Tunnel blasting vibration monitoring method and system based on image recognition

Abstract

The application provides a tunnel blasting vibration monitoring method and system based on image recognition, and relates to the technical field of tunnel safety monitoring. The method comprises the following steps: obtaining a geological parameter set of a target area of a tunnel and an image sequence at a blasting moment, performing wave analysis on the geological parameters to obtain vibration characteristic data reflecting the non-uniform characteristics of the rock mass, simultaneously performing optical flow analysis and regional identification processing on the image sequence, generating a displacement vector field and an identified deformation region respectively, correlating the vector field with the micro-deformation region in space to form vibration field data with the structural characteristics of the rock mass, coupling the vibration field data with the vibration characteristic data to reconstruct a three-dimensional propagation path, and finally identifying a damage risk region of the target area of the tunnel based on the energy distribution pattern on the path. The application improves the damage risk identification accuracy of the tunnel rock mass under blasting vibration.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel safety monitoring, and particularly relates to a tunnel blasting vibration monitoring method and system based on image recognition. BACKGROUND

[0002] In tunnel engineering construction, the vibration caused by blasting operation can affect the stability of the surrounding rock mass, especially when there are adjacent buildings on the surface or in the ground under complex geological conditions, a technical means is needed to monitor the vibration speed of the affected building structure in real time and accurately identify the potential risk area.

[0003] The prior art technical solution adopts the method of arranging a vibration sensor array, analyzes the vibration signal amplitude and frequency characteristics recorded by the sensor, and combines the geological survey data to infer the vibration propagation trend, so as to evaluate the influence of blasting vibration on the rock mass. This method can only monitor the vibration speed at the sensor arrangement position, and in order to obtain the vibration characteristics of a certain area of the structure, a large number of sensors need to be arranged, which is too high in economic cost.

[0004] The technical solution has limitations in presenting the vibration propagation path, and the changes in vibration energy distribution caused by the structural differences of the rock mass are not intuitive enough, and the judgment of the damage area under complex rock conditions deviates greatly from the actual situation. SUMMARY

[0005] The present application provides a tunnel blasting vibration monitoring method and system based on image recognition to solve the problem of low damage risk identification accuracy of the tunnel rock mass under blasting vibration in the prior art.

[0006] To solve the above technical problems, in a first aspect, the present application provides a tunnel blasting vibration monitoring method based on image recognition, comprising:

[0007] Obtaining a set of geological parameters of a target area of a tunnel and an image sequence of the target area of the tunnel;

[0008] Performing wave analysis on the set of geological parameters to obtain vibration characteristic data;

[0009] Synchronously performing optical flow analysis and region identification processing on the image sequence, wherein the displacement vector field is generated by pixel displacement calculation in the optical flow analysis process, and the deformation region is identified by spatial feature extraction in the region identification processing process;

[0010] Spatially correlating the displacement vector field with the deformation region to form vibration field data, and coupling the vibration field data with the vibration characteristic data to reconstruct a three-dimensional propagation path;

[0011] Based on the energy distribution of the three-dimensional propagation path, a damage risk area in a tunnel target area is identified.

[0012] Optionally, the coupling of the vibration field data and the vibration feature data to reconstruct a three-dimensional propagation path comprises:

[0013] Fusion of the motion features extracted from the vibration field data and the wave features extracted from the vibration feature data;

[0014] An impedance model is established based on rock mass strength information of the set of geological parameters;

[0015] The fused features are simulated by the impedance model in combination with a physical information neural network to calculate a propagation trajectory of vibration energy in the rock stratum space;

[0016] Deformation degree information is extracted from the deformation area, and the propagation trajectory is directionally corrected in combination with the motion direction of the pixel points and the deformation degree information to generate a three-dimensional propagation path.

[0017] Optionally, the simulation of the fused features by the impedance model in combination with the physical information neural network to calculate the propagation trajectory of the vibration energy in the rock stratum space comprises:

[0018] The fused features are input to a physical information neural network, and the fused features are regularity-constrained by a physical constraint layer of the physical information neural network to obtain a preliminary propagation field;

[0019] The preliminary propagation field is coupled and calculated with the impedance model by an impedance coupling layer of the physical information neural network, and based on the coupling calculation result, an attenuation process of the vibration energy in different lithology areas is simulated to obtain an energy attenuation field;

[0020] Spatial feature extraction of the energy attenuation field is performed by a feature extraction layer of the physical information neural network to obtain a spatial distribution field of the vibration energy;

[0021] A main propagation path of the vibration energy is extracted from the spatial distribution field by a path generation layer of the physical information neural network, and based on the main propagation path, a propagation trajectory of the vibration energy in the rock stratum space is generated.

[0022] Optionally, the image sequence is synchronously subjected to optical flow analysis and region identification processing, wherein a displacement vector field is generated by pixel displacement calculation in the optical flow analysis process, and a deformation area is identified by spatial feature extraction in the region identification processing, comprising:

[0023] Analyze consecutive image frames in the image sequence to establish a mapping relationship;

[0024] Based on the mapping relationship, the motion speed and direction of each pixel are determined to form an initial displacement set;

[0025] The initial displacement set is detected to identify abnormal and missing vectors. The abnormal vectors are removed and the missing vectors are filled in to generate a displacement vector field.

[0026] By using multiple convolutional layers of a convolutional neural network, feature extraction is performed on the image sequence to obtain image feature maps at different scales;

[0027] By using the feature fusion layer of a convolutional neural network, feature maps of different scales are fused to extract texture variation and deformation features of the rock surface;

[0028] The classification output layer of the convolutional neural network identifies and outputs deformed regions based on the texture change features and the deformation features.

[0029] Optionally, identifying the damage risk area in the tunnel target region based on the energy distribution of the three-dimensional propagation path includes:

[0030] Analyze the energy distribution in the three-dimensional propagation path to obtain the energy concentration degree and energy propagation direction;

[0031] The energy concentration region is determined based on the energy concentration degree and the energy propagation direction;

[0032] By combining the joint and fracture information in the set of geological parameters, the structurally weak points in the energy concentration region are identified;

[0033] Based on the degree of spatial overlap between the weak structural parts and the energy concentration area, a damage risk index is calculated.

[0034] Based on the distribution of the damage risk index, the damage risk areas in the tunnel target area are determined.

[0035] Optionally, the step of spatially associating the displacement vector field with the deformation region to form vibration field data includes:

[0036] Motion vector information is extracted from the displacement vector field, and deformation position information is extracted from the deformation region;

[0037] Align the motion vector information with the deformation position information to establish a mapping relationship between the motion vector and the deformation position;

[0038] Based on the joint and fracture information in the set of geological parameters, the mapping relationship is structurally constrained by a graph convolutional network to generate vibration field data.

[0039] Optionally, the step of performing wave analysis on the set of geological parameters to obtain vibration characteristic data includes:

[0040] Extract rock mass strength information and joint and fracture information from the set of geological parameters;

[0041] Under each known analysis scale in multiple analysis scale tables, the fluctuation trend characteristics of the rock mass strength information and the joint and fracture information are calculated respectively, and the fluctuation trend characteristics are removed to obtain the scale fluctuation signals corresponding to each type of information under the analysis scale.

[0042] Extract the amplitude and morphological change information of the scale fluctuation signal corresponding to each type of information at all analysis scales, and perform scale correlation processing on the amplitude and morphological change information of all information at different analysis scales to form a correlation feature set;

[0043] Vibration feature data are obtained through feature fusion processing based on the aforementioned associated feature set.

[0044] Secondly, this application provides a tunnel blasting vibration monitoring system based on image recognition, comprising:

[0045] The acquisition module is used to acquire a set of geological parameters and an image sequence of the tunnel target area.

[0046] The analysis module is used to perform wave analysis on the set of geological parameters to obtain vibration characteristic data;

[0047] The generation module is used to simultaneously perform optical flow analysis and region recognition processing on the image sequence. In the optical flow analysis process, a displacement vector field is generated by calculating pixel displacement, and in the region recognition processing process, deformed regions are identified by extracting spatial features.

[0048] The association module is used to spatially associate the displacement vector field with the deformation region to form vibration field data, and to couple the vibration field data with the vibration feature data to reconstruct the three-dimensional propagation path;

[0049] The identification module is used to identify the damage risk area in the target area of ​​the tunnel based on the energy distribution of the three-dimensional propagation path.

[0050] Thirdly, this application provides an electronic device, comprising:

[0051] Memory, used to store computer programs;

[0052] A processor is configured to execute the computer program to implement the steps of the image recognition-based tunnel blasting vibration monitoring method described in the first aspect above.

[0053] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the image recognition-based tunnel blasting vibration monitoring method described in the first aspect above.

[0054] This application provides a method for monitoring tunnel blasting vibration based on image recognition. The method includes: acquiring a set of geological parameters and an image sequence of the tunnel target area; performing wave analysis on the geological parameter set to obtain vibration characteristic data; simultaneously performing optical flow analysis and region identification processing on the image sequence, wherein, during optical flow analysis, a displacement vector field is generated by calculating pixel displacement, and during region identification processing, deformation regions are identified by extracting spatial features; spatially associating the displacement vector field with the deformation regions to form vibration field data, and coupling the vibration field data with the vibration characteristic data to reconstruct a three-dimensional propagation path; and identifying damage risk areas within the tunnel target area based on the energy distribution of the three-dimensional propagation path.

[0055] The technical solution provided in this application has the following beneficial effects:

[0056] This application first establishes a comprehensive data foundation for subsequent analysis by acquiring a set of geological parameters and image sequences of the target tunnel area. Then, fluctuation analysis is performed on the geological parameter set to accurately reflect the non-uniform characteristics of the rock mass at different scales. Based on this, optical flow analysis and region identification processing are simultaneously carried out to fully capture the blasting vibration response characteristics from kinematic and deformation perspectives, respectively. Subsequently, the displacement vector field is spatially correlated with the deformation region to form vibration field data that reflects the structural characteristics of the rock mass. Next, this vibration field data is coupled with vibration characteristic data to accurately reconstruct the three-dimensional propagation path of vibration energy in the rock strata. Finally, based on the energy distribution of the three-dimensional propagation path, the damage risk area is accurately identified.

[0057] Furthermore, this application also simulates the vibration propagation attenuation process by fusing the motion characteristics of vibration field data and the fluctuation characteristics of vibration characteristic data, combining impedance models and physical information neural networks, and using deformation degree information and motion direction to correct the propagation trajectory, ultimately generating an accurate three-dimensional propagation path.

[0058] Furthermore, by fully considering the impact of the non-uniformity and structural characteristics of the rock mass on vibration propagation, and combining multi-source data fusion with physical constraints, this scheme effectively improves the accuracy and reliability of vibration energy propagation path reconstruction, thus providing more precise technical support for the safety monitoring of vibrations of nearby buildings induced by tunnel blasting.

[0059] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 A flowchart illustrating an image recognition-based method for monitoring tunnel blasting vibration, provided as an embodiment of this application;

[0062] Figure 2 A schematic diagram illustrating a specific implementation of an image recognition-based method for monitoring tunnel blasting vibration, as provided in this application embodiment;

[0063] Figure 3 This is a schematic diagram of the structure of a tunnel blasting vibration monitoring system based on image recognition, provided in an embodiment of this application. Detailed Implementation

[0064] Existing vibration sensor array solutions have significant limitations in presenting vibration propagation paths, such as implementation difficulties and high economic costs. Their reliance on discrete point measurement data for trend inference makes it difficult to intuitively reflect the changes in vibration energy distribution caused by differences in the internal structure of the rock mass. Furthermore, the judgment of the damaged area is prone to errors under complex rock strata conditions.

[0065] To address the aforementioned issues, this application proposes an image recognition-based method for monitoring tunnel blasting vibration. This method constructs a three-dimensional propagation path by integrating geological parameter analysis and image sequence processing. Specifically, it first simultaneously acquires geological parameters and dynamic displacement images of the target area of ​​the tunnel at the moment of blasting, and then combines wave analysis and neural network processing to obtain the vibration response characteristics and micro-deformation area information of the rock mass. Next, it spatially correlates the displacement vector field with the micro-deformation area to form vibration field data that reflects the structural characteristics of the rock mass. This data is then coupled with the vibration response characteristics to reconstruct the three-dimensional propagation path. This method effectively overcomes the limitations of traditional point-based measurements. Relying on the collaborative analysis of image recognition and geological features, it achieves a visualized presentation of the vibration propagation process, significantly improving the accuracy of identifying damage risk areas under complex rock strata conditions, thus providing a new technical approach for tunnel blasting safety monitoring.

[0066] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0067] The core of this application is to provide a method for monitoring tunnel blasting vibration based on image recognition, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0068] Step 101: Obtain the set of geological parameters and the image sequence of the tunnel target area.

[0069] In step 101, the target area of ​​the tunnel refers to the working face of the tunnel excavation; the set of geological parameters includes rock mass strength information and joint and fissure information. Rock mass strength information represents the ability of rock to resist external force damage, and joint and fissure information represents the degree of development and distribution characteristics of fissures in the rock mass; the image sequence is a combination of motion images of the surface of nearby buildings affected by blasting vibration, continuously acquired by high-speed photography equipment at the moment of blasting.

[0070] In this embodiment of the application, geological exploration equipment is used to measure the rock mass strength parameters and joint distribution parameters of the tunnel target area, thereby forming a set of geological parameters; at the same time, high-speed photography equipment is used to continuously acquire multiple frames of rock mass surface images at fixed time intervals during the blasting operation, and these images are arranged in chronological order to form an image sequence; the two types of data acquired are used together as the basic input for subsequent analysis.

[0071] For example, during the blasting construction of a tunnel project, a rock strength meter was first used to measure the rock strength at 42 measuring points in the target area of ​​the tunnel, with the measured values ​​ranging from 35.5 to 58.1 MPa. At the same time, joint and fracture information in the area was obtained through geological logging, recording the joint density distribution of 2 to 6 joints per meter. In addition, a high-speed photography device with a frame rate of 25,000 frames per second was used to simultaneously acquire continuous images of the moment of blasting, thus obtaining an image sequence containing 256 frames. These measured data together provided the basic support for subsequent vibration analysis.

[0072] Step 102: Perform wave analysis on the set of geological parameters to obtain vibration characteristic data.

[0073] In step 102, vibration characteristic data is a quantitative indicator that characterizes the rock mass's response to blasting vibrations, reflecting the differences in the rock mass's wave behavior at different scales.

[0074] In this embodiment of the application, the rock mass strength information and joint and fracture information in the geological parameter set are divided into multiple scales, and the local fluctuation characteristics of the data are calculated at each scale while removing the trend component; then the fluctuation amplitude characteristics and morphological characteristics at each scale are extracted, and then these characteristics at different scales are correlated and integrated to finally form vibration characteristic data that can characterize the non-uniformity of the rock mass.

[0075] For example, the rock mass strength information and joint and fracture information obtained in step 101 are divided into 8 analysis scales, and the local trend characteristics are calculated using the sliding window method at each scale. Then, the trend value is subtracted from the original data to obtain the corresponding wave signal. Next, the amplitude range and zero-crossing rate characteristics of the wave signal at each scale are extracted, and then the principal component analysis method is used to fuse these different scale characteristics to finally obtain vibration characteristic data containing 5 main components.

[0076] Step 103: Perform optical flow analysis and region recognition processing on the image sequence simultaneously. In the optical flow analysis process, a displacement vector field is generated by calculating pixel displacement. In the region recognition process, deformed regions are identified by extracting spatial features.

[0077] In step 103, the displacement vector field is a two-dimensional field composed of pixel motion vectors and directions, and the deformation region is the small deformation region on the rock surface caused by blasting vibration.

[0078] In this embodiment, pixel matching is performed on consecutive frames in the image sequence to establish pixel displacement relationships between adjacent frames, and the motion velocity and direction of each pixel are calculated to form an initial displacement set. Then, motion consistency detection is used to remove abnormal vectors and fill in missing regions, thereby generating an optimized displacement vector field. At the same time, a convolutional neural network is used to extract multi-scale features of the image, identify texture changes and deformation regions on the rock surface, and thus determine the spatial distribution of micro-deformation regions.

[0079] For example, for the image sequence acquired in step 101, the optical flow algorithm is first used to calculate the pixel displacement between adjacent frames to obtain an initial set containing 100,000 motion vectors; then, 5% of the abnormal vectors are removed by neighborhood consistency detection and 3% of the missing regions are interpolated to generate a displacement vector field; at the same time, a 5-layer convolutional neural network is used to extract image features and identify 3 deformation regions with deformation feature values ​​exceeding 0.15; these results together provide the kinematic and deformation data required for subsequent analysis.

[0080] Step 104: Spatially correlate the displacement vector field with the deformation region to form vibration field data, and couple the vibration field data with the vibration characteristic data to reconstruct the three-dimensional propagation path.

[0081] In step 104, the vibration field data is a comprehensive dataset that integrates motion vectors and deformation characteristics. The three-dimensional propagation path is a trajectory model of vibration energy propagating in the rock strata space. Vibration energy refers to the energy carried by the mechanical vibration generated by blasting as it propagates in the rock mass. It originates from the blasting action and is transmitted through the rock mass medium. It is related to the blasting parameters, rock mass characteristics, and propagation distance. The rock strata refer to the underground rock geological body that includes the tunnel target area and its rock mass. In terms of structural relationship, the tunnel target area is a local working face that has been excavated and exposed in the rock strata. The tunnel target area is the specific rock medium within the working face.

[0082] In this embodiment, the motion vector in the displacement vector field is aligned with the deformation position of the deformation region to establish a mapping relationship between motion and deformation. Structural constraints are applied to this mapping relationship based on joint and fracture information to generate vibration field data. Subsequently, the vibration field data is fused with vibration characteristic data, and a corresponding impedance model is established in combination with the rock mass strength distribution to simulate the attenuation process of vibration propagation and finally reconstruct the three-dimensional propagation path.

[0083] For example, the displacement vector field obtained in step 103 is registered with the deformation region, and the mapping relationship is optimized by a graph convolutional network based on the joint distribution information obtained in step 101 to generate vibration field data. Then, the vibration field data is fused with the vibration feature data obtained in step 102, and a corresponding impedance model is established by combining the rock mass strength distribution. A physical information neural network is used to simulate the vibration propagation process, and finally, a three-dimensional propagation path that can clearly show the diffraction of energy in the joint region is reconstructed.

[0084] Step 105: Based on the energy distribution of the three-dimensional propagation path, identify the damage risk area in the tunnel target area.

[0085] In step 105, the energy distribution of the three-dimensional propagation path refers to the propagation trajectory and intensity distribution characteristics of vibration energy in the three-dimensional space of the rock mass. It is obtained by fusing geological parameters, vibration response characteristics, and blasting vibration field data and reconstructing them after lithological constraints. The damage risk area is the spatial range in which there is a potential possibility of rock mass failure.

[0086] In this embodiment, the energy accumulation characteristics and propagation direction in the three-dimensional propagation path are analyzed to determine the energy concentration area. At the same time, the distribution of rock mass joints is combined to identify the weak parts of the structure. Then, the spatial overlap between the energy concentration area and the weak parts of the structure is calculated, and finally the specific location and range of the damage risk area are determined based on the overlap index.

[0087] For example, based on the three-dimensional propagation path obtained in step 104, two energy concentration areas are identified by analyzing the energy distribution pattern, with maximum energy intensities of 45 kJ / m³ and 38 kJ / m³, respectively. Combined with the joint and fracture information obtained in step 101, weak structural parts with dense joints are further identified. Subsequently, the damage risk index R = 0.6 × (E / 25) + 0.4 × (J / 3) is used for calculation, where E is the energy intensity and J is the joint density. Finally, damage risk areas with a risk index greater than 0.75 are determined, thus providing precise guidance for engineering protection.

[0088] This method integrates geological parameter analysis and image sequence processing to achieve three-dimensional visualization monitoring of the tunnel blasting vibration propagation process, thereby accurately identifying damage risk areas under complex rock strata conditions and ultimately providing effective technical support for tunnel construction safety.

[0089] To further improve the accuracy of vibration energy propagation path reconstruction, in some embodiments, step 104 involves coupling the vibration field data with the vibration characteristic data to reconstruct the three-dimensional propagation path, such as... Figure 2 As shown, it includes:

[0090] Step 201: Fuse the motion features extracted from the vibration field data with the wave features extracted from the vibration feature data.

[0091] In step 201, the structural characteristics of the tunnel target area refer to the inherent geological structural characteristics of the rock mass, and the motion characteristics are the dynamic response characteristics exhibited based on the structural characteristics of the rock mass. The fluctuation characteristics in the vibration characteristic data are obtained by performing fluctuation analysis on the geological parameters, reflecting the response characteristics and fluctuation patterns of the rock mass to external vibration excitation at different scales.

[0092] In this embodiment, the velocity magnitude and direction angle of the motion vector are extracted from the vibration field data as motion features, and the wave amplitude and morphological indices at various scales are extracted from the vibration feature data as wave features. Then, through feature splicing and normalization processing, these two types of features are integrated into a feature vector of a unified dimension, thereby forming a fused feature set.

[0093] Step 202: Based on the rock mass strength information of the geological parameter set, establish an impedance model.

[0094] In step 202, the rock mass strength information of the geological parameter set is obtained through on-site rock mechanics testing, representing the spatial variation and distribution characteristics of the rock mass strength in the target area of ​​the tunnel; the impedance model is a spatial distribution model describing the magnitude of resistance encountered by vibration waves when propagating in the rock mass.

[0095] In this embodiment of the application, based on the rock mass strength measurement data in the geological parameter set, a continuous strength distribution map is first generated by spatial interpolation method. Then, an impedance calculation model is established based on the proportional relationship between strength and wave impedance, thereby converting the strength distribution into the corresponding impedance distribution, and finally forming an impedance model covering the entire target area of ​​the tunnel.

[0096] Step 203: Using the impedance model and a physical information neural network, the fused features are simulated to calculate the propagation trajectory of vibration energy in the rock strata.

[0097] In step 203, the propagation trajectory is the initial path of vibrational energy propagation in the rock strata.

[0098] In this embodiment, the fused feature set is input into the physical information neural network, and the wave equation constraint is applied using the physical constraint layer in the network to ensure that the propagation process conforms to the wave theory. At the same time, the attenuation behavior of vibration energy in different impedance regions is simulated by combining the impedance model, and the vibration propagation field that satisfies the physical laws is obtained through iterative calculation. Finally, the main propagation trajectory of vibration energy is extracted from it.

[0099] Step 204: Extract deformation degree information from the deformed region, combine the movement direction of the pixel with the deformation degree information, and perform directional correction on the propagation trajectory to generate a three-dimensional propagation path.

[0100] In step 204, the deformation degree information is a parameter extracted from the deformed region that reflects the magnitude of the deformation.

[0101] In this embodiment, deformation data at each location is extracted from the deformation region and the motion direction information in the displacement vector field is fused to calculate the spatial correlation between deformation and motion direction; then the curvature of the propagation trajectory is adjusted and the direction is optimized to generate a three-dimensional propagation path that is more consistent with the actual rock mass deformation characteristics.

[0102] Here is a specific example:

[0103] In the specific implementation process, based on the vibration field data and vibration feature data obtained in the aforementioned embodiments, motion features are first extracted from the vibration field data. These motion features include the magnitude of the motion speed and the direction angle, with the maximum motion speed reaching 25 millimeters per second and the main motion direction concentrated in the range of 30 to 60 degrees. At the same time, wave features are extracted from the vibration feature data, including wave amplitude and morphological indicators at 8 scales. These two types of features are integrated into a 256-dimensional feature vector through fusion processing.

[0104] Next, based on the rock mass strength information from the geological parameter set, where the strength values ​​range from 35.6 to 58.1 MPa, a continuous impedance model is established through spatial interpolation. In this model, the impedance value Z is related to the rock density ρ and the wave velocity c as follows: The rock density is taken as 2650 kg / m³. 3 The wave velocity is calculated using an empirical formula for intensity.

[0105] The fused 256-dimensional feature vector is then input into a physical information neural network, which contains four hidden layers. The network simulates the attenuation process of vibration propagation using wave equation constraints and an impedance model. After 500 iterations of training, the propagation trajectory of vibration energy in the rock strata is calculated, with the wave equation in two-dimensional form. ,in The displacement field is expressed in meters. The time unit is seconds. The unit for wave speed is meters per second;

[0106] Finally, deformation degree information was extracted from the deformation region. The deformation feature values ​​of the three micro-deformation regions were 0.18, 0.22, and 0.15, respectively. Combined with the motion direction information of pixels in the displacement vector field, the three-dimensional propagation path generated by the correction formula clearly shows the diffraction characteristics of vibration energy in the dense joint region, thus providing an accurate basis for the identification of damage risk areas. The correction formula is as follows: ,in This indicates that the corrected direction angle is in degrees. The original direction angle is expressed in degrees. This is a correction factor, representing the adjustment amount of a unit degree of deformation to the directional angle, expressed in degrees. The value representing the degree of deformation is dimensionless.

[0107] In this embodiment, the accurate reconstruction of the vibration energy propagation path is achieved through the synergistic processing of multi-feature fusion and physical constraints, thereby improving the reliability of damage risk identification under complex rock strata conditions.

[0108] To further improve the accuracy of vibration energy propagation trajectory calculation, in some embodiments, step 203: simulating the fused features using the impedance model combined with a physical information neural network to calculate the vibration energy propagation trajectory in the rock strata includes:

[0109] Step 301: Input the fused features into the physical information neural network, and apply regular constraints to the fused features through the physical constraint layer of the physical information neural network to obtain the preliminary propagation field.

[0110] In step 301, the physical constraint layer is a network layer in the physical information neural network used to apply physical law constraints; the preset physical law refers to the physical law that elastic waves follow in the propagation of solid media, which specifically includes the wave propagation law described by the wave equation and the law of conservation of energy; the preliminary propagation field is the vibration field data that conforms to the wave propagation theory after being constrained by physical laws.

[0111] In this embodiment of the application, after the fused features are input into the physical information neural network, the feature data is processed by the wave equation terms in the physical constraint layer to ensure that the vibration propagation process satisfies the propagation law of elastic waves in the medium and outputs a preliminary propagation field that conforms to the physical law.

[0112] Step 302: Through the impedance coupling layer of the physical information neural network, the preliminary propagation field is coupled with the impedance model for calculation. Based on the coupling calculation results, the attenuation process of vibration energy in different lithological regions is simulated to obtain the energy attenuation field.

[0113] In step 302, the impedance coupling layer is a network layer in the physical information neural network used to process the influence of rock mass impedance; different lithological regions refer to the division of regions with different rock mechanical properties in the rock strata space. These regions are divided based on the rock mass strength information and joint and fracture information in the geological parameter set. Different lithological regions together constitute the complete rock strata space; the energy attenuation field is the energy distribution field where the vibration energy is attenuated after considering the rock mass impedance characteristics.

[0114] In this embodiment, the initial propagation field is coupled with the impedance model through an impedance coupling layer, thereby enabling the simulation of the attenuation degree of vibration energy in different lithological regions based on the impedance value. That is, the higher the impedance, the more obvious the attenuation, and finally obtains an energy attenuation field that reflects the actual rock mass conditions.

[0115] Step 303: Through the feature extraction layer of the physical information neural network, spatial features of the energy attenuation field are extracted to obtain the spatial distribution field of vibration energy.

[0116] In step 303, the feature extraction layer is a network layer in the physical information neural network used to extract spatial features, and the spatial distribution field is the distribution pattern data of vibration energy in three-dimensional space.

[0117] In this embodiment, the energy decay field is extracted by convolutional feature extraction layer, thereby capturing the distribution pattern and trend of energy in space, and generating a spatial distribution field that can clearly reflect the energy concentration area and propagation direction.

[0118] Step 304: Extract the main propagation path of vibration energy from the spatial distribution field through the path generation layer of the physical information neural network, and generate the propagation trajectory of vibration energy in the rock strata space based on the main propagation path.

[0119] In step 304, the path generation layer is a network layer in the physical information neural network used to generate propagation paths. The main criteria for determining the propagation path are the continuous distribution characteristics of vibration energy intensity in the spatial distribution field and the consistency of energy propagation direction. The path is characterized by high energy intensity value, continuous spatial extension characteristics and clear propagation directionality. It is defined as an energy conduction channel that connects energy concentration areas in the rock strata space and conforms to the law of vibration wave propagation.

[0120] In this embodiment, the energy intensity distribution in the spatial distribution field is analyzed by the path generation layer, and regions with continuous energy and high intensity are identified as the main propagation paths. Finally, these paths are connected and integrated to form a complete vibration energy propagation trajectory.

[0121] Here is a specific example:

[0122] In the specific implementation process, based on the 256-dimensional fused feature vector obtained in the aforementioned embodiment, it is first input into a physical information neural network containing 4 hidden layers, and the fused features are constrained by the wave propagation law through a physical constraint layer, wherein a two-dimensional wave equation is used as the constraint condition, and a preliminary propagation field conforming to the physical law is obtained after the constraint processing.

[0123] Next, the initial propagation field is coupled with the rock mass impedance distribution model through an impedance coupling layer for calculation. The impedance value Z in the rock mass impedance distribution model is obtained through the formula... Calculations show that The density of the rock is taken as 2650 kg / m³. 3 , The wave velocity, calculated using an empirical formula based on rock mass strength, ranges from 1870 to 3220 meters per second, with the impedance value in the hard rock region reaching approximately 8.5 × 10⁻⁶. 6 (N·s) / m³, impedance value of soft rock region is 4.96×10 6 (N·s) / m³, based on the coupled calculation results, simulates the attenuation process of vibration energy, which follows the formula ,in The energy intensity after decay is expressed in kJ / m³. 3 , This represents the initial energy intensity, expressed in kJ / m³. 3 , This indicates that the attenuation coefficient is expressed in units of per meter. The propagation distance is expressed in meters, thus obtaining the energy attenuation field;

[0124] Subsequently, a feature extraction layer is used to extract spatial features of the energy attenuation field. A 3×3 convolutional kernel is used for feature mapping to extract spatial gradient and regional clustering features of the energy distribution, thus obtaining the spatial distribution field of the vibration energy. Finally, a path generation layer extracts the main propagation paths of the vibration energy from the spatial distribution field, setting an energy intensity threshold of 25 kJ / m². 3 The system identifies blocks with energy values ​​exceeding the threshold and spatial continuity as the main propagation paths, and then connects and integrates these paths to generate a complete propagation trajectory of vibration energy in the rock strata. This trajectory accurately shows the linear propagation characteristics of vibration energy in hard rock areas and the diffraction phenomenon in soft rock areas.

[0125] In this embodiment, the precise calculation of the vibration energy propagation trajectory is achieved through multi-layer collaborative processing of physical information neural networks, providing reliable technical support for damage risk identification.

[0126] To further improve the accuracy and efficiency of image sequence analysis, in some embodiments, step 103: simultaneously performing optical flow analysis and region identification processing on the image sequence, wherein, during optical flow analysis, a displacement vector field is generated by calculating pixel displacement, and during region identification processing, deformed regions are identified by extracting spatial features, including:

[0127] Step 401: Analyze the consecutive image frames in the image sequence and establish a mapping relationship.

[0128] In step 401, the continuity of consecutive image frames is obtained by the image acquisition device acquiring them sequentially at a preset fixed time interval. There is a strict temporal order between adjacent image frames, and there is no time interval gap between any two adjacent frames, thus ensuring the continuity of the image sequence in the time dimension. The mapping relationship describes the correspondence of the positional changes of each pixel between consecutive frames.

[0129] In this embodiment of the application, the brightness values ​​of consecutive image frames in the image sequence are compared pixel by pixel. By calculating the brightness difference of pixels at the same spatial position between adjacent frames, a displacement mapping relationship reflecting the change of pixel position is established.

[0130] Step 402: Based on the mapping relationship, determine the motion speed and direction of each pixel to form an initial displacement set.

[0131] In step 402, the motion speed is the displacement of a pixel per unit time, the motion direction is the angular direction of the pixel's movement, and the initial displacement set is a preliminary motion data set composed of the motion speed and direction of all pixels.

[0132] In this embodiment of the application, based on the mapping relationship, the displacement of each pixel in a unit of time is calculated as the motion velocity, and the angle of the displacement vector is calculated as the motion direction. Finally, the motion data of all pixels are summarized to form an initial displacement set.

[0133] Step 403: Detect the initial displacement set, identify abnormal vectors and missing vectors, remove the abnormal vectors and fill in the missing vectors to generate a displacement vector field.

[0134] In step 403, anomaly vectors are anomalous data points that differ too much from surrounding vectors; missing vectors refer to missing data in the initial displacement set that fail to calculate effective motion information at certain spatial locations due to image noise, occlusion, or matching failure.

[0135] In this embodiment, spatial continuity analysis is performed on the initial displacement set to identify and remove abnormal vectors that differ too much from the direction or velocity of neighboring vectors. At the same time, the positions of missing vectors are interpolated to complete the displacement vector field, and finally a complete and consistent displacement vector field is generated.

[0136] Step 404: Extract features from the image sequence using multiple convolutional layers of a convolutional neural network to obtain image feature maps at different scales.

[0137] In step 404, the multi-layer convolutional layer is a combination of multiple convolutional layers in a convolutional neural network used to extract image features. The image feature map is a visual representation of the image features extracted through convolution operations.

[0138] In this embodiment, the image sequence is processed by multiple convolutional layers of a convolutional neural network. Each convolutional operation extracts image features at different levels of abstraction, thereby ultimately obtaining a set of image feature maps containing multi-scale information.

[0139] Step 405: Through the feature fusion layer of the convolutional neural network, feature maps of different scales are fused to extract texture change features and deformation features of the rock surface.

[0140] In step 405, the feature fusion layer is a network layer in a convolutional neural network used to integrate features at different scales, the texture change feature is a parameter reflecting the change in the texture pattern of the rock mass surface, and the deformation feature is a parameter characterizing the change in the geometric shape of the rock mass.

[0141] In this embodiment, image feature maps of different scales are fused through a feature fusion layer, and the detailed information of shallow features and the semantic information of deep features are combined to extract comprehensive features that can reflect the changes in texture and geometric deformation of the rock surface. The specific implementation process is as follows: First, the feature maps output by each layer of the convolutional neural network are upsampled or downsampled to make their size uniform. Then, multi-scale information is integrated through feature splicing or weighted fusion. Finally, the texture change pattern and deformation distribution pattern that can reflect both local details and global structure are extracted from the fused features through convolution operation.

[0142] Step 406: Based on the texture change features and the deformation features, identify and output the deformation region through the classification output layer of the convolutional neural network.

[0143] In step 406, the classification output layer is a network layer in the convolutional neural network used for the final classification decision.

[0144] In this embodiment of the application, the texture change features and deformation features are analyzed and processed by the classification output layer. Based on the feature combination pattern, the micro-deformation area on the rock surface caused by blasting vibration is identified, and its spatial location and range information are output.

[0145] In this embodiment, by simultaneously performing optical flow analysis and region identification processing, the dynamic response characteristics of the rock mass are fully captured, providing reliable image data support for vibration propagation analysis.

[0146] To further improve the accuracy of damage risk area identification, in some embodiments, step 105: identifying damage risk areas in the tunnel target area based on the energy distribution of the three-dimensional propagation path includes:

[0147] Step 501: Analyze the energy distribution in the three-dimensional propagation path to obtain the energy concentration degree and energy propagation direction.

[0148] In step 501, energy concentration is a quantitative data describing the degree of concentration of vibration energy in spatial distribution, and energy propagation direction is characteristic data reflecting the main propagation path direction of vibration energy.

[0149] In this embodiment of the application, the vibration energy intensity value at each spatial location in the three-dimensional propagation path is analyzed to calculate the statistical index of local energy concentration as the energy concentration degree, and the main directional trend of the energy propagation path is extracted as the energy propagation direction.

[0150] Step 502: Determine the energy concentration area based on the energy concentration degree and the energy propagation direction.

[0151] In step 502, the energy concentration area refers to the area in the three-dimensional propagation path where the vibration energy intensity is higher than that of the surrounding area and has a continuous distribution characteristic. The judgment conditions include the energy intensity value exceeding the set threshold, the area reaching the minimum limit range, and the energy distribution having spatial continuity.

[0152] In this embodiment, regions with energy intensity values ​​significantly higher than the surrounding background values ​​are identified based on energy concentration. The spatial continuity of these regions along the propagation path is determined by combining the energy propagation direction, and finally the boundary range of the energy concentration region is delineated.

[0153] Step 503: Identify the structurally weak points in the energy concentration region by combining the joint and fracture information in the set of geological parameters.

[0154] In step 503, the structurally weak part is the area in the rock mass where the mechanical properties are reduced due to joint development and other reasons.

[0155] In this embodiment of the application, by combining the joint and fracture information in the geological parameter set, the parts of the energy concentration area where geological defects are developed, such as dense joint zones and joint intersection areas, are identified, and these parts are marked as structurally weak parts. The specific implementation process is as follows: the spatial coordinates of the energy concentration area are superimposed and analyzed with the joint distribution map, so as to identify the areas in the overlapping area where the joint density is greater than the critical value or the angle between the joint direction and the energy propagation direction is less than a specific angle, and these areas that meet the mechanical weakening conditions are marked as structurally weak parts.

[0156] Step 504: Calculate the damage risk index based on the degree of spatial overlap between the weak structural part and the energy concentration area.

[0157] In step 504, the damage risk index is a comprehensive indicator that quantitatively assesses the likelihood of damage occurring.

[0158] In this embodiment of the application, based on the spatial relationship between the weak structural part and the energy concentration area, the degree of overlap between the two in space is calculated. Combining the energy intensity value and the degree of joint development, a damage risk index is obtained through weighted calculation.

[0159] Step 505: Based on the distribution of the damage risk index, determine the damage risk area in the target area of ​​the tunnel.

[0160] In step 505, the distribution of the damage risk index is a spatial distribution map formed in the entire target area of ​​the tunnel after calculating the risk index value of each spatial unit. It means that the relative probability of rock mass damage occurring at different locations is reflected. The higher the value, the greater the risk of damage occurring in that area.

[0161] In this embodiment of the application, based on the spatial distribution of the damage risk index, a continuous area where the risk index exceeds a set threshold is identified as a damage risk area, and its spatial location and range information are recorded.

[0162] In this embodiment of the application, the accurate location of the damage risk area is achieved through multi-source information fusion analysis, providing effective technical protection for tunnel construction safety.

[0163] To further improve the accuracy of vibration field data generation, in some embodiments, step 104: spatially associating the displacement vector field with the deformation region to form vibration field data, includes:

[0164] Step 601: Extract motion vector information from the displacement vector field and extract deformation position information from the deformation region.

[0165] In step 601, the motion vector information is data containing motion velocity and direction extracted from the displacement vector field. The deformation location information is data reflecting the spatial distribution of deformation extracted from the deformation region.

[0166] In this embodiment, the motion velocity value and motion direction angle of each measurement point are read from the displacement vector field as motion vector information, and the spatial coordinates and distribution range of the deformation region are obtained from the deformation region as deformation position information.

[0167] Step 602: Align the motion vector information with the deformation position information to establish a mapping relationship between the motion vector and the deformation position.

[0168] In step 602, the mapping relationship is a data structure that establishes the corresponding relationship between motion vectors and deformation positions.

[0169] In this embodiment, firstly, the two-dimensional coordinate system of motion vector information and the three-dimensional coordinate system of deformation position information are registered and transformed. On this basis, the spatial position is unified through the coordinate transformation matrix, thereby establishing a one-to-one mapping relationship between motion vector and deformation position.

[0170] Step 603: Based on the joint and fracture information in the set of geological parameters, the mapping relationship is subjected to structural constraint processing through a graph convolutional network to generate vibration field data.

[0171] In step 603, the joint and fracture information of the geological parameter set is obtained through geological survey and measurement, which represents the spatial distribution and development characteristics of joints and fractures in the rock mass of the tunnel target area.

[0172] In this embodiment, based on the joint and fracture information in the geological parameter set, each joint in the geological parameter set is first used as an independent graph node, and the spatial connectivity between joints is used as the graph edges connecting these nodes, thereby constructing a spatial graph structure that characterizes the rock mass structure. Then, a graph convolutional network is used to perform deep feature extraction and spatial smoothing on the mapping relationship contained in the spatial graph structure, so that the correspondence between the motion vector information transmitted in the network and the actual deformation position information of the surrounding rock can better conform to the real structural characteristics of the rock mass. Finally, after the above processing, vibration field data with a clear expression of rock mass structural characteristics is generated.

[0173] In this embodiment, the structural constraint processing of graph convolutional networks enables the organic integration of motion data and rock mass structural features, thereby improving the reliability and practicality of vibration field data.

[0174] To further improve the accuracy of vibration characteristic data in characterizing the non-uniformity of rock masses, in some embodiments, step 102: performing wave analysis on the set of geological parameters to obtain vibration characteristic data includes:

[0175] Step 701: Extract rock mass strength information and joint and fracture information from the set of geological parameters.

[0176] In step 701, rock mass strength information is data characterizing the rock's ability to resist external force damage. Joint and fracture information is data reflecting the degree of fracture development and distribution characteristics in the rock mass.

[0177] In this embodiment of the application, rock mass strength measurement values ​​and rock mass joint measurement values ​​are read from the geological parameter set to form rock mass strength information and joint fracture information.

[0178] Step 702: Under each analysis scale in the known multiple analysis scale tables, calculate the fluctuation trend characteristics of the rock mass strength information and the joint and fracture information respectively, and remove the fluctuation trend characteristics to obtain the scale fluctuation signals corresponding to each type of information under the analysis scale.

[0179] In step 702, the analysis scale table is a list of multiple pre-defined analysis scales; the fluctuation trend feature is the change trend of the data sequence within a local range; removing the fluctuation trend involves removing the calculated fluctuation trend components from the corresponding original rock mass strength information and joint and fracture information to obtain the scale fluctuation signal; the scale fluctuation signal is the remaining fluctuation component after removing the trend.

[0180] In this embodiment of the application, according to each scale value in the analysis scale table, local trend fitting is performed on the rock mass strength information sequence and the joint and fracture information sequence to obtain their respective fluctuation trend characteristics. Then, the corresponding trend characteristics are subtracted from the original data sequence to obtain the scale fluctuation signals of various types of information at each analysis scale.

[0181] The specific process is as follows: At each selected analysis scale, the rock mass strength information sequence and the joint and fracture information sequence are first divided into local intervals; then, trend fitting is performed on the data points within each local interval to obtain a local trend line describing the data change pattern of that interval; subsequently, the original rock mass strength information sequence and the joint and fracture information sequence are subtracted from their corresponding local trend lines to obtain the rock mass strength fluctuation signal and the rock mass joint fluctuation signal after removing the long-term trend, respectively; finally, these two fluctuation signals together constitute the scale fluctuation signal at this specific analysis scale.

[0182] Step 703: Extract the amplitude change information and morphological change information of the scale fluctuation signal corresponding to each type of information at all analysis scales, and perform scale correlation processing on the amplitude change information and morphological change information of all information at different analysis scales to form a correlation feature set.

[0183] In step 703, amplitude change information is a statistical measure of the amplitude of the scale fluctuation signal, morphological change information is a descriptive measure of the waveform characteristics of the scale fluctuation signal, and the associated feature set is a data set that integrates multi-scale features.

[0184] In this embodiment, at each set analysis scale, the rock mass strength information sequence and the joint and fracture information sequence are divided into local intervals. Then, trend fitting is performed on the data points within each local interval to obtain the corresponding local trend line. Next, the original rock mass strength information sequence and the original joint and fracture information sequence are subtracted from their corresponding local trend lines to obtain the rock mass strength fluctuation signal and the rock mass joint fluctuation signal after removing the influence of the trend. Finally, these two fluctuation signals together constitute the scale fluctuation signal at this analysis scale.

[0185] Step 704: Based on the associated feature set, vibration feature data is obtained through feature fusion processing.

[0186] In this embodiment of the application, principal component analysis is performed on the associated feature set, the main components with higher contribution rates are selected, and the original multi-dimensional features are fused into low-dimensional vibration feature data through linear transformation.

[0187] In this embodiment of the application, multi-scale wave analysis and feature fusion processing are used to accurately characterize the non-uniform properties of the rock mass, providing reliable feature data support for vibration propagation analysis.

[0188] Figure 3 A schematic diagram of a tunnel blasting vibration monitoring system based on image recognition is provided in an embodiment of this application, as shown below. Figure 3 As shown, the detailed implementation section describes:

[0189] The acquisition module 31 is used to acquire the set of geological parameters of the tunnel target area and the image sequence of the tunnel target area.

[0190] Analysis module 32 is used to perform wave analysis on the set of geological parameters to obtain vibration characteristic data.

[0191] The generation module 33 is used to perform optical flow analysis and region recognition processing on the image sequence simultaneously. In the optical flow analysis process, a displacement vector field is generated by calculating pixel displacement. In the region recognition process, deformed regions are identified by extracting spatial features.

[0192] The association module 34 is used to spatially associate the displacement vector field with the deformation region to form vibration field data, and to couple the vibration field data with the vibration feature data to reconstruct the three-dimensional propagation path.

[0193] The identification module 35 is used to identify the damage risk area in the tunnel target area based on the energy distribution of the three-dimensional propagation path.

[0194] Without altering the overall concept of this scheme, its core principles can also be applied to the monitoring and assessment of other nearby structures. Specifically, the core of this scheme lies in capturing the dynamic displacement of the structural surface caused by blasting vibrations through non-contact image sensing technology, and coupling geological conditions with the structural characteristics themselves to reconstruct the actual propagation path of the vibration energy. This principle is applicable to any nearby structure that may be affected by blasting vibrations because the propagation and attenuation laws of vibration waves in soil and rock media, as well as the physical nature of the interaction between vibration waves and structures, are consistent.

[0195] Therefore, this method provides a unified technical framework for assessing the overall impact of blasting operations on the surrounding environment. Specifically, this scheme can monitor the following typical adjacent structures: 1. Surface buildings, such as residential buildings, office buildings, factories, and historical buildings, mainly monitoring the vibration response of their foundations, load-bearing walls, or columns; 2. Underground structures, such as operating subway tunnels, underground utility tunnels, various pipelines, and civil defense projects, mainly monitoring the vibration of their lining structures or pipe surfaces; 3. Other structures, such as bridge piers and foundations, roadbed slopes, and retaining walls.

[0196] The image recognition-based tunnel blasting vibration monitoring system of this application embodiment is used to implement the aforementioned image recognition-based tunnel blasting vibration monitoring method. Therefore, the specific implementation of the image recognition-based tunnel blasting vibration monitoring system can be found in the embodiment section of the image recognition-based tunnel blasting vibration monitoring method above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0197] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described image recognition-based tunnel blasting vibration monitoring methods.

[0198] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described image recognition-based tunnel blasting vibration monitoring methods.

[0199] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0200] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the image recognition-based tunnel blasting vibration monitoring method.

[0201] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0202] The foregoing has provided a detailed description of the tunnel blasting vibration monitoring method and system based on image recognition provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for monitoring tunnel blasting vibration based on image recognition, characterized in that, include: Acquire a set of geological parameters and an image sequence of the tunnel target area; Wave analysis was performed on the set of geological parameters to obtain vibration characteristic data; Optical flow analysis and region recognition processing are performed synchronously on the image sequence. In the optical flow analysis process, a displacement vector field is generated by calculating pixel displacement. In the region recognition processing process, deformed regions are identified by extracting spatial features. The displacement vector field is spatially correlated with the deformation region to form vibration field data, and the vibration field data is coupled with the vibration characteristic data to reconstruct the three-dimensional propagation path; Based on the energy distribution of the three-dimensional propagation path, damage risk areas in the tunnel target area are identified. The coupling of the vibration field data with the vibration feature data to reconstruct the three-dimensional propagation path includes: The motion features extracted from the vibration field data are fused with the wave features extracted from the vibration feature data; An impedance model is established based on the rock mass strength information of the aforementioned set of geological parameters; The impedance model is used to simulate the fused features in combination with a physical information neural network to calculate the propagation trajectory of vibration energy in the rock strata. Deformation degree information is extracted from the deformed region, and the direction of movement of the pixels and the deformation degree information are combined to correct the direction of the propagation trajectory and generate a three-dimensional propagation path. The process of simulating the fused features using the impedance model and a physical information neural network to calculate the propagation trajectory of vibration energy in the rock strata includes: The fused features are input into a physical information neural network. The physical constraint layer of the physical information neural network applies regular constraints to the fused features to obtain a preliminary propagation field. The initial propagation field is coupled with the impedance model through the impedance coupling layer of the physical information neural network. Based on the coupling calculation results, the attenuation process of vibration energy in different lithological regions is simulated to obtain the energy attenuation field. The spatial feature extraction layer of the physical information neural network is used to extract the spatial features of the energy attenuation field to obtain the spatial distribution field of the vibration energy. The path generation layer of the physical information neural network extracts the main propagation path of vibration energy from the spatial distribution field, and generates the propagation trajectory of vibration energy in the rock strata space based on the main propagation path.

2. The method according to claim 1, characterized in that, The simultaneous optical flow analysis and region recognition processing of the image sequence includes, during optical flow analysis, generating a displacement vector field through pixel displacement calculation; and during region recognition processing, identifying deformed regions through spatial feature extraction. Analyze consecutive image frames in the image sequence to establish a mapping relationship; Based on the mapping relationship, the motion speed and direction of each pixel are determined to form an initial displacement set; The initial displacement set is detected to identify abnormal and missing vectors. The abnormal vectors are removed and the missing vectors are filled in to generate a displacement vector field. By using multiple convolutional layers of a convolutional neural network, feature extraction is performed on the image sequence to obtain image feature maps at different scales; By using the feature fusion layer of a convolutional neural network, feature maps of different scales are fused to extract texture variation and deformation features of the rock surface; The classification output layer of the convolutional neural network identifies and outputs deformed regions based on the texture change features and the deformation features.

3. The method according to claim 1, characterized in that, The identification of damage risk areas in the tunnel target region based on the energy distribution of the three-dimensional propagation path includes: Analyze the energy distribution in the three-dimensional propagation path to obtain the energy concentration degree and energy propagation direction; The energy concentration region is determined based on the energy concentration degree and the energy propagation direction; By combining the joint and fracture information in the set of geological parameters, the structurally weak points in the energy concentration region are identified; Based on the degree of spatial overlap between the weak structural parts and the energy concentration area, a damage risk index is calculated. Based on the distribution of the damage risk index, the damage risk areas in the tunnel target area are determined.

4. The method according to claim 1, characterized in that, The step of spatially associating the displacement vector field with the deformation region to form vibration field data includes: Motion vector information is extracted from the displacement vector field, and deformation position information is extracted from the deformation region; Align the motion vector information with the deformation position information to establish a mapping relationship between the motion vector and the deformation position; Based on the joint and fracture information in the set of geological parameters, the mapping relationship is structurally constrained by a graph convolutional network to generate vibration field data.

5. The method according to claim 1, characterized in that, The wave analysis of the geological parameter set to obtain vibration characteristic data includes: Extract rock mass strength information and joint and fracture information from the set of geological parameters; Under each known analysis scale in multiple analysis scale tables, the fluctuation trend characteristics of the rock mass strength information and the joint and fracture information are calculated respectively, and the fluctuation trend characteristics are removed to obtain the scale fluctuation signals corresponding to each type of information under the analysis scale. Extract the amplitude and morphological change information of the scale fluctuation signal corresponding to each type of information at all analysis scales, and perform scale correlation processing on the amplitude and morphological change information of all information at different analysis scales to form a correlation feature set; Vibration feature data are obtained through feature fusion processing based on the aforementioned associated feature set.

6. A tunnel blasting vibration monitoring system based on image recognition, characterized in that, include: The acquisition module is used to acquire a set of geological parameters and an image sequence of the tunnel target area. The analysis module is used to perform wave analysis on the set of geological parameters to obtain vibration characteristic data; The generation module is used to simultaneously perform optical flow analysis and region recognition processing on the image sequence. In the optical flow analysis process, a displacement vector field is generated by calculating pixel displacement, and in the region recognition processing process, deformed regions are identified by extracting spatial features. The association module is used to spatially associate the displacement vector field with the deformation region to form vibration field data, and to couple the vibration field data with the vibration feature data to reconstruct the three-dimensional propagation path; The identification module is used to identify the damage risk area in the tunnel target area based on the energy distribution of the three-dimensional propagation path; The coupling of the vibration field data with the vibration feature data to reconstruct the three-dimensional propagation path includes: The motion features extracted from the vibration field data are fused with the wave features extracted from the vibration feature data; An impedance model is established based on the rock mass strength information of the aforementioned set of geological parameters; The impedance model is used to simulate the fused features in combination with a physical information neural network to calculate the propagation trajectory of vibration energy in the rock strata. Deformation degree information is extracted from the deformed region, and the direction of movement of the pixels and the deformation degree information are combined to correct the direction of the propagation trajectory and generate a three-dimensional propagation path. The process of simulating the fused features using the impedance model and a physical information neural network to calculate the propagation trajectory of vibration energy in the rock strata includes: The fused features are input into a physical information neural network. The physical constraint layer of the physical information neural network applies regular constraints to the fused features to obtain a preliminary propagation field. The initial propagation field is coupled with the impedance model through the impedance coupling layer of the physical information neural network. Based on the coupling calculation results, the attenuation process of vibration energy in different lithological regions is simulated to obtain the energy attenuation field. The spatial feature extraction layer of the physical information neural network is used to extract the spatial features of the energy attenuation field to obtain the spatial distribution field of the vibration energy. The path generation layer of the physical information neural network extracts the main propagation path of vibration energy from the spatial distribution field, and generates the propagation trajectory of vibration energy in the rock strata space based on the main propagation path.

7. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the image recognition-based tunnel blasting vibration monitoring method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the image recognition-based tunnel blasting vibration monitoring method as described in any one of claims 1 to 5.

Images