Intelligent monitoring system for earth discharge quantity of shield tunneling based on image acquisition and processing

Abstract

This invention relates to the field of image processing technology, and discloses an intelligent monitoring system for soil discharge volume in shield tunneling based on image acquisition and processing. It includes an image fusion module for acquiring images of a specific area and constructing a pile model from these images; an index extraction module for dividing the pile model into model layers and extracting three-dimensional indices; a soil discharge volume calculation module for calculating the actual soil discharge volume of the model layers; and a monitoring and prompting module for calculating the real-time soil discharge volume. This invention can perform a virtual-to-real conversion operation on the actual soil pile structure, facilitating further subdivision of the soil pile in the conveyor section. It can also perform two progressively layered subdivision operations on the soil pile being tunneled by the shield on the conveyor belt, thereby accurately refining and distinguishing the continuously output soil pile on the conveyor belt, facilitating subsequent soil discharge volume calculation operations, and thus improving the accuracy of the shield tunneling soil discharge volume monitoring results.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically, to an intelligent monitoring system for the amount of soil removed during tunnel boring machine (TBM) excavation based on image acquisition and processing. Background Technology

[0002] The amount of soil removed during shield tunneling is a key parameter for measuring tunneling efficiency, assessing ground disturbance, ensuring construction safety, and controlling surface settlement. If the amount of soil removed during shield tunneling exceeds the theoretical excavation volume, it can easily lead to ground loss and ground settlement. If the amount of soil removed during shield tunneling is less than the theoretical value, it may cause soil pressure imbalance in front of the shield machine, inducing the risk of gushing or machine jamming. Therefore, it is necessary to conduct real-time intelligent monitoring of the amount of soil removed during shield tunneling.

[0003] The patent application with publication number CN119963480A discloses an intelligent monitoring method for the amount of soil discharged during shield tunneling based on image acquisition and laser scanning equipment. It establishes a neural network learning system based on the optimized YOLOv5 algorithm to conduct in-depth analysis of the collected data. This method can more accurately and conveniently analyze and record data that is difficult to detect manually in real time during engineering projects, so as to have a more accurate and in-depth understanding of the construction environment, timely detection of risks, and rapid and accurate response decisions. Existing monitoring systems typically use a holistic approach to monitor and calculate the amount of excavated soil on the conveyor belt when monitoring the amount of soil discharged during tunnel boring machine (TBM) excavation. However, due to the long length of the conveyor belt and the large volume of excavated soil being transported, this holistic approach results in a vague and ambiguous monitoring location and target for the excavated soil on the conveyor belt. It fails to perform precise monitoring of local locations within the excavated soil, leading to low accuracy in subsequent soil discharge volume monitoring calculations and reducing the overall accuracy of the soil discharge volume monitoring results.

[0004] In view of this, the present invention proposes an intelligent monitoring system for the amount of soil discharged during shield tunneling based on image acquisition and processing to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an intelligent monitoring system for the amount of soil removed during shield tunneling based on image acquisition and processing, comprising: The image fusion module is used to divide the conveyor belt into continuously distributed conveying sections along the conveying direction, acquire section images of the conveying sections at the same time, and the section images are the upper image, the left-down image and the right-down image. Based on the overlay fusion mechanism, the section images are constructed into a stacked model. The index extraction module is used to divide the stacking model into continuously distributed model layers along the conveying direction. Taking the model layer as the first object, it extracts three-dimensional indexes, including the drop excess ratio and the local concavity-convexity ratio. Taking the interval image as the second object, it extracts planar indexes, including the excess particle ratio, color value balance ratio, length value, width value, and height value. The soil discharge calculation module is used to summarize the three-dimensional and two-dimensional indicators of the same model layer into a comprehensive indicator. The theoretical soil discharge volume corresponding to the comprehensive indicator is predicted by the soil discharge volume prediction model, and the actual soil discharge volume of the model layer is calculated by combining the looseness coefficient of the model layer. The monitoring and prompting module is used to calculate the real-time soil discharge volume by weighted summation of the actual soil discharge volume of the model layer in all transport sections, and to dynamically update the real-time soil discharge volume and adaptively formulate shield tunneling prompt information.

[0006] Furthermore, the method for acquiring interval images is as follows: Mark a random point on the conveying interval, and continuously adjust the position of the random point until the distance from the random point to the two long sides of the conveying interval is the same, and the distance to the two short sides of the conveying interval is the same. Then, record the adjusted random point as the midpoint of the interval. Measure the length of the shorter side in the conveying section, and record 80% of the length of the shorter side as the unit amplitude; The point located directly above the midpoint of the interval and one unit amplitude away from the midpoint is designated as the upper point; the point located 45 degrees to the left of the midpoint and one unit amplitude away from the midpoint is designated as the left downward point; and the point located 45 degrees to the right of the midpoint and one unit amplitude away from the midpoint is designated as the right downward point. The distances from the top point, left top point, and right top point to the midpoint of the interval are adjusted synchronously until there is no overlap in the images captured at the same point between two adjacent transmission intervals. At the same time, the top image, left top image, and right top image are captured.

[0007] Furthermore, the overlay and fusion mechanism is as follows: the images in the intervals are overlaid and fused from the middle to both sides.

[0008] Furthermore, the method for constructing the stacking model is as follows: The projection area of ​​the mound is identified in the upper-level image by edge detection technology. The projection area is discretized into an array of grids based on the calibrated grid side length. The common point of two adjacent grids is recorded as a grid point, and B grid points are obtained. Initialize the initial height of the grid points on the Z-axis, convert the grid points into world coordinate points, and project the world coordinate points onto the left and right top images respectively to obtain the left and right index points; A window is selected on the left top view with the left punctuation point as the center, and is denoted as the reference window. A window is randomly selected on the right top view and is denoted as the adjustment window. The position of the adjustment window is moved horizontally on the right top view with the epipolar line as the baseline. The matching cost between the adjustment window and the reference window is calculated by the matching cost function. The adjustment window corresponding to the minimum matching cost is recorded as the target window, and the absolute value of the difference between the coordinates of the left index point in the reference window and the right index point in the target window on the X-axis is taken to obtain the disparity. Stereo correction is performed on the cameras at the left and right top points. The baseline length and focal length of the cameras are retrieved. The stereo height is calculated by combining the parallax, baseline length, and focal length, and the stereo height is used to replace the initial height. The coordinate values ​​of the grid points on the X, Y, and Z axes are combined to generate point cloud coordinates. Then, 3D point cloud modeling technology is used to construct a stacked model of all the point cloud coordinates to obtain A stacked models.

[0009] Furthermore, the method for extracting the percentage of vertical drop exceeding the limit is as follows: Mark the points on the upper surface of each of the C model layers one by one, draw auxiliary lines between any two points to form point lines, and measure the angle formed by any two point lines to obtain the drop angle value. The drop angle value that is less than or equal to the calibrated angle value is recorded as the excess angle value. The number of excess angle values ​​is compared with the number of drop angle values ​​to calculate the excess drop ratio.

[0010] Furthermore, the method for extracting the local concavity / convexity ratio is as follows: The initial height of all points on the C model layers is calculated one by one, and the average height is calculated by summing the initial heights of all points. Points with heights greater than the average height are designated as convex points, and points with heights less than the average height are designated as concave points. The number of convex points and the number of concave points were counted separately, and the local convex-concave ratio was calculated by comparing the number of convex points and the number of concave points.

[0011] Furthermore, the method for extracting the proportion of excessive particles is as follows: The upper-level image is preprocessed to remove noise and enhance contrast. Based on the model layer division method, the preprocessed upper-level image is divided into C sub-images that correspond one-to-one with the model layer. The boundary line between two adjacent stones in the sub-image is identified by the edge detection algorithm, and the area inside the boundary line of the same closed structure is recorded as the stone region. Two boundary points are randomly marked on the boundary line of the stone area. One boundary point is marked as the stationary point and kept still, while the other boundary point is marked as the moving point and keeps moving. The distance between the stationary point and the moving point is measured in real time and recorded as the point-to-point distance value. Points whose distance from one point to another is greater than the standard distance are marked as out-of-limit points. The number of out-of-limit points on the boundary line is divided by the number of boundary points to calculate the sub-proportion. The stone areas with a sub-ratio greater than the standard ratio threshold are recorded as out-of-limit areas. The coverage area of ​​the out-of-limit areas is divided by the coverage area of ​​the sub-image to calculate the out-of-limit particle ratio.

[0012] Furthermore, the method for extracting the color value balance ratio is as follows: Convert the sub-image to a grayscale image, mark the grayscale value of all pixels in the grayscale image, record pixels with grayscale values ​​within the same grayscale range as points of the same type, and after summarizing points of the same type, generate D point sets; Count the number of points of the same type and the total number of pixels in the D point sets one by one, and record them as the value of the same type and the total value. Then, compare the maximum value of the same type with the total value and calculate the color value balance ratio.

[0013] Furthermore, the method for calculating the actual amount of soil discharged is as follows: The humidity moment when the model layer first appeared on the conveyor belt was found by querying the timestamp one by one. The humidity value at the time of the humidity was detected by the humidity sensor to obtain the soil humidity value. The humidity difference was calculated by subtracting the soil humidity value from the calibrated humidity threshold. The average value of the distances between all points exceeding the limit in the model layer is obtained by summing the distances between them. The difference between the average value of the exceeding limit and the calibrated exceeding threshold is then calculated to obtain the exceeding difference value. The humidity difference and excess difference of the C model layers are assigned corresponding proportional coefficients and then summed to calculate the looseness coefficient; The actual soil discharge volume of the C model layers is calculated by multiplying the theoretical soil discharge volume of each model layer by the corresponding loosening coefficient.

[0014] Furthermore, the shield tunneling prompts include information on continued shield tunneling and information on termination of shield tunneling; the method for formulating the shield tunneling prompts is as follows: Compare the dynamically updated real-time soil discharge volume with the upper limit of soil discharge volume; When the real-time soil discharge volume is less than the upper limit of the soil discharge volume, continuous shield tunneling information is generated; When the real-time soil discharge volume is greater than or equal to the upper limit of soil discharge volume, the tunnel boring machine is terminated.

[0015] The technical advantages of the intelligent monitoring system for soil removal volume in shield tunneling based on image acquisition and processing of the present invention are as follows: (1): By dividing the conveyor belt into conveying sections and simultaneously acquiring multi-angle section images of the conveying sections, the present invention can break down the large-span conveyor belt into smaller parts. Combined with the simultaneous acquisition of multi-point section images, it can achieve multi-point and refined image monitoring of the soil piles on the conveyor belt, thereby avoiding the general problem of the overall monitoring method.

[0016] (2): This invention constructs the interval image into a stacking model and combines the operation of collecting indicators after the stacking model is segmented. On the one hand, it can perform virtual-to-real conversion operation on the real soil pile structure, which facilitates the further subdivision of the soil pile in the transport interval and the simulation operation of the physical soil pile. On the other hand, it can also provide a direct and clear object for extracting three-dimensional indicators, avoiding the uncertainty brought about by directly extracting indicators from the complex and huge stacking model, and thus laying the foundation for the accurate calculation of the subsequent soil discharge volume.

[0017] (3) By continuously dividing the conveyor belt and the pile model, the present invention can perform two progressively layered operations to break down the soil pile of the shield tunneling on the conveyor belt into smaller parts, thereby accurately refining and distinguishing the soil pile continuously output from the conveyor belt, which facilitates the subsequent calculation of the soil discharge volume and improves the accuracy of the monitoring results of the soil discharge volume of the shield tunneling. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of a module of an intelligent monitoring system for shield tunneling soil discharge based on image acquisition and processing, provided in Embodiment 1 of the present invention. Figure 2 This is a flowchart illustrating an intelligent monitoring method for the amount of soil removed during shield tunneling based on image acquisition and processing, as provided in Embodiment 2 of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1: Please refer to Figure 1 As shown in this embodiment, a smart monitoring system for shield tunneling soil removal based on image acquisition and processing includes: The image fusion module divides the conveyor belt into continuous conveying sections along the conveying direction, synchronously acquires section images of soil piles on the conveying sections, and constructs an accumulation model of soil piles within the conveying sections based on the overlay fusion mechanism. The large amount of excavated soil generated by the tunnel boring machine during tunnel excavation is transported and discharged by the conveyor belt. Due to the long length of the conveyor belt, directly calculating and monitoring the amount of excavated soil discharged on the entire conveyor belt has negative effects such as large calculation workload and general and inaccurate monitoring. Therefore, it is necessary to break down the conveyor belt into smaller sections to achieve segmented monitoring of the excavated soil on the conveyor belt. Specifically, when dividing the conveyor belt into conveying sections, along the conveying direction of the conveyor belt, the conveyor belt is divided into A continuously distributed conveying sections based on the preset section length.

[0021] In this embodiment, when the conveyor belt is divided into conveying sections, the number of conveying sections is at least 3, and the multiple conveying sections are distributed sequentially according to the conveying direction of the conveyor belt, and two adjacent conveying sections are in a continuous state in terms of position.

[0022] Section images refer to multi-angle images of the conveyor belt and soil piles within the conveying section captured by image acquisition equipment. These images serve as the primary basis for subsequent calculation and monitoring of soil discharge volume. Since the soil piles within the conveying section have a three-dimensional structure, a single-angle image cannot fully represent the soil piles. Therefore, it is necessary to acquire images from multiple different angles. Specifically, the interval images are the upper-view image, the left-view image, and the right-view image; A top-view image is an image taken from directly above the mound at a bird's-eye view. A left-view image and a right-view image are images taken from a bird's-eye view slightly to the left and slightly to the right above the mound, respectively.

[0023] It should be noted that outward-sloping side guards are usually installed on both sides of the conveyor belt in the tunnel boring machine. Therefore, it is not possible to take pictures directly from the left and right sides of the conveyor belt. Instead, it is necessary to take pictures from the top left and top right angles to obtain left and right top images.

[0024] Specifically, the method for acquiring interval images is as follows: Mark a random point on the conveying section, and continuously adjust the position of the random point until the distance from the random point to the two long sides of the conveying section is the same, and the distance to the two short sides of the conveying section is also the same. Record the adjusted random point as the midpoint of the section; ensure that the midpoint of the section is located at the centroid of the conveying section. Measure the length of the shorter side in the conveying section, and record 80% of the length of the shorter side as the unit amplitude; The point located directly above the midpoint of the interval and one unit amplitude away from the midpoint is designated as the upper point; the point located 45 degrees to the left of the midpoint and one unit amplitude away from the midpoint is designated as the left downward point; and the point located 45 degrees to the right of the midpoint and one unit amplitude away from the midpoint is designated as the right downward point. Observe the camera's images at the top, left, and right viewpoints, and simultaneously adjust the distances from the top, left, and right viewpoints to the midpoint of the interval until there is no overlap in the images captured by adjacent transmission intervals at the same point. At the same time, capture the top, left, and right view images.

[0025] The obtained top-view, left-down, and right-down images can only display the transport section from multiple angles in a planar manner. In order to better display the three-dimensional situation of the soil pile in the transport section, it is necessary to superimpose and combine the images of the same transport section, and combine them with three-dimensional modeling technology to construct a virtual pile model, so that the pile model can display the soil pile from a virtual three-dimensional perspective. When superimposing and combining the upper, left-down, and right-down images of the transport section, since the angles of each image are not consistent, it is necessary to superimpose and combine the images of the section under the superposition and fusion mechanism to ensure that the constructed pile model can be consistent with the actual soil pile.

[0026] Specifically, the overlay and fusion mechanism is as follows: the interval images are overlaid and fused from the middle to both sides; this ensures that the interval images of the same transmission interval can be overlaid in sequence, laying the foundation for the construction of the subsequent stacking model.

[0027] The method for constructing a stacking model is as follows: Edge detection technology is used to identify the edge line of the location of the mound in the upper image, and the area inside the edge line is recorded as the projection area. Using the calibrated grid side length as a standard, the projected area is discretized into an array of grids, and the common point of two adjacent grids is recorded as a grid point, resulting in B grid points; the calibrated grid side length is a value used to limit the grid side length to ensure that each grid can maintain a consistent size; Initialize the initial height of the grid points on the Z-axis, convert the grid points to world coordinate points, and project the world coordinate points onto the left and right top images respectively to obtain the left and right index points; when initializing the initial height, it is usually recorded as 0, so as to facilitate subsequent conversion and calculation of the grid points to world coordinate points; Centered on the left-hand top view, select a window and call it the reference window. Then, randomly select a window on the right-hand top view and call it the adjustment window. The size of the window is not limited, as long as it can completely enclose the spatial positions of the left and right top views, thus providing the position and coordinate basis for further calculation of the initial height. Using the epipolar line as the baseline, the position of the adjustment window is horizontally moved on the right-downward image, and the matching cost between the adjustment window and the baseline window is calculated using a matching cost function (such as SAD, SSD, NCC). The adjustment window corresponding to the minimum matching cost is recorded as the target window, and the absolute value of the difference between the coordinates of the left index point in the reference window and the right index point in the target window on the X-axis is taken to obtain the disparity. Stereo correction is performed on the cameras at the left and right top points. The baseline length and focal length of the cameras are retrieved. The stereo height is calculated by combining the parallax, baseline length, and focal length, and the stereo height is used to replace the initial height. The formula for calculating the height of a three-dimensional structure is: In the formula, For three-dimensional height, Focal length Baseline length For parallax; The coordinate values ​​of the grid points on the X, Y, and Z axes are combined to generate point cloud coordinates. Then, 3D point cloud modeling technology is used to construct a stacked model of all the point cloud coordinates to obtain A stacked models.

[0028] It should be noted that 3D point cloud modeling technology can recover the 3D structure of an object from a set of discrete point cloud data. This process usually includes steps such as point cloud acquisition, preprocessing, feature extraction, and 3D model reconstruction. As this technology is existing in the field, it will not be described in detail here.

[0029] The indicator extraction module divides the stacked model into continuous model layers, extracts the three-dimensional indicators of the model layers, and extracts the planar indicators of the model layers from the interval image. The aforementioned pile model is used to provide a virtual overall display of the structure and shape of the soil pile within the transport section. However, it cannot provide a detailed local representation of a specific location, making it impossible for the pile model to achieve continuous multi-point, detailed monitoring of soil discharge within the soil pile. Therefore, it is necessary to segment the pile model to achieve the effect of breaking down the pile model into smaller parts.

[0030] The model layer is a partially stacked model with a smaller thickness obtained by cutting the stacked model at equal intervals according to the conveying direction; Specifically, when dividing the stacked model into model layers, the dividing interval is first set, and according to the conveying direction, the stacked model is divided into C continuous model layers based on one dividing interval.

[0031] Three-dimensional indicators are used to represent the relevant structural and morphological characteristics of the model layer in three-dimensional space, that is, to represent the soil mound referred to by the model layer from a three-dimensional perspective. Specifically, the three-dimensional indicators include the percentage of elevation difference exceeding the limit and the percentage of local concavity and convexity.

[0032] The percentage of height difference exceeding the limit refers to the proportion of the number of points in the model layer whose angle exceeds the calibrated angle when connected by the line connecting any two points. This can be used to represent the magnitude of the height difference in the model layer. Specifically, the method for extracting the percentage of elevation difference exceeding the limit is as follows: Mark the points on the upper surface of each of the C model layers one by one. Draw auxiliary lines between any two points to form point lines. Measure the angle formed by any two point lines to obtain the drop angle value. The angle formed by two point lines can be any angle between 0 and 180 degrees. The smaller the angle, the greater the height difference between the two points. When the drop angle value is less than or equal to the calibrated angle value, the drop angle value is recorded as the over-limit angle value. The number of over-limit angle values ​​is compared with the number of drop angle values ​​to calculate the over-limit ratio. The calibrated angle value refers to the maximum drop angle value when it is recorded as the over-limit angle value. The calibrated angle value is obtained by collecting a large number of historical maximum drop angle values ​​when they are recorded as over-limit angle values ​​and then averaging them. The formula for calculating the percentage of vertical drop exceeding the limit is: In the formula, The percentage of drop exceeding the limit. The number of angle values ​​exceeding the limit. This represents the number of elevation angle values.

[0033] The local convexity / concave ratio refers to the ratio of the area of ​​the convex region to the area of ​​the concave region in the model layer, which can be used to represent the relative size of the convex and concave areas in the model layer. The method for extracting the local concavity / convexity ratio is as follows: The initial height of all points on the C model layers is calculated one by one, and the average height is calculated by summing the initial heights of all points. The initial height of each point is compared with the average height. Points with a height greater than the average height are marked as convex points, and points with a height less than the average height are marked as concave points. The number of convex points and the number of concave points were counted separately, and the local convex-concave ratio was calculated by comparing the number of convex points and the number of concave points. The formula for calculating the local concavity / convexity ratio is: In the formula, For the local concavity and convexity ratio, The number of protruding points. This represents the number of depressions.

[0034] Planar indices are indices used to represent the relevant structural and morphological features of the model layer in planar space, that is, to represent the soil mound referred to by the model layer from a planar perspective; Specifically, planar indicators include the proportion of particles exceeding the limit, the proportion of color value balance, length value, width value, and height value; The proportion of particles exceeding the standard particle diameter refers to the ratio between the area of ​​the slag soil in the model layer with particle diameters exceeding the standard particle diameter and the total area, which can be used to represent the particle diameter of the slag soil in the model layer. The method for extracting the proportion of particles exceeding the limit is as follows: The upper-level image is denoised using a filter, and the contrast of the denoised upper-level image is enhanced to achieve the preprocessing effect of the upper-level image. Based on the model layer division method, the preprocessed upper image is divided into C sub-images that correspond one-to-one with the model layer. By dividing the upper image into sub-images, on the one hand, the one-to-one correspondence between the sub-image and the model layer can be achieved, ensuring that the planar indicators collected on the sub-image can correspond to the same positional structure as the three-dimensional indicators. On the other hand, the upper image can also be broken down into smaller parts to reduce the burden of each planar indicator extraction. The boundary line between two adjacent stones in the sub-image is identified by the edge detection algorithm, and the area inside the boundary line of the same closed structure is recorded as the stone region. Two boundary points are randomly marked on the boundary line of the stone area. One boundary point is marked as the stationary point and kept still, while the other boundary point is marked as the moving point and keeps moving. The distance between the stationary point and the moving point is measured in real time and recorded as the point-to-point distance value. Moving points whose point distance value is greater than the standard distance value are marked as out-of-limit points. The number of out-of-limit points on the boundary line is divided by the number of boundary points to calculate the sub-proportion. The standard distance value refers to the maximum point distance value when it is not marked as an out-of-limit point, so that it can provide a limiting basis for stone areas with relatively small particle diameters. Areas of stones with a sub-scale greater than the standard scale threshold are designated as out-of-limit regions. The coverage area of ​​these out-of-limit regions is divided by the coverage area of ​​the sub-image to calculate the out-of-limit particle scale. The standard scale threshold refers to the maximum sub-scale value that is not designated as an out-of-limit region, thus providing the basis for calculating the out-of-limit particle scale.

[0035] Color balance ratio refers to the proportion of stones and soil in the model layer that maintain a consistent color, which can represent the color similarity of stones and soil in the model layer. The method for extracting the color value balance ratio is as follows: Convert the sub-image to a grayscale image and mark the grayscale value of all pixels in the grayscale image; Pixels with gray values ​​within the same gray range are categorized as points of the same type. After grouping together points of the same type, D point sets are generated. The gray range refers to the range from the minimum to the maximum gray value of pixels corresponding to the same color, which provides the numerical basis for the summary set of pixels of different colors. Count the number of points of the same type and the total number of pixels in the D point sets one by one, and record them as the value of the same type and the total value. Then, compare the maximum value of the same type with the total value and calculate the color value balance ratio.

[0036] The length value refers to the length of the model layer parallel to the conveying direction, and serves as the basis for subsequent calculations of the amount of soil discharged.

[0037] The width value refers to the horizontal width of the model layer perpendicular to the conveying direction, and serves as the basis for subsequent calculations of the soil discharge volume.

[0038] The height value refers to the average height between different points in the model layer that are vertically perpendicular to the conveying direction. It serves as the height basis for subsequent calculations of the soil discharge volume. The height value is obtained by averaging the initial heights of different points.

[0039] The soil discharge calculation module combines three-dimensional and two-dimensional indicators into a comprehensive indicator. It predicts the theoretical soil discharge volume of the model layer through the soil discharge volume prediction model, and calculates the actual soil discharge volume of the model layer by combining the looseness coefficient of the model layer. The comprehensive index is a comprehensive representation of multiple indicators in the three-dimensional and two-dimensional dimensions of the model layer, which can summarize and integrate scattered and diverse indicators. The soil discharge volume prediction model is a neural network model based on machine learning technology, with comprehensive indicators as input data and theoretical soil discharge volume as output data, so that the soil discharge volume prediction model can serve as a direct basis for the corresponding soil pile flow rate in the model layer. In this embodiment, the theoretical discharge volume is a value automatically predicted by the discharge volume prediction model based on comprehensive index analysis. It can only represent the discharge volume corresponding to the model layer from a theoretical perspective.

[0040] The aforementioned soil dumping volume prediction model is not obtained directly, but is based on the existing neural network model, combined with a large number of comprehensive indicators and corresponding theoretical soil dumping volume, and obtained through repeated optimization and iterative training. Specifically, the training method for the soil dumping volume prediction model is as follows: Multiple sets of data on the model layer during historical periods, including the proportion of drop exceeding the limit, the proportion of local unevenness, the proportion of particles exceeding the limit, the proportion of color value balance, the length value, the width value, the height value, and the corresponding theoretical soil discharge volume, are collected in advance. After summarizing the proportion of drop exceeding the limit, the proportion of local unevenness, the proportion of particles exceeding the limit, the proportion of color value balance, the length value, the width value, and the height value of the same set, a comprehensive index is generated, resulting in multiple sets of comprehensive indexes and multiple sets of theoretical soil discharge volumes. Each set of comprehensive indicators is labeled as a training feature, and the theoretical soil discharge volume corresponding to each set of training features is labeled. The labeled training features are divided into a training set and a test set; 80% of the training features are used as the training set and 20% of the training features are used as the test set; the neural network model is trained using the training set and tested using the test set. A preset error threshold is set. When the mean of the prediction errors of all training features in the test set is less than the preset error threshold, a soil discharge prediction model that predicts the theoretical soil discharge volume based on comprehensive indicators is obtained.

[0041] By inputting the comprehensive indices of the C model layers collected in real time into the soil discharge prediction model, the theoretical soil discharge volume of the C model layers at the current moment can be predicted.

[0042] Because the density of the soil piles in the transport section is not uniform, and the moisture content of the soil in the tunneling geology is also different, the actual amount of soil discharged from the soil piles in the transport section is not consistent with the theoretical amount of soil discharged. The looseness coefficient is a numerical representation of the degree of compactness between soil particles in a mound. The larger the looseness coefficient, the more compact the mound, and the greater the amount of soil to be removed from the model layer; conversely, the smaller the looseness coefficient, the greater the compactness of the mound. Therefore, when calculating the actual amount of soil to be removed from the model layer, the looseness coefficient of each model layer needs to be considered. Specifically, the method for calculating the actual amount of soil discharged is as follows: The time when the model layer first appeared on the conveyor belt was retrieved by querying the timestamp one by one to obtain the humidity time. The humidity value at the humidity time was then detected by the humidity sensor to obtain the soil moisture value. The humidity difference is calculated by subtracting the soil moisture value from the calibrated humidity threshold. The average of the point distances of all out-of-limit points in the model layer is obtained by summing up the out-of-limit average values. The out-of-limit difference is calculated by subtracting the out-of-limit average value from the calibrated out-of-limit threshold value. The calibrated out-of-limit threshold value is the maximum value of the out-of-limit average value when it has no effect on the looseness coefficient. This provides an analytical basis for the degree of influence of the out-of-limit average value on the looseness coefficient. The humidity difference and excess difference of the C model layers are assigned corresponding proportional coefficients and then summed to calculate the looseness coefficient; The formula for calculating the looseness coefficient is: In the formula, The looseness coefficient, This is the humidity difference value. The difference exceeding the limit, , These are the proportional coefficients for humidity difference and excess difference, respectively; The actual soil discharge volume of the C model layers is calculated by multiplying the theoretical soil discharge volume of each model layer by the corresponding loosening coefficient. The formula for calculating the actual amount of soil discharged is: In the formula, This represents the actual amount of soil disposed of. This represents the theoretical volume of soil to be disposed of.

[0043] It should be noted that the formulas mentioned above are all dimensionless calculations, and are derived from software simulations using a large amount of data to obtain a formula that is closest to the real situation. The proportionality coefficients in the formulas and the preset thresholds in the analysis process are set by those skilled in the art based on the actual situation or obtained through large-scale data simulations. The size of the proportionality coefficient is a specific value obtained by quantifying each parameter to facilitate subsequent comparisons. The size of the proportionality coefficient depends on the amount of sample data and the processing coefficients initially set by those skilled in the art for each set of sample data; as long as it does not affect the proportional relationship between the parameter and the quantified value, it is acceptable.

[0044] The monitoring and prompting module calculates the real-time soil discharge volume of the conveyor belt by weighted summation of the actual soil discharge volume of the model layer within the conveying section, and dynamically updates the real-time soil discharge volume to adaptively generate shield tunneling prompt information. After obtaining the actual soil discharge volume of the model layer, the real-time soil discharge volume of all conveying sections on the entire conveyor belt at the current moment can be calculated based on the actual soil discharge volume, so that the real-time soil discharge volume can be used as the final result of the shield tunneling operation at the current moment. Specifically, when calculating the real-time soil discharge volume, the actual soil discharge volumes of the C model layers belonging to the same stacking model are first summed with weights to obtain the interval soil discharge volume. Then, the interval soil discharge volumes of the A transport intervals are summed with weights to calculate the real-time soil discharge volume.

[0045] When calculating the real-time soil discharge volume, the real-time soil discharge volume at this time is only the result of the soil discharge volume at the current moment. Since shield tunneling is continuous, the real-time soil discharge volume will also be updated dynamically and will increase continuously over time. Specifically, when dynamically updating the real-time soil discharge volume, as the conveyor belt continues to move forward, the camera will capture images of the conveyor belt's transport section in real time at subsequent moments. Based on the newly acquired images, the calculation standard for the real-time soil discharge volume is repeatedly executed, thus dynamically updating the real-time soil discharge volume.

[0046] After the real-time soil discharge volume is dynamically updated, the updated real-time soil discharge volume can be analyzed and compared. The real-time soil discharge volume can then be used as the final basis for intelligent monitoring of the soil discharge volume during tunnel boring machine (TBM) excavation, and adaptive TBM excavation prompts can be generated to match the real-time soil discharge volume. Shield tunneling information includes information on continued shield tunneling and information on termination of shield tunneling; Specifically, the method for formulating shield tunneling prompts is as follows: The real-time soil discharge volume, which is dynamically updated, is compared with the upper limit value of the soil discharge volume. The upper limit value of the soil discharge volume is a pre-set target value for the soil discharge volume of this shield tunneling operation, and serves as the numerical basis for formulating subsequent shield tunneling prompt information. When the real-time soil discharge volume is less than the upper limit of soil discharge volume, it means that the soil discharge volume during shield tunneling has not reached the preset soil discharge volume, and shield tunneling operation still needs to be carried out. In this case, continuous shield tunneling information is generated. When the real-time soil discharge volume is greater than or equal to the upper limit of soil discharge volume, it means that the soil discharge volume during shield tunneling has reached the preset soil discharge volume, and shield tunneling operation is not required. In this case, the shield tunneling is terminated.

[0047] In this embodiment, when the continuous tunneling information is set, the operating parameters of the tunnel boring machine remain unchanged and the tunneling operation continues until the real-time soil discharge volume is greater than or equal to the upper limit of the soil discharge volume. Only then will the tunneling information be set to terminate the tunneling operation and the tunnel boring machine will stop.

[0048] Example 2: Please refer to Figure 2 As shown, the parts not described in detail in this embodiment are described in Embodiment 1. This embodiment provides an intelligent monitoring method for shield tunneling soil disposal volume based on image acquisition and processing. It is implemented based on an intelligent monitoring system for shield tunneling soil disposal volume based on image acquisition and processing, and includes: S01: Divide the conveyor belt into continuously distributed conveying sections along the conveying direction, collect section images of the conveying sections at the same time, and construct a stacking model of the section images based on the overlay and fusion mechanism; S02: Divide the stacking model into continuously distributed model layers along the conveying direction. Take the model layer as the first object and extract the three-dimensional index. Take the interval image as the second object and extract the planar index. S03: Combine the three-dimensional and two-dimensional indices of the same model layer into a comprehensive index. Predict the theoretical soil discharge volume corresponding to the comprehensive index using the soil discharge volume prediction model. Combine the looseness coefficient of the model layer to calculate the actual soil discharge volume of the model layer. S04: Calculate the real-time soil discharge volume by weighted summation of the actual soil discharge volume of the model layers in all transport sections, and dynamically update the real-time soil discharge volume to adaptively generate shield tunneling prompts.

[0049] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent monitoring system for earth removal quantity of shield tunneling based on image acquisition processing, characterized in that, include: The image fusion module is used to divide the conveyor belt into continuously distributed conveying sections along the conveying direction, acquire section images of the conveying sections at the same time, and the section images are the upper image, the left-down image and the right-down image. Based on the overlay fusion mechanism, the section images are constructed into a stacked model. The overlay and fusion mechanism is as follows: the images in the intervals are overlaid and fused from the middle to both sides; The method for constructing a stacking model is as follows: The projection area of ​​the mound is identified in the upper-level image by edge detection technology. The projection area is discretized into an array of grids based on the calibrated grid side length. The common point of two adjacent grids is recorded as a grid point, and B grid points are obtained. Initialize the initial height of the grid points on the Z-axis, convert the grid points into world coordinate points, and project the world coordinate points onto the left and right top images respectively to obtain the left and right index points; A window is selected on the left top view with the left punctuation point as the center, and is denoted as the reference window. A window is randomly selected on the right top view and is denoted as the adjustment window. The position of the adjustment window is moved horizontally on the right top view with the epipolar line as the baseline. The matching cost between the adjustment window and the reference window is calculated by the matching cost function. The adjustment window corresponding to the minimum matching cost is recorded as the target window, and the absolute value of the difference between the coordinates of the left index point in the reference window and the right index point in the target window on the X-axis is taken to obtain the disparity. Stereo correction is performed on the cameras at the left and right top points. The baseline length and focal length of the cameras are retrieved. The stereo height is calculated by combining the parallax, baseline length, and focal length, and the stereo height is used to replace the initial height. The coordinate values ​​of the grid points on the X, Y, and Z axes are combined to generate point cloud coordinates. Then, a stacked model is constructed from all the point cloud coordinates using 3D point cloud modeling technology to obtain A stacked models. The index extraction module is used to divide the stacking model into continuously distributed model layers along the conveying direction. Taking the model layer as the first object, it extracts three-dimensional indexes, including the drop excess ratio and the local concavity-convexity ratio. Taking the interval image as the second object, it extracts planar indexes, including the excess particle ratio, color value balance ratio, length value, width value, and height value. The method for extracting the percentage of vertical drop exceeding the limit is as follows: Mark the points on the upper surface of each of the C model layers one by one, draw auxiliary lines between any two points to form point lines, and measure the angle formed by any two point lines to obtain the drop angle value. The drop angle value less than or equal to the calibrated angle value is recorded as the excess angle value. The number of excess angle values ​​is compared with the number of drop angle values ​​to calculate the excess drop ratio. The method for extracting the local concavity / convexity ratio is as follows: The initial height of all points on the C model layers is calculated one by one, and the average height is calculated by summing the initial heights of all points. Points with heights greater than the average height are designated as convex points, and points with heights less than the average height are designated as concave points. The number of convex points and the number of concave points were counted separately, and the local convex-concave ratio was calculated by comparing the number of convex points and the number of concave points. The soil discharge calculation module is used to summarize the three-dimensional and two-dimensional indicators of the same model layer into a comprehensive indicator. The theoretical soil discharge volume corresponding to the comprehensive indicator is predicted by the soil discharge volume prediction model, and the actual soil discharge volume of the model layer is calculated by combining the looseness coefficient of the model layer. The monitoring and prompting module is used to calculate the real-time soil discharge volume by weighted summation of the actual soil discharge volume of the model layer in all transport sections, and to dynamically update the real-time soil discharge volume and adaptively formulate shield tunneling prompt information.

2. The intelligent monitoring system for earth volume of shield tunneling based on image acquisition and processing according to claim 1, characterized in that, The method for acquiring interval images is as follows: Mark a random point on the conveying interval, and continuously adjust the position of the random point until the distance from the random point to the two long sides of the conveying interval is the same, and the distance to the two short sides of the conveying interval is the same. Then, record the adjusted random point as the midpoint of the interval. Measure the length of the shorter side in the conveying section, and record 80% of the length of the shorter side as the unit amplitude; The point located directly above the midpoint of the interval and one unit amplitude away from the midpoint is designated as the upper point; the point located 45 degrees to the left of the midpoint and one unit amplitude away from the midpoint is designated as the left downward point; and the point located 45 degrees to the right of the midpoint and one unit amplitude away from the midpoint is designated as the right downward point. The distances from the top point, left top point, and right top point to the midpoint of the interval are adjusted synchronously until there is no overlap in the images captured at the same point between two adjacent transmission intervals. At the same time, the top image, left top image, and right top image are captured.

3. The intelligent monitoring system for earth volume of shield tunneling based on image acquisition and processing according to claim 2, characterized in that, The method for extracting the proportion of particles exceeding the limit is as follows: The upper-level image is preprocessed to remove noise and enhance contrast. Based on the model layer division method, the preprocessed upper-level image is divided into C sub-images that correspond one-to-one with the model layer. The boundary line between two adjacent stones in the sub-image is identified by the edge detection algorithm, and the area inside the boundary line of the same closed structure is recorded as the stone region. Two boundary points are randomly marked on the boundary line of the stone area. One boundary point is marked as the stationary point and kept still, while the other boundary point is marked as the moving point and keeps moving. The distance between the stationary point and the moving point is measured in real time and recorded as the point-to-point distance value. Points whose distance from one point to another is greater than the standard distance are marked as out-of-limit points. The number of out-of-limit points on the boundary line is divided by the number of boundary points to calculate the sub-proportion. The stone areas with a sub-ratio greater than the standard ratio threshold are recorded as out-of-limit areas. The coverage area of ​​the out-of-limit areas is divided by the coverage area of ​​the sub-image to calculate the out-of-limit particle ratio.

4. The intelligent monitoring system for shield tunneling soil removal volume based on image acquisition and processing according to claim 3, characterized in that, The method for extracting the color value balance ratio is as follows: Convert the sub-image to a grayscale image, mark the grayscale value of all pixels in the grayscale image, record pixels with grayscale values ​​within the same grayscale range as points of the same type, and after summarizing points of the same type, generate D point sets; Count the number of points of the same type and the total number of pixels in the D point sets one by one, and record them as the value of the same type and the total value. Then, compare the maximum value of the same type with the total value and calculate the color value balance ratio.

5. The intelligent monitoring system for shield tunneling soil removal volume based on image acquisition and processing according to claim 4, characterized in that, The actual amount of soil discharged is calculated as follows: The humidity moment when the model layer first appeared on the conveyor belt was found by querying the timestamp one by one. The humidity value at the time of the humidity was detected by the humidity sensor to obtain the soil humidity value. The humidity difference was calculated by subtracting the soil humidity value from the calibrated humidity threshold. The average value of the distances between all points exceeding the limit in the model layer is obtained by summing the distances between them. The difference between the average value of the exceeding limit and the calibrated exceeding threshold is then calculated to obtain the exceeding difference value. The humidity difference and excess difference of the C model layers are assigned corresponding proportional coefficients and then summed to calculate the looseness coefficient; The actual soil discharge volume of the C model layers is calculated by multiplying the theoretical soil discharge volume of each model layer by the corresponding loosening coefficient.

6. The intelligent monitoring system for shield tunneling soil removal volume based on image acquisition and processing according to claim 5, characterized in that, Shield tunneling prompts include information on continued shield tunneling and information on terminated shield tunneling; the method for formulating shield tunneling prompts is as follows: Compare the dynamically updated real-time soil discharge volume with the upper limit of soil discharge volume; When the real-time soil discharge volume is less than the upper limit of the soil discharge volume, continuous shield tunneling information is generated; When the real-time soil discharge volume is greater than or equal to the upper limit of soil discharge volume, the tunnel boring machine is terminated.

Images