Tunnel leakage water intelligent identification system based on mobile infrared thermal imaging

Abstract

The utility model discloses a tunnel leakage water intelligent identification system based on mobile infrared thermal imaging, and the system comprises a plurality of observation modules and an upper housing installation assembly, and the observation modules are rotatably disposed on the upper housing installation assembly. The observation module comprises a visible light camera, an infrared thermal imager, a front cover plate, a rear cover plate, a shell and rotating shaft limiting pieces, the front cover plate and the rear cover plate are installed on the front side and the rear side of the shell respectively, the two rotating shaft limiting pieces are arranged on the left side and the right side of the shell respectively, and the visible light camera is fixed to the inner side of the front cover plate; the thermal infrared imager is fixed on the shell; a rotating shaft limiting piece of the observation module is connected with the upper shell installation assembly through a locking screw and a self-locking stop pin. The tunnel water leakage detection device has the advantages that the angles of the visible light camera and the thermal infrared imager can be adjusted, so that images of the inner surface of a tunnel are acquired, the acquired image information is collected and processed, and the tunnel water leakage is identified.

Description

A smart identification system for tunnel water leakage based on mobile infrared thermal imaging Technical Field

[0001] This utility model relates to the technical field of tunnel leakage detection, and in particular to an intelligent identification system for tunnel water leakage based on mobile infrared thermal imaging. Background Technology

[0002] In the field of tunnel structural health monitoring, leakage detection is a crucial link in ensuring the safe operation of tunnels. Traditional detection methods mostly rely on manual inspections or fixed sensors, which suffer from low detection efficiency and limited coverage. In recent years, multimodal fusion detection technology based on visible light cameras and infrared thermal imagers has been gradually applied. It identifies surface defects such as cracks and peeling through visible light images, and combines them with infrared thermal imagers to capture areas with abnormal temperatures to determine potential leakage risks. However, in existing technologies, sensor modules are usually installed at a fixed angle, leading to the following problems: First, the internal structure of tunnels is complex and has large curvature variations, making it difficult for fixed viewing angles to adapt to the surface morphology of different sections, easily creating blind spots; Second, when multiple sensors collect data collaboratively, the position or angle of the equipment needs to be repeatedly adjusted to ensure spatial consistency of data, significantly increasing operation time; Third, although some adjustable devices can achieve angle changes, they rely on complex mechanical structures or manual intervention, resulting in low adjustment accuracy and difficulty in stable operation within the confined space of tunnels. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of the existing technology by providing a smart tunnel leakage identification system based on mobile infrared thermal imaging. This system comprises an observation module, an upper shell mounting assembly, a lower frame assembly, and a caster wheel assembly. Multiple observation modules are rotatably mounted on the upper shell mounting assembly. Each observation module includes a visible light camera and an infrared thermal imager. The angle of the observation module on the upper shell mounting assembly is adjustable, as are the angles of the visible light camera and the infrared thermal imager. This allows for the acquisition of images of the tunnel's inner surface, and the processing and aggregation of the acquired image information to identify tunnel leakage.

[0004] The objective of this utility model is achieved through the following technical solution:

[0005] A smart identification system for tunnel water leakage based on mobile infrared thermal imaging is disclosed. The system includes observation modules and an upper shell mounting assembly. Multiple observation modules are rotatably mounted on the upper shell mounting assembly. Each observation module includes a visible light camera, an infrared thermal imager, a front cover plate, a rear cover plate, a housing, and pivot limiting components. The front and rear cover plates are respectively mounted on the front and rear sides of the housing. Two pivot limiting components are provided, with each component located on the left and right sides of the housing. The visible light camera is fixed inside the front cover plate, and the infrared thermal imager is fixed to the housing. The pivot limiting components of the observation modules are locked to the upper shell mounting assembly. The locking screw is connected to the self-locking stop pin. The front and rear sides of the rotating shaft limiting component are respectively provided with a fixed post with internal threads and a sliding groove arranged in the front-rear direction. The left and right sides of the upper shell mounting assembly are provided with arc-shaped sliding grooves and through holes. Nuts are provided at the through holes. The locking screw passes through the arc-shaped sliding groove and is connected to the fixed post. The self-locking stop pin includes a sliding pin, a spring, a threaded rod and a handle. One end of the threaded rod is connected to the handle and the other end has a mounting hole. One end of the spring is fixed in the mounting hole and the other end is connected to the sliding pin. The threaded rod is threaded to the nut and passes through the through hole of the upper shell mounting assembly so that the sliding pin is embedded in the sliding groove.

[0006] The identification system also includes a lower frame assembly and a caster wheel assembly. The upper shell mounting assembly is installed on the top of the lower frame assembly, and the caster wheel assembly is installed at the four corners of the bottom of the lower frame assembly.

[0007] The upper shell mounting assembly has an arched structure.

[0008] The observation module is provided in four parts, with two of the observation modules located on the bottom sides of the upper shell mounting assembly, and the other two observation modules located on the top sides of the upper shell mounting assembly.

[0009] The arc of the arc-shaped groove is 42°.

[0010] The advantages of this invention are: the angles of the visible light camera and the infrared thermal imager are adjustable, enabling the acquisition of images of the tunnel's inner surface, and the collection and processing of the acquired image information to identify tunnel water leakage. Attached Figure Description

[0011] Figure 1 is a perspective view of the intelligent identification system for tunnel water leakage based on mobile infrared thermal imaging of this utility model.

[0012] Figure 2 is a front view of the intelligent identification system for tunnel water leakage based on mobile infrared thermal imaging of this utility model.

[0013] Figure 3 is a side view of the intelligent identification system for tunnel leakage based on mobile infrared thermal imaging of this utility model.

[0014] Figure 4 is a top view of the intelligent identification system for tunnel leakage based on mobile infrared thermal imaging of this utility model.

[0015] Figure 5 is a perspective view of the observation module of this utility model;

[0016] Figure 6 is a schematic diagram of the upper shell mounting assembly of this utility model;

[0017] Figure 7 is a cross-sectional view of the upper shell mounting assembly of this utility model;

[0018] Figure 8 is a structural schematic diagram of the self-locking stop pin of this utility model;

[0019] Figure 9 is an architecture diagram of the intelligent identification system for tunnel water leakage based on mobile infrared thermal imaging of this utility model.

[0020] Figure 10 is a schematic diagram of the collaborative design of vibration suppression and infrared detector based on MPC in this utility model;

[0021] Figure 11 is a schematic diagram of the infrared design architecture of this utility model;

[0022] Figure 12 is a data processing flowchart of this utility model;

[0023] Figure 13 is a diagram of the dual-modal feature association model of this utility model;

[0024] Figure 14 is a diagram showing the location detection of subway water seepage according to this utility model;

[0025] As shown in Figures 1-14, the markings in the figures represent:

[0026] Observation module 1, visible light camera 101, infrared thermal imager 102, front cover plate 103, rear cover plate 104, housing 105, pivot limiter 106, fixing post 1061, slide groove 1062, upper housing mounting assembly 2, arc-shaped slide groove 201, through hole 202, nut 203, lower frame assembly 3, universal wheel assembly 4, locking screw 5, self-locking stop pin 6, sliding pin 601, threaded rod 602, handle 603;

[0027] The curvature α of the arc-shaped groove. Detailed Implementation

[0028] The features and other related features of this utility model will be further described in detail below with reference to the accompanying drawings and embodiments, so as to facilitate the understanding of those skilled in the art:

[0029] Example: As shown in Figures 1-4, this example relates to an intelligent identification system for tunnel water leakage based on mobile infrared thermal imaging. The identification system mainly includes an observation module 1, an upper shell mounting assembly 2, a lower frame assembly 3, and a caster wheel assembly 4. Multiple observation modules 1 are provided and rotatably mounted on the upper shell mounting assembly 2. In this example, the upper shell mounting assembly 2 has an arched structure. There are four observation modules 1, two of which are located on the bottom sides of the upper shell mounting assembly 2, and the other two are located on the top sides of the upper shell mounting assembly 2, so as to observe various positions of the tunnel. The upper shell mounting assembly 2 is mounted on the top of the lower frame assembly 3, and the caster wheel assembly 4 is mounted at the four corners of the bottom of the lower frame assembly 3.

[0030] As shown in Figures 5-8, the observation module 1 includes a visible light camera 101, an infrared thermal imager 102, a front cover plate 103, a rear cover plate 104, a housing 105, and a pivot limiting component 106. The front cover plate 103, the rear cover plate 104, and the housing 105 are all made of aluminum alloy. The front cover plate 103 and the rear cover plate 104 are respectively installed on the front and rear sides of the housing 105. The front cover plate 103 is detachable. There are two pivot limiting components 106, and the two pivot limiting components 106 are respectively located on the left and right sides of the housing 105. The visible light camera 101 is fixed inside the front cover plate 103, and the infrared thermal imager 102 is fixed on the housing 105. The lenses of the visible light camera 101 and the infrared thermal imager 102 are both aimed at the outside of the observation module 1 to collect images of the tunnel. The parameters of the visible light camera 101 and the infrared thermal imager 102 are shown in Table 1 below. The rotating shaft limiting component 106 of the observation module 1 is connected to the upper shell mounting assembly 2 by a locking screw 5 and a self-locking stop pin 6. The rotating shaft limiting component 106 has a fixed post 1061 with internal threads on its front side and a sliding groove 1062 arranged in the front-rear direction on its rear side. The left and right sides of the upper shell mounting assembly 2 are provided with arc-shaped sliding grooves 201 and through holes 202. Nuts 203 are provided at the through holes 202. The locking screw 5 passes through the arc-shaped sliding groove 201 and is connected to the fixed post 1061. The device can move along the arc of the arc-shaped slide 201 (at this time, the head of the locking screw 5 is not locked and fixed with the upper shell mounting component 2, and the self-locking stop pin 6 is required to limit the observation module 1) to adjust the angle of the observation module 1. After the position of the locking screw 5 is adjusted to the correct position, tighten the locking screw 5 so that the head of the locking screw 5 is locked and fixed with the upper shell mounting component 2. In this embodiment, the arc α of the arc-shaped slide 201 is 42°, that is, the angle adjustment range of the observation module 1 is 42°.The self-locking stop pin 6 (made of stainless steel) includes a sliding pin 601, a spring, a threaded rod 602, and a handle 603. One end of the threaded rod 602 is connected to the handle 603, and the other end has a mounting hole. The mounting hole is axially arranged along the threaded rod 602. One end of the spring is fixed in the mounting hole, and the other end is connected to the sliding pin 601. In the normal state, the spring is entirely within the mounting hole, and the sliding pin 601 is partially located within the mounting hole. When the spring is compressed, the sliding pin 601 may be partially or entirely located within the mounting hole. The threaded rod 602 is threadedly connected to the nut 203 and passes through the through hole 202 of the upper housing mounting assembly 2, so that the sliding pin 601 is embedded in the sliding groove 1062. Specifically, the working principle of the self-locking stop pin 6 is as follows: The threaded rod 602 is installed on the nut 203. As the threaded rod 602 is continuously tightened, the sliding pin 601 approaches the rotating shaft limiting member 10. 6. If the sliding pin 601 is exactly aligned with the groove 1062 of the rotating shaft limiting member 106, the sliding pin 601 is directly embedded in the groove 1062. If the sliding pin 601 is not exactly aligned with the groove 1062 of the rotating shaft limiting member 106, the sliding pin 601 will squeeze the rotating shaft limiting member 106, and the spring will be compressed. By adjusting the position of the rotating shaft limiting member 106, the position of the groove 1062 of the rotating shaft limiting member 106 is aligned with the position of the sliding pin 601. The spring rebounds, and the sliding pin 601 is embedded in the groove 1062. This achieves the limitation (non-locking) of the sliding pin 601 (self-locking stop pin 6) on the rotating shaft limiting member 106 (observation module 1) to ensure the angle adjustment of the observation module 1. The sliding pin 601 can move relative to the groove 1062 (actually, the groove 1062 moves) so that the fixed post 1061 is aligned with the arc-shaped groove 201.

[0031] Table 1: Parameters of Infrared Thermal Imager and Visible Light Camera;

[0032]

[0033] In addition, the identification system also includes a router module, a control and display module, and a data processing module. The router module establishes a local area network (LAN) to enable communication between devices. It adopts a LAN system architecture design based on multi-protocol adaptive communication to build a highly reliable, low-latency heterogeneous device interconnection platform. The control and display module allows real-time image viewing and device control via a tablet computer. The hardware interaction terminal is an industrial-grade ruggedized tablet computer equipped with a 10.1-inch IPS fully laminated touchscreen. It achieves bidirectional real-time interaction between control command transmission and infrared / visible video streams through dual-band Wi-Fi 6 / 6E (supporting 2×2 MU-MIMO) and gigabit Ethernet adaptive access to the core routing module. The data processing module analyzes the acquired images, identifies leak points, and calculates the leakage area.

[0034] As shown in Figure 9, after data acquisition, it needs to be processed and analyzed. First, dynamic scanning images are deblurred to optimize image clarity for subsequent detection. Then, visible light and infrared images are stitched together, followed by infrared temperature calibration to complete preliminary image processing. Subsequently, multispectral image registration is performed to fuse infrared and visible light images, facilitating the final identification of leakage areas.

[0035] To address equipment vibration and synchronization issues among multiple devices, a vibration suppression and infrared detector co-design based on MPC, as well as a multi-modal sensor fusion and standardized interface design, were adopted.

[0036] As shown in Figure 10, the vibration suppression and infrared detector are designed in conjunction with MPC:

[0037] To address the image blurring issue in infrared thermal imaging systems during mobile scenarios, an active image stabilization control system with an FPGA+ARM dual-core architecture is adopted, which combines the dynamic characteristics of the infrared detector to achieve vibration compensation.

[0038] Hardware architecture: Utilizing Xilinx Zynq-7000 series chips, the FPGA handles infrared image preprocessing (non-uniformity correction, multi-frame fusion) and parallel acceleration of the MPC algorithm, while the ARM runs the adaptive parameter adjustment algorithm. The infrared detector is an uncooled vanadium oxide (VOx) microbolometer with a pixel size of 17μm and a NETD ≤ 50mK, achieving 14-bit quantization output through a 24-bit Σ-Δ ADC.

[0039] Vibration suppression strategy: A six-degree-of-freedom vibration spectrum model is constructed by integrating a MEMS accelerometer (range ±16g) and a fiber optic gyroscope (zero-bias stability 0.1° / h). The FPGA fuses sensor data through Kalman filtering, predicts the resonant frequency, and drives a piezoelectric ceramic actuator (displacement accuracy 0.1μm), reducing the optical axis offset by more than 85%.

[0040] Thermal management optimization: An NTC temperature sensing module (±0.3℃ accuracy) is integrated into the detector package, and a high-efficiency custom heat sink is used to achieve closed-loop control of the operating temperature, avoiding the impact of dark current drift on the ADC sampling accuracy.

[0041] As shown in Figure 11 and Table 2 below, the design of multimodal sensor fusion and standardized interface is as follows:

[0042] To achieve synchronous acquisition of signals from multiple devices: 12 sensor data are synchronized through the global clock network (jitter <200ps) technology of FPGA. This method has a huge improvement compared with the traditional discrete ADC + independent clock source solution. The infrared detector adopts the ROIC (readout integrated circuit) with built-in TIA amplifier, and the signal-to-noise ratio is improved to 72dB.

[0043] Table 2: Comparison of FPGA global clock network and vertical ADC-independent clock source;

[0044]

[0045] The entire data processing flow is shown in Figure 12, and mainly includes the following aspects:

[0046] 1. Image Acquisition and Preprocessing

[0047] (1) Dynamic scanning image deblurring

[0048] During high-speed movement of the vehicle, the captured images are inevitably affected by motion blur. To address this issue, the instantaneous velocity field of the camera is first obtained by integrating the acceleration data over time. The integral process considers the influence of the rotational component on the trajectory, and its mathematical expression is:

[0049] Formula 1;

[0050] In the formula, The initial speed of the car and , It is determined by the rotation matrix (obtained by integrating the gyroscope data) and the lever vector r.

[0051] After obtaining accurate motion estimation, an improved Wiener filtering algorithm is used to restore the image. This method introduces adaptive regularization constraints on the basis of traditional frequency domain filtering, effectively balancing the trade-off between image deblurring and noise amplification. The frequency domain response function of the filter is:

[0052] Formula 2;

[0053] In the formula, This represents the frequency domain representation (estimated value) of the restored image. The frequency domain transfer function representing the fuzzy kernel (point spread function PSF). Then it means The complex conjugate, and These represent the power spectral densities of the signal and noise, respectively. Experiments show that this method can improve image sharpness metrics (such as MTF50) by more than 40% at a speed of 30 km / h.

[0054] (2) Visible light image stitching

[0055] To achieve continuous detection of the tunnel lining surface, accurate stitching of visible light images acquired from multiple perspectives is required. This system employs an improved SIFT (Scale Invariant Feature Transform) algorithm for feature matching. This algorithm enhances the stability of feature points by constructing a nonlinear scale space.

[0056] Formula 3;

[0057] These are the pixel values ​​of the original image (grayscale image). It is a two-dimensional Gaussian function, defined as:

[0058] Equation 4;

[0059] It is the standard deviation of the Gaussian kernel, which controls the smoothness (scale parameter).

[0060] Is the image at scale? The Gaussian smoothing result is used to construct the scale space. This is a convolution operation, representing a Gaussian kernel and a locally weighted average of the image. This formula can generate multi-scale image pyramids using Gaussian filtering with different σ values, for detecting features of different sizes.

[0061] , , It is the second-order partial derivative of the image. middle Second derivative of direction (marginal curvature). Mixed partial derivatives in the x and y directions (corner response).

[0062] After obtaining reliable feature matching, determining the optimal suture line is a crucial issue. This system constructs a multi-objective energy function that incorporates geometric consistency, photometric consistency, and feature similarity.

[0063] Formula 5;

[0064] a. Geometric constraints:

[0065] Mathematical form: I;

[0066] It is the image gradient (first derivative), which reflects changes in edges and textures. yes The norm (usually p=1 or 2) is used to enhance robustness to noise.

[0067] Physical meaning: Forcing two images ( and Alignment is achieved structurally, such as consistent edge positioning.

[0068] b. Photometric constraints:

[0069] Mathematical form: ;

[0070] It is a weighted least squares method, where the weights are usually related to the local variance of the image.

[0071] Physical meaning: It ensures that the pixel intensity of two images is similar, and is suitable for scenes with small changes in lighting.

[0072] c. Feature constraints:

[0073] Mathematical form: ;

[0074] d1 and d2 are feature descriptors.

[0075] Physical significance: It improves alignment accuracy through high-level feature matching, and is especially suitable for large displacements or non-rigid deformations.

[0076] By solving for the minimum value of the energy function using dynamic programming, the optimal stitching line can be obtained in O(nlogn) time complexity.

[0077] (3) Infrared temperature calibration

[0078] The radiation data acquired by the infrared thermal imager needs to be accurately converted into temperature values. This system uses a dual blackbody reference source for real-time calibration and establishes a complete temperature inversion model based on Planck's radiation law:

[0079] Formula 6;

[0080] In the formula, It is the sensor gain coefficient (calibration parameter). This represents the sensor bias (dark current or background noise). c1 and c2 are both Planck's radiation constants.

[0081] 3.7418× ;

[0082] 1.4388× ;

[0083] (h is Planck's constant, k is Boltzmann's constant), λ is the sensor's operating wavelength (unit: meters).

[0084] N(T) is the first-order partial derivative with respect to A:

[0085] Formula 7;

[0086] This model fully considers the nonlinear response characteristics of the detector and the influence of environmental radiation. Solving the above equations using the Levenberg-Marquardt optimization algorithm achieves a temperature measurement accuracy of ±0.5℃. Furthermore, to eliminate the detector's inherent non-uniformity noise, the system also implements a joint correction algorithm based on time-domain high-pass filtering and spatial-domain median filtering.

[0087] To achieve accurate fusion of infrared and visible light images, this system proposes an improved SURF-PIIFD (Accelerated Robust Feature-Phase Consistent Infrared Feature Description) registration algorithm. This algorithm first extracts scale-invariant feature points from the dual-modal image, and then constructs a composite descriptor containing gradient direction statistics and phase consistency information.

[0088] Formula 8;

[0089] Input variables It is a two-dimensional function. The set of coordinates of a local region within the domain. It is a function exist The first-order partial derivative (gradient component) in the direction. The calculation method is as follows:

[0090] Equation 9;

[0091] This is the gradient summation term, reflecting the overall trend of the gradient within region P (such as the directional bias of the edges). If the result is 0, it indicates that the region is symmetrical (positive and negative gradients cancel each other out); if it is non-zero, there may be unidirectional edges. This is the gradient sum of squares term, used to describe the total energy of the gradient. Similar to the previous one, but it is necessary to ensure that the result is non-negative.

[0092] Regarding registration error control, this system establishes a complete error propagation model, analyzing the impact of feature point positioning errors and descriptor matching errors on the final registration accuracy:

[0093] Formula 10;

[0094] Jacobian matrix, representing Partial derivative with respect to input x: = , It is the covariance term of additional noise or unmodeled dynamics.

[0095] Experimental results show that the algorithm can achieve a registration accuracy of better than 0.5 pixels in a tunnel environment, which fully meets the requirements of defect detection.

[0096] By fusing the temperature distribution presented in infrared images with the gradient information extracted by the local temperature gradient operator, accurate identification of target regions is achieved, thus constructing a dual-modal feature association model as shown in Figure 13. This model fully integrates two complementary feature data and, through joint analysis, effectively improves the accuracy of target detection and description.

[0097] Leakage areas typically appear as temperature anomalies in infrared images. This system proposes a Local Temperature Gradient (LTG) operator to quantify this anomaly:

[0098] Formula 11;

[0099] The maximum temperature gradient intensity at pixel (x,y) is used to quantify the degree of local temperature anomalies. It is the size of the multi-scale neighborhood window (unit: pixels), used to detect leakage areas of different sizes (small, medium, and large scales). Indicates in The temperature gradient vector calculated within the neighborhood (approximated by the central difference method). It calculates the magnitude of the gradient vector, taking into account temperature changes in both the horizontal and vertical directions.

[0100] This operator, through multi-scale gradient calculation, can effectively capture leakage areas of different sizes. Combined with statistical methods, an adaptive threshold is set to determine anomalies.

[0101] Equation 12;

[0102] In the formula, Q3 is the third quartile, IQR is the interquartile range, which measures the dispersion of the data and can eliminate the interference of outliers. k is the sensitivity coefficient. The larger the value, the stricter the judgment standard. It is usually taken as a value between 1.5 and 3.0.

[0103] In visible light images, leakage areas are often accompanied by specific texture features. This system employs an improved CLAHE (Contrast-Limited Adaptive Histogram Equalization) algorithm for texture enhancement:

[0104] Equation 13;

[0105] The original image is in position Pixel value at that location, The maximum pixel value in the image, used for normalization operations. It is the average pixel intensity of the entire image, reflecting the overall brightness level. Therefore It captures local brightness changes by measuring the average intensity of pixels within a neighborhood window centered on the center. It represents the dispersion of image pixel intensity, measuring the contrast range. This is an adjustment coefficient that controls the sensitivity of the gamma value to local brightness differences (the larger the value, the more aggressive the enhancement), typically between 0.1 and 0.3. This algorithm enhances contrast through local adaptive gamma correction while avoiding noise amplification caused by over-enhancement.

[0106] To comprehensively utilize temperature and texture information, this system constructs a multi-feature fusion classifier based on random forest. Feature selection is optimized by calculating a feature importance index.

[0107] Equation 14;

[0108] In the formula, It is a feature importance indicator: measuring the importance of the first feature. The degree to which a feature (temperature) contributes to classification accuracy. The higher the value, the more important the feature. This represents the total number of trees in the random forest, indicating the stability of the computation; the more trees, the better. The more robust the estimate (as shown in Table 1) =500). The measurement indicates the first The training sample set for each tree is typically a subset of the Bootstrap sample (random sampling with replacement). It is an indicator function used to benchmark accuracy when using all features on a sample. Is the classification correct? For individual assessment features The predictive ability of fused features is shown in Table 3 below. The classification performance of fused features is significantly better than that of single features.

[0109] Table 3: Classifier Performance Comparison Table;

[0110]

[0111] The contour lines of the seepage area are extracted using the Marching Squares algorithm shown in Equation 15:

[0112] Formula 15;

[0113] In the formula, It is the gradient vector, representing the scalar field. exist and The rate of change of direction, pointing in the direction of the fastest change. , yes The partial derivative with respect to the y-direction is approximated by the central difference method, with an accuracy of [missing value]. . This represents the value of the scalar field at grid point (i,j). The grid spacing determines the discretization accuracy. 2 This represents the actual distance between adjacent grid points. The algorithm generates a sub-pixel-level smooth vector profile by calculating the gradient of the scalar field in each direction (using the central difference method to obtain the directional partial derivatives; the gradient vector reflects the rate and direction of temperature change, and the grid spacing determines the discretization accuracy), thereby accurately capturing the irregular boundaries of the leakage area.

[0114] Next, to accurately calculate the area, considering the effects of tunnel curvature and camera perspective distortion, ground sampling distance (GSD) correction (as shown in Equation 16) and compensation for the angle between the surface normal vector and the line of sight were introduced. Combined with surface gradient information obtained from stereo matching, a geometric correction model was constructed to control the measurement error within 5%.

[0115] Formula 16;

[0116] Finally, using the information entropy weighting method of Equation 17, the thermal radiation characteristics of the infrared image are effectively fused with the high-resolution spatial details of the visible light image, achieving accurate labeling and quantitative display of the location, area, and severity of the leakage area, as shown in Figure 14:

[0117] Formula 17.

[0118] In this embodiment, a mobile infrared water leakage application test was conducted in a subway tunnel. The entire test selected 6 sections of 3 subway lines, with a total of 11,159 effective tunnel scan rings and a total scan length of more than 10 kilometers.

[0119] As shown in Table 4, the test employed a manual comparison method, which involved using a mobile infrared sweeping system and combining manual inspection with handheld infrared manual verification to verify the system's accuracy. In the six test sections with mobile scanning, manual inspection identified 98 leaks, while the mobile infrared system detected 141 leaks. Both methods detected 93 leaks, resulting in 5 false positives from the mobile infrared system. Handheld infrared was also used to manually verify 151 leaks. Of the 141 leaks detected by the mobile infrared system, 133 were confirmed to be present, while the remaining 11 were verified on-site to be leak-free. Particularly noteworthy is the mobile infrared system's accuracy in detecting roof leaks, accurately capturing 20 roof leaks, 5 of which were missed by manual inspection, demonstrating its significant advantage in detecting this type of leak.

[0120] Table 4: Detection Rate Statistics;

[0121]

[0122] In terms of detection accuracy, mobile infrared detection performed best, reaching 92.3%, significantly higher than the 68.5% of manual inspection. This advantage is mainly due to the multimodal data fusion technology and intelligent algorithm analysis employed by the system, which can more comprehensively capture leakage characteristics. The low accuracy of manual inspection is mainly limited by the subjectivity of human eye recognition and fatigue factors.

[0123] In terms of detection efficiency, the mobile infrared detection system can complete the detection of a 1-kilometer tunnel in just 20 minutes, far superior to the 3 hours / kilometer required for manual inspection. This high efficiency is due to the system's automated detection process and parallel processing capabilities, which greatly reduces the time cost required by traditional detection methods. Manual inspection is inefficient because it requires checking and recording each section, and is limited by human walking speed and judgment time.

[0124] As shown in Table 5 below, the mobile infrared detection system can detect leaks as small as 5 cm² in terms of the smallest identifiable leak area. This advantage stems from the high-resolution imaging and sub-pixel-level analysis algorithms employed by the system. Manual inspections are limited by the resolution of the naked eye, making it difficult to detect even minute leaks.

[0125] Table 5: Comparison of Infrared Motion Detection and Manual Detection Indicators;

[0126]

[0127] There are four main reasons for false positives from moving infrared sensors:

[0128] 1) Lighting interference: Although large light interferences can be eliminated, small areas of lighting can still interfere with judgment.

[0129] 2) Impact of fire hydrant pipes: Due to varying pipe temperatures, some pipes may be falsely reported as leak points.

[0130] 3) Special shapes: In rectangular segments, there is a possibility of misjudgment at the included corners of the rectangle. Due to the cavity effect, the temperature at the included corners of the rectangle will be lower during infrared detection, leading to misjudgment.

[0131] 4) Obstruction: When workers or other objects block the view inside the tunnel, it can cause omissions.

[0132] Taking advantage of the significant temperature difference between seepage water and the tunnel structure, infrared thermal imaging is used to assist in the identification of seepage water. This can effectively solve the problem of misjudgment of seepage water by manual and image-based methods, and greatly improve the accuracy of identifying tunnel seepage defects.

[0133] The beneficial technical effects of this embodiment are as follows: the angles of the visible light camera and the infrared thermal imager are adjustable, enabling the acquisition of images of the inner surface of the tunnel, and the acquisition of image information is collected and processed to identify tunnel water leakage.

[0134] Although the above embodiments have described the concept and embodiments of the present invention in detail with reference to the accompanying drawings, those skilled in the art will recognize that various improvements and modifications can still be made to the present invention without departing from the scope of the claims, and therefore will not be elaborated here.

Claims

1. A tunnel water leakage intelligent identification system based on mobile infrared thermal imaging, characterized in that The identification system includes an observation module and an upper shell mounting assembly. Multiple observation modules are rotatably mounted on the upper shell mounting assembly. Each observation module includes a visible light camera, an infrared thermal imager, a front cover plate, a rear cover plate, a housing, and a pivot limiting component. The front and rear cover plates are respectively mounted on the front and rear sides of the housing. Two pivot limiting components are provided, with each component located on the left and right sides of the housing. The visible light camera is fixed inside the front cover plate, and the infrared thermal imager is fixed to the housing. The pivot limiting component of the observation module is connected to the upper shell mounting assembly via locking screws and self-locking stop pins. The pivot limiting component has a fixed post with internal threads on its front side and a sliding groove arranged in the front-rear direction on its rear side. The left and right sides of the upper shell mounting assembly are provided with arc-shaped sliding grooves and through holes. Nuts are provided at the through holes. The locking screw passes through the arc-shaped sliding grooves and is connected to the fixed post. The self-locking stop pin includes a sliding pin, a spring, a threaded rod and a handle. One end of the threaded rod is connected to the handle and the other end has a mounting hole. One end of the spring is fixed in the mounting hole and the other end is connected to the sliding pin. The threaded rod is threaded to the nut and passes through the through hole of the upper shell mounting assembly so that the sliding pin is embedded in the sliding groove.

2. The intelligent identification system for tunnel water leakage based on mobile infrared thermal imaging as described in claim 1, characterized in that... The identification system also includes a lower frame assembly and a caster wheel assembly. The upper shell mounting assembly is installed on the top of the lower frame assembly, and the caster wheel assembly is installed at the four corners of the bottom of the lower frame assembly.

3. The tunnel water leakage intelligent identification system based on mobile infrared thermal imaging of claim 1, wherein The upper shell mounting assembly has an arched structure.

4. The intelligent identification system for tunnel water leakage based on mobile infrared thermal imaging as described in claim 3, characterized in that... The observation module is provided in four parts, with two of the observation modules located on the bottom sides of the upper shell mounting assembly, and the other two observation modules located on the top sides of the upper shell mounting assembly.

5. The tunnel water leakage intelligent identification system based on mobile infrared thermal imaging of claim 1, wherein The arc of the arc-shaped groove is 42°.

Images