtain water leakage images, and then training a water leakage recognition model using a pre-trained YOLOv8 network model. An evaluation index system for the water leakage model is constructed and evaluated to obtain the water leakage measurement results. Non-contact thermal imaging is then performed on the leakage points to supplement the missing water leakage features in visible light images under poor lighting conditions. This provides a basis for detecting water leakage in caverns, effectively improving the saliency features and imaging quality of the fused images, and significantly increasing detection efficiency. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 Flowchart of the method for measuring water leakage during the construction of a large underground cavern provided by the present invention; Figure 2 The flowchart of the evaluation index system for constructing a leakage water model provided by the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0018] See Figure 1 This invention provides a method for measuring water leakage during the construction of large underground caverns, comprising the following steps: S1: Acquire images of the underground cavern during its construction process, and perform image preprocessing on the images to obtain image data. Image preprocessing includes image registration, image enhancement, and image noise reduction. S2: Extract multiple visible light and infrared images from the image data, and fuse the visible light and infrared images to obtain an image fusion model. The image fusion includes: first, denoising the acquired infrared and visible light images to eliminate the interference of image noise on the target information; second, scaling the high-resolution visible light image; third, grayscale processing the visible light and infrared images; and finally, inputting the preprocessed infrared and visible light images into the image fusion network. S3: Extract and segment features from the image fusion model to obtain the leakage image, and input the leakage image into the trained YOLOV8 network model to train the leakage recognition model. S4: Construct an evaluation index system for the leakage model and evaluate the leakage measurement results. The construction of the evaluation index system for the leakage model includes determining the index weights and constructing the evaluation model.

[0019] In this embodiment, images of a large underground cavern during its construction are acquired, and these images are preprocessed to obtain image data. This preprocessing includes: image registration, which involves feature detection, feature matching, spatial transformation model estimation, resampling, and interpolation. Feature detection extracts common features representing the source images. Feature matching establishes a correspondence between the image to be registered and the reference image by comprehensively using various feature descriptors, similarity measurement standards, and feature spatial correlation. Spatial transformation estimation is used to calculate the mapping function based on feature matching. Resampling and interpolation are used when performing spatial geometric transformations on the image to be registered; since the mapping function will generate non-integer pixel coordinates, an appropriate interpolation algorithm is needed to determine the grayscale value of the coordinate points. Image enhancement includes infrared image normalization and histogram equalization, converting the cavern image into a digital image. The expression for linear normalization is: ,in, Represents the coordinates of the converted digital image. Temperature value at that location, This represents the minimum temperature value in the temperature matrix. This represents the maximum temperature value in the temperature matrix; the expression for nonlinear normalization is... , where m represents the scaling factor.

[0020] It should be noted that the histogram equalization process includes: dividing the hole image into equal parts. Rectangular sub-blocks, each sub-block having a size of [size missing]. The image is segmented sequentially from top to bottom and from left to right, with no overlap. Histogram equalization is applied to all sub-blocks to obtain the transformation function. Limiting the local histogram height requires calculating the cutoff value for each rectangular sub-block, based on a pre-set threshold. The rectangular sub-blocks are passed through a threshold. If the number of gray levels of the nth rectangular sub-block is During the rectangular sub-block segmentation, the average number of pixels allocated to each gray level is: 'c' represents the restriction coefficient, and the cutoff value is obtained. , This indicates the number of pixels in the rectangular sub-block in the x-direction. This represents the number of pixels in the rectangular sub-block in the y-direction; based on the truncation value, the grayscale histograms of all rectangular sub-blocks are truncated sequentially, and the truncation value is redistributed to the grayscale values ​​of each level of the hole-bank histogram. The number of truncated pixels allocated to each pixel level is... , Let represent the total number of truncated pixels. Then, the pixel distribution of the histogram after pixel redistribution is: Meanwhile, global histogram equalization is used to process the truncated rectangular sub-blocks to improve image contrast and achieve uniform grayscale distribution; the grayscale value of the center pixel of the rectangular sub-block is used as a reference to perform bilinear interpolation on the grayscale value of each pixel in the image.

[0021] Specifically, due to spatial differences in images captured by infrared and visible light sensors at different times and viewing angles, appropriate registration techniques must be selected based on the specific application and image characteristics before image fusion to ensure image quality. Infrared imager-generated image data contains temperature information, which needs to be converted into digital images for later annotation and training, ensuring that the temperature data retains its original information and characteristics during visualization. Linear normalization can convert a two-dimensional temperature matrix into a digital image, while nonlinear normalization can also achieve digital image conversion. To address the issues of excessive enhancement and noise amplification in local areas of leaky water images, limiting contrast adaptive histogram equalization can enhance contrast and suppress noise by setting an upper limit on the local histogram peak value. Since the correspondence between pixels in the rectangular sub-block truncation equalization differs, using the gray value of the center pixel of the rectangular sub-block as a benchmark, bilinear interpolation of the gray value of each pixel in the image can improve image contrast and achieve a more uniform gray-level distribution.

[0022] It should be understood that by acquiring images of the underground cavern during its construction process, and preprocessing these images to obtain image data, multiple visible light and infrared images are extracted from the image data. These images are then fused to obtain an image fusion model. Feature extraction and segmentation are performed on the image fusion model to obtain leakage images. These leakage images are then input into a trained YOLOv8 network model to train a leakage recognition model. An evaluation index system for the leakage model is constructed, and the leakage measurement results are evaluated. Non-contact thermal imaging is then performed on the leakage points to supplement the missing cavern leakage features in visible light images under poor lighting conditions. This provides a basis for cavern leakage detection, effectively improves the saliency features and imaging quality of leakage in the fused images, and significantly improves the efficiency of leakage detection.

[0023] Optionally, the infrared and visible light images 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 infrared image normalization process is as follows: ,in, This represents the infrared thermal image of the original two-dimensional matrix. This represents a grayscale image of a mapped two-dimensional matrix, where k represents the scaling factor; the expression for visible light grayscale processing is... 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 then subjected to noise reduction processing. (Input image data) p As a guide image l Texture preservation and guiding image l resemblance ， Output image q Then, similar to the input image p To reduce the size of the input image data p and output image data q Differences between ， Guided filtering requires determining linear constant coefficients. and linear constant coefficients The coefficients are determined by minimizing the cost function and using the least squares method. Sum of coefficients The corresponding expression is ,in, This represents a local window centered at pixel k. Indicates input image data p In the window Linear constant coefficients within, ; Indicates a guide image l In the window Linear constant coefficients w represents the penalty coefficient, preventing linear constant coefficients. and linear constant coefficients It tends towards infinity; Represents linear constant coefficients and linear constant coefficients The guided filter representation; Represents a local window The number of pixels within; and These represent the guide images. l Mean and variance of pixels within a local window; This represents the output image obtained after filtering, with the following constraints: .

[0024] In this embodiment, the image fusion model is obtained by fusing visible light and infrared images, including: performing fusion according to specific rules in the grayscale pixel space of the original image using a spatial domain image fusion algorithm, and calculating the average pixel value of the image using a weighted averaging algorithm. The expression for generating the fused image after weighting is as follows: ,in, and All of these represent weighting coefficients. The YOLOv8 network model includes a Backbone layer, a Neck layer, a Head layer, and a Loss layer. The Backbone layer comprises a CBS convolutional module, a C2f module, and an SPPF module. The CBS convolutional module includes a Conv convolutional layer, a Batch Normalization (BN) layer, and a SiLU activation function. The Neck layer uses a path feature pyramid network to construct the 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, the calculation of which is as follows: ,in, This represents the true value of the category. This represents the predicted value for that category. This represents the Sigmoid function; the CIoU loss function incorporates the height ratio of the predicted bounding box to the ground truth bounding box in the loss calculation, and the corresponding expression is: ,in, This represents the length of the line connecting the center point of the ground truth bounding box and the center point of the predicted bounding box. The diagonal distance of the minimum outer matrix between the predicted and actual bounding boxes is represented by ; IoU represents the intersection-over-union ratio of the predicted and actual bounding boxes, and v represents the aspect ratio influence factor; the expression for the aspect ratio influence factor v is: , This indicates the actual width of the border. This indicates the actual height of the border. This indicates the predicted border width. This indicates the height of the predicted border.

[0025] It should be noted that the DF loss function models the distribution of the target object's bounding box as a generalized distribution, enabling the neural network to quickly focus on the values ​​around continuous positions of the target bounding box. The corresponding expression is: Where y represents the coordinates of the manually labeled true bounding box. , This represents the values ​​of two points adjacent to y. , This represents the Sigmoid function value at the corresponding point; when near When the DFL loss value decreases, the DFL loss value decreases; when the DFL reaches its minimum value, and The value is ; The mask loss function is the pixel-wise binary cross-entropy between the ground truth mask and the predicted mask, and the corresponding calculation expression is: ,in, Indicates the prediction mask, This represents the actual mask.

[0026] Specifically, the convolution operation uses a pre-defined convolution kernel of a fixed size to traverse the image from top left to bottom right with a specific stride. During this process, the weights on the convolution kernel are multiplied by the values ​​of the pixels it covers, and the results are summed to obtain the convolution kernel calculation result. The matrix obtained after the traversal is the feature map. During model training, parameter updates change the output distribution of subsequent network layers, which in turn affects subsequent iterations, slowing down the model training speed. Batch normalization layers standardize the input data by calculating the mean and variance of the smallest batch of samples, ensuring a stable input data distribution for each network layer, thereby accelerating the model's gradient convergence speed and preventing gradient minimization and explosion. When building a neural network model, if feature information between layers is transmitted linearly, the model's ability to fit multidimensional data will be affected. Therefore, it is necessary to establish connections in different networks using activation functions. By incorporating non-linear attributes into the neural network model, the model's descriptive power is enhanced. In the YOLOv8 model, the activation function is the SiLU function.

[0027] Optionally, see Figure 2 An evaluation index system for the leakage model was constructed, and the leakage measurement results were evaluated, including: Constructing the initial matrix X: Given m existing evaluation objects, each involving n evaluation indicators, then... Where n represents the number of evaluation indicators, and m represents the number of research subjects. This 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 using a unified standard. For positive indicators: ,in, , For negative indicators: , This represents the value of the j-th index in the i-th image; The variability of the calculated index is , , This represents the sample mean of the j-th indicator. This represents the standard deviation of the j-th indicator; Calculate the conflict of indicators, construct an indicator correlation matrix, and calculate the correlation coefficients in each column of indicators. The sum of: , ,in, This represents the correlation coefficient between the q-th indicator and the j-th indicator. This represents the number of conflicts for the j-th indicator; Calculate the objective weight of each evaluation indicator .

[0028] In this embodiment, the leakage measurement results include at least one of leakage area, leakage intensity, leakage shape, or leakage location. To utilize the leakage identification results, i.e., the measurement results, more efficiently and improve the effectiveness of assessing the severity of leakage in the cavern, a leakage severity assessment model is constructed. This mainly includes two parts: establishing a severity assessment index system and constructing the leakage severity assessment model. In establishing the index system, firstly, the influencing factors of leakage severity are identified based on the leakage identification results; secondly, the adverse effects of each factor on the cavern are comprehensively analyzed to determine the criteria for classifying different severity levels, thereby determining the assessment indicators and establishing the assessment index system.

[0029] It is important to note 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 reflects the scope and severity of the seepage problem. A larger seepage area means that more areas of the cavern are eroded, leading to more widespread material damage and structural weakening. This not only increases the risk of local or overall structural failure of the cavern but may also threaten its long-term stability. Seepage intensity is usually measured as the flow rate of water passing through a specific area per unit time, expressed as flow rate per square meter per hour. The temperature of the seepage area in infrared images is used as a characterization value for seepage intensity, reflecting the seepage intensity based on temperature changes. The seepage shape plays a crucial role in assessing the severity of cavern seepage, directly affecting the way and extent in which seepage affects the cavern structure. Different seepage shapes imply different water flow distributions and erosion patterns, thus having varying degrees of impact on the structural stability and safety of the cavern.

[0030] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.

[0031] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0032] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for measuring water leakage during the construction of a large underground cavern, characterized in that, Includes the following steps: Images of a large underground cavern during its construction process are acquired, and these images are preprocessed to obtain image data. The image preprocessing includes image registration, image enhancement, and image noise reduction. Multiple visible light and infrared images are extracted from image data, and the visible light and infrared images are fused to obtain an image fusion model. The image fusion includes: first, denoising the acquired infrared and visible light images to eliminate the interference of image noise on the target information; second, scaling the high-resolution visible light image; third, grayscale processing of the visible light and infrared images; and finally, inputting the preprocessed infrared and visible light images into the image fusion network. The image fusion model is used to extract and segment features to obtain the leakage image. The leakage image is then input into the trained YOLOV8 network model to train the leakage recognition model. An evaluation index system for a leakage model is constructed, and the leakage measurement results are evaluated. The construction of the evaluation index system for the leakage model includes determining the index weights and constructing the evaluation model. The YOLOv8 network model includes a Backbone layer, a Neck layer, a Head layer, and a Loss layer. The Backbone layer includes a CBS convolutional module, a C2f module, and an SPPF module. The CBS convolutional module includes a Conv convolutional layer, a Batch Normalization (BN) layer, and a SiLU activation function. The Neck layer uses a path feature pyramid network to construct the feature pyramid. 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 as follows: ,in, This represents the true value of the category. This represents the predicted value for that category. Represents the Sigmoid function; The CIoU loss function incorporates the height ratio of the predicted bounding box to the ground truth bounding box in the loss calculation; the corresponding expression is: ,in, This represents the length of the line connecting the center point of the ground truth bounding box and the center point of the predicted bounding box. The diagonal distance between the minimum outermost matrix of the predicted bounding box and the actual bounding box is represented; IoU represents the intersection-union ratio of the predicted bounding box and the actual bounding box; and v represents the aspect ratio influence factor. The expression for the aspect ratio influence factor v is: , This indicates the actual width of the border. This indicates the actual height of the border. This indicates the predicted border width. Indicates the height of the predicted border; The DF loss function models the distribution of the target object's bounding box as a generalized distribution, enabling the neural network to quickly focus on the values ​​around continuous positions of the target bounding box. The corresponding expression is: Where y represents the coordinates of the manually labeled true bounding box. , This represents the values ​​of two points adjacent to y. , This represents the Sigmoid function value at the corresponding point; when near When the DFL loss value decreases, the DFL loss value decreases; when the DFL reaches its minimum value, and The value is ; The mask loss function is the pixel-wise binary cross-entropy between the ground truth mask and the predicted mask, and the corresponding calculation expression is: ,in, Indicates the prediction mask, This represents the actual mask.

2. The method for measuring water leakage during the construction of a large underground cavern as described in claim 1, characterized in that, Acquire images of the large underground cavern during its construction process, and perform image preprocessing on these 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 images. Feature matching establishes the correspondence between the image to be registered and the reference image by comprehensively using various feature description operators, similarity measurement standards, and feature spatial correlation. Spatial transformation estimation is used to calculate the mapping function parameters based on feature matching. Resampling and interpolation are used when performing spatial geometric transformations on the image to be registered. The mapping function will generate non-integer pixel coordinates, and interpolation algorithms are needed to determine the gray values ​​of the coordinate points. Image enhancement includes infrared image normalization and histogram equalization, converting cave images into digital images. The expression for linear normalization is: ,in, Represents the coordinates of the converted digital image. Temperature value at that location, This represents the minimum temperature value in the temperature matrix. This represents the maximum temperature value in the temperature matrix; the expression for nonlinear normalization is... , where m represents the scaling factor.

3. The method for measuring water leakage during the construction of a large underground cavern as described in claim 2, characterized in that, The histogram equalization process includes: The cave image is divided into equal parts. Rectangular sub-blocks, each sub-block having a size of [size missing]. The image is segmented in the order from top to bottom and from left to right, with no overlap between segments; Perform histogram equalization on all sub-blocks to obtain the transformation function; Limiting the height of a local histogram requires calculating the cutoff value for each rectangular sub-block, based on a pre-set limit threshold. The rectangular sub-blocks are passed through a threshold. If the number of gray levels of the nth rectangular sub-block is During the rectangular sub-block segmentation, the average number of pixels allocated to each gray level is: 'c' represents the restriction coefficient, and the cutoff value is obtained. , This indicates the number of pixels in the rectangular sub-block in the x-direction. This indicates the number of pixels in the rectangular sub-block along the y-direction; Based on the truncation value, the grayscale histograms of all rectangular sub-blocks are truncated sequentially, and the truncation value is redistributed to the grayscale values ​​of each level of the hole-bank histogram. The number of truncated pixels assigned to each pixel level is [number missing]. , Let represent the total number of truncated pixels. Then, the pixel distribution of the histogram after pixel redistribution is: Meanwhile, global histogram equalization is used to process the truncated rectangular sub-blocks to improve image contrast and achieve uniform grayscale distribution. Using the gray value of the center pixel of the rectangular sub-block as a reference, bilinear interpolation is performed on the gray value of each pixel in the image.

4. The method for measuring water leakage during the construction of a large underground cavern as described in claim 3, characterized in that, Also includes: Infrared and visible light images are converted to grayscale images. Normalization maps the infrared temperature matrix to the grayscale image. The visible light image is converted to a grayscale image by calculating the weighted average of the RGB channels. The expression for infrared image normalization is as follows: ,in, This represents the infrared thermal image of the original two-dimensional matrix. This represents a grayscale image of a mapped two-dimensional matrix, where k represents the scaling factor; the expression for visible light grayscale processing is... 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 then subjected to noise reduction processing. (Input image data) p As a guide image l Texture preservation and guiding image l resemblance ， Output image q Then, similar to the input image p To reduce the size of the input image data p and output image data q Differences between ， Guided filtering requires determining linear constant coefficients. and linear constant coefficients The coefficients are determined by minimizing the cost function and using the least squares method. Sum of coefficients The corresponding expression is ,in, This represents a local window centered at pixel k. Indicates input image data p In the window Linear constant coefficients within, ; Indicates a guide image l In the window Linear constant coefficients w represents the penalty coefficient, preventing linear constant coefficients. and linear constant coefficients It tends towards infinity; Represents linear constant coefficients and linear constant coefficients The guided filter representation; Represents a local window The number of pixels within; and These represent the guide images. l Mean and variance of pixels within a local window; This represents the output image obtained after filtering, with the following constraints: .

5. The method for measuring water leakage during the construction of a large underground cavern as described in claim 1, characterized in that, The image fusion model is obtained by fusing visible light images and infrared images, including: A spatial domain image fusion algorithm is used to perform fusion in the grayscale pixel space of the original image according to specific rules, and a weighted averaging algorithm is used to calculate the average pixel value of the image. The expression for the weighted fused image is as follows: ,in, and All represent weighting coefficients. .

6. The method for measuring water leakage during the construction of a large underground cavern according to claim 1, characterized in that, An evaluation index system for the leakage model was constructed, and the leakage measurement results were evaluated, including: Constructing the initial matrix X: Given m existing evaluation objects, each involving n evaluation indicators, then... Where n represents the number of evaluation indicators, and m represents the number of research subjects. This 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 using a unified standard. For positive indicators: ,in, , For negative indicators: , This represents the value of the j-th index in the i-th image; The variability of the calculated index is , , This represents the sample mean of the j-th indicator. This represents the standard deviation of the j-th indicator; Calculate the conflict of indicators, construct an indicator correlation matrix, and calculate the correlation coefficients in each column of indicators. The sum of: , ,in, This represents the correlation coefficient between the q-th indicator and the j-th indicator. This represents the number of conflicts for the j-th indicator; Calculate the objective weight of each evaluation indicator .

7. The method for measuring water leakage during the construction of a large underground cavern as described in claim 6, characterized in that, Leakage measurement results include at least one of the following: leakage area, leakage intensity, leakage shape, or leakage location.

Images