Tunnel face water gushing hazard identification method based on image fusion

Abstract

The present application relates to the cross technical field of tunnel engineering safety monitoring and image information processing, and discloses a tunnel face water gushing hidden danger identification method based on image fusion, which solves the problem that the existing tunnel face water gushing hidden danger identification technology cannot accurately identify early concealed water gushing hidden dangers and objectively quantify risks. The method comprises the following steps: first, synchronously collecting infrared images and visible light images of the tunnel face and performing pretreatment, then, extracting temperature field features and texture field features; then, using a weighted fusion method to fuse the temperature field features and the texture field features, and generating a fused image; finally, constructing a water gushing risk model, normalizing the temperature field features and the texture field features, calculating the water gushing risk value according to the water gushing risk model, dividing the risk level according to the risk value, and performing corresponding early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the cross technical field of tunnel engineering safety monitoring and image information processing, and in particular to a tunnel face water inrush hidden danger identification method based on image fusion. BACKGROUND

[0002] As a key infrastructure in the fields of transportation, water conservancy and energy, tunnel engineering is prone to face water inrush when passing through water-rich strata such as karst strata, sandy pebble strata and fault fracture zones. The root cause of the water inrush is the hidden water-bearing fissures or karst caves behind the tunnel face. If not identified in time, it may lead to accidents such as collapse of the tunnel face, water and mud inrush, which not only delays the construction period, but also may cause equipment damage and personnel casualties. Therefore, early and accurate identification of water inrush hidden dangers at the tunnel face is a core requirement for safety control in tunnel construction.

[0003] Currently, the identification technology for water inrush hidden dangers at the tunnel face mainly includes geological survey technology and image recognition technology.

[0004] (1) Geological survey technology:

[0005] Geological survey technology is a traditional method for tunnel advanced detection, such as advanced geological radar and borehole detection. Among them, advanced geological radar detects water-bearing structures within a certain range behind the tunnel face by emitting electromagnetic signals, but its resolution is easily affected by the electromagnetic environment in the tunnel, and the detection resolution is low, which cannot effectively distinguish dry fissures from water-filled fissures. Borehole detection requires drilling multiple boreholes at the tunnel face, which can directly verify whether there is water, but it is a destructive detection method with low efficiency and high risk of water inrush.

[0006] (2) Image recognition technology:

[0007] Image recognition technology usually uses infrared image recognition or visible light image recognition. Among them, infrared image recognition technology identifies abnormal areas by capturing temperature distribution differences at the tunnel face, but it can only reflect temperature characteristics and cannot associate with structural fissures at the tunnel face, which may misjudge temperature difference anomalies caused by ventilation cooling, equipment heat dissipation and other non-water inrush factors as water inrush hidden dangers. Visible light image recognition extracts structural texture features (such as fissure length, width and density) to judge structural stability, but it cannot identify whether the fissure is water-filled. For water-filled fissures hidden in rock bodies and not fully connected, visible light images cannot detect them.

[0008] In addition, the existing identification technology for water inrush hidden dangers at the tunnel face also has the problem of poor field applicability: many devices require manual parameter adjustment, which is unsafe in environments with high dust and narrow space; and data processing relies on a back-end computer, which cannot achieve real-time early warning on site and cannot meet the dynamic monitoring needs of tunnel construction.

[0009] In summary, there is an urgent need for a solution that can accurately extract temperature and texture features, establish a quantitative correlation model, and adapt to complex tunnel environments, to solve the core problems of early hidden water gushing hazard identification difficulty and subjective risk judgment, and to ensure the safety of tunnel construction in water-rich strata. SUMMARY

[0010] The technical problem to be solved by the present application is to provide a tunnel face water gushing hazard identification method based on image fusion, which solves the problem of difficulty in accurately identifying early hidden water gushing hazards and objectively quantifying risks in existing tunnel face water gushing hazard identification technologies.

[0011] The technical solution adopted by the present application to solve the above technical problem is:

[0012] The tunnel face water gushing hazard identification method based on image fusion comprises the following steps:

[0013] S1. Synchronously collecting infrared images and visible light images of the tunnel face;

[0014] S2. Preprocessing the collected infrared images and visible light images respectively;

[0015] S3. Extracting temperature field features based on the preprocessed infrared images, and extracting texture field features based on the preprocessed visible light images;

[0016] S4. Using a weighted fusion method to fuse the temperature field features and the texture field features, and generating a fused image based on the fused features, combining the temperature information of the infrared images and the texture information of the visible light images;

[0017] S5. Constructing a water gushing risk model, normalizing the temperature field features and the texture field features, calculating the water gushing risk value according to the water gushing risk model, dividing the risk level according to the risk value and performing corresponding early warning.

[0018] Further, in step S1, the way of synchronously collecting infrared images and visible light images of the tunnel face comprises:

[0019] An infrared-visible light dual-camera synchronous acquisition system is built behind the tunnel face, after parameter calibration and spatial registration of the dual cameras are completed, infrared images and visible light images of the tunnel face are synchronously collected at a set frequency;

[0020] The parameter calibration includes: adjusting the emissivity of the infrared camera according to the surface characteristics of the rock mass in the tunnel; adjusting the exposure time of the visible light camera according to the illumination intensity of the face, and turning on the fill light when the illumination intensity is insufficient;

[0021] The space registration comprises: taking the light-reflecting mark pasted in advance at the center of the working face as a target, obtaining the internal and external parameters of the dual cameras through calibration, establishing a pixel mapping relationship between the infrared image and the visible light image, and ensuring that the images collected by the dual cameras are aligned under the same space coordinate system.

[0022] Further, in step S2, the manner of pre-processing the collected infrared image comprises:

[0023] An adaptive Wiener filtering algorithm is used for image denoising, and the formula is:

[0024] ;

[0025] wherein g(x,y) is the pixel value of the denoised infrared image; f(x,y) is the pixel value of the original infrared image; μ(x,y) is the image mean value in a 3*3 local window centered at pixel (x,y), is the image variance in the window; is the noise variance, which is obtained by image dark area statistics.

[0026] Further, in step S2, the manner of pre-processing the collected visible light image comprises:

[0027] A Retinex algorithm is used for image enhancement, and the formula is:

[0028] ;

[0029] wherein, is the enhanced visible light image; is the pixel value of the original visible light image; is a Gaussian filter kernel; denotes convolution operation.

[0030] Further, in step S2, it further comprises:

[0031] The pre-processed infrared image and visible light image are subjected to dual image registration optimization: a plurality of feature points on the pre-processed infrared image and visible light image are extracted respectively; the feature points are matched through a matcher, and the random sample consensus algorithm is used to eliminate false matches; the pixel mapping relationship is updated based on the correctly matched feature points, and the registration accuracy is ensured to be ≤0.5 pixels.

[0032] Further, in step S3, the temperature field features are extracted based on the pre-processed infrared image, comprising:

[0033] The temperature value of each pixel point is extracted from the pre-processed infrared image, and the overall average temperature of the working face is calculated:

[0034] ;

[0035] wherein, is the overall average temperature of the tunnel face; M and N are respectively the number of rows and columns of the infrared image; is the temperature value of the i-th pixel point;

[0036] An environmental temperature correction term obtained by a tunnel environment monitoring sensor is introduced , to calculate the corrected single-point temperature difference:

[0037] ;

[0038] wherein, is the corrected single-point temperature difference, when <0, take =0, indicating that the pixel point has no temperature difference anomaly, when >0, indicating that the pixel point has a temperature difference anomaly;

[0039] By counting the pixel points with >0, the temperature difference anomaly area is obtained.

[0040] Further, in step S3, the texture field features are extracted based on the preprocessed visible light image, including:

[0041] An edge detection algorithm is used to extract edges from the preprocessed visible light image, and after Gaussian filter smoothing processing, gradient amplitude and direction calculation, non-edge pixel suppression processing, the edge breakpoints are connected to obtain an edge image;

[0042] The edge image is subjected to morphological opening operation, and the edges are thinned to single-pixel-width crack skeletons through a skeleton extraction algorithm to obtain a crack image;

[0043] The crack density D and the average crack width are calculated:

[0044] ;

[0045] ;

[0046] wherein, P and Q are respectively the number of rows and columns of the visible light image; d is the actual physical size corresponding to the image pixel, which is obtained through camera calibration; S is the actual area of the tunnel face, which is calculated through the tunnel section size; F(x, y) is the crack image pixel value, 1 indicating a crack pixel and 0 indicating a non-crack pixel;

[0047] ;

[0048] ;

[0049] is the total area of the fissure region in the fissure image; is the total length of the fissure skeleton.

[0050] Further, in step S4, the weighted fusion method is used to fuse the temperature field features and the texture field features, and based on the fused features, the temperature information of the infrared image and the texture information of the visible light image are combined to generate a fusion image, including:

[0051] The weight of the temperature field features is set to 0.5, and the weight of the texture field features is set to 0.5, and a fusion feature matrix is constructed:

[0052] ;

[0053] Wherein, is the fusion feature matrix; is the feature matrix of the temperature field features, In the temperature difference anomaly area, the pixel value is set to 1, and the pixel value of the non-temperature difference anomaly area is set to 0; is the feature matrix of the texture field features, In the fissure, the pixel value is set to 1, and the pixel value of the non-fissure is set to 0;

[0054] Based on the fusion feature matrix , the temperature information of the infrared image and the texture information of the visible light image are combined to generate a fusion image, wherein the temperature difference anomaly area is marked by color, and the fissure texture is highlighted by gray scale contrast.

[0055] Further, in step S5, the water gushing risk model is:

[0056] ;

[0057] Wherein, R is the water gushing risk value; is the average normalized temperature difference of the temperature difference anomaly area, that is, the temperature difference of each pixel point of the temperature difference anomaly area After normalization, the mean value is calculated; is the normalized value of the fissure density; is the normalized value of the average fissure width; 、 、 is a preset weight coefficient, which is determined according to engineering experience.

[0058] Further, in step S5, the risk value is divided into risk levels and corresponding warning is carried out, including:

[0059] R≤0.3, safe, measures: normal construction, monitoring once every 30 minutes;

[0060] 0.3 < R < 0.5, low risk, measures: increase the monitoring frequency to once every 15 minutes, and increase the number of face inspections;

[0061] 0.5 < R < 0.7, medium risk, measures: suspend the excavation construction, use advanced small pipe grouting to reinforce the face, and increase the monitoring frequency to once every 5 minutes;

[0062] R > 0.7, high risk, measures: immediately evacuate the face construction personnel and equipment, start the emergency plan, use full-face grouting to block the water channel, and continuously monitor for 24 hours.

[0063] The beneficial effects of the present application are:

[0064] (1) Breakthrough in identifying hidden water inrush, improving identification accuracy:

[0065] The temperature difference abnormal area extracted based on infrared images can reflect the internal water filling signal of the rock mass, and the fissure texture extracted based on visible light images can locate the water inrush channel. After the fusion of the two, it can be determined whether the temperature difference abnormal area coincides with the fissure space, avoiding the misjudgment of temperature difference abnormality as a hidden danger caused by ventilation cooling, equipment heat dissipation and other non-water inrush factors, and accurately identifying the hidden water filling fissure that does not penetrate to the surface of the face, greatly reducing the probability of misjudgment and omission.

[0066] (2) Realize the quantitative evaluation of water inrush risk, and avoid the deviation of subjective decision:

[0067] The present application eliminates the dimensional differences of temperature difference, fissure density and average fissure width through normalization processing, introduces a weight coefficient to balance the contribution of each feature to water inrush risk, accurately calculates the quantitative risk value, objectively reflects the synergistic effect of water body existence probability, water inrush channel development degree and water inrush capacity, and avoids the risk misjudgment caused by a single index. In addition, the present application sets up multiple risk levels based on the risk value, and matches different early warning measures for each risk level, so that the risk prevention and control has clear standards and is targeted, improving the scientificity and reliability of construction decision.

[0068] (3) Adapt to the complex environment of the tunnel site, and ensure the real-time and safety of monitoring:

[0069] The infrared emissivity and visible light exposure parameters of the dual-camera acquisition system in the present application automatically adapt to environmental changes, and the whole image processing process is automatically executed without manual on-site intervention, which not only avoids the safety hazards of personnel approaching the face, but also greatly shortens the early warning response time, providing sufficient window for emergency disposal of sudden water inrush. In addition, the acquisition frequency can be flexibly adjusted according to the construction progress, avoiding monitoring redundancy or missing key periods caused by fixed frequency.

[0070] (4) Reduce monitoring cost and improve engineering economy:

[0071] The existing geological exploration technology (such as advanced geological radar, drilling detection) has high equipment purchase and operation and maintenance cost, drilling detection also needs to destroy the integrity of the rock mass, which may induce additional water gushing risk, and the single detection efficiency is low, which is difficult to adapt to the continuous construction demand of the tunnel. The infrared and visible light cameras used in the application are conventional industrial equipment, which do not need special customization and have much lower cost than professional geological exploration equipment; the system can be disassembled and migrated, and can be repeatedly applied to multiple tunnel projects, and only needs simple cleaning and maintenance in the later period, which greatly reduces the whole life cycle cost. In addition, without drilling an advanced drill hole, the drilling construction time is saved, and the risk of water gushing caused by the destruction of rock mass balance by drilling is avoided, and the loss caused by the stoppage of work due to accidents is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 The flow chart of the tunnel face water gushing hidden danger identification method based on image fusion in the embodiment of the application.

[0073] Figure 2 The flow chart of the pre-processing of the infrared image and the visible light image in the embodiment of the application.

[0074] Figure 3 The feature layer fusion flow chart based on weighted fusion in the embodiment of the application.

[0075] Figure 4 The water gushing risk model calculation and risk grade division flow chart in the embodiment of the application. DETAILED DESCRIPTION

[0076] The application aims to provide a tunnel face water gushing hidden danger identification method based on image fusion, which solves the problem that the existing tunnel face water gushing hidden danger identification technology is difficult to accurately identify early hidden water gushing hidden dangers and objectively quantify risks. The core idea is: for the core contradiction that the water-rich stratum tunnel face water gushing hidden danger is strong, difficult to identify and lacks basis for risk quantification, the dual-mode image feature deep fusion + multi-dimensional risk quantification modeling is taken as the technical core, and the whole process identification system of image synchronous acquisition, environment anti-interference preprocessing, temperature difference anomaly and fracture texture double feature extraction, feature weighted fusion, risk value calculation and grading early warning is constructed, which provides a systematic solution for early accurate identification and safety control of the tunnel face water gushing hidden danger.

[0077] The technical means adopted by the application to realize the above-mentioned core idea includes:

[0078] (1) The essence of water inrush hidden danger is a concealed water-filled fissure, which needs to meet two prerequisites of existing water body (temperature difference) and having a channel (fissure development), and both are indispensable. The fundamental defect of the existing technology is that only a single feature is captured, such as infrared image can only identify temperature difference but cannot associate fissure (easy to misjudge non-water inrush temperature difference anomaly), and visible light image can only extract fissure but cannot judge whether it is water-filled (easy to miss concealed water-filled fissure).

[0079] The present application takes infrared temperature field feature (reflecting water body) and visible light texture field feature (reflecting fissure) as two core bases for identifying water inrush hidden danger, and through technical design, the spatial and logical relationship between the two is associated to solve the misjudgment and missed judgment problem from the source. Only when the temperature difference anomaly area and the fissure texture space coincide, and the features of both meet the risk threshold, is it determined as a potential water inrush hidden danger, ensuring the accuracy of the identification.

[0080] (2) Dust, uneven lighting and vibration in the tunnel can cause image feature distortion, so first, the Gaussian noise of the infrared image is suppressed by adaptive filtering, the light and texture components of the visible light image are separated by Retinex algorithm, and the camera offset is eliminated by dynamic registration optimization, to ensure that the "temperature difference anomaly area" extracted later is a real water body signal, and the "fissure texture" is a reliable channel feature, avoiding feature distortion caused by environmental interference.

[0081] In the fusion of the feature layer, the temperature difference anomaly area feature matrix and the fissure texture feature matrix are weighted and fused, and the balanced weight is set based on engineering experience, so that the fusion result can clearly present the spatial coincidence area of the temperature difference anomaly area and the fissure, directly positioning the specific position of the water-filled fissure, and providing visual and associated feature support for subsequent risk judgment.

[0082] (3) The present application builds a temperature difference-fissure density correlation risk model, combines the engineering logic of "water body is a prerequisite for water inrush, fissure is a channel for water inrush, and width affects water inrush capacity", reasonably allocates weights to temperature difference, fissure density and average fissure width, calculates the risk value by linear combination, and then divides the risk value into multiple risk grades, each grade matches clear warning measures, so that the construction party has clear decision basis.

[0083] The scheme of the present application will be further described below in combination with the drawings and examples.

[0084] The present embodiment provides a tunnel face water inrush hidden danger identification method based on image fusion. Before implementation, first, an infrared-visible light dual-camera synchronous acquisition system is built behind the tunnel face, providing a basis for subsequent image acquisition, processing, feature extraction and analysis.

[0085] In an exemplary embodiment, the dual-camera synchronous acquisition system is deployed 6-10 m behind the tunnel face, which needs to avoid the construction interference area. The system mainly includes an infrared thermal imager, a high-definition visible light camera, an integrated support and a field processing terminal; wherein the infrared thermal imager and the high-definition visible light camera are fixed through the integrated support to ensure that the lens axes are parallel and face the center of the tunnel face, avoiding subsequent registration errors caused by angle deviation; the infrared thermal imager and the high-definition visible light camera are wirelessly connected with the field processing terminal.

[0086] Before acquisition, first, the dual-camera is calibrated and spatially registered, wherein the parameter calibration is to adjust the emissivity of the infrared camera according to the surface characteristics of the rock mass in the tunnel, and to adjust the exposure time of the visible light camera according to the illumination intensity of the tunnel face, and to turn on the fill light when the illumination intensity is insufficient; the spatial registration includes: taking the reflective mark pasted in advance at the center of the tunnel face as the target point, obtaining the internal and external parameters of the dual-camera through calibration, the internal parameters including focal length, principal point coordinates, etc., and the external parameters including relative position, attitude angle, etc.; based on the internal and external parameters, the pixel mapping relationship between the infrared image and the visible light image is established to ensure that the images acquired by the dual-camera are aligned in the same spatial coordinate system.

[0087] Based on the above-mentioned dual-camera synchronous acquisition system, the embodiment provides an implementation process of the tunnel face water gushing hidden danger identification method based on image fusion, which is shown in Figure 1 , which includes the following implementation process:

[0088] S1. Synchronously acquiring infrared images and visible light images:

[0089] In this step, the infrared images and visible light images of the tunnel face are synchronously acquired according to the set acquisition frequency. The acquisition frequency is set according to the tunnel construction progress: when the tunnel is normally excavated, the daily progress is 1-2 m, and the acquisition is performed once every 30 minutes; when the water-rich stratum is passed through, such as the karst development area, the acquisition is performed once every 10 minutes; when a small amount of water seeps out of the tunnel face, the acquisition is performed once every 5 minutes. The acquired two kinds of images are wirelessly transmitted to the field processing terminal, and the subsequent processing steps are realized through the software in the field processing terminal.

[0090] S2. Image preprocessing:

[0091] In this step, the acquired infrared images and visible light images are respectively preprocessed; the core goal is to eliminate the environmental interference (dust, noise, uneven illumination) of the tunnel to improve the image quality and provide reliable data for the subsequent dual-feature extraction.

[0092] In an exemplary embodiment, the image preprocessing specifically includes three sub-steps of infrared image denoising, visible light image enhancement and dual-image registration optimization, as shown in Figure 2 .

[0093] 1. Infrared image denoising:

[0094] Dust in the tunnel can cause Gaussian noise in the infrared image, with a mean of 0 and a variance of 0.02-0.05. This embodiment uses an adaptive Wiener filter algorithm for denoising, which adjusts the filter coefficient by the image mean and variance in the local window, while retaining the temperature details and suppressing the noise. The filter formula is:

[0095]

[0096] where g(x, y) is the denoised infrared image pixel value; f(x, y) is the original infrared image pixel value; μ(x, y) is the image mean in the 3x3 local window centered at pixel (x, y), is the image variance in the window; is the noise variance, obtained by image dark area statistics.

[0097] 2. Visible light image enhancement:

[0098] Uneven lighting in the tunnel can cause bright and dark differences in the visible light image. This embodiment uses the Retinex algorithm for enhancement. This algorithm separates the reflection component and the lighting component of the image, the reflection component reflects the object texture, and the lighting component reflects the environment lighting. The reflection component is enhanced to highlight the crack texture. The core formula is:

[0099]

[0100] where I(x, y) is the enhanced visible light image; is the original visible light image pixel value; is the Gaussian filter kernel, with a standard deviation of 80; denotes convolution operation.

[0101] 3. Dual-image registration optimization:

[0102] Considering that vibration during tunnel construction can cause slight camera offset, on the basis of spatial registration during parameter setting of the dual-camera before the implementation of the scheme, feature point matching algorithm is used for registration optimization, specifically: extracting the feature points of the pre-processed infrared and visible light images, extracting hundreds of feature points for each image; performing feature point matching through the matcher, and using the random sample consensus algorithm to remove incorrect matches; updating the mapping relationship of the dual images based on the correctly matched feature points, ensuring that the registration accuracy is ≤0.5 pixels.

[0103] S3. Temperature field and texture field feature extraction:

[0104] ​​​In this step, based on the pre-processed image, the temperature field features (temperature difference abnormal area) of the infrared image and the texture field features (crack texture) of the visible light image are extracted respectively to realize the accurate quantification of double features.

[0105] In an exemplary embodiment, the process of double feature extraction is as follows:

[0106] 1. Infrared image temperature field feature extraction:

[0107] First, the temperature value of each pixel point is extracted from the denoised infrared image , unit: ℃; the overall average temperature of the tunnel face is calculated , the formula is:

[0108] ;

[0109] Wherein, M and N are the number of rows and columns of the infrared image, in this embodiment, the number of rows of the infrared image is set to 640, and the number of columns is set to 512.

[0110] Second, considering the influence of environmental factors such as ventilation and equipment heat dissipation in the tunnel on temperature, an environmental temperature correction term is introduced, which is obtained by an environmental monitoring sensor in the tunnel, with a value range of 0-0.5℃;

[0111] The corrected single-point temperature difference T(x,y) is calculated, and the formula is:

[0112] ;

[0113] When T(x,y) < 0, take T(x,y) = 0, which means that there is no temperature difference anomaly at this pixel point.

[0114] 2. Visible light image texture field feature extraction:

[0115] In this embodiment, the extraction of visible light image texture field features is realized through the processes of edge detection, crack extraction and feature quantification, which are as follows:

[0116] Edge detection: edge detection algorithm is used to extract the edges of the visible light image, which contains the edges of the cracks. The image is smoothed by Gaussian filtering, and the standard deviation of Gaussian filtering is set to 1.0; the gradient amplitude and direction of the image are calculated; the non-edge pixels are suppressed by using double threshold, and the high threshold is set to 80 and the low threshold is set to 40; the edge breakpoints are connected to obtain the edge image E(x,y).

[0117] Crack extraction: for the edge image E(x,y), morphological opening operation is used to remove small noise edges, and the structure element of morphological opening operation is 3x3 rectangle; then the wide edges are thinned to single-pixel width crack skeleton by skeleton extraction algorithm, and the crack image F(x,y) is obtained.

[0118] Feature quantification: Calculate the core feature of fissure texture, fissure density D, unit: m / m², formula:

[0119] ;

[0120] Where P, Q are the number of rows and columns of the visible light image respectively; d is the actual physical size corresponding to the image pixel, which is obtained by camera calibration; S is the actual area of the tunnel face, which is calculated by the tunnel section size; F(x, y) is the pixel value of the fissure image, 1 represents fissure pixel, and 0 represents non-fissure pixel.

[0121] At the same time, the average width of the fissure is calculated , unit: mm, formula:

[0122] ;

[0123] Where, is the total area of the fissure region in the fissure image, unit: m².

[0124] ;

[0125] is the total length of the fissure skeleton, unit: m.

[0126] ;

[0127] After the above double feature extraction, the features obtained in this step include temperature field features: temperature difference abnormal area (combination of pixel points where T(x, y) > 0); texture field features: fissure density D and fissure average width .

[0128] S4. Weighted fusion of temperature field features and texture field features:

[0129] In this step, a weighted fusion method is used to realize the feature layer fusion of infrared and visible light images. This method combines the features by determining the weights of the features of the two kinds of images, and fully retains the detailed features of the temperature difference abnormal area and the fissure texture.

[0130] In an exemplary embodiment, the feature layer fusion process based on weighted fusion is shown in Figure 3 , which includes:

[0131] 1. Feature weight determination:

[0132] The weight is determined according to the importance of the infrared temperature feature and the visible light crack feature in identifying water gushing. The infrared temperature feature reflects the existence of water body and is one of the core basis for judging water gushing. The visible light crack feature reflects the development degree of water gushing channel and is also one of the core basis for judging water gushing. According to engineering experience, the weight of the infrared temperature feature is set to 0.5, the weight of the visible light crack feature is set to 0.5, and the sum of the weights of the two features is 1, so as to ensure the balance of the fused features.

[0133] 2. Feature fusion calculation:

[0134] The temperature difference anomaly area feature extracted from the infrared image and the crack texture feature extracted from the visible light image are linearly weighted and fused.

[0135] Let the feature matrix of the infrared image be , the feature matrix of the visible light image be , and the fused feature matrix be The calculation formula is:

[0136] ;

[0137] In the formula, the pixel value of the temperature difference anomaly area is set to 1, and the pixel value of the non-temperature difference anomaly area is set to 0. In the formula, the pixel value of the crack is set to 1, and the pixel value of the non-crack is set to 0.

[0138] 3. Fusion image generation:

[0139] Based on the fused feature matrix , the temperature information of the infrared image and the texture information of the visible light image are combined to generate a fusion image . In the fusion image, the temperature difference anomaly area is distinguished by color marking, for example, the temperature difference anomaly area is marked with blue, and the crack texture is highlighted by gray contrast, so that the fusion image clearly presents the position of the temperature difference anomaly area, the distribution of the crack, and the overlapping area of the two, which is convenient for construction personnel to intuitively identify hidden dangers.

[0140] Specifically, when generating the fusion image, the spatial correlation mask is first taken as the fusion feature matrix to clearly define the correlation area of the temperature difference anomaly and the fissure and various types of non-correlation areas, to ensure that they are consistent in size and spatially aligned with the infrared image and the visible light image; for the correlation area, the temperature value of the infrared image is extracted and color-coded according to the interval, and then superimposed with the texture information of the visible light image according to the set weight, to retain the fissure details and mark the temperature difference; for the non-correlation area, the core information of the corresponding image is retained and differentiated (such as reducing the brightness and saturation) to highlight the correlation area; then, the edge details are enhanced through channel reorganization and Laplacian sharpening, and the brightness contrast is optimized through adaptive histogram equalization, and finally the key information is superimposed and marked, to output the PNG format fusion image that clearly presents the temperature difference anomaly area and the fissure distribution.

[0141] S5. Quantitative calculation of water inrush risk and grade division:

[0142] In this step, a water inrush risk model is constructed, the temperature field features and the texture field features are normalized, the water inrush risk value is calculated according to the water inrush risk model, the risk grade is divided according to the risk value, and the corresponding warning is given.

[0143] In an exemplary embodiment, the water inrush risk model calculation and risk grade division process is shown in Figure 4 , which includes:

[0144] 1. Feature normalization processing:

[0145] Since the temperature difference T(x, y), the fissure density D, and the average fissure width have different dimensions, normalization processing is first performed to map the feature values to the [0, 1] interval. In this embodiment, the min-max normalization method is adopted, and the formula is:

[0146] ;

[0147] ;

[0148] ;

[0149] wherein, is the normalized temperature difference, is the normalized fissure density, is the normalized average fissure width; is the minimum value of the single-point temperature difference of all pixel points in the infrared image of the working face; is the maximum value of the single-point temperature difference of all pixel points in the infrared image of the working face; is the minimum value of the extracted fissure density in the visible light image of the working face; is the maximum value of the extracted fissure density in the visible light image of the working face; Minimum value of average fracture width extracted from the visible light image of the working face; Maximum value of average fracture width extracted from the visible light image of the working face.

[0150] 2. Water inrush risk value calculation:

[0151] By introducing a weight coefficient, a temperature difference-fracture density correlation risk model, i.e., a water inrush risk model, is constructed:

[0152]

[0153] wherein R is the water inrush risk value, ranging from 0 to 1, and the greater the value, the higher the risk, is the average normalized temperature difference of the temperature difference abnormal area, i.e., the temperature difference of each pixel point in the temperature difference abnormal area is the mean value obtained after normalization; is the normalized value of the fracture density; is the normalized value of the average fracture width; is a preset weight coefficient, which is determined according to engineering experience, for example: = 0.4, the temperature difference anomaly reflects the existence of water body; = 0.4, the fracture density reflects the development degree of water inrush channel; = 0.2, the fracture width reflects the water inrush capacity.

[0154] 3. Risk level division and early warning measures:

[0155] In this embodiment, the working face water inrush risk is divided into four levels according to the size of the risk value R, and corresponding early warning measures are developed:

[0156] First-level risk (safe, R≤0.3):

[0157] For example: the temperature difference abnormal area is small, ≤0.2, the fracture density is low, ≤0.3, the average fracture width is narrow, ≤0.2, and there is no water inrush hidden danger.

[0158] Measures: normal construction, monitoring once every 30 minutes.

[0159] Second-level risk (low risk, 0.3

[0160] For example: there is a small range of temperature difference abnormal area, 0.2 ≤0.4, the fracture density is moderate, 0.3 ≤0.5, the average fracture width is moderate, 0.2 ​​​≤0.4, potential slight water inrush risk.

[0161] Measures: Strengthen the monitoring frequency to once every 15 minutes, and increase the number of face inspections.

[0162] Third level risk (medium risk, 0.5 < R≤0.7):

[0163] For example: the abnormal area of temperature difference is large, 0.4 < R≤0.6, the fissure density is high, 0.5 < R≤0.7, the average fissure width is wide, 0.4 < R≤0.6, and there is a moderate water inrush risk. ≤0.6, fissure density is relatively high, 0.5 < R≤0.7, ≤0.7, average fissure width is wide, 0.4 < R≤0.6, ≤0.6, there is a moderate water inrush risk.

[0164] Measures: Suspend the driving construction, use advanced small pipe grouting to reinforce the face, and increase the monitoring frequency to once every 5 minutes.

[0165] Fourth level risk (high risk, R>0.7):

[0166] For example: the abnormal area of temperature difference is large, R>0.6, the fissure density is high, R>0.7, the average fissure width is wide, R>0.6, and sudden water inrush is easy to occur. >0.6, fissure density is high, >0.7, average fissure width is wide, >0.6, and sudden water inrush is easy to occur.

[0167] Measures: Immediately evacuate the face construction personnel and equipment, start the emergency plan, use full-face grouting to block the water inrush channel, and continuously monitor for 24 hours.

[0168] Although embodiments of the present application have been described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirits of the present application, and all of them do not depart from the protection scope of the present application.

Claims

1. A method for identifying water gushing hidden troubles of a tunnel face based on image fusion, characterized in that, It comprises the following steps: S1. Synchronously collecting infrared images and visible light images of a tunnel face; S2. Preprocessing the collected infrared images and visible light images respectively; S3. Extracting temperature field features based on the preprocessed infrared images; Extracting texture field features based on the preprocessed visible light images; The extraction of temperature field features based on the preprocessed infrared images comprises: Extracting temperature values of each pixel point from the preprocessed infrared images, and calculating the overall average temperature of the face: ； wherein, is the overall average temperature of the face; M and N are the number of rows and columns of the infrared image, respectively; is the temperature value of the i-th pixel. Introducing an environmental temperature correction term obtained by a tunnel environment monitoring sensor calculating a corrected single-point temperature difference: ； wherein, is the corrected single-point temperature difference, when < 0, take = 0, indicating that the pixel point has no temperature difference anomaly, when > 0, indicating that the pixel point has a temperature difference anomaly; By statistics The pixel points with temperature difference greater than 0 are obtained to obtain the temperature difference abnormal area. The extraction of texture field features based on the preprocessed visible light images comprises: Using an edge detection algorithm to extract edges from the preprocessed visible light images, and through Gaussian filter smoothing processing, gradient amplitude and direction calculation, non-edge pixel suppression processing, connecting edge breakpoints, an edge image is obtained; Performing morphological opening operation on the edge image, and through skeleton extraction algorithm, the edges are thinned to single-pixel-width crack skeletons, and a crack image is obtained; calculating the fracture density D and the average fracture width : ； ； Wherein, P and Q are the number of rows and columns of the visible light image respectively; d is the actual physical size corresponding to the image pixel, which is obtained through camera calibration; S is the actual area of the face, which is calculated through the size of the tunnel section; F(x, y) is the pixel value of the crack image, 1 represents a crack pixel, and 0 represents a non-crack pixel; ； ； is the total area of the fissure region in the fissure image; is the total length of the fissure skeleton; S4. Using a weighted fusion method to fuse the temperature field features and the texture field features, and based on the fused features, combining the temperature information of the infrared image and the texture information of the visible light image to generate a fused image: Setting the weight of the temperature field features as 0.5 and the weight of the texture field features as 0.5, a fusion feature matrix is constructed: ； wherein, is a feature matrix of the temperature field feature, is a feature matrix of the temperature field feature, In the temperature difference anomaly area, the pixel value is set to 1, and the pixel value of the non-temperature difference anomaly area is set to 0. is a feature matrix of the texture field feature, In the crack, the pixel value is set to 1, and the pixel value of the non-crack is set to 0. based on the fusion feature matrix , the infrared image temperature information and the visible light image texture information are combined to generate a fusion image, wherein a temperature difference abnormal area is marked by color, and a crack texture is highlighted by grayscale contrast; S5. Constructing a water gushing risk model, after normalizing the temperature field features and the texture field features, calculating the water gushing risk value according to the water gushing risk model, dividing the risk level according to the risk value and making corresponding early warning; The water gushing risk model is: ； Wherein, R is the water inrush risk value; is the average normalized temperature difference of the temperature difference anomaly area, that is, the temperature difference of each pixel point of the temperature difference anomaly area After normalization, the mean value is calculated again; is the normalized value of the fissure density; is the normalized value of the average fissure width; , , is a preset weight coefficient, which is determined according to engineering experience.

2. The tunnel face water gushing hidden danger identification method based on image fusion according to claim 1, characterized in that, In step S1, the way of synchronously collecting infrared images and visible light images of the tunnel face comprises: Building an infrared-visible light dual-camera synchronous acquisition system behind the tunnel face, after completing parameter calibration and space registration of the dual cameras, synchronously collecting infrared images and visible light images of the tunnel face according to a set frequency; The parameter calibration comprises: adjusting the emissivity of the infrared camera according to the surface characteristics of the rock mass in the tunnel; adjusting the exposure time of the visible light camera according to the illumination intensity of the face, and turning on the fill light when the illumination intensity is insufficient; The space registration comprises: taking the reflective mark pasted in advance at the center of the face as a target, obtaining the internal and external parameters of the dual cameras through calibration, establishing the pixel mapping relationship between the infrared image and the visible light image, and ensuring that the images collected by the dual cameras are aligned in the same space coordinate system.

3. The tunnel face water gushing hidden danger identification method based on image fusion according to claim 1, characterized in that, In step S2, the way of preprocessing the collected infrared images comprises: Using a self-adaptive Wiener filter algorithm for image denoising, the formula is: ； where g(x, y) is the pixel value of the de-noised infrared image; f(x, y) is the pixel value of the original infrared image; μ(x, y) is the mean value of the image in the 3x3 local window with the pixel (x, y) as the center, is the image variance in the window; is the noise variance, which is obtained by statistics of the dark area of the image.

4. The tunnel face water gushing hidden trouble identification method based on image fusion according to claim 1, characterized in that, In step S2, the pre-processing manner of the collected visible light image comprises: The Retinex algorithm is used for image enhancement, and the formula is: ； wherein, is the enhanced visible light image; is the original visible light image pixel value; is a Gaussian filter kernel; * denotes a convolution operation.

5. The tunnel face water gushing hidden trouble identification method based on image fusion according to claim 1, characterized in that, In step S2, it further comprises: The pre-processed infrared image and visible light image are subjected to double-image registration optimization: a plurality of feature points on the pre-processed infrared image and visible light image are extracted respectively; the feature points are matched through a matcher, and the random sample consensus algorithm is used to eliminate the error matching; the pixel mapping relationship is updated based on the correctly matched feature points, and the registration accuracy is ensured to be less than or equal to 0.5 pixels.

6. The tunnel face water gushing hidden trouble identification method based on image fusion according to claim 1, characterized in that, In step S5, the risk value is divided into risk levels and corresponding early warnings are made, which comprises: R≤0.3, safe, measures: normal construction, monitoring once every 30 minutes; 0.3 0.5 0.5 R>0.7, high risk, measures: immediately evacuate the construction personnel and equipment at the face, start the emergency plan, use full-face grouting to block the water gushing channel, and continuously monitor for 24 hours.

Images