Shield tunneling deslagging non-contact real-time measurement and over-discharge prevention intelligent control method

Abstract

The invention discloses a non-contact real-time measurement and super-discharge prevention intelligent control method for shield tunneling deslagging. The method comprises the following steps that 1, equipment is selected and installed; 2, acquiring an image; 3, preprocessing the image; 4, constructing a CNN machine learning model; 5, setting a CNN machine learning model and an alarm mechanism; step 6, establishing a real-time monitoring and feedback system; and 7, recording and analyzing data. The CNN can automatically extract local features in complex data, and is especially suitable for image processing. In the tunneling process, the CNN can recognize useful features such as the working state of the shield tunneling machine from image data, and therefore more accurate reference is provided for super-discharge risk prediction and control. As the CNN can process a large amount of complex data and capture precision changes in the complex data, the prediction precision of the CNN is generally higher than that of a traditional method, and the precision and stability of a control system can be effectively improved by adopting the CNN in anti-super-discharge control.

Description

Technical Field

[0001] The invention relates to a tunnel shield construction control method, in particular to a shield tunneling slag discharge non-contact real-time measurement and over-discharge prevention intelligent control method. Background Art

[0002] Shield machine soil over-discharge refers to the process in which the tunnel boring machine discharges the excavated soil or rock soil to the ground or tunnel during the shield excavation process, exceeding the predetermined discharge volume or discharge rate, resulting in increased pressure in the underground working environment or other adverse effects. Shield tunnel boring machine (TBM) is an important equipment widely used in underground engineering construction, especially in urban underground tunnel construction, and is widely used in tunneling operations of subways, pipelines, underground passages and other projects. Due to the complexity and uncertainty of underground soil layers, shield tunneling operations often face problems such as uneven soil layers, tunnel boring machine power overload, unstable soil pressure, and cutterhead damage. Among them, over-discharge is one of the common safety hazards in shield excavation. If it is not properly controlled, it will lead to construction progress delays, equipment damage, and even accidents. In serious cases, it may affect the stability of surrounding buildings and geological environment. At present, the anti-over-discharge control of shield tunnel boring machines mostly relies on traditional rule-based control systems. Traditional control systems monitor by setting thresholds for parameters such as soil pressure and propulsion force. When the monitoring value exceeds the threshold, the system triggers an alarm and makes a predetermined adjustment. However, this control method based on preset rules has problems such as slow response, lack of intelligence and low control accuracy. Summary of the invention

[0003] In order to solve the problem of over-discharge of slag during the tunneling process of the shield machine, which may cause large settlement or even instability of the stratum, the present invention provides a non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling and slag discharge. The method introduces real-time monitoring of slag over-discharge based on image recognition and CNN neural network.

[0004] The objective of the present invention is achieved through the following technical solutions:

[0005] A non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling slag discharge includes the following steps:

[0006] Step 1: Select device and install it

[0007] Step 1.1: Equipment selection: Equipment includes high-definition cameras, edge computing devices and sensors;

[0008] Step 1.2: Equipment installation and positioning: Install the HD camera at the front of the shield machine to ensure that the field of view covers the entire excavation area and avoids obstacles. The edge computing device is close to the HD camera to reduce data transmission delay and ensure real-time performance.

[0009] Step 1.3: Wiring and configuration: Ensure smooth connections between high-definition cameras, edge computing devices, and sensors, use high-speed data transmission interfaces, and configure device drivers and related software to ensure normal operation. After the equipment is installed, perform functional tests to verify image acquisition, data transmission, and measurement accuracy. Adjust and optimize based on the test results to ensure the system achieves optimal performance.

[0010] Step 2: Acquire the image

[0011] When the shield machine is running, start the high-definition camera to collect real-time images and save them as videos or image sequences to ensure stable data transmission to the computer or server;

[0012] Step 3: Image Preprocessing

[0013] Step 3.1: Use Gaussian filtering to remove noise from the image to ensure a clearer image in subsequent processing;

[0014] Step 3.2: Apply the adaptive thresholding method to calculate the threshold in the local area of ​​the image and dynamically adjust it according to the lighting conditions in different areas to further improve the image quality, thereby enhancing the important features in the image and removing background noise;

[0015] Step 3.3: Enhance the image contrast through gamma correction to make the soil features more obvious;

[0016] Step 3.4: Use the Canny edge detection algorithm to extract the soil boundary, which is helpful for the subsequent soil volume calculation;

[0017] Step 3.5: Convert the image from RGB color space to HSV color space to better distinguish different types of soil;

[0018] Step 3.6: Set a threshold in the HSV space to extract areas with a specific color range to identify the muck;

[0019] Step 4: Build a Convolutional Neural Network (CNN) machine learning model

[0020] Step 4.1: Collect various types of soil images, including soil images in normal state and super-discharge state, to ensure sample diversity;

[0021] Step 4.2: Label the collected images to distinguish different soil conditions: normal, slightly over-discharged, and severely over-discharged, providing labels for training;

[0022] Step 4.3: Use data augmentation techniques to generate additional training samples to increase the robustness of the model;

[0023] Step 4.4: Normalize the image pixel values ​​to between 0 and 1 to ensure the consistency of the input data;

[0024] Step 4.5: Divide the data set into training set, validation set and test set, and perform model training, validation and final testing to ensure that the performance of the model on different data sets meets the expected results;

[0025] Step 4.6: Build and train a CNN machine learning model;

[0026] Step 4.7: Use the trained CNN machine learning model on the test set to extract features and identify the state of the preprocessed images. The model will automatically identify different soil states: normal, slightly over-discharged, and severely over-discharged;

[0027] Step 5: Apply CNN machine learning model and set up alarm mechanism

[0028] Step 5.1: Integrate the trained CNN machine learning model into the real-time monitoring system to classify the images captured by the HD camera;

[0029] Step 5.2: Based on the classification results output by the CNN machine learning model, monitor the soil status in real time and issue an over-discharge alarm in time;

[0030] Step 5.3: Use CNN to identify and segment the muck area from the two-dimensional image, extract the boundary, shape and area information of the muck, and combine the conveyor belt speed and the muck accumulation height to calculate the three-dimensional volume of the muck and the muck discharge rate to improve the recognition accuracy;

[0031] Step 6: Establish a real-time monitoring and feedback system

[0032] Step 6.1: Establish a real-time monitoring and feedback system that can continuously receive the status recognition results from CNN and provide real-time feedback. If an over-discharge situation is detected, the system will quickly generate an alarm;

[0033] Step 6.2: Based on the test results, the system automatically adjusts the working parameters of the shield machine, reduces the excavation speed, and adjusts the earthwork output to prevent over-discharge of slag;

[0034] Step 7: Data Recording and Analysis

[0035] Step 7.1: Record the monitored images, recognition results and system feedback in the database for subsequent analysis and review;

[0036] Step 7.2: Regularly analyze monitoring data, evaluate the system's performance under different environments, and optimize the model and control strategy based on the feedback results.

[0037] Compared with the prior art, the present invention has the following advantages:

[0038] 1. CNN can automatically extract local features from complex data, and is particularly suitable for processing images. During the tunneling process, CNN can identify useful features such as the working status of the shield machine from image data, thereby providing a more accurate reference for over-excavation risk prediction and control.

[0039] 2. CNN can process large-scale data and make real-time predictions, which is crucial for the rapid response of shield tunneling to prevent over-discharge control. Through real-time monitoring and early warning, the tunneling strategy can be adjusted in time to avoid over-discharge.

[0040] 3. Through training, CNN is able to identify complex patterns in large amounts of data, classify different tunneling situations, and predict possible over-discharge risks.

[0041] 4. Since CNN can process a large amount of complex data and capture the accuracy changes therein, its prediction accuracy is usually higher than that of traditional methods. In anti-overload control, the use of CNN can effectively improve the accuracy and stability of the control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of the intelligent control method for preventing over-discharge in shield tunneling. DETAILED DESCRIPTION

[0043] The technical solution of the present invention is further described below in conjunction with the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention should be included in the protection scope of the present invention.

[0044] The present invention provides a non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling and slag discharge, such as Figure 1 As shown, the method comprises the following steps:

[0045] Step 1: Select the device and install it.

[0046] Step 1.1: Equipment selection and installation are key steps to ensure efficient operation of the system. First, the equipment selection mainly includes high-definition cameras, edge computing devices, and sensors. The high-definition camera needs to be waterproof, dustproof, and shock-resistant industrial-grade high-definition camera suitable for the construction site environment. Ensure that the camera has sufficient resolution (1920×1080) and frame rate (30 frames per second). The camera has a wide-angle lens to cover a larger monitoring area, capture clear images, and provide good imaging in low-light environments. The edge computing device has strong computing power and sufficient storage to support real-time image processing and model reasoning.

[0047] Step 1.2: Installation and positioning of equipment. Select a suitable installation location and install the HD camera on the front of the shield machine to ensure that the field of view covers the entire excavation area and avoids obstructions. Adjust the angle of the camera during installation so that it can clearly capture the state of the slag. At the same time, use shock-proof brackets and anti-vibration materials to fix the HD camera to ensure that the HD camera remains stable during operation and reduce the impact of vibration on image quality during construction. The edge computing device should be close to the camera to reduce data transmission delay and ensure real-time performance.

[0048] Step 1.3: Wiring and configuration. Ensure smooth connections between the HD camera, edge computing device, and sensor, using a high-speed data transmission interface (Ethernet). Ensure that the HD camera has a stable power supply, which may need to be connected through the shield machine's power system. Select a suitable data transmission method, such as HDMI, USB, or network connection, to ensure real-time transmission of image data to the processing system. At the same time, configure the device's driver and related software to ensure its normal operation. After the equipment is installed, perform functional testing to verify image acquisition, data transmission, and measurement accuracy, and make adjustments and optimizations based on the test results to ensure the system achieves optimal performance.

[0049] Step 2: Acquire images.

[0050] When the shield machine is running, start the high-definition camera to collect real-time images and save them as videos or image sequences to ensure stable data transmission to the computer or server.

[0051] Step 2.1: According to the model and specifications of the HD camera, determine that its connection method is Ethernet, select a suitable connection interface, and ensure that the computer or embedded system has the corresponding port. Connect the HD camera to the computer or embedded system and ensure that the power supply and data transmission are normal.

[0052] Step 2.2: Install the corresponding driver according to the model of the HD camera to ensure that the system can recognize the camera.

[0053] Step 2.3: Use the image acquisition software Python to perform functional testing and try to capture images or record videos to ensure that the HD camera can work properly and transmit data.

[0054] Step 2.4: Use Python as the image acquisition software and configure the image acquisition parameters. Set the image resolution, frame rate, exposure time and other parameters as needed to ensure image quality.

[0055] Step 3: Image preprocessing.

[0056] Step 3.1: Use Gaussian filtering to remove noise from the image to ensure a clearer image in subsequent processing.

[0057] In two-dimensional space, the Gaussian function can be expressed as:

[0058]

[0059] Where G(x,y) is the Gaussian auxiliary function, which describes the filtering at the point (x,y); σ is the standard deviation of the Gaussian function, which controls the width of the filter; x,y are the coordinates of each pixel relative to the center of the filter.

[0060] Step 3.1.1: Generate a Gaussian kernel of size 5×5 with the specified standard deviation σ.

[0061] Step 3.1.2: Normalize the Gaussian kernel to ensure that the sum of all elements of the Gaussian kernel is 1 to maintain the overall brightness of the image.

[0062] Step 3.1.3: Convolve the Gaussian kernel with the input image and calculate the new value of each pixel.

[0063] Step 3.2: Apply the adaptive thresholding method, which calculates the threshold in the local area of ​​the image and dynamically adjusts it according to the lighting conditions in different areas, thereby enhancing the important features in the image and removing background noise.

[0064] Step 3.2.1: Load the image and convert the color image to grayscale.

[0065] Step 3.2.2: Use Python to perform adaptive thresholding and select the threshold type as "gaussian".

[0066] Step 3.2.3: Adjust the window size according to the characteristics of the image and the illumination changes. Smaller windows are better for processing details, and larger windows are suitable for processing illumination changes in larger areas.

[0067] Step 3.3: Enhance the image contrast through gamma correction to make the soil features more obvious.

[0068] Step 3.3.1: Load the image to be processed and process it using the gamma correction formula:

[0069]

[0070] In the formula, I out is the pixel value of the image after correction; I in is the image pixel value of the original image; γ is the gamma value, which controls the brightness adjustment degree of the image.

[0071] Step 3.3.2: Write the code in Python and compare the images before and after processing to evaluate the effect of gamma correction.

[0072] Step 3.3.3: According to the performance of the soil characteristics under different lighting conditions, the gamma value is adjusted accordingly to find the optimal setting. In this embodiment, the gamma value is 0.6.

[0073] Step 3.3.4: Evaluate the gamma correction results to see if the dirt features in the image are more prominent and to ensure that the contrast enhancement does not result in the loss of important details.

[0074] Step 3.4: Use the Canny edge detection algorithm to extract the soil boundary, which will help with the subsequent soil volume calculation.

[0075] Step 3.4.1: Use the Sobel operator to calculate the gradient of the image, and obtain the horizontal and vertical gradients G through convolution operations. x and G y .

[0076] Step 3.4.2: Calculate the gradient strength and direction of each pixel in the image; the gradient is the direction and magnitude of the change in image brightness, obtained by calculating the derivative of the image in the horizontal and vertical directions. The gradient strength and direction can be calculated using the following formula:

[0077] Gradient strength G:

[0078]

[0079] Gradient direction θ:

[0080]

[0081] Step 3.4.3: When displaying gradient magnitudes, normalize them for easier visualization.

[0082] Step 3.4.4: Convert the angle. The gradient direction in the formula is expressed in radians. Convert it to an angle:

[0083]

[0084] In the formula, θ degree is the angle value of the gradient direction; θ is the radian value of the gradient direction.

[0085] Step 3.5: Convert the image from RGB color space to HSV color space to better distinguish different types of dirt. HSV space is more robust to lighting changes and helps with color recognition.

[0086] Step 3.5.1: HSV expresses color by performing nonlinear conversion on hue (H), saturation (S), and value (V). The specific conversion formula is as follows:

[0087] Hue (H):

[0088]

[0089] Saturation(S):

[0090]

[0091] Value (V):

[0092] V=max(R,G,B)

[0093] In the formula, C is the color difference, C = C max -C min ; R, G, B represent the intensity values ​​of the red, green, and blue channels respectively; H, S, V represent the hue, saturation, and brightness respectively.

[0094] Step 3.5.2: Use the built-in function rgb2hsv to easily convert RGB to HSV.

[0095] Step 3.5.3: After converting to HSV space, it is easier to analyze and distinguish different types of slag. The color of the slag is classified according to the hue value, and the saturation and brightness are adjusted to increase the contrast, so that different types of slag are more prominent.

[0096] Step 3.6: Set a threshold in the HSV space to extract areas of a specific color range to identify muck.

[0097] Step 3.6.1: According to the color of the slag to be identified, set the threshold ranges of H, S, and V. In this embodiment, the upper and lower limits of H are set to 0.1-0.3, the upper and lower limits of S are set to 0.4-1.0, and the upper and lower limits of V are set to 0.3-1.0;

[0098] Step 3.6.2: Use logical operations to create a mask to extract the areas that meet the threshold conditions.

[0099] Step 3.6.3: Extract specific color regions through masking and visualize the results.

[0100] Step 3.6.4: According to the actual image and soil characteristics, the thresholds of H, S, and V need to be continuously adjusted to achieve the best recognition effect. By observing the extraction effect, these parameters can be gradually optimized.

[0101] Step 3.6.5: The extracted muck areas can be used for further analysis and as training data for machine learning models.

[0102] Step 4: Build a convolutional neural network (CNN) machine learning model.

[0103] Step 4.1: Collect various types of soil images, including soil images in normal state and super-discharge state, to ensure sample diversity.

[0104] Step 4.1.1: Considering various factors such as different soil types, different lighting conditions, and different locations during the construction process, select different construction environments and conditions to collect images.

[0105] Step 4.1.2: Make sure the sample size collected is large enough, usually at least hundreds of images in each state, to cover a variety of possible situations, which helps improve the generalization ability of the model and reduce overfitting.

[0106] Step 4.2: Annotate the collected images to distinguish different soil conditions (normal, slightly over-discharged, and severely over-discharged) and provide labels for training.

[0107] Step 4.2.1: Determine the labeling category. According to the project requirements, the slag status can be divided into multiple categories. In this embodiment, it is divided into normal status (slag discharge is in line with expectations and there are no abnormal conditions), slight over-discharge (slag discharge is slightly higher than the preset standard, but still within an acceptable range) and severe over-discharge (slag discharge obviously exceeds the standard and may affect construction safety).

[0108] Step 4.2.2: Import the collected images into the annotation tool and prepare to start annotation.

[0109] Step 4.2.3: For each image, select appropriate areas for annotation based on the appearance and characteristics of the slag. In the annotation tool, select the corresponding slag state (normal, slightly over-discharged, severely over-discharged) to add labels to the annotated areas. If needed, the slag in different states can be annotated more carefully, such as the size and color of the slag particles, for subsequent analysis.

[0110] Step 4.2.4: After completing the annotation, save the annotation results in CSV format. These files will be used for model training together with the original images.

[0111] Step 4.3: Use data augmentation techniques (rotation, scaling, flipping) to generate additional training samples to increase the robustness of the model.

[0112] Step 4.3.1: Randomly rotate the image by a certain angle to help the model adapt to the soil images in different directions. The rotation angle is set between -30° and 30° in this embodiment.

[0113] Step 4.3.2: Randomly scale the image to simulate the effects of shooting at different distances. A range of scaling factors can be set, which is 0.8 to 1.2 in this embodiment.

[0114] Step 4.3.3: Flipping the image horizontally or vertically helps the model learn left-right or top-bottom symmetric features.

[0115] Step 4.3.4: Randomly translate the image to help the model adapt to the different positions of the object in the image.

[0116] Step 4.3.5: Randomly adjust the brightness and contrast of the image to enhance the model’s adaptability to lighting changes.

[0117] Step 4.3.6: Combine different data augmentation methods to create a data generator. During the model training phase, the data generator dynamically generates augmented samples.

[0118] Step 4.3.7: After the model training is completed, use the validation set to evaluate the model performance, compare the results with and without data augmentation, and observe the performance of the model under different test conditions.

[0119] Step 4.4: Normalize the image pixel values ​​to between 0 and 1 to ensure the consistency of the input data.

[0120] H channel normalization:

[0121]

[0122] S channel normalization:

[0123]

[0124] V channel normalization:

[0125]

[0126] In the formula, H normalized , S normalized 、V normalized Respectively represent the normalized hue, saturation, and brightness values;

[0127] Step 4.5: Divide the dataset into training set, validation set and test set, and train, validate and finally test the model to ensure that the performance of the model on different datasets reaches the expected effect. Use 70% for training, 15% for validation and 15% for testing.

[0128] Step 4.5.1: In order to ensure the randomness of the division, it is recommended to randomly shuffle the data set before division, so as to avoid the deviation that may be caused by the sample order.

[0129] Step 4.5.2: When dividing the data set, the class distribution of the samples should also be considered, especially in classification tasks. You can use stratified sampling to ensure that the proportion of each class in the training set, validation set, and test set is similar.

[0130] Step 4.5.3: After the division is completed, check the number of samples and category distribution of each data set to ensure that the division meets expectations and can effectively support subsequent model training and evaluation.

[0131] Step 4.5.4: After completing the division of the data set, the model can be trained, verified and finally tested to ensure that the performance of the model on different data sets meets the expected results.

[0132] Step 4.6: Build and train a CNN machine learning model.

[0133] Step 4.7: Use the trained CNN machine learning model on the test set to perform feature extraction and state recognition on the preprocessed images. The model will automatically identify different soil states: normal, slightly over-discharged, and severely over-discharged.

[0134] Step 5: Apply CNN machine learning model and set up alarm mechanism.

[0135] Step 5.1: Integrate the trained CNN machine learning model into the real-time monitoring system in step 6 to classify the images captured by the HD camera.

[0136] Step 5.2: Based on the classification results output by the CNN machine learning model, the soil status is monitored in real time and an over-discharge alarm is issued in time.

[0137] Step 5.2.1: After classifying the real-time captured images using the trained convolutional neural network (CNN), the CNN machine learning model outputs the probabilities of different soil states;

[0138] Step 5.2.2: Set the threshold value to 0.7 and judge the state of the slag. In this embodiment, if the probability of "severe over-discharge" exceeds the set threshold value (0.7), it is determined that the current state of the slag is severe over-discharge.

[0139] Step 5.2.3: Set up an over-discharge alarm mechanism. No alarm will be issued in normal state, only record but no alarm will be issued for a minor over-discharge, and an alarm will be issued immediately for a serious over-discharge.

[0140] Step 5.3: Use CNN to identify and segment the muck area from the two-dimensional image, and extract the boundary, shape, area and other information of the muck. Combined with the conveyor belt speed and the muck pile height (obtained by the laser rangefinder), the three-dimensional volume of the muck and the slag discharge rate can be calculated to improve the accuracy of recognition.

[0141] Step 5.3.1: Use a laser rangefinder to obtain the average accumulation height of the slag area.

[0142] Step 5.3.2: Calculate the volume of the muck in each frame using the following formula:

[0143] V 渣土 =A 渣土 ×H 渣土

[0144] Step 5.3.3: Calculate the volume of soil discharge per unit time by combining the conveyor belt running speed and the image frame rate (N is the total number of frames)

[0145]

[0146] Step 6: Establish a real-time monitoring and feedback system.

[0147] Step 6.1: Establish a real-time monitoring and feedback system that can continuously receive the status recognition results from CNN and provide real-time feedback. If an over-discharge situation is detected, the system will quickly generate an alarm.

[0148] Step 6.1.1: Build a real-time feedback loop so that the output of the CNN can quickly influence the control strategy of the shield machine;

[0149] Step 6.1.2: Adopt adaptive control algorithm to ensure automatic adjustment of excavation parameters such as propulsion speed, torque control, earthwork output, etc. according to the soil status output by CNN: In the case of slight or severe over-discharge, appropriately reduce propulsion speed to reduce soil discharge, adjust the torque of shield machine according to soil characteristics and soil status to ensure smooth excavation and avoid severe over-discharge, and in the case of severe over-discharge, adjust the earthwork output rate to reduce discharge.

[0150] Step 6.1.3: By analyzing the long-term soil discharge data during the excavation process, a statistical model between soil discharge and ground deformation is established. If there is a significant fluctuation in the discharge volume and there is a certain correlation with the settlement or displacement changes of the surrounding soil layers, it can be inferred whether the ground has exceeded the standard deformation.

[0151] If the amount of soil discharge suddenly increases and lasts for a period of time, it means that the tunnel boring machine has entered a soft and easily deformed soil layer. This is because the ground is loose, causing the tunnel boring machine to dig more soil to complete the excavation task, thus causing the soil layer to sink or partially collapse.

[0152] If the amount of soil discharged decreases, it may mean that the shield machine has entered a harder soil layer or that the soil layer has become too stable. It may also mean that there are problems such as insufficient thrust of the tunnel boring machine and excessive support of the soil layer.

[0153] Step 6.2: Based on the test results, the system automatically adjusts the working parameters of the shield machine, reduces the excavation speed, and adjusts the earth output to prevent over-discharge of debris, which is achieved through the servo control system.

[0154] Step 7: Data recording and analysis.

[0155] Step 7.1: Record the monitored images, recognition results and system feedback in the database for subsequent analysis and review.

[0156] Step 7.2: Regularly analyze monitoring data, evaluate the system's performance under different environments, and optimize the model and control strategy based on the feedback results.

Claims

1. A non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling, characterized in that The method comprises the following steps: Step 1: Select device and install it Step 1.1: Equipment selection: Equipment includes high-definition cameras, edge computing devices and sensors; Step 1.2: Equipment installation and positioning: Install the HD camera at the front of the shield machine to ensure that the field of view covers the entire excavation area and avoids obstacles. The edge computing device is close to the HD camera to reduce data transmission delay and ensure real-time performance. Step 1.3: Wiring and configuration: Ensure smooth connections between high-definition cameras, edge computing devices, and sensors, use high-speed data transmission interfaces, and configure device drivers and related software to ensure normal operation. After the equipment is installed, perform functional tests to verify image acquisition, data transmission, and measurement accuracy. Adjust and optimize based on the test results to ensure the system achieves optimal performance. Step 2: Acquire the image When the shield machine is running, start the high-definition camera to collect real-time images and save them as videos or image sequences to ensure stable data transmission to the computer or server; Step 3: Image Preprocessing Step 3.1: Use Gaussian filtering to remove noise from the image to ensure a clearer image in subsequent processing; Step 3.2: Apply the adaptive thresholding method to calculate the threshold in the local area of ​​the image and dynamically adjust it according to the lighting conditions in different areas to further improve the image quality, thereby enhancing the important features in the image and removing background noise; Step 3.3: Enhance the image contrast through gamma correction to make the soil features more obvious; Step 3.4: Use the Canny edge detection algorithm to extract the soil boundary, which is helpful for the subsequent soil volume calculation; Step 3.5: Convert the image from RGB color space to HSV color space to better distinguish different types of soil; Step 3.6: Set a threshold in the HSV space to extract areas with a specific color range to identify the muck; Step 4: Build a CNN machine learning model Step 4.1: Collect various types of soil images, including soil images in normal state and super-discharge state, to ensure sample diversity; Step 4.2: Label the collected images to distinguish different soil conditions: normal, slightly over-discharged, and severely over-discharged, providing labels for training; Step 4.3: Use data augmentation techniques to generate additional training samples to increase the robustness of the model; Step 4.4: Normalize the image pixel values ​​to between 0 and 1 to ensure the consistency of the input data; Step 4.5: Divide the data set into training set, validation set and test set, and perform model training, validation and final testing to ensure that the performance of the model on different data sets meets the expected results; Step 4.6: Build and train a CNN machine learning model; Step 4.7: Use the trained CNN machine learning model on the test set to extract features and identify the state of the preprocessed images. The model will automatically identify different soil states: normal, slightly over-discharged, and severely over-discharged; Step 5: Apply CNN machine learning model and set up alarm mechanism Step 5.1: Integrate the trained CNN machine learning model into the real-time monitoring system to classify the images captured by the HD camera; Step 5.2: Based on the classification results output by the CNN machine learning model, monitor the soil status in real time and issue an over-discharge alarm in time; Step 5.3: Use CNN to identify and segment the muck area from the two-dimensional image, extract the boundary, shape and area information of the muck, and combine the conveyor belt speed and the muck accumulation height to calculate the three-dimensional volume and muck discharge rate of the muck to improve the accuracy of recognition; Step 6: Establish a real-time monitoring and feedback system Step 6.1: Establish a real-time monitoring and feedback system that can continuously receive the status recognition results from CNN and provide real-time feedback. If an over-discharge situation is detected, the system will quickly generate an alarm; Step 6.2: Based on the test results, the system automatically adjusts the working parameters of the shield machine, reduces the excavation speed, and adjusts the earthwork output to prevent over-discharge of slag; Step 7: Data Recording and Analysis Step 7.1: Record the monitored images, recognition results and system feedback in the database for subsequent analysis and review; Step 7.2: Regularly analyze monitoring data, evaluate the system's performance under different environments, and optimize the model and control strategy based on the feedback results.

2. The non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling and slag discharge according to claim 1 is characterized in that The specific steps of step 3.1 are as follows: Step 3.1.1: Generate a Gaussian kernel of size 5×5 according to the specified standard deviation σ; Step 3.1.2: Normalize the Gaussian kernel to ensure that the sum of all elements of the Gaussian kernel is 1 to maintain the overall brightness of the image; Step 3.1.3: Convolve the Gaussian kernel with the input image and calculate the new value of each pixel; The specific steps of step 3.2 are as follows: Step 3.2.1: Load the image and convert the color image to grayscale; Step 3.2.2: Use Python to perform adaptive threshold processing and select the threshold type as "gaussian"; Step 3.2.3: Adjust the window size according to the image characteristics and lighting changes; The specific steps of step 3.3 are as follows: Step 3.3.1: Load the image to be processed and process it using the gamma correction formula: In the formula, I out is the pixel value of the image after correction; I in is the image pixel value of the original image; γ is the gamma value, which controls the brightness adjustment degree of the image; Step 3.3.2: Write the code in Python and compare the images before and after processing to evaluate the effect of gamma correction; Step 3.3.3: Adjust the gamma value accordingly according to the performance of soil characteristics under different lighting conditions; Step 3.3.4: Evaluate the gamma correction results to see if the soil features in the image are more prominent and ensure that the contrast enhancement does not result in the loss of important details. The specific steps of step 3.4 are as follows: Step 3.4.1: Use the Sobel operator to calculate the gradient of the image, and obtain the horizontal and vertical gradients G through convolution operations. x and G y ; Step 3.4.2: Calculate the gradient strength and direction for each pixel in the image; Step 3.4.3: When displaying the gradient magnitude, normalize it for easier visualization; Step 3.4.4: Convert the angle. The gradient direction in the formula is expressed in radians. Convert it to an angle: In the formula, θ degree is the angle value of the gradient direction; θ is the radian value of the gradient direction; The specific steps of step 3.5 are as follows: Step 3.5.1: HSV expresses color by performing nonlinear transformation on hue H, saturation S, and lightness V; Step 3.5.2: Use the built-in function rgb2hsv to easily convert RGB to HSV; Step 3.5.3: Classify the colors of the slag according to the hue value, and adjust the saturation and brightness to increase the contrast, so that different types of slag are more prominent; The specific steps of step 3.6 are as follows: Step 3.6.1: Set the threshold ranges of H, S, and V according to the color of the slag to be identified; Step 3.6.2: Use logical operations to create a mask and extract the areas that meet the threshold conditions; Step 3.6.3: Extract specific color regions through masks and visualize the results; Step 3.6.4: According to the actual image and soil characteristics, the thresholds of H, S, and V need to be continuously adjusted to achieve the best recognition effect. By observing the extraction effect, these parameters can be gradually optimized; Step 3.6.5: The extracted muck areas can be used for further analysis and as training data for machine learning models.

3. The non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling and slag discharge according to claim 2 is characterized in that In step 3.4.2, the gradient strength and direction are calculated by the following formula: Gradient strength G: Gradient direction θ:

4. The non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling and slag discharge according to claim 2 is characterized in that In step 3.5.1, the specific conversion formula is as follows: V=max(R,G,B) In the formula, C is the color difference, C = C max -C min ; R, G, B represent the intensity values ​​of the red, green, and blue channels respectively; H, S, V represent the hue, saturation, and brightness respectively.

5. The non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling and slag discharge according to claim 1 is characterized in that The specific steps of step 4.1 are as follows: Step 4.1.1: Considering different soil types, different lighting conditions, and different locations during the construction process, select different construction environments and conditions to collect images; Step 4.1.2: Make sure the sample size is large enough, usually at least hundreds of images per state, to cover a wide range of possible situations. The specific steps of step 4.2 are as follows: Step 4.2.1: Determine the labeling category. According to the project requirements, the soil status is divided into multiple categories: normal status, slight over-discharge and severe over-discharge; Step 4.2.2: Import the collected images into the annotation tool and prepare to start annotation; Step 4.2.3: For each image, select appropriate areas for annotation based on the appearance and characteristics of the slag. In the annotation tool, select the corresponding slag state to add a label to the annotated area; Step 4.2.4: After completing the annotation, save the annotation results in CSV format. These files will be used for model training together with the original images. The specific steps of step 4.3 are as follows: Step 4.3.1: Randomly rotate the image by a certain angle to help the model adapt to the muck images in different directions; Step 4.3.2: Randomly scale the image to simulate the effect of shooting at different distances; Step 4.3.3: Flipping the image horizontally or vertically helps the model learn left-right or top-bottom symmetric features; Step 4.3.4: Randomly translate the image to help the model adapt to the different positions of the object in the image; Step 4.3.5: Randomly adjust the brightness and contrast of the image to enhance the model's adaptability to lighting changes; Step 4.3.6: Combine different data augmentation methods to create a data generator. During the model training phase, the data generator dynamically generates augmented samples. Step 4.3.7: After the model training is completed, use the validation set to evaluate the model performance, compare the results with and without data augmentation, and observe the performance of the model under different test conditions.

6. The non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling and slag discharge according to claim 1 is characterized in that In step 4.4, the normalization formulas for the H, S, and V channels are: In the formula, H normalized , S normalized 、V normalized Represent the normalized hue, saturation, and brightness values ​​respectively.

7. The non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling and slag discharge according to claim 1 is characterized in that The specific steps of step 5.2 are as follows: Step 5.2.1: After using the trained CNN to classify the real-time captured images, the CNN machine learning model will output the probabilities of different soil states; Step 5.2.2: Set a threshold to determine the state of the slag. If the probability of "severe over-discharge" exceeds the set threshold, the current slag state is determined to be severe over-discharge; Step 5.2.3: Set up an over-discharge alarm mechanism. In normal conditions, no alarm will be sounded. In case of a minor over-discharge, only record the discharge but do not sound an alarm. In case of a serious over-discharge, an alarm will be sounded immediately.

8. The non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling and slag discharge according to claim 1 is characterized in that The specific steps of step 5.3 are as follows: Step 5.3.1: Obtain the average accumulation height of the slag area by using a laser rangefinder; Step 5.3.2: Calculate the volume of the muck in each frame using the following formula: In 渣土 =A 渣土 ×H 渣土 Step 5.3.3: Calculate the volume of soil discharged per unit time by combining the conveyor belt running speed and the image frame rate: Where N is the total number of frames.

9. The non-contact real-time measurement and over-discharge prevention intelligent control method for shield tunneling and slag discharge according to claim 1 is characterized in that The specific steps of step 6.1 are as follows: Step 6.1.1: Build a real-time feedback loop so that the output of the CNN can quickly influence the control strategy of the shield machine; Step 6.1.2: Adopt adaptive control algorithm to ensure automatic adjustment of excavation parameters such as propulsion speed, torque control, and earthwork output according to the soil status output by CNN: in the case of slight or severe over-discharge, appropriately reduce propulsion speed to reduce soil discharge; adjust shield machine torque according to soil characteristics and soil status to ensure smooth excavation and avoid severe over-discharge; in the case of severe over-discharge, adjust earthwork output rate to reduce discharge; Step 6.1.3: By analyzing the long-term soil discharge data during the excavation process, a statistical model between soil discharge and ground deformation is established. If there is a significant fluctuation in the discharge volume and there is a certain correlation with the settlement or displacement changes of the surrounding soil layers, it can be inferred whether the ground has exceeded the standard deformation.

Images