A method for intelligently identifying water leakage risk of a tunnel face in a water-rich clay layer

Abstract

This application relates to the field of tunnel construction safety monitoring technology, and discloses an intelligent identification method for water seepage risk at the tunnel face in water-rich clay layers. This method controls an active infrared excitation module to project radiation in an oblique incidence manner, uses polarization imaging to acquire sequences and calculates Stokes vectors; calculates the degree of linear polarization based on the Stokes vectors and performs threshold segmentation to generate a water film distribution mask; uses the Stokes vectors to construct a polarization angle field reflecting local normal changes as texture features, solves the optical flow constraint equation to obtain the instantaneous velocity field; performs frequency domain transformation on the velocity field, extracts the proportion of low-frequency creep energy through power spectral density analysis, and determines the soil rheological softening factor; finally, combines the water film distribution, rheological softening factor, and instantaneous velocity field to construct a mudslide risk index. This invention utilizes polarization texture reconstruction to solve the problem of weak texture tracking and eliminates elastic vibration interference through frequency domain energy decoupling, effectively improving the accuracy of mudslide risk identification.

Description

Technical Field

[0001] This invention relates to the field of tunnel construction safety monitoring technology, specifically to an intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers. Background Technology

[0002] In tunnel construction with complex geological conditions such as water-rich clay, sudden water and mud inrushes at the tunnel face are typical geological hazards caused by the interaction between groundwater and weak surrounding rock. Real-time monitoring of water seepage at the tunnel face and the rheological characteristics of the soil is a crucial step in preventing these disasters and ensuring construction safety.

[0003] Infrared thermal imaging technology, due to its non-contact nature, wide field of view, and sensitivity to water bodies, is often used to monitor the water content at tunnel faces. Traditional infrared monitoring methods mainly rely on the radiation contrast formed by the temperature difference or thermal emissivity difference between the water body and the surrounding rock to identify seepage areas. However, in the deep tunnel construction environment, the seeping water is often in thermal equilibrium with the surrounding rock, with a very small temperature difference between the two. This causes the radiation intensity of the target and the background in the infrared thermal image to tend to be the same, making it difficult to effectively distinguish the seepage area from the normal rock and soil area.

[0004] In addition to identifying static water distribution, monitoring minute displacements and deformation trends of the soil at the tunnel face is equally important for early warning of mudslide risks. Currently used machine vision-based optical flow methods or digital image correlation methods require rich texture features or grayscale gradients on the observed object surface for feature point tracking. Water-rich clay tunnel faces typically exhibit weak texture features with uniform thermal radiation and smooth surfaces, making it difficult for conventional image algorithms to accurately calculate the displacement field due to the lack of feature gradients. Simultaneously, tunnel construction sites experience strong mechanical vibrations from drilling and blasting, excavator operations, etc. Existing displacement monitoring methods usually rely on threshold judgments based on the total displacement over time, making it difficult to separate high-frequency elastic vibrations caused by construction from low-frequency plastic creep, a precursor to soil disasters, easily leading to false alarms or missed alarms. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers. This method solves the problems of existing tunnel face monitoring technologies, such as the difficulty in distinguishing isothermal water leakage by infrared thermal imaging, the failure of displacement tracking due to weak textured surfaces, and the difficulty in separating construction elastic vibration from soil plastic creep.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: an intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers. This method solves the technical problems of conventional infrared thermal imaging technology in isothermal cement differentiation and weak texture optical flow tracking by using active polarized infrared imaging and time-frequency micro-motion analysis technology.

[0007] In this method, the active infrared excitation module is activated to project infrared radiation, and the module is controlled to continuously radiate the tunnel face in an oblique incidence manner, so that the optical axis of the radiation center forms a preset angle with the normal direction of the observation area. This step utilizes the Fresnel reflection principle to increase the specular reflectivity of the water-bearing area with the increase of the incident angle, thereby enhancing the difference in physical characteristics between water and soil in radiation image sequences with different polarization directions.

[0008] After acquiring the image sequence, the Stokes vector, including the total infrared radiation intensity, is calculated using a Stokes solver unit to fully characterize the energy and polarization state of the target region. The horizontal and vertical polarization difference components, and the 45-degree and 135-degree polarization difference components are obtained using a feature segmentation unit. The total intensity modulus of linearly polarized light is synthesized, and the ratio of this modulus to the total infrared radiation intensity is calculated to obtain the degree of linear polarization. Based on the numerical distribution characteristics of the degree of linear polarization and the maximum inter-class variance method, a segmentation threshold is calculated to identify and mark high-polarization free water film coverage areas, generating a water film distribution mask.

[0009] To address the problem of optical flow tracing failure caused by missing texture in thermal infrared images of the tunnel face, this invention utilizes a micro-motion analysis unit to construct a polarization angle field based on Stokes vectors. Since the polarization angle characterizes the principal direction of the incident photoelectric vector vibration, and its distribution is closely related to the local normal direction of the object surface, the constructed polarization angle field can reconstruct texture features reflecting the microscopic deformation of the tunnel face. Based on these texture features, the instantaneous velocity field of the tunnel face changing with time is obtained by solving the optical flow constraint equation that conserves the polarization angle.

[0010] To further differentiate the rheological risks of soil, a short-time time-frequency analysis was performed on the velocity modulus sequence of the instantaneous velocity field using a risk assessment unit to calculate the power spectral density. By setting a cutoff frequency to separate elastic and plastic modes, the proportion of low-frequency creep energy density in the total energy density was calculated to obtain the soil rheological softening factor. This step can effectively eliminate high-frequency elastic vibration interference and quantify the plastic flow trend of the soil.

[0011] Finally, a mud inrush risk index is constructed by combining a risk assessment unit with a water film distribution mask, a rheological softening factor, and an instantaneous velocity field, and a risk warning is output. The mud inrush risk index is obtained by calculating the product of the water film distribution mask, the rheological softening factor, and a velocity-sensitive term; whereby the velocity-sensitive term is obtained by transforming the natural constant by a negative exponent, the exponent of which depends on the product of the velocity sensitivity coefficient and the instantaneous velocity modulus. This calculation model only determines high risk when the region simultaneously meets the conditions of water abundance, rheological softening, and significant displacement velocity characteristics, achieving accurate early warning through multi-physics coupling.

[0012] This invention provides an intelligent method for identifying the risk of water seepage at the tunnel face in water-rich clay layers. It has the following beneficial effects: 1. This invention utilizes a combination of active oblique incidence infrared excitation and polarization imaging to solve the problem that conventional infrared thermography is difficult to distinguish between water bodies and background rock and soil under isothermal conditions. By controlling the incident angle to induce the Fresnel reflection effect, the linear polarization degree of the water surface is significantly higher than that of the rough rock and soil surface. This allows for the generation of a high-contrast water film distribution mask using the difference in polarization degree, enabling accurate segmentation of the free water region even under conditions where the temperature difference between the working face and the seepage water is small.

[0013] 2. This invention proposes to construct a polarization angle field as a texture feature for optical flow tracing, overcoming the problem of weak texture optical flow calculation failure caused by the uniform distribution of radiation intensity in thermal infrared images of the tunnel face; the polarization angle field directly maps the microscopic normal direction change of the geological body surface, and can reconstruct rich structured textures in areas lacking gray-level gradients. Combined with the optical flow constraint equation of polarization angle conservation, it significantly improves the perception accuracy of small deformations and displacement fields of the tunnel face.

[0014] 3. This invention effectively reduces the interference of construction machinery vibration on mudslide risk identification by performing frequency domain energy decoupling analysis on the instantaneous velocity field; by separating high-frequency stiffness energy and low-frequency creep energy, the rheological softening state of the soil is quantified, and a multi-physics field coupled risk index is constructed by combining water film distribution and displacement velocity, realizing the transformation from simple displacement monitoring to geological disaster rheological mechanism identification, and improving the accuracy and reliability of early warning. Attached Figure Description

[0015] Figure 1 This is an overall flowchart of the method of the present invention; Figure 2 This is a schematic diagram illustrating an application scenario of the active polarization monitoring system of the present invention; Figure 3 This is a schematic diagram illustrating the principle of polarization reflection enhancement of the present invention. Figure 4 This is a schematic diagram illustrating the rheological feature decoupling principle of the present invention; Figure 5 This is a risk assessment diagram for multi-source fusion according to the present invention.

[0016] Among them, 110 is the polarized infrared imaging module; 120 is the active infrared excitation module; 130 is the data processing module; 131 is the Stokes calculation unit; 132 is the feature segmentation unit; 133 is the micro-motion analysis unit; and 134 is the risk assessment unit. Detailed Implementation

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see the appendix Figure 1 Appendix Figure 2 This invention provides an intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers, comprising the following steps: S10, the active infrared excitation module 120 is activated to project broadband infrared radiation onto the tunnel face, and the polarized infrared imaging module 110 is used to simultaneously acquire the radiation intensity image sequence of the tunnel face in four polarization directions: 0 degrees, 45 degrees, 90 degrees and 135 degrees. S20, Stokes vectors of each pixel are calculated using Stokes solution unit 131 based on radiation intensity images of four polarization directions, including total radiation intensity, horizontal and vertical polarization difference and diagonal polarization difference, and an effective signal mask for filtering background noise is generated based on the signal-to-noise ratio characteristics of total radiation intensity. S30, the feature segmentation unit 132 calculates the linear polarization degree of each pixel based on the Stokes vector, compares the linear polarization degree with the preset water body segmentation threshold, and generates a water film distribution mask indicating the free water distribution area of ​​the working face. S40, the polarization angle field distribution of the tunnel face is calculated by the micro-motion analysis unit 133 based on the Stokes vector. The polarization angle field is used as the texture feature for optical flow tracing. By solving the optical flow constraint equation based on the conservation of polarization angle, the instantaneous velocity field of the tunnel face under environmental excitation is obtained. S50, using risk assessment unit 134 to perform short-time Fourier transform on the instantaneous velocity field, extract low-frequency creep energy density and high-frequency stiffness energy density, and determine the soil rheological softening factor based on the proportion of low-frequency creep energy density in the total energy. S60 utilizes the risk assessment unit 134 to construct a mud inrush risk index by combining the water film distribution mask, rheological softening factor, and total radiation intensity. The mud inrush risk index is compared with the preset graded alarm threshold, and the corresponding risk level prompt signal is output.

[0019] Please see the appendix Figure 2 As shown, the present invention provides an intelligent identification system for water leakage risk at the tunnel face in water-rich clay layers. This system is applied to the monitoring scenario at the tunnel face during tunnel construction.

[0020] The intelligent identification system for water seepage risk at the tunnel face in water-rich clay layers includes: a polarized infrared imaging module 110, an active infrared excitation module 120, and a data processing module 130. The modules communicate and are connected for data exchange via physical cables or an industrial bus.

[0021] The polarization infrared imaging module 110 is used for non-contact acquisition of thermal infrared radiation data and polarization state data of the tunnel face. This polarization infrared imaging module 110 includes a focal plane long-wave infrared detector, and a micro-polarizer array is integrated on the detector's pixel array. The micro-polarizer array contains four different polarization transmission directions: 0 degrees, 45 degrees, 90 degrees, and 135 degrees. The polarization infrared imaging module 110 is used to simultaneously output radiation intensity image data for the above four polarization directions in a single exposure.

[0022] An active infrared excitation module 120 is disposed on one or both sides of the polarization infrared imaging module 110, and is used to project broadband infrared radiation onto the tunnel face. The emission optical axis of the active infrared excitation module 120 forms a preset incident angle with the normal direction of the tunnel face. This incident angle is used to generate a Fresnel reflection effect in the water-bearing area of ​​the tunnel face to enhance the polarization characteristics of the reflected infrared light. The active infrared excitation module 120 uses an infrared radiating plate or infrared illumination lamp group with constant radiant power, and its radiation band covers the long-wave infrared region.

[0023] The data processing module 130 is electrically connected to the polarized infrared imaging module 110 and is used to receive image data and execute risk identification algorithms. The data processing module 130 includes a Stokes algorithm unit 131, a feature segmentation unit 132, a micro-motion analysis unit 133, and a risk assessment unit 134.

[0024] The Stokes calculation unit 131 is used to calculate the Stokes vector of each pixel on the face based on the radiation intensity images of the four polarization directions output by the polarized infrared imaging module 110, and output intermediate data including the total radiation intensity map, the horizontal and vertical polarization difference map and the diagonal polarization difference map.

[0025] The feature segmentation unit 132 is used to calculate the linear polarization degree of each pixel and generate a water film distribution mask on the face of the tunnel based on the linear polarization degree threshold. The feature segmentation unit 132 is also used to generate an effective signal mask based on the signal-to-noise ratio characteristics of the total radiation intensity map to filter out background noise regions.

[0026] The micro-motion analysis unit 133 is used to calculate the polarization angle field distribution of the tunnel face and to use the polarization angle field as a texture feature for optical flow field calculation. The micro-motion analysis unit 133 obtains the instantaneous velocity field of the tunnel face in the time series by solving the constraint equation based on the conservation of polarization angle. This velocity field reflects the small displacement and deformation of the tunnel face under environmental excitation.

[0027] The risk assessment unit 134 is used to perform frequency domain transformation on the velocity field output by the micro-motion analysis unit 133 to extract the low-frequency creep energy density and high-frequency stiffness energy density. Based on the water film distribution mask and frequency domain energy characteristics output by the feature segmentation unit 132, the risk assessment unit 134 calculates the mud inrush risk index and outputs an alarm signal according to the preset risk classification standard.

[0028] In actual deployment, the polarized infrared imaging module 110 and the active infrared excitation module 120 are jointly mounted on a gimbal bracket with a protective shell. The gimbal bracket is fixed to the tunnel's initial support structure or a mobile trolley to maintain continuous observation of the tunnel face. The data processing module 130 is implemented using an industrial-grade edge computing terminal and is deployed at the observation site or transmitted to the monitoring room server via fiber optic cable.

[0029] Step S10: Active polarization image sequence acquisition and environmental stimulus construction This step aims to construct an active radiation field with specific geometric constraints, artificially enhance the signal differences between water and soil in the polarization dimension, and acquire high-dimensional data containing spatiotemporal information, specifically including: Constructing an active polarization reflection enhancement environment: An active infrared excitation module 120 is configured to continuously radiate radiation onto the tunnel face in an oblique incidence manner. Utilizing the Fresnel reflection principle, a preset incident angle is formed between the optical axis of the radiation center and the normal direction of the observation area on the tunnel face. Under this geometric constraint, the reflectivity of the water surface increases with the increase of the incident angle and exhibits obvious linear polarization characteristics, while the rough clay surface mainly undergoes diffuse reflection and shows a significant depolarization effect, thereby improving the signal-to-noise ratio in the water area. Simultaneously, the radiation power density of the active infrared excitation module 120 is ensured to be stable, and the radiation band covers the long-wave infrared band; furthermore, a closed-loop feedback adjustment mechanism for the radiation power is established; the system monitors in real time. If the average gray value of the region of interest in the image deviates from the preset linear response range of the detector (e.g., saturation or excessive darkness), the output power of the active infrared excitation module 120 is automatically adjusted by modulating the pulse width or current to adapt to the signal attenuation caused by changes in observation distance due to the movement of the trolley or changes in the concentration of smoke and dust on site, ensuring that the imaging quality is always in the optimal dynamic range.

[0030] Simultaneous acquisition of multi-angle polarized radiation intensity images: The polarized infrared imaging module 110 is used to continuously acquire time-series data of the tunnel face. This module adopts a focal plane detector architecture and uses a micro polarizer array to achieve synchronous sensing of different linear polarization transmission directions (such as 0 degrees, 45 degrees, 90 degrees, and 135 degrees).

[0031] Generate a time-series radiance image sequence: The system converts the original electrical signal into a digital image matrix, analyzes and outputs the radiation intensity component sequences corresponding to different polarization directions, and performs non-uniformity correction processing to form the basic dataset.

[0032] Step S20: Calculation of polarization state information and construction of the Stokes vector field To obtain a physically meaningful polarization description, the system performs a linear transformation on the image sequence based on the definition of the Stokes vector. By combining the radiation intensity components of different polarization directions, it calculates that: before performing the Stokes vector calculation, to address the spatial phase difference between the polarization channels of the focal plane detector, gradient-based demosaic interpolation is first performed on the original image array. An edge-aware algorithm is then used to map pixel data at 0 degrees, 45 degrees, 90 degrees, and 135 degrees to the same spatial center coordinates, eliminating false polarization artifacts at high-frequency edges caused by non-coordinated pixel physical positions, ensuring pixel-level alignment accuracy for subsequent polarization degree and polarization angle calculations. Total infrared radiation intensity ( ): Characterizes the total radiant energy of the target, corresponding to a traditional thermal infrared image; Horizontal and vertical polarization difference components ( ): Characterizes the difference between the horizontal linear polarization component and the vertical linear polarization component; The polarization difference components between 45 degrees and 135 degrees ( : Characterizes the difference between the linear polarization components in the 45-degree direction and the 135-degree direction.

[0033] The system counts the current total intensity of far-infrared radiation ( The system uses a histogram distribution of infrared radiation intensity to mark dark pixels with intensity values ​​below the dynamic noise floor (e.g., 0.5 times the lower quartile of the histogram) as invalid regions. In subsequent polarization degree and optical flow calculations, the results for invalid regions are forcibly set to zero to prevent background thermal noise from being misidentified as highly polarized water bodies or highly rheological regions. Simultaneously, an effective signal mask is generated based on the total infrared radiation intensity distribution, marking effective signal regions to eliminate noise interference. Furthermore, to address the infrared depolarization phenomenon caused by smoke and dust scattering commonly found in tunnel construction environments, a polarization defogging algorithm based on dark channel priors is introduced. The system uses the average polarization degree of the background region to estimate atmospheric optical polarization parameters, inverts the atmospheric scattering physical model, subtracts the scattering component caused by suspended particles (i.e., air curtain interference) from the original radiation intensity image, restores the intrinsic polarization contrast of the target at the tunnel face, and enhances the system's penetration imaging capability under dense smoke and dust conditions.

[0034] Step S30: Material segmentation based on polarization physical differences By utilizing the physical differences in the reflection process at different material interfaces based on the polarization characteristics of light waves, high-precision segmentation of water seepage areas can be achieved.

[0035] Solution of linear polarization degree field: The system calculates the degree of linear polarization (DoLP) based on the Stokes vector field. This physical quantity characterizes the proportion of linearly polarized energy in the total radiant energy. Its calculation logic is as follows: take the polarization difference component ( and The total intensity modulus of linearly polarized light synthesized from the total intensity of infrared radiation ( ) Perform ratio calculations.

[0036] Physical difference analysis based on Fresnel reflection mechanism: Under active infrared oblique incidence, the surface of a free water film undergoes specular reflection, exhibiting a high degree of linear polarization; while the surface of rock and soil undergoes diffuse reflection, exhibiting a low degree of linear polarization.

[0037] Generation of the water film distribution mask: Based on the bimodal characteristic of the linear polarization degree numerical distribution, an adaptive threshold strategy (such as the maximum inter-class variance method) is used to calculate the optimal segmentation threshold. Regions with linear polarization degrees higher than this threshold are marked as free water film coverage areas, generating a water film distribution mask. The system can perform spatial optimization through morphological operations; further, it introduces spatial consistency constraints based on the angle of polarization (AoP) to distinguish between free-flowing water bodies and static wet rock surfaces. Due to the surface tension of free water films, they possess smooth mirror properties, and the AoP of their reflected light exhibits a smooth transition characteristic in spatial distribution (low local standard deviation). In contrast, wet rock and soil surfaces, although having high DoLP, are microscopically rough, and their AoP exhibits random speckle characteristics (high local standard deviation). The system calculates the local gradient or standard deviation of the AoP field, eliminating regions with high DoLP but chaotic AoP spatial distribution, thereby accurately locating flowing water bodies.

[0038] In addition, to eliminate transient reflective interference caused by mobile construction equipment or personnel, the system performs temporal consistency verification on the initially generated water film distribution mask, establishes a sliding window of length N (e.g., 5 frames), and retains only the areas whose spatial position remains stable in multiple consecutive frames (with an overlap higher than a preset ratio) as the final effective water film distribution areas; at the same time, it uses thermal radiation differences to eliminate false polarization interference from metal components; given that although metal construction equipment has a high degree of polarization, the radiation temperature caused by its surface smoothness and material emissivity is usually significantly different from that of the surrounding rock and seepage water in thermal equilibrium (around 18°C), the system calculates the contrast between the average radiation intensity of the candidate water film area and the surrounding background. If obvious hot spots (mechanical heating) or cold spots (low emissivity of metal reflecting low ambient temperature) are detected, they are identified as artificial metal interference objects and removed from the water film distribution mask.

[0039] Step S40: Micro-motion field tracking based on polarization angle texture By introducing polarization angle features, surface micro-deformation is transformed into traceable high-frequency texture information, solving the problem of optical flow failure caused by weak texture in thermal infrared images.

[0040] Construction of high-frequency polarization angle texture field: The system uses the Stokes component to calculate the polarization angle (AoP), which represents the angle between the principal direction of the incident photoelectric vector vibration and the reference axis. To eliminate the influence of the periodic abrupt change in the polarization angle value at the boundary between 0 and 180 degrees on the gradient calculation, the system performs vectorization reconstruction of the original polarization angle field, decomposing it into a sine component field (sin(2·AoP)) and a cosine component field (cos(2·AoP)). Based on these two continuously changing scalar fields, the optical flow vector is calculated separately, and then the final velocity field is synthesized. Due to the strong coupling between the change in the local normal direction of the face and the polarization angle, the system constructs a polarization angle texture field with high-frequency characteristics. To address the numerical jump problem (i.e., phase entanglement) of the polarization angle at the boundary between 0 and 180 degrees, the system does not directly perform gradient calculation on the original polarization angle field, but first performs vectorization reconstruction: the polarization angle of each pixel is... It is decomposed into two continuous orthogonal component fields, respectively and Subsequently, respectively and The velocity vector is solved by applying the optical flow constraint equation and then synthesized. This process eliminates false edges caused by angular periodicity and ensures the continuity of the micro-motion field calculation.

[0041] Establish a polarization angle-conserving optical flow constraint model: Unlike the traditional brightness conservation assumption, this embodiment is based on the polarization angle conservation assumption, that is, within an extremely short sampling interval, the polarization angle texture attached to the fluid surface is translated with the fluid movement.

[0042] Numerical solution of the small velocity field: Based on the spatiotemporal gradient information of the polarization angle field, a system of partial differential equations is solved using a local window optimization algorithm (such as the Lucas-Kanade method). Combined with a multi-scale Gaussian pyramid strategy, a downsampling sequence of images is constructed, and optical flow is solved layer by layer in a coarse-to-fine order. The coarse displacement calculated at the upper low-resolution level is used as the initial estimate for the lower high-resolution level calculation, thus addressing the problem of optical flow constraint failure caused by excessive instantaneous soil displacement at a single scale. A dense velocity field sequence reflecting the dynamic deformation of the tunnel face is obtained, and the instantaneous velocity modulus is calculated. Meanwhile, in order to eliminate global motion errors caused by the vibration of the monitoring platform or the small displacement of the trolley, the system selects a low linear polarization region (i.e., a relatively static soil and rock background region) as a static reference benchmark, uses a robust estimation method to calculate the global affine transformation matrix of the background, and subtracts the global motion component from the original velocity field to obtain the absolute displacement velocity of the soil at the tunnel face relative to the stable background.

[0043] Furthermore, to address the perspective distortion problem caused by 45-degree oblique incidence (i.e., inconsistent spatial resolution between the upper and lower parts of the image), the system introduces an inverse perspective transformation (IPM) module. Based on the preset incident angle and camera intrinsic parameters, a homography matrix is ​​constructed to map and correct the pixel / second velocity field of the image plane to the true millimeter / second velocity field of the physical plane of the face, ensuring the consistency of physical dimensions in the spatial distribution of subsequent kinetic energy density calculations. In this process, the system introduces linear polarization degree (DoLP) as a confidence weight for optical flow calculation. Given that polarization angle data in low-polarization regions often contains significant random noise, the system constructs a DoLP-based weighting function to spatially filter the instantaneous velocity field, automatically attenuating or eliminating spurious optical flow vectors in low-polarization regions (i.e., unreliable texture regions), thereby improving the robustness of the overall velocity field calculation. Further, a structure tensor analysis mechanism is introduced to address the aperture problem in weakly textured regions. The system calculates the structure tensor matrix of the polarization angle texture field and solves for its minimum eigenvalue (…). ); It characterizes the richness and isotropy of local textures; the system sets a solvability threshold, only... Optical flow calculation is performed in regions exceeding the threshold. For extremely smooth regions with excessively small eigenvalues, the velocity field is forcibly set to zero, thus mathematically eliminating algorithm divergence and spurious motion estimation caused by texture loss.

[0044] Step S50: Time-frequency decoupling of rheological characteristics and evaluation of softening degree Frequency domain analysis of the time-varying optical flow velocity field is performed to separate the rheological characteristics characterizing soil instability from the background of environmental vibration.

[0045] Time-frequency transformation of the micro-motion velocity field: The system performs short-time time-frequency analysis (such as short-time Fourier transform) on the instantaneous velocity modulus sequence, calculates the power spectral density, and obtains the distribution of micro-motion kinetic energy along the frequency axis.

[0046] Rheological modal energy decoupling: A cutoff frequency is set as the boundary between elastic and plastic modes. This cutoff frequency is not a fixed value but is dynamically set through a spectrum scanning algorithm. The system first searches for the maximum peak value of the power spectral density within a preset high-frequency range (e.g., 10Hz-50Hz), identifies it as the dominant environmental vibration frequency (e.g., the operating frequency of a rock drill), and sets the cutoff frequency to 1 / 2 to 1 / 3 of this dominant frequency to ensure effective separation of mechanical elastic vibration energy. Furthermore, to prevent low-frequency harmonics or beat frequency signals generated by mechanical vibration from mixing into the low-frequency creep energy band, the system constructs an adaptive comb filter. Based on the searched dominant vibration frequency fundamental wave, it automatically generates a filter containing the fundamental frequency. The system uses a stopband sequence of the wave and its preceding Nth harmonics to perform multi-point notch filtering on the power spectral density, precisely filtering out the energy projection of mechanical vibration across the entire frequency band and retaining only the non-periodic soil plastic flow spectral characteristics. During this process, the system executes construction status determination logic: if the maximum peak energy found is lower than the preset background noise baseline threshold, it determines that the current period is a quiet period of mechanical shutdown, and the system automatically switches to silent monitoring mode, forcibly locking the cutoff frequency to a preset extremely low value (e.g., 0.5Hz) to prevent erroneous mixing of high-frequency thermal noise into creep energy when there is no mechanical vibration reference; the power spectral density is integrated in the low-frequency band to obtain the low-frequency creep energy density. ), corresponding to the plastic flow energy of the soil; the power spectral density is integrated in the high-frequency band to obtain the high-frequency stiffness energy density ( This corresponds to the elastic oscillation energy of the soil.

[0047] Calculation of rheological softening factor: To normalize the assessment of risks in different regions, a rheological softening factor was constructed. The calculation formula is as follows: ; In the formula, This represents the rheological softening factor, with a value range of [0,1]. Indicates low-frequency creep energy density; Indicates high-frequency stiffness energy density; To prevent the use of a tiny constant with a denominator of zero, when this factor approaches 1, it indicates an extremely high risk of softening and mudslide at the working face.

[0048] Step S60: Multi-source feature fusion and risk quantification assessment The spatial-temporal alignment and logical coupling of water body distribution characteristics, kinematic velocity characteristics, and rheological characteristics generate quantitative indicators.

[0049] Construct a pixel-level water and mud inrush risk index model: Based on the physical causal chain of material basis (water presence), mechanical state (softening), and motion performance (displacement), the features obtained in the aforementioned steps are fused at the pixel level to calculate the risk index. The calculation formula is as follows: ; In the formula, Indicates a risk index; The water film distribution mask generated during the generation of the water film distribution mask (1 for water and 0 for no water) is used as a logic gate signal. The rheological softening factor is the rheological softening factor calculated in the calculation and is used as a weighting coefficient. This represents the instantaneous velocity modulus calculated in the numerical solution of the micro-motion velocity field; The velocity sensitivity coefficient is given. This model indicates that the risk index only increases significantly when three conditions are simultaneously met: a water-rich environment, rheological softening, and continuous displacement.

[0050] Regional risk aggregation and alerts: High-risk areas are extracted using connected component analysis, and a comprehensive risk score including area weighting is calculated. After time-domain smoothing filtering, this score is compared with a preset graded alarm threshold, and corresponding early warning or control actions are executed. Furthermore, the system calculates the time rate of change (i.e., the first derivative) of the comprehensive risk score. If the risk score shows a continuous upward trend and the growth rate exceeds a preset acceleration threshold, the system will automatically raise the alarm level even if the current score has not reached the highest alarm line, to warn of the typical exponentially accelerating destructive characteristics before a mudslide disaster. Simultaneously, the system integrates a rheological dynamics-based reciprocal velocity prediction module, which tracks the reciprocal of the instantaneous velocity modulus of high-risk areas in real time. The system analyzes the evolution curve over time and performs linear regression analysis. When the reciprocal velocity curve shows a trend of convergence towards the time axis, it calculates the intercept of the regression line with the time axis. This intercept is the theoretical estimated time to failure (TTF). The system outputs the TTF value in the form of a countdown, providing on-site construction personnel with a quantified evacuation time window, thus achieving a leap from status warning to trend forecast.

[0051] Specific application example: Monitoring of sudden surge at the working face of a water-rich soft rock tunnel. Project Background: This embodiment is applied to the entrance section of a railway tunnel under construction in southwestern my country. The surrounding rock of this section is classified as Class IV, and the geological structure is mainly weathered phyllite interbedded with mudstone. Groundwater is well-developed, the tunnel face is often in a moist state, and there are local water-rich and weak zones, making it extremely prone to water and mud inrush disasters.

[0052] Hardware deployment (corresponding to S10): Data acquisition equipment: A long-wave infrared polarization imaging system is installed on the secondary lining trolley 15 meters away from the working face.

[0053] Excitation light source: an active infrared thermal radiation source with a power of 500W (wavelength 8-14μm) was configured. After preheating for 30 minutes, the power fluctuation stabilized at 0.05% / min.

[0054] Geometric constraints: Adjust the angle of the light source so that its optical axis forms a 45-degree angle with the normal of the face (near Brewster's angle), and maximize the polarization effect of the water body by utilizing the Fresnel reflection principle.

[0055] Parameter settings: Image acquisition frame rate is 25fps, resolution is 640×512.

[0056] Detailed explanation of the monitoring process: Phase 1: Hazard Identification and Illustrated Physical Feature Analysis (S20-S30) The system first initiates data acquisition and calculates the Stokes vector in real time. The corresponding monitoring image at this time is shown in Part 1: Schematic diagram of polarization reflection enhancement principle. In the left sub-image (traditional thermal imaging monitoring effect), the system output... Component profile. The black wavy lines appear generally chaotic and flat, with the radiation intensity value fluctuating around 0.5 without showing any obvious characteristic peaks. This indicates that although water seepage exists, the temperature of both the rock and the seeping mud is around 18℃, with a temperature difference of less than 0.5℃, resulting in almost no difference in their radiation energy. This signal aliasing phenomenon directly reflects the technical challenge of the thermal equilibrium state described in step S10, namely, that neither the naked eye nor traditional thermal imagers can distinguish between water and rock / soil in this state.

[0057] Subsequently, the system entered polarization domain analysis to calculate the degree of linear polarization (DoLP). Due to the use of a 45-degree oblique incidence excitation, the generated right sub-image (the polarization enhancement monitoring effect of this invention) showed a significant change: The curve remains at a low level (close to 0) on both sides, indicating that the rough soil and rock mainly undergo diffuse reflection and the depolarization effect is obvious; while a significant peak (value of about 0.65) appears in the middle part of the curve, which corresponds to the seepage area where specular reflection occurs.

[0058] This high signal-to-noise ratio waveform that rises from the ground corresponds to the effect of successfully segmenting the water body through differences in physical characteristics in step S30.

[0059] Based on this difference, the system generated a clear water film distribution mask (Mwater). The evolution of this feature over time is recorded in the first row of the curve (feature 1: water film distribution mask) in Part 3: Multi-source fusion risk assessment map: the curve remains at 0 (no water) in the early stage of monitoring, and jumps to 1 (water) in a step-like manner around the 10th minute, which indicates that the system has successfully locked the material basis of the sudden water disaster.

[0060] Phase Two: Micro-motion Capture and Velocity Field Construction (S40) For the locked water-rich area, although the pixel grayscale is uniform in the thermal image, the system calculates its polarization angle (AoP) field and finds a high-frequency texture similar to a fingerprint. The Lucas-Kanade algorithm is used to track the movement of the AoP texture and construct a micro-motion velocity field.

[0061] The result is shown in the third row of the multi-source fusion risk assessment graph. The image is no longer a static distribution map, but rather shows the dynamic process of velocity changing over time.

[0062] Monitoring data shows that from the 10th to the 40th minute, the curve only fluctuated slightly (background noise), but after the 45th minute, the curve showed an exponential and sharp rise, entering the accelerated deformation stage, with the instantaneous speed rapidly climbing to more than 20 mm / min. This quantitative physical index change trend indicates that substantial instability and sliding are occurring inside the soil.

[0063] Phase 3: Rheological property analysis and frequency domain decoupling (S50) At this time, a rock drill was operating inside the tunnel, causing vibrations throughout the site. To distinguish between mechanical vibrations and precursors to water inrush, the system performed frequency domain analysis, and the generated analysis spectrum is shown in the rheological characteristic decoupling principle diagram: The horizontal axis (frequency) of the spectrum represents the speed of the face micro-motion (Hz), and the vertical axis (normalized energy density) represents the vibration intensity. The dashed line (cutoff frequency) in the graph corresponds to the limit set in step S50 (e.g., 2 Hz). The left side is the low-frequency plastic rheological region, and the right side is the high-frequency elastic vibration region.

[0064] For the background rock area, its spectral characteristics match the left subplot (stable state) in the figure: there are peaks only in the high-frequency region on the right (around 20Hz), which means that although the face is moving, it is elastic vibration caused by the rock drill, and the soil itself has not been displaced. The rheological softening factor is 0, and it is in a safe state.

[0065] However, for the water-rich area in the lower right corner, its spectral characteristics match the right subplot (instability state) in the figure: in addition to mechanical vibration, a huge black peak (labeled as rheological creep energy) appears near 0 Hz on the left, indicating that the soil is undergoing irreversible slow flow (the 0Hz-1Hz component rises sharply).

[0066] The system calculates the rheological softening factor based on the formula ( ): ; In the formula, This represents the rheological softening factor, with a value range of [0,1]. Indicates low-frequency creep energy density; Indicates high-frequency stiffness energy density; To prevent tiny constants with a denominator of zero.

[0067] The calculation results show that the area The value climbed to 0.85 and approached 1, indicating extreme danger and suggesting that the soil at that location had transformed from an elastic body into a viscous fluid.

[0068] Phase Four: Risk Warning and Response (S60) The system finally performs logical fusion of the aforementioned features, and the resulting curve corresponds to the fourth row of the multi-source fusion risk assessment chart (final output: multi-source fusion mudslide risk index): This curve is a comprehensive representation of the logical operations of the three characteristics (water, softening, and speed) mentioned above.

[0069] Within 0-10 minutes, due to the absence of water, the risk index is close to 0.

[0070] Within 10-40 minutes, although there was water and the soil was softened, the speed was not out of control, and the risk index rose slowly but did not exceed the limit.

[0071] Around the 46th minute, with the explosive growth of the velocity field, the three malignant conditions of abundant water, high softening, and accelerated deformation were simultaneously met, and the risk index curve rose sharply and broke through the alarm threshold (0.75) shown by the dotted line.

[0072] The gray shaded areas in the diagram indicate the time periods when a Level IV alarm is triggered.

[0073] An alarm sounded immediately, and personnel were evacuated. Actual conditions indicated that approximately 40 cubic meters of mudslide occurred in the area about 30 minutes after the alarm was triggered. Thanks to the timely warning (about half an hour before the disaster), no casualties were reported.

Claims

1. A method for intelligent identification of water leakage risk at the tunnel face in water-rich clay layers, characterized in that, Includes the following steps: The active infrared excitation module (120) is activated to project infrared radiation, and the polarized infrared imaging module (110) is used to collect radiation image sequences with different polarization directions. The Stokes vector, including the total intensity of infrared radiation, is calculated based on the radiation image sequence using the Stokes solver unit (131). The feature segmentation unit (132) calculates the degree of linear polarization based on the Stokes vector and performs threshold comparison to generate a water film distribution mask; Using the micro-motion analysis unit (133) to construct a polarization angle field as a texture feature based on the Stokes vector, the instantaneous velocity field is obtained by solving the optical flow constraint equation that conserves the polarization angle; The instantaneous velocity field is transformed in the frequency domain using the risk assessment unit (134) to extract the low-frequency creep energy density and the high-frequency stiffness energy density. The soil rheological softening factor is determined based on the proportion of the low-frequency creep energy density in the total energy. The risk assessment unit (134) is used to construct a mud inrush risk index by combining the water film distribution mask, the rheological softening factor and the instantaneous velocity field, and the mud inrush risk index is compared with the preset graded alarm threshold to output the corresponding risk level prompt signal.

2. The intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers according to claim 1, characterized in that, The step of activating the active infrared excitation module (120) to project infrared radiation includes: The active infrared excitation module (120) is controlled to continuously radiate the tunnel face in an oblique incidence manner, so that the optical axis of the radiation center of the active infrared excitation module (120) forms a preset incident angle with the normal direction of the observation area of ​​the tunnel face; The incident angle is used to generate Fresnel reflection in the water-bearing area, so that the reflectivity of the water surface increases with the increase of the incident angle.

3. The intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers according to claim 1, characterized in that, The steps of acquiring radiation image sequences with different polarization directions using the polarized infrared imaging module (110) include: Using a focal plane detector integrated with a micro-polarizer array, radiation signals from the linearly polarized transmission directions at 0°, 45°, 90° and 135° are simultaneously sensed in a single exposure. The Stokes vector includes the total infrared radiation intensity representing the total energy of the target radiation, the horizontal and vertical polarization difference components representing the difference between the horizontal and vertical linear polarization components, and the 45-degree and 135-degree polarization difference components representing the difference between the linear polarization components in the 45-degree and 135-degree directions.

4. The intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers according to claim 3, characterized in that, The step of calculating the degree of linear polarization using the feature segmentation unit (132) based on the Stokes vector includes: Obtain the total intensity modulus of the linearly polarized light synthesized from the horizontal linear polarization component, the vertical linear polarization component, and the polarization difference components at 45 degrees and 135 degrees; The degree of linear polarization is obtained by calculating the ratio of the total intensity modulus of the linearly polarized light to the total intensity of the infrared radiation.

5. The intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers according to claim 1, characterized in that, The step of generating a water film distribution mask that indicates the free water distribution area at the tunnel face includes: Based on the numerical distribution characteristics of the linear polarization degree, the optimal segmentation threshold is calculated using the maximum inter-class variance method. The regions with linear polarization degrees higher than the optimal segmentation threshold are marked as free water film coverage regions, thereby generating the water film distribution mask.

6. The intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers according to claim 1, characterized in that, The step of calculating the polarization angle field distribution of the tunnel face using the micro-motion analysis unit (133) based on the Stokes vector includes: The polarization angle is calculated using the polarization difference component contained in the Stokes vector, and the polarization angle represents the angle between the principal direction of the incident photoelectric vector vibration and the reference axis. A polarization angle field is constructed to reflect the local normal direction change at the tunnel face.

7. The intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers according to claim 1, characterized in that, The step of using the risk assessment unit (134) to perform frequency domain transformation on the instantaneous velocity field includes: Short-time time-frequency analysis is performed on the instantaneous velocity modulus sequence of the instantaneous velocity field to calculate the power spectral density; The cutoff frequency is set as the boundary between the elastic and plastic modes; The power spectral density is integrated in the low-frequency band below the cutoff frequency to obtain the low-frequency creep energy density; The power spectral density is integrated over a high-frequency band above the cutoff frequency to obtain the high-frequency stiffness energy density.

8. The intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers according to claim 7, characterized in that, The steps for determining the soil rheological softening factor include: The ratio of the low-frequency creep energy density to the denominator is calculated to obtain the rheological softening factor; The denominator is the sum of the low-frequency creep energy density, the high-frequency stiffness energy density, and a preset small constant.

9. The intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers according to claim 1, characterized in that, The step of constructing a mudslide risk index using the risk assessment unit (134) in conjunction with the water film distribution mask, the rheological softening factor, and the instantaneous velocity field includes: The mudslide risk index is obtained by calculating the product of the water film distribution mask, the rheological softening factor, and the velocity-sensitive term. The velocity sensitivity term is obtained by calculating the difference between a constant and a negative power of the natural constant, where the exponent of the negative power is the negative value of the product of the velocity sensitivity coefficient and the instantaneous velocity modulus of the instantaneous velocity field.

10. The intelligent identification method for water leakage risk at the tunnel face in water-rich clay layers according to claim 1, characterized in that, Following the step of outputting the risk warning, the following also includes: High-risk areas where the mudslide risk index exceeds a threshold are extracted using connected component analysis techniques. Calculate the comprehensive risk score that includes area weighting; The comprehensive risk score is subjected to time-domain smoothing filtering, and an early warning action is executed based on the processed score.

Images