A tunnel intelligent monitoring system and method based on data collection hierarchical processing

Abstract

The application discloses a tunnel intelligent monitoring system and method based on data collection hierarchical processing, pre-processes visual data to obtain clear structured data, judges the data amount of the clear structured data based on a hierarchical threshold, judges the stage to which the data amount of the current structured data belongs, processes the structured data according to the belonging stage, when the stage is a large sample stage, transmits the clear structured data to a large model data processing module for deep analysis, processes the clear structured data by using a multi-level and multi-dimensional data analysis framework constructed based on a large model, analyzes the data to obtain the maximum deformation condition in the maximum strain area, compares the obtained deformation condition with a warning threshold, and performs risk warning. The hierarchical data layout transmission mode and the triggered small sample incremental learning mechanism are adopted, and the machine vision and the large model are combined to dynamically learn the model, so that the limitations in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering safety monitoring technology, and in particular to a tunnel intelligent monitoring system and method based on data acquisition and hierarchical processing. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] In the construction and operation of tunnels in karst terrain, traditional detection methods rely heavily on manual experience and simple threshold alarms due to unstable geological structures, karst cave development, groundwater, and surrounding rock deformation. These methods cannot meet the demands of real-time data transmission and analysis in the face of the dynamic changes in complex karst terrain. The rapid development of artificial intelligence and machine vision technologies has provided new options for automatic monitoring systems in tunnel engineering. Especially considering the need for synchronization and accuracy in data acquisition and processing during the construction of complex karst tunnels, the application of machine vision data acquisition and large-scale model data processing in engineering construction has become a trend.

[0004] Traditional monitoring systems suffer from limitations in effectively integrating data collected from different monitoring modules (such as displacement, stress, temperature and humidity sensors, and gas concentration sensors) to comprehensively analyze risks, and their data processing efficiency is low. Traditional monitoring system algorithms also lack the ability to model complex nonlinear relationships (such as the coupling of surrounding rock and multiple influencing factors in complex karst tunnels), resulting in high error rates. Furthermore, traditional intelligent detection systems rely on network transmission for data transfer; data loss is easily caused by tunnel network signal interruptions, and model updates are often delayed, requiring manual retraining and hindering the ability to dynamically adjust early warning strategies based on data accumulation. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this invention provides a tunnel intelligent monitoring system and method based on hierarchical data acquisition and processing. It adopts a hierarchical data deployment and transmission method and a triggered small sample incremental learning mechanism, and uses a combination of machine vision and large model to perform dynamic model learning, thus solving the limitations of the prior art.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, the present invention provides a tunnel intelligent monitoring system based on hierarchical data acquisition and processing, comprising:

[0008] The data acquisition module is used to acquire visual data inside the tunnel; the visual data includes raw RGB images and real-time video streams.

[0009] A data preprocessing module is configured to preprocess the visual data to obtain clear structured data.

[0010] A data cache transmission module is configured to judge the data volume of the clear structured data based on a hierarchical threshold, judge the stage to which the data volume of the current structured data belongs, and process the structured data according to the stage; when the stage is a large sample stage, the clear structured data is transmitted to a large model data processing module for deep analysis.

[0011] A large model data processing module is configured to process the clear structured data by using a multi-level and multi-dimensional data analysis framework constructed based on a large model, analyze the data to obtain the maximum deformation condition in the maximum strain region, and perform risk warning after comparing the obtained deformation condition with a warning threshold.

[0012] A chip architecture module based on RISC-V is configured to structure the data acquisition module, the data preprocessing module, the data cache transmission module, and the large model data processing module into a chip.

[0013] In a further technical solution, the preprocessing of the visual data specifically includes filtering the visual data based on a CycleGAN defogging network, correcting the blur by using an algorithm in combination with inertial measurement unit data, filtering the visual data, and finally converting the filtered data into structured data based on the spatial scale of a target.

[0014] In a further technical solution, the stage includes a small sample stage, a medium sample stage, and a large sample stage. When the data volume of the structured data is less than a first threshold, the stage is the small sample stage. When the data volume of the structured data is greater than or equal to the first threshold and less than a second threshold, the stage is the medium sample stage. When the data volume of the structured data is greater than or equal to the second threshold, the stage is the large sample stage.

[0015] In a further technical solution, when the stage is the small sample stage, the structured data is cached locally, and the structured data is used to optimize a local monitoring model.

[0016] In a further technical solution, when the stage is the medium sample stage, a rule engine is used to determine whether the structured data is abnormal. If the structured data is abnormal, an abnormality flag is triggered.

[0017] In a further technical solution, the multi-level and multi-dimensional data analysis framework divides the structured data into an original signal layer, a feature abstraction layer, and a semantic understanding layer. Each layer is further divided into a space dimension, a time dimension, and an attribute dimension.

[0018] In a further technical solution, after the structured data is divided into different levels and different dimensions, a deformation propagation dynamics equation is established and expressed as:

[0019]

[0020] wherein, 、 denotes the displacement of the target point and its adjacent target points , denotes time, denotes the mechanical coupling coefficient of the target point at the target point position, is an external load influence factor, denotes a time-dependent external excitation function.

[0021] In a second aspect, the present application provides a tunnel intelligent monitoring method based on data acquisition hierarchical processing, comprising:

[0022] obtaining visual data inside the tunnel; the visual data includes original RGB images and real-time video streams;

[0023] preprocessing the visual data to obtain clear structured data;

[0024] judging the data amount of the clear structured data based on hierarchical thresholds, judging the stage to which the data amount of the current structured data belongs, and processing the structured data according to the stage; when the stage is a large sample stage, the clear structured data is transmitted to a large model data processing module for deep analysis;

[0025] processing the clear structured data by using a multi-level and multi-dimensional data analysis framework constructed based on a large model, analyzing the data to obtain the maximum deformation condition in the maximum strain area, and performing risk early warning after comparing the obtained deformation condition with a warning threshold.

[0026] In a third aspect, the present application provides a computer readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the steps of the tunnel intelligent monitoring method based on data acquisition hierarchical processing according to the second aspect.

[0027] In a fourth aspect, the present application provides a computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the tunnel intelligent monitoring method based on data acquisition hierarchical processing according to the second aspect when executing the program.

[0028] The above one or more technical solutions have the following beneficial effects:

[0029] The tunnel intelligent monitoring system provided by the application carries out hierarchical processing on the collected machine vision data, is divided into three levels according to different data quantities, when the data quantity is < threshold A, the data is locally cached, and the local monitoring model is optimized by using limited data to avoid overfitting, when threshold A <= data quantity < threshold B, the rule engine can be combined to realize rapid local early warning and reduce calculation delay, when the data quantity is >= threshold B, the remote transmission module is triggered, and the data is uploaded to the large model data processing platform through the network for deep analysis. When the machine vision data is collected, the data quantity is constantly changing, according to the data grading provided by the application, the efficiency of data processing can be improved, and the memory occupation under different data quantities is effectively considered, and the memory occupation when the data is more is effectively reduced.

[0030] The application provides a large model data processing module, and the construction of the large model data processing module is based on a large model open source on the market to build a multi-level and multi-dimensional data analysis framework. After the data is divided in different levels and different dimensions, the collected multi-source data is coupled to avoid independent analysis of the data and cause the error rate of risk prediction.

[0031] In the application, the large model data processing module can generate a three-dimensional tunnel model according to existing data, dynamically mark abnormal points according to the data collected by machine vision, and click to view historical data curves and associated parameters. Meanwhile, the future risk visualization module is provided, which can display the historical data curve (such as displacement change with time) of the point, support zooming and panning, quickly locate a data segment of a specific time period, support exporting CSV or PDF format report, contain a curve diagram and key statistical values (maximum value, average value, variance), and further mark a risk level (red / yellow / green) in the three-dimensional tunnel model and display a future risk hot area. BRIEF DESCRIPTION OF DRAWINGS

[0032] The drawings constituting a part of the specification of the application are used to provide further understanding of the application, and the illustrative embodiments of the application and the description thereof are used to explain the application, and do not constitute improper limitation on the application.

[0033] Figure 1 It is a flowchart of the tunnel intelligent monitoring method of the embodiment of the application. DETAILED DESCRIPTION

[0034] It should be pointed out that the following detailed description is exemplary and is intended to provide further explanation of the application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the application belongs.

[0035] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, devices, components and / or combinations thereof, but do not preclude the presence or addition of one or more other features, steps, operations, devices, components, and / or combinations thereof.

[0036] The embodiments in the present application and the features in the embodiments can be combined with each other in the case of no conflict.

[0037] Embodiment one

[0038] As shown in the figure, the embodiment discloses a tunnel intelligent monitoring system based on data collection hierarchical processing, comprising: Figure 1 A data collection module is configured to acquire visual data inside the tunnel; the visual data includes original RGB images and real-time video streams;

[0039] A data preprocessing module is configured to preprocess the visual data to obtain clear structured data;

[0040] A data cache transmission module is configured to judge the data volume of the clear structured data based on hierarchical thresholds, judge the stage to which the data volume of the current structured data belongs, and process the structured data according to the stage; when the stage is a large sample stage, the clear structured data is transmitted to a large model data processing module for deep analysis;

[0041] The large model data processing module is configured to process the clear structured data by using a multi-level and multi-dimensional data analysis framework constructed based on a large model, analyze the data to obtain the maximum deformation in the maximum strain area, and perform risk warning after comparing the obtained deformation with a warning threshold;

[0042] A chip architecture module based on RISC-V is configured to structure the data collection module, the data preprocessing module, the data cache transmission module and the large model data processing module into a chip.

[0043] In the embodiment, the data collection module is configured to acquire visual data inside the tunnel. The data collection module includes a target (made of a specially-made high-reflective waterproof material) and a machine vision measuring instrument, and an electrical cabinet. The visual data includes high-resolution original RGB image data and real-time video streams.

[0044]

[0045] ​The machine vision measuring instrument adopts a static machine vision measuring instrument for multi-scenario monitoring, a POE is used for power supply, and the measurement distance is within 0-400 m. The sampling rate is less than 1HZ, the measurement direction is transverse and longitudinal, and the calibration mode is achieved by AI algorithm to automatically correct the angle and distance influence, without distance measurement and leveling, greatly improving the monitoring efficiency and reducing the waste of human resources. The precision of the static machine vision measuring instrument can reach millimeter level, and the working temperature is-40℃ to 80℃, which ensures high precision and multi-scenario application. The machine vision measuring instrument is internally provided with a professional sensor to obtain IMU (inertial measurement unit) data, and the professional sensor includes an accelerometer, a gyroscope and a magnetometer. The accelerometer can detect the displacement of the mass block under the action of acceleration through a micro-electro-mechanical system (MEMS), the gyroscope measures the rotational angular velocity by using the Coriolis effect or a vibrating structure, and the magnetometer determines the direction of the device relative to the geomagnetic north pole by the Hall effect.

[0046] The target adopts a common target which can be attached to the measured object. In order to reduce the interference of environmental light, a dust scattering resistant ring LED array or a near infrared light source is used for active light compensation. The light intensity is adjusted according to the dust concentration in the tunnel to avoid overexposure or underexposure.

[0047] A plurality of targets are arranged on the tunnel structure, and a machine vision measuring instrument is installed at a position with relatively stable structure. The machine vision measuring instrument identifies the target image on the structure, and when the measured structure moves, the target coordinates change, so that the horizontal and vertical bidirectional displacement of the measured object is measured. The machine vision measuring instrument inputs 4K@30fps RAW12 data through a MIPI-CSI2 interface, which is convenient for subsequent chip data storage.

[0048] An electrical cabinet is arranged to protect the electrical equipment for storing the tunnel monitoring site, including a power supply system (including an air switch, an earth leakage protection, a power converter), a router, a switch, a lightning protection system (optional), and other accessories (wiring harness, antenna support, etc.).

[0049] The application has applicability to tunnels constructed by different methods, and can collect vision and arrange hardware modules according to the construction method and the site condition of the tunnel. For example, when the traditional drill and blast method is used for construction, a static machine vision measuring instrument can be added to the rock drilling jumbo, and the camera on the working face is retracted into the explosion-proof cabin before blasting, which not only plays a real-time monitoring role, but also protects the facilities from being damaged in blasting; when the TBM (full-face tunnel boring machine) is used for construction, a track type sliding table can be arranged in the tunnel to carry the machine vision measuring instrument for movement, covering the full-face detection of the tunnel, and a mobile platform is installed at the tail of the TBM for real-time monitoring.

[0050] In the embodiment, the data preprocessing module is used for preprocessing the visual data to obtain clear structured data. Before the collected visual data is cached and transmitted, the visual data is filtered based on the CycleGAN defogging network to remove dust or fog and restore clear images, and IMU (inertial measurement unit) data obtained by the sensor arranged in the machine vision measuring instrument device is combined to obtain the motion state information of the collected device, including real-time measurement of acceleration, angular velocity and attitude change of the object, professional algorithms are used to correct the blur caused by mechanical vibration, and the visual data is filtered locally; finally, the filtered data is converted based on the spatial scale of the target to obtain the coordinate, size, shape description and other structured data obtained by preprocessing, and cached and transmitted. The filtering of the visual data based on the CycleGAN defogging network in the module is performed by the VPU to generate YUV420 format data stream for storage, and the chip integration architecture is realized by RISC-V.

[0051] In the embodiment, the data caching and transmission module is used for judging the data amount of the clear structured data based on the hierarchical threshold, judging the stage to which the data amount of the current structured data belongs, and processing the structured data according to the stage; when the stage is the large sample stage, the clear structured data is transmitted to the large model data processing module for deep analysis.

[0052] In order to realize low delay and high reliability of data transmission, and considering that there is a signal blind area in the tunnel, the data amount is divided into a small sample stage, a medium sample stage and a large sample stage, and different storage and transmission strategies are adopted in each stage. Specifically, two thresholds are set as threshold A and threshold B, and the three stages of data amount are classified.

[0053] Threshold A: the minimum data amount threshold for triggering switching from small sample incremental learning (small sample stage) to medium sample local rule engine (medium sample stage). The minimum sample amount required for incremental learning is determined according to the pre-experiment, and the prediction error of the model in the sample amount is verified by cross-validation: , , wherein is the allowable error.

[0054] Threshold B: the critical data amount threshold for triggering switching from the medium sample local rule engine to remote cloud analysis (large sample stage). It is determined according to the network bandwidth and the allowable transmission time of single transmission data amount:

[0055]

[0056] , wherein ​represents the initial critical data volume threshold, represents the minimum data volume requirement to trigger data transmission; represents the network bandwidth, which refers to the maximum rate of the data transmission channel; represents the maximum allowed upload time, which is the system's time limit for a single data transmission; represents the sample transmission cost, including data compression rate, network resource occupation, and other comprehensive costs.

[0057] It can be dynamically adjusted according to the network situation in different tunnels and the allowed transmission time, which is more beneficial to the application in different situations.

[0058] (1) When the amount of collected data is less than the minimum sample size required for local monitoring model fine-tuning, i.e., the amount of collected data < threshold A, the structured data is cached locally, and the local monitoring model is optimized using limited data to avoid overfitting. The local monitoring model uses a multi-source transfer-multi-task network (MT-MTNet) to pre-train three similar karst tunnel monitoring data, load the pre-trained model, and input the locally pre-processed structured data, such as displacement value (μ), surrounding rock grade (I~V), groundwater pressure (MPa), and crack parameters calculated by sliding window. The displacement statistics (μ, σ), geological parameters, and crack parameters are concatenated into a mixed feature vector and adjusted in real time by a dynamic weighting mechanism to determine the proportion of each parameter. The local monitoring model is trained to improve the accuracy of the local monitoring model in correlating displacement and settlement curves and locating high-risk areas with medium sample size.

[0059] The initial monitoring data collected in the tunnel is insufficient to directly train a high-precision model. The local monitoring model is a pre-trained model based on general monitoring tasks (such as crack identification and settlement prediction). The model is pre-trained using monitoring data from other tunnels (such as similar geological conditions or structural types) and transferred to the target tunnel. According to the collected data and meta-learning, the model quickly adapts to a small amount of local data, achieving the goal of pre-training the model and quickly adapting and optimizing the local collected data.

[0060] Local new data , where represents the number of samples of the new data, represents the input features (or independent variables) of the th data point, represents the output label (or dependent variable) of the th data point. The local monitoring model selects the algorithm based on the fine-tuning of transfer learning and online gradient descent (Online Gradient Descent). The weighted loss is defined for the local new data, which is represented as:

[0061]

[0062] wherein, , represents a weight coefficient, balancing new data fitting and local monitoring model stability; represents a predicted output of the model to input , represents an optimized model parameter, represents a parameter of a pre-trained base model.

[0063] (2) When threshold A ≤ data volume < threshold B, it is a medium sample stage. When the collected data volume reaches the medium sample stage, it reaches the stage that can be prewarned. The rule engine can be combined to realize fast local prewarning, reduce calculation delay, and the prewarning of the medium sample stage is preliminary prewarning. The current moment of the structured data (displacement value), the sliding window mean and variance are dynamically calculated:

[0064] ,

[0065] wherein, represents a sliding window mean, represents a size of a sliding window, represents an index of a current time point, represents a summation index variable (the end point ), represents a displacement value collected at a time point , represents a sliding window variance, represents a sliding window standard deviation.

[0066] When , an abnormality flag is triggered, at which time preliminary prewarning can be performed according to the existing data volume (triggering the abnormality flag), and after reaching the large sample data volume, key attention can be paid.

[0067] ​(3) When the data volume is greater than or equal to threshold B, it is a large sample stage, at which time the remote transmission module is triggered to upload the structured data to the large model data processing module through the network for deep analysis. The remote transmission module provides two transmission modes, wired transmission and wireless transmission. The wired transmission is optical fiber transmission, which is fixed along the side wall or top of the tunnel to prevent corrosion, and is connected with the sensor network (such as strain gauge, displacement meter) through an optical transceiver; an optical fiber junction box is set every 500-1000 meters for easy segmented maintenance; this transmission mode is suitable for long tunnel (such as high-speed rail tunnel, cross-mountain tunnel) or whole line monitoring far away from the city network with poor signal quality. The wireless transmission is cellular network transmission, which is directly connected through the 4G / 5G base station deployed around the tunnel and the leakage cable or small base station in the underground section to extend the signal to upload data. This transmission mode is suitable for urban tunnels with good operator signal coverage. When remote transmission, the H.265 encoding rate (4Mbps-20Mbps) is adjusted according to the network state (such as 5G or Wi-Fi), and the RISC-V NoC bus is used to transmit key area data preferentially.

[0068] Through the large model data processing module, the risk prediction of subsequent construction projects can be realized, and the solution is automatically generated according to the constructed solution library. Data transmission is performed through the signal base station to ensure the transmission of high-definition video stream (bandwidth > 100Mbps), end-to-end delay < 50ms, so that the data is uploaded to the large model data processing module (large model data processing platform) with low delay.

[0069] In the embodiment, the large model data processing module is used to process the clear structured data by using a multi-level and multi-dimensional data analysis framework constructed based on a large model to obtain an analysis result. When the data volume is greater than or equal to threshold B, the data is transmitted to a data processing platform constructed based on a large model, i.e., a large model data processing module. The platform is directly connected with the data platform output by the machine vision measuring instrument, avoiding the problems of data loss or low efficiency when the data is transmitted twice.

[0070] The big model data processing module is based on the open source big model on the market to build a multi-level and multi-dimensional data analysis framework. It divides data into three different levels. According to the data collection source and physical signal form, it is divided into the original signal layer. According to the information abstraction degree and task relevance, it is divided into the feature abstraction layer. According to the engineering semantics and decision requirements, it is divided into the semantic understanding layer. The hierarchical processing chain is used to process data at different levels. The original signal layer uses lightweight CNN to process images and Kalman filter to process sensor data. The feature abstraction layer extracts spatial features of tunnel point cloud through 3D ResNet. The semantic understanding layer fuses time and space features through multi-modal Transformer and outputs semantic labels. Each level contains specific information. In addition, not only single-dimensional problems are focused on, but also information is classified from multiple dimensions. According to the geographical structure distribution, time evolution law and engineering characteristics of data, it is divided into three dimensions of space, time and attribute for decoupling analysis. After the big model data processing module divides data into different levels and dimensions, it couples the collected data in different aspects based on the weighting coefficient and establishes the deformation propagation dynamics equation:

[0071]

[0072] wherein, , denotes the displacement of the target point and its adjacent target point , denotes time, denotes the mechanical coupling coefficient at the position of the target point and the target point , is an external load influence factor, denotes a time-dependent external excitation function.

[0073] The module completes the INT8 quantization and layer fusion of YOLOv5-Tiny on the PC end to generate a binary model suitable for RISC-Tensor instructions.

[0074] In this embodiment: the hierarchical early warning module establishes an abnormal score function in the big model data processing platform according to all the coupled data collected, which is represented as:

[0075]

[0076] wherein, denotes the abnormal score at time , the larger the value, the more significant the system abnormality; denotes the number of feature dimensions, denotes the predicted value of the i-th feature, denotes the predicted value of the i-th feature, denotes the predicted value of the i-th feature. actual observation value of the feature, denotes the decay coefficient, denotes the current time interval from data generation time.

[0077] set a monitoring deformation threshold, when trigger a hierarchical early warning ( adjust dynamically according to the rock mass category), which is conducive to improving efficiency, reducing waste of human resources, and improving the accuracy of the monitoring process.

[0078] The early warning is divided into three levels in total. The triggering condition of the first level early warning (emergency) is that the above-mentioned abnormal score exceeds the preset safety threshold (such as the displacement value breaks through the upper limit for 3 times within 10 minutes) for many times in succession, or the deviation between the prediction result of the large model and the measured value continuously expands. Triggering the first level early warning immediately starts the sound and light alarm device, and highlights the abnormal position through the local display screen; if the network is available, the emergency notification is pushed to the management personnel's mobile phone and the monitoring platform at the same time.

[0079] The triggering condition of the second level early warning (potential risk) is that the comprehensive abnormal score exceeds the limit for a single data but is not sustained, or the trend analysis shows that the parameter change rate exceeds the historical average level. Triggering the second level early warning pops up a yellow warning box on the local screen, records the abnormal log and generates a daily report for subsequent manual checking.

[0080] The triggering condition of the third level early warning (long-term trend anomaly) is that the abnormal score fluctuates within the safety range, but presents a one-way slow change (such as continuous subsidence) for a long time (such as 72 hours). Triggering the third level early warning generates a periodic health evaluation report, prompting the operation and maintenance personnel to pay attention.

[0081] In addition, the large model data processing platform sets a false alarm suppression mechanism. Through time window verification, the instantaneous abnormal signal, i.e. the abnormal score exceeding the threshold for a moment, is repeatedly detected within the time window (such as appearing twice within 5 minutes), the manual confirmation interface supports the on-site personnel to manually mark the false alarm through touch screen or mobile terminal, and the system automatically learns and corrects the rule weight.

[0082] Future risk visualization module: The large model data processing platform uses a lightweight three-dimensional engine or professional BIM software (such as Revit) to parse the tunnel structure design file provided by the project, and converts it into an interactive OpenGL model. After generating a three-dimensional tunnel model, the abnormal signal obtained based on the large model is dynamically marked on the three-dimensional tunnel model, and the historical data curve and associated parameters can be viewed by clicking. The future risk is visualized, and the visualization panel can display the historical data curve (such as displacement change over time) of the point. Zooming, panning, and quick positioning of specific time period data segments are supported. CSV or PDF format reports containing graphs and key statistical values (maximum, mean, variance) can be exported. The risk level (red / yellow / green) can also be marked in the three-dimensional tunnel model to display the future risk hot zone.

[0083] In this embodiment, the large model data processing platform simultaneously constructs a solution library by integrating industry standards, engineering cases, and expert experience to form standardized response strategies (such as crack repair solutions and grouting reinforcement processes). Based on actual treatment effect feedback, the solution library is continuously optimized (such as updating the priority of a certain type of leakage treatment solution after improving its efficiency).

[0084] Matching and recommending solutions, according to the risk type (such as structural deformation and water leakage), warning severity (first / second / third level), and site conditions (workability, resource availability) prompted by the large model data platform, the optimal solution is selected from the solution library. Manual intervention is also supported to adjust the priority (such as selecting a temporary support solution when the construction period is tight). The risk warning module exchanges data through shared memory and executes decision rules by the main core.

[0085] In this embodiment, the tunnel intelligent monitoring system also includes a RISC-V-based chip architecture module. Based on the RISC-V reduced instruction set, the above-mentioned module architecture is integrated into a chip. The core layer of the hardware part uses a 64-bit C906 core RISC-V processor. The main core is responsible for the scheduling and system management of each module, and the real-time core uses E24 to handle sensor interrupts and low-latency responses. Specifically, the RISC-V-based chip design integrates the following modules: a data acquisition module responsible for acquiring raw visual data from external sensors or other data sources; a data preprocessing module for preprocessing collected visual data to improve data quality; a data cache transmission module for providing temporary storage for preprocessed data and efficiently transmitting preprocessed data to the large model data processing module; and a large model data processing module supporting complex large model data processing tasks.

[0086] The visual processing unit and the large model data processing acceleration module respectively adopt integrated SIMD vector instruction extension and deployed Tensor instruction extension, support image filtering, geometric correction and other operations, directly connect the camera MIPI interface and support INT8 quantization inference, process the crack identification model. The interface layer architecture adopts a dual-channel DMA controller to realize direct transmission of camera data to memory, avoiding CPU load, and the local cache part occupies 2MB shared L2 cache, which is divided into an image preprocessing area and a local monitoring model pre-training model parameter area. Based on the RISC-V reduced instruction set, the data acquisition module, the data preprocessing module, the data cache transmission module, the large model data processing module, and the hierarchical early warning and future risk visualization module, the solution intelligent generation module are structured into a chip, which can realize the integration and optimization of the detection system.

[0087] Embodiment Two

[0088] The embodiment discloses a tunnel intelligent monitoring method based on data acquisition hierarchical processing, comprising:

[0089] Obtain visual data inside the tunnel;

[0090] Preprocess the visual data to obtain clear visual data;

[0091] Judge the data amount of the clear visual data based on a hierarchical threshold, judge the stage to which the data amount of the current visual data belongs, and process the visual data according to the stage; when the stage is a large sample stage, the clear visual data is transmitted to a large model data processing module for deep analysis;

[0092] Process the clear visual data by using a multi-level and multi-dimensional data analysis framework constructed based on a large model to obtain an analysis result.

[0093] Embodiment Three

[0094] The purpose of the embodiment is to provide a computing device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to realize the steps of the method of embodiment two.

[0095] Embodiment Four

[0096] The purpose of the embodiment is to provide a computer-readable storage medium, a computer-readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to execute the steps of the method of embodiment two.

[0097] The steps involved in the apparatuses of the above embodiments three and four correspond to the method of embodiment two, and the specific implementation can refer to the relevant description of embodiment two. The term "computer readable storage medium" should be understood as including a single medium or multiple media of one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying the instruction set for execution by a processor and causing the processor to perform any of the methods in the present application.

[0098] Those skilled in the art should understand that each module or step of the present application described above can be realized by a general computer device, alternatively, they can be realized by program codes executable by a computing device, so that they can be stored in a storage device for execution by a computing device, or they can be respectively manufactured into each integrated circuit module, or a plurality of modules or steps among them can be manufactured into a single integrated circuit module to realize. The present application is not limited to any specific combination of hardware and software.

[0099] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0100] The above describes the specific embodiments of the present application in combination with the accompanying drawings, but is not intended to limit the protection scope of the present application. Those skilled in the art should understand that various modifications or changes made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the protection scope of the present application.

Claims

1. A tunnel intelligent monitoring system based on hierarchical data acquisition and processing, characterized in that, include: The data acquisition module is used to acquire visual data inside the tunnel; The visual data includes raw RGB images and real-time video streams; The data preprocessing module is used to preprocess the visual data to obtain clear structured data; The data caching and transmission module is used to determine the amount of the clear structured data based on a hierarchical threshold, determine the stage to which the current amount of structured data belongs, and process the structured data according to the stage. When the stage is a large sample stage, the clearly structured data is transmitted to the large model data processing module for in-depth analysis; The large model data processing module is used to process the clear structured data using a multi-level and multi-dimensional data analysis framework based on the large model, analyze the data to obtain the maximum deformation in the maximum strain area, and conduct risk warning based on the obtained deformation and the warning threshold. The RISC-V-based chip architecture module is used to integrate the data acquisition module, data preprocessing module, data buffering and transmission module, and large model data processing module into the chip.

2. The tunnel intelligent monitoring system based on hierarchical data acquisition and processing as described in claim 1, characterized in that, The preprocessing of the visual data specifically involves: filtering the visual data based on the CycleGAN dehazing network, and using an algorithm to correct the blurring by combining the inertial measurement unit data. Finally, the filtered data is converted based on the spatial proportion of the target to obtain structured data.

3. The tunnel intelligent monitoring system based on hierarchical data acquisition and processing as described in claim 1, characterized in that, The stages include a small sample stage, a medium sample stage, and a large sample stage. When the amount of structured data is less than the first threshold, it belongs to the small sample stage; when the amount of structured data is greater than or equal to the first threshold and less than the second threshold, it belongs to the medium sample stage; and when the amount of structured data is greater than or equal to the second threshold, it belongs to the large sample stage.

4. The tunnel intelligent monitoring system based on hierarchical data acquisition and processing as described in claim 3, characterized in that, During the small sample stage, the structured data is cached locally, and the local monitoring model is optimized using the structured data.

5. The tunnel intelligent monitoring system based on hierarchical data acquisition and processing as described in claim 3, characterized in that, During the mid-sample stage, the rule engine is used to determine whether there are anomalies in the structured data. If anomalies are found, an anomaly marker is triggered.

6. The tunnel intelligent monitoring system based on hierarchical data acquisition and processing as described in claim 3, characterized in that, The multi-level and multi-dimensional data analysis framework divides structured data into a raw signal layer, a feature abstraction layer, and a semantic understanding layer. Each layer is further divided into three dimensions: space, time, and attributes.

7. The tunnel intelligent monitoring system based on hierarchical data acquisition and processing as described in claim 6, characterized in that, After dividing the structured data into different levels and dimensions, a deformation propagation dynamics equation is established, expressed as: in, , Indicate target point and its adjacent target points The displacement, Indicates time, Indicate target point With target The mechanical coupling coefficient at the location, External load influence factor Represents the time-dependent external excitation function.

8. A tunnel intelligent monitoring method based on hierarchical data acquisition and processing, characterized in that, include: Acquire visual data of the tunnel interior; The visual data includes raw RGB images and real-time video streams; The visual data is preprocessed to obtain clear structured data; The amount of the clear structured data is determined based on the hierarchical threshold, the stage to which the current structured data belongs is determined, and the structured data is processed according to the stage. When the stage is a large sample stage, the clearly structured data is transmitted to the large model data processing module for in-depth analysis; A multi-level, multi-dimensional data analysis framework based on a large model is used to process the clear structured data, analyze the data to obtain the maximum deformation in the maximum strain region, and then conduct risk warning based on the obtained deformation and the warning threshold.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the steps in the tunnel intelligent monitoring method based on data acquisition and hierarchical processing as described in claim 8.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the tunnel intelligent monitoring method based on hierarchical data acquisition and processing as described in claim 8.

Images