Nuclear magnetic resonance and image intelligent bimodal unfrozen water dynamic identification method and system

Abstract

The invention provides a nuclear magnetic resonance and image intelligent bimodal unfrozen water dynamic identification method and system, and belongs to the technical field of frozen soil engineering and geophysical exploration, and the method comprises the steps: sealing a frozen soil sample in a nuclear magnetic sample tube, and achieving the gradient cooling of the sample in a preset low temperature range through temperature control; scanning by using a nuclear magnetic resonance technology to obtain transverse relaxation time T2 spectrums and nuclear magnetic images of the frozen soil sample at different temperature nodes; after image preprocessing is carried out on the nuclear magnetic image, a temperature-signal intensity curve in the nuclear magnetic image is automatically extracted by utilizing an image recognition technology; carrying out integral calculation on a liquid water signal in the T2 spectrum, and dynamically calculating the content of the unfrozen water at each temperature based on a nuclear magnetic signal intensity formula; according to the method, a pore-fracture co-evolution matrix is constructed, accurate analysis of unfrozen water dynamic and pore evolution in the freezing and thawing process is realized, the method is suitable for hydrothermal coupling characteristic analysis of Qinghai-Tibet soil in a complex freezing and thawing environment, and reliable technical support is provided for stability evaluation of frozen soil engineering.

Description

Technical Field

[0001] The present invention relates to the technical fields of frozen soil engineering and geophysical exploration, and particularly relates to a nuclear magnetic resonance and image intelligence dual-modal unfrozen water dynamic identification method and system. Background Art

[0002] As the largest permafrost distribution area in the mid-latitudes of the world, the stability of permafrost engineering in the Qinghai-Tibet Plateau directly affects the safe operation of major infrastructure such as railways, highways, and energy pipe networks. In the permafrost soil with high salinity and multiple fractures unique to this region, during the freeze-thaw cycle, the dynamic change of the unfrozen water content significantly affects the formation mechanism of frost heave deformation and thaw settlement disasters through the latent heat of phase change and seepage pressure. Traditional detection methods have significant limitations: differential scanning calorimetry (DSC) needs to destroy the sample structure and cannot achieve in-situ monitoring; time domain reflectometry (TDR) is limited by the single-point measurement mode and is difficult to capture the three-dimensional moisture distribution characteristics of the soil; while the resistivity method has a detection error rate as high as 15%-20% due to the interference of high-salt ions. Although nuclear magnetic resonance (NMR) technology has gradually been applied to this field with the advantages of non-destructive detection and high resolution, there are still key bottlenecks: the temperature-signal relationship curve in the nuclear magnetic resonance image depends on manual extraction, with low data processing efficiency, significant subjective errors, and deteriorated signal-to-noise ratio; existing methods fail to integrate T2 spectrum analysis and image feature recognition technology and cannot quantify the co-evolution law of the pore-fracture system during the freeze-thaw process. In addition, the current image processing technology has insufficient analytical accuracy for complex curves and coordinate scales, severely restricting the construction accuracy of the temperature-unfrozen water correlation model. Summary of the Invention

[0003] The purpose of the present invention is to provide a dynamic identification and detection method for unfrozen water in multi-fractured permafrost, so as to solve the core problems existing in the monitoring of unfrozen water in multi-fractured permafrost, such as low data analysis efficiency, lack of quantification of pore evolution, and insufficient adaptability to complex environments.

[0004] To achieve the above purpose, the present invention proposes a dynamic and precise identification method for unfrozen water driven by nuclear magnetic resonance (LF-NMR) and image intelligence dual-modal, including the following steps: S1: Seal the frozen soil sample in a nuclear magnetic resonance sample tube, and realize gradient cooling of the sample within a preset low temperature range through temperature control; S2: Use nuclear magnetic resonance technology to scan and obtain the transverse relaxation time T2 spectrum and nuclear magnetic resonance image of the frozen soil sample at different temperature nodes; S3: After preprocessing the nuclear magnetic resonance image, use image recognition technology to automatically extract the temperature-signal intensity curve in the nuclear magnetic resonance image; S4: Integrate the liquid water signal in the T2 spectrum, and dynamically calculate the content of unfrozen water at each temperature based on the following nuclear magnetic signal intensity formula; , In the formula, is the nuclear magnetic resonance signal intensity at temperature T, which is obtained by integrating the peak area of the T2 spectrum; is the nuclear magnetic resonance signal intensity corresponding to the initial liquid water; w0 is the initial total water content.

[0005] Furthermore, it also includes steps S5 and S6: S5: Invert the pore radius using the transverse relaxation time T2 spectrum, extract the pore distribution characteristics, calculate the proportion of pore volume in each pore size interval, and establish a linear relationship between the gray value (0 - 255) of the nuclear magnetic resonance image and the pore radius; S6: Construct a pore - fracture co - evolution matrix, and define the evolution matrix to describe the coupling relationship between the pore structure, unfrozen water, and temperature during the freeze - thaw process, and achieve precise analysis of the dynamics of unfrozen water and pore evolution during the freeze - thaw process: ; Among them, is the change in radius of micro - pores with a pore diameter less than 0.2μm due to the growth stress of ice crystals during the freeze - thaw process, which quantifies the mesoscopic frost heaving characteristics; represents the change in width of fractures with a width greater than 1μm due to ice - wedge action during the freeze - thaw process, which characterizes the macroscopic expansion rate; ∇w u is the unfrozen water gradient, and ∇T is the temperature gradient.

[0006] Furthermore, step S5 is specifically as follows: S5.1: Invert the pore radius using the linear relationship between the transverse relaxation time T2 spectrum and the pore radius R: ; Among them, the surface relaxation rate takes the empirical value , and converts the T2 distribution curve into a pore radius distribution; S5.2: Pore classification and quantification: According to the pore size, the pores are divided into ultra - micropores ( ), micropores ( ), small pores ( ), medium pores ( ), and large pores ( ). By integrating the peak area of the T2 spectrum, calculate the proportion of pore volume in each pore size interval: ; Among them, is the pore volume in a certain pore size interval, is the total pore volume; When the proportion of micropore volume exceeds 60%, the change in micro - pore radius The main evolutionary process reflects the mesoscopic frost heaving mechanism; when the proportion of macropore volume exceeds 10%, the change in fracture width becomes the main variable, corresponding to the macroscopic failure process; S5.3: By measuring the pore radius distribution of the soil sample through mercury intrusion porosimetry, extracting the gray values (0 - 255) of the pore regions in the nuclear magnetic resonance (NMR) image of the same soil sample, establishing the mapping relationship between the gray value of the NMR image and the pore radius, and thus constructing a linear relationship (linear regression model) between the gray value of the NMR image and the pore radius.

[0007] ; Calibrate the parameters through mercury intrusion porosimetry (MIP) and to achieve the quantitative conversion from the gray value of the NMR image to the pore radius, and thus construct a pore - fracture co - evolutionary matrix.

[0008] Furthermore, in step S1, the preset low - temperature range is from - 30°C to 20°C; In step S2, a 0.5T permanent magnet low - field nuclear magnetic resonance spectrometer is used to perform NMR scans on the frozen samples, triggering an NMR scan every 5°C, so as to obtain the transverse relaxation time T2 spectra and NMR images of the frozen soil samples at a total of 9 temperature nodes: 20°C, 10°C, 0°C, - 2°C, - 5°C, - 7°C, - 10°C, - 15°C, - 30°C, and comprehensively capture the unfrozen water phase change characteristics during the entire freeze - thaw process.

[0009] During the NMR data acquisition process, the CPMG sequence is used, the echo time TE = 0.2ms, and the number of accumulations is 32 times. The system can ensure that the signal - to - noise ratio ≥ 90dB in the temperature range from - 30°C to 20°C.

[0010] Furthermore, in step S3, the process of the image pre - processing includes: Grayscale processing: Convert the NMR image from the RGB color space to a grayscale image; Binarization processing: Use the adaptive threshold method with the threshold set to 120 to convert the grayscale image to a binary image; Median filtering processing: Use a 3×3 median filter to remove the salt - and - pepper noise in the image to improve the image quality.

[0011] Furthermore, in step S3, the image recognition technology includes: Use the improved Hough transform to identify the temperature axis and the signal intensity axis, and combine the Canny edge detection algorithm and the least - squares fitting to achieve the correction of the coordinate axis tilt (error ≤ 0.5°) with a false detection rate < 5%; Integrate the Tesseract OCR engine with regular expressions (supporting scientific notation, fractions, and exponential formats) to extract scale values, parse them, and automatically calibrate scale offset errors, accurately converting the scale information in the image into actual physical quantity values, with a scale recognition rate ≥ 95%; the OCR parsing supports scientific notation and fraction formats, and matches numerical values through regular expressions; Based on HSV color space segmentation (red curve threshold: (H = 0 - 10, S ≥ 120, V ≥ 70)) and morphological closing operation, realize pixel-level extraction of the temperature-signal curve.

[0012] Adopt a multi-threaded parallel architecture, with a single-frame image processing time < 0.5 seconds, support continuous monitoring at 10 frames per second, and meet the real-time requirements of the project.

[0013] Furthermore, the parameter settings of the improved Hough transform are: ρ = 1, θ = π / 180, threshold = 100, minimum line segment length = 100 pixels; The specific method for realizing pixel-level extraction of the temperature-signal curve based on HSV color space segmentation and morphological closing operation is as follows: First, perform color segmentation on the red curve through the HSV color space, and set the HSV threshold range as lower = [0, 120, 70], upper = [10, 255, 255]; Second, use the OpenCV library to extract the curve contour and filter out noise points with an area less than 100 pixels; Finally, convert the pixel coordinates into actual numerical values, and use linear interpolation to calculate the temperature and signal intensity values corresponding to the curve points. This extraction method effectively establishes the relationship between temperature and signal intensity, providing a basis for subsequent data analysis and processing.

[0014] Furthermore, the method of the present invention has an implementation closed-loop property, specifically manifested as: Data flow: Starting from nuclear magnetic signal acquisition, through image processing, feature extraction, and model calculation, finally provide a basis for engineering decisions, and the engineering practice feedback further optimizes the entire technical process.

[0015] Verification closed-loop: First, perform parameter calibration in the laboratory, then go to the actual field environment for verification, perform parameter iteration according to the verification results, and finally realize large-scale application, forming a complete technical implementation closed-loop, continuously improving the reliability and practicality of the technology.

[0016] The present invention also proposes a nuclear magnetic resonance and image intelligent dual-modal unfrozen water dynamic recognition system for implementing the above recognition method. The system includes: Multi-modal data acquisition subsystem: including a nuclear magnetic acquisition module and a temperature control module, used to obtain the transverse relaxation time T2 spectrum and nuclear magnetic images of frozen soil samples at different temperature nodes; Intelligent Image Processing Unit: Equipped with OpenCV and Tesseract OCR libraries, it is used for detecting coordinate axes, extracting curve fitting, scale analysis, and data conversion of nuclear magnetic resonance images; Data Analysis Module: Based on the Python platform, it has built-in unfrozen water content calculation algorithms (dynamic unfrozen water content calculation model) and pore evolution models (pore-fracture co-evolution model).

[0017] Furthermore, the data analysis module includes a special analysis model for Qinghai-Tibet soil and a freeze-thaw coupling analysis module: The special analysis model for Qinghai-Tibet soil includes a dynamic unfrozen water content calculation model and a pore-fracture co-evolution model.

[0018] Among them, the pore-fracture co-evolution model introduces pore characteristic factors (micropores <0.2μm, fractures > 1μm): ; In the formula, is the volume of micropores with a pore diameter less than 0.2μm, which is calculated by integrating the peak area in the corresponding relaxation time interval of the T2 spectrum, reflecting the mesoscopic frost heaving mechanism. is the volume of fractures with a width greater than 1μm, which is determined by mapping the gray scale of nuclear magnetic resonance images and the pore radius calibrated by mercury intrusion porosimetry, characterizing the macroscopic fracture expansion.

[0019] is calibrated with the surface relaxation rate to establish a multi-scale evolution equation set: ; Among them, the evolution coefficients and are obtained by measuring the mercury intrusion radii of micropores and fractures and by mercury intrusion porosimetry and fitting.

[0020] Within the temperature range ΔT, the gradients and ∇T are approximately constant. Integrating with respect to time gives the state change: ; After normalizing the time variable ( = 1), the pore-fracture co-evolution model has the following relationship with the pore-fracture co-evolution matrix: ; For the calibration process, first, data collection is carried out. The pore radius is measured by mercury intrusion porosimetry (MIP), and the micropore volume and fracture volume are calculated using nuclear magnetic resonance to measure the T2 spectrum, and ; Then, parameter fitting is carried out. The data is substituted into the evolution equation, and k1 and k2 are adjusted so that ; Finally, verification is carried out. In the freeze-thaw experiment, ensure that the crack expansion amount calculated by the model and the scanning deviation ≤ 0.05μm.

[0021] The freeze-thaw coupling analysis module is used to construct a pore-fracture co-evolution matrix. Through matrix eigenvalue analysis, the frost heave risk levels (low / medium / high risk) are divided, and the prediction error is better than that of traditional empirical models. It is linked with the cloud platform to generate a frost heave risk heat map, supporting real-time early warning and decision-making optimization for projects such as the Qinghai-Tibet Railway and cold region tunnels.

[0022] Furthermore, in the multi-modal data acquisition subsystem, Nuclear magnetic resonance acquisition module: A low-field nuclear magnetic resonance instrument equipped with a 0.5T permanent magnet is used. The minimum echo time TEmax = 0.1ms, supporting CPMG sequence acquisition. The low-field nuclear magnetic resonance instrument is equipped with an anti-interference double-layer shielded probe coil, and the magnetic field uniformity ≤ 10ppm. Through dynamic shimming technology and frequency adaptive filtering algorithm, the signal-to-noise ratio ≥ 90dB is ensured, ensuring the reliability of the signal under extreme low temperature (-30°C) and complex pore structure conditions, and is suitable for frozen soil detection; Temperature control module: Integrating a liquid nitrogen circulation pipeline and a PID temperature controller, the working temperature range is -30°C to 20°C, and the temperature control accuracy is ±0.01°C. The temperature closed-loop control is realized through the PID temperature controller. The temperature sensor used is a PT100 platinum resistance, and the temperature measurement accuracy is ±0.01°C; Through the gradient cooling design (rate 1°C / min) and the 9-temperature node scanning strategy, the whole process of soil sample freeze-thaw is fully covered.

[0023] The intelligent image processing unit supports multi-threaded parallel computing. The intelligent image processing unit includes an adaptive color segmentation module, supporting the recognition of red and blue double-color curves. The HSV threshold ranges are as follows: Red: lower = [0, 120, 70], upper = [10, 255, 255], Blue: lower = [100, 120, 70], upper = [130, 255, 255]; The data analysis module has a built-in data verification function. By comparing experimental data with the theoretical model, the calculation error is automatically corrected.

[0024] Furthermore, the data analysis module integrates a cubic spline interpolation algorithm, and the interpolation error ≤ 2%. It also supports the visual output of pore distribution and the division of frost heave risk levels.

[0025] The present invention also provides a computer-readable storage medium storing program codes for executing the above recognition method. The codes include an image preprocessing unit, a Hough transform detection unit, an OCR parsing unit, and an unfrozen water content calculation unit. Among them, the image preprocessing unit supports batch processing and can process multiple nuclear magnetic resonance images simultaneously, significantly improving the processing speed; the unfrozen water content calculation unit supports multi-core parallel computing, and the computing speed is significantly improved compared with single-core computing.

[0026] The technical upgrading directions of the present invention include multi-physical field coupling, edge computing and 5G transmission, and artificial intelligence optimization. By integrating temperature, seepage, and stress data, a full-dimensional model of the frost heaving mechanism is constructed. Through real-time analysis of millisecond-level data, rapid decision-making at the engineering site is realized. The introduction of deep learning algorithms (such as CNN) optimizes curve segmentation and feature extraction, and the contour integrity is significantly improved.

[0027] Compared with the prior art, the advantages of the present invention are as follows: 1. The present invention combines low-field nuclear magnetic resonance (LF-NMR) and image intelligent processing technology for the first time, constructs a dynamic correlation model of temperature-nuclear magnetic resonance signal, breaks through the limitations of traditional single-modal detection, and realizes the synchronous monitoring of unfrozen water content and pore evolution through the dual drive of dynamic nuclear magnetic resonance signal analysis and image feature extraction, significantly improving the data dimension and analysis accuracy, filling the technical gap of multi-modal non-destructive monitoring, avoiding the problems of strong destructiveness, limited single-point measurement, and high-salt interference of traditional methods (such as DSC, TDR), and also solving the core problems such as low data analysis efficiency, lack of quantification of pore evolution, and insufficient adaptability to complex environments in the monitoring of unfrozen water in multi-fractured frozen soil, overcoming the technical bottleneck of dynamic monitoring of unfrozen water in frozen soil and pore structure analysis, and having high precision, real-time performance and engineering applicability, providing a revolutionary tool for the stability assessment of frozen soil engineering, disaster warning, and optimization of artificial freezing technology.

[0028] 2. Through the unfrozen water dynamic recognition method of the present invention, the test accuracy is effectively improved, the detection error of unfrozen water content is significantly reduced, the inversion accuracy error of pore radius ≤ 0.05 μm, and the resolution can reach 0.1 μm, meeting the requirements of microstructural analysis. At the same time, the unfrozen water dynamic recognition system of the present invention can operate stably in a low-temperature environment, and its temperature change and data fluctuation are kept within a small range, having good environmental adaptability; through anti-interference design, the stability of the nuclear magnetic resonance signal of the system under complex conditions is ensured, the signal-to-noise ratio ≥ 90 dB, and the salt ion interference suppression rate > 90%.

[0029] 3. The present invention realizes non-destructive and accurate monitoring of the unfrozen water content of multi-fractured Qinghai-Tibet soil by integrating dynamic nuclear magnetic resonance signal analysis, intelligent image processing and freeze-thaw coupling modeling, providing technical support for engineering safety in alpine regions, and is particularly suitable for the stability assessment of frozen soil engineering, frost heaving disaster warning and quality control of artificial freezing construction in alpine regions.

[0030] 4. The present invention has extensive application scenario scalability and can be applied to the ground-based monitoring of polar research stations, the ecological restoration of frozen soil areas, and underground energy storage projects. For the first time, the nuclear magnetic resonance-image dual-modal technology is extended to the fields of frozen soil thermal stability assessment and clean energy optimization. It is adapted to the low-temperature environment of -30°C and can monitor the unfrozen water content and pore structure changes of the foundation in real time to ensure the structural safety under extreme conditions. Through the collaborative analysis of moisture and pores, the vegetation restoration strategy is optimized, the ecological restoration efficiency is improved, the frozen soil thermal stability is evaluated, the design of the ground source heat pump system is optimized, and the efficient utilization of clean energy is facilitated.

[0031] 5. The present invention uses an intelligent image processing unit to automatically and intelligently process nuclear magnetic resonance images. By using an improved Hough transform (parameters: ρ = 1, θ = π / 180) combined with Canny edge detection and least squares fitting, the axis tilt correction is achieved (error ≤ 0.5°), and the subjective error problem of traditional manual extraction is solved; through intelligent OCR parsing, the data conversion efficiency is significantly improved; also, through the HSV color space segmentation technology and morphological closing operation, the pixel-level accurate extraction of the temperature-signal curve is achieved, and the noise points with an area < 100 pixels are effectively filtered.

[0032] 6. The present invention establishes a temperature-signal dynamic correlation model that dynamically maps temperature and nuclear magnetic resonance signal intensity through gradient cooling and nuclear magnetic resonance scanning at multiple temperature nodes, and successfully captures the unfrozen water phase change characteristics during the entire freeze-thaw process. By combining the gray value of the nuclear magnetic resonance image and the mercury intrusion method to verify the data, a pore-fracture collaborative evolution model is formed, and an evolution matrix is constructed to quantitatively evaluate the frost heave risk level. Description of the Drawings

[0033] Figure 1 It is a schematic flow chart of the nuclear magnetic resonance and image intelligent dual-modal unfrozen water dynamic recognition method proposed in Embodiment 1 of the present invention. Detailed Embodiments

[0034] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described below.

[0035] Taking a typical frozen soil area on the Qinghai-Tibet Plateau as an example, Embodiment 1 proposes a nuclear magnetic resonance and image intelligent dual-modal unfrozen water dynamic recognition method. As Figure 1 shown, the method includes the following steps: 1. Sample preparation and pretreatment: Using the hollow drilling method, undisturbed soil samples are collected from a typical frozen soil area on the Qinghai-Tibet Plateau to obtain cylindrical samples with a diameter of 39.1 mm and a height of 80 mm. Then, through the pretreatment of the cooperation of screening and centrifugation operations, gravel or impurity particles with a diameter > 2 mm are removed, and the pretreated soil samples are sealed in nuclear magnetic resonance sample tubes.

[0036] In this embodiment, when collecting undisturbed soil samples from typical permafrost areas on the Qinghai-Tibet Plateau, the hollow coring method can effectively ensure that the internal structure of the soil mass is not damaged. The obtained cylindrical sample has a size of 39.1 mm in diameter and 80 mm in height. This specification not only facilitates experimental operations but also maximally retains the characteristics of the soil sample in its natural state. Additionally, in the process of removing gravel with a diameter > 2 mm, screening and centrifugation operations are coordinated. Through a 2-mm aperture sieve, larger particle impurities can be initially removed. Subsequently, centrifugation is carried out at a speed of 3000 rpm for 5 minutes to further homogenize the soil sample, making the coefficient of variation (CV) of the soil sample moisture content < 5%, ensuring the stability and reliability of subsequent experimental data. The sample is sealed in a polytetrafluoroethylene (PTFE) sample tube. This material has stable chemical properties, and the wall thickness ≤ 1 mm, minimizing interference with nuclear magnetic signals as much as possible. Ceramic sealing caps are installed at both ends, with a porosity < 0.1%, which can effectively prevent the sample from exchanging substances with the external environment and maintain the consistency of the sample state.

[0037] After preprocessing the Qinghai-Tibet permafrost samples, initial parameter calibration is also required. The specific method is as follows: The initial total water content w0 is measured by the drying method. Keeping the sample in a constant temperature environment of 105°C for 24 hours can ensure that the water in the soil sample is completely evaporated, providing accurate basic data for the subsequent calculation of unfrozen water content.

[0038] 2. Nuclear magnetic data acquisition and temperature control setting: After sealing the Qinghai-Tibet permafrost sample in a nuclear magnetic sample tube, the temperature of the sample is gradually decreased in the low-temperature range from -30°C to 20°C through the temperature control module. A low-field nuclear magnetic resonance spectrometer is used to scan and obtain the transverse relaxation time T2 spectrum and nuclear magnetic images of the permafrost sample at different temperature nodes.

[0039] In this embodiment, the specific operations and configurations are as follows: 1) Configuration of nuclear magnetic resonance spectrometer In this embodiment, a low-field nuclear magnetic resonance spectrometer equipped with a 0.5T permanent magnet is selected. The magnetic field uniformity ≤ 10 ppm ensures the stability of the magnetic field environment during the experiment. The minimum echo time TEmax = 0.1 ms, supporting CPMG sequence acquisition, meeting the requirement for precise acquisition of sample signals. The probe coil adopts a double-layer copper shielding design, combined with a frequency adaptive filtering algorithm, and the signal-to-noise ratio (SNR) can ≥ 90 dB, greatly suppressing high-frequency noise interference. In terms of pulse sequence parameter settings, the echo time TE = 0.2 ms, the number of echoes N = 1024, and the number of accumulations is 32 times. Through dynamic shimming technology (adjusted by gradient coils), the magnetic field uniformity is further improved, enabling the T2 spectrum resolution to reach ±0.01 ms, so as to more accurately obtain the distribution information of the sample transverse relaxation time.

[0040] 2) Operation of temperature control module In this embodiment, the temperature control module integrates a liquid nitrogen circulation pipeline and a PID temperature controller. After starting the liquid nitrogen circulation pipeline, with the help of the PID temperature controller, the temperature can be accurately set to linearly decrease from 20°C to -30°C at a rate of 1°C / min, and the temperature control accuracy is as high as ±0.5%. A nuclear magnetic resonance (NMR) scan is triggered every 5°C, covering a total of 9 temperature nodes: 20°C, 10°C, 0°C, -2°C, -5°C, -7°C, -10°C, -15°C, and -30°C, comprehensively covering the entire process of soil sample freezing and thawing. A PT100 platinum resistance temperature sensor is used, with a temperature measurement accuracy of ±0.01°C. The real-time collected temperature data is transmitted to the main control system, realizing the synchronous recording of temperature and NMR signals, providing reliable data correlation for subsequent analysis of the changes in soil sample characteristics at different temperatures.

[0041] 3) Data acquisition and storage At each temperature point, the system synchronously acquires T2 spectrum data (i.e., transverse relaxation time distribution) and NMR images (resolution of 512×512 pixels, gray scale range 0 - 255). The data storage adopts a hierarchical architecture. Among them, the T2 spectrum data is saved in CSV format, and the relaxation time (ms), signal amplitude (a.u.), and corresponding temperature labels are detailedly recorded in the file, facilitating subsequent data retrieval and analysis. The NMR images are stored in DICOM format, which not only retains the spatial resolution but also saves metadata such as magnetic field strength and sampling time, providing necessary information for the accurate interpretation of the images. The database is real-time backed up to the cloud, supporting multi-terminal access. Researchers can retrieve historical data for retrospective analysis anytime and anywhere, greatly improving the utilization efficiency and security of the data.

[0042] 3. Image processing and feature extraction: After preprocessing the obtained NMR images, image recognition technology is used to automatically extract the temperature-signal intensity curve in the NMR images through coordinate axis detection and correction, scale parsing and numerical conversion, and curve extraction and coordinate conversion. The specific process is as follows: 1) Image preprocessing The OpenCV library is used for grayscale processing to convert the RGB NMR images into 8-bit grayscale images, removing the color information in the images and reducing interference factors. Adaptive thresholding is used for binarization, with the threshold set to 120, effectively enhancing the contrast of the coordinate axes and curves in the images. A median filter is applied to remove salt-and-pepper noise, and at the same time, combined with edge-preserving filtering, while removing noise, the details of the curves are protected to the greatest extent, laying a good foundation for subsequent operations such as coordinate axis detection and curve extraction.

[0043] 2) Coordinate axis detection and correction The Canny algorithm (low threshold 50, high threshold 150) is used to extract image edges and generate a clear edge intensity map. The improved Hough transform (parameters ρ=1, θ=π / 180, threshold=100) is used to detect horizontal and vertical lines in the image, and interfering line segments with a length of less than 100 pixels are filtered out to improve detection accuracy. The axis equation is fitted based on the least squares method to accurately calculate the inclination angle of the coordinate axis (tolerance ≤0.5°), and the image is geometrically corrected through affine transformation to ensure that the coordinate axis in the image is in a standard position, which is convenient for subsequent scale analysis and numerical conversion.

[0044] 3) Scale analysis and numerical conversion The 50-pixel height range at the bottom of the image is intercepted as the x-axis scale area, and the 50-pixel width range on the left is used as the y-axis scale area, and the scale information is extracted specifically. Configure the TesseractOCR engine parameters to accurately extract the scale text. Use regular expressions to parse scientific notation (such as "2.5e-3" is converted to 0.0025) and fractional formats (such as "1 / 293.15" is converted to 0.00341), and automatically calibrate the scale offset error to accurately convert the scale information in the image into actual physical quantity values.

[0045] 4) Curve extraction and coordinate transformation The target curve is segmented based on the HSV color space. For the red curve, H=0-10, S≥120, V≥70 is set; for the blue curve, H=100-130, S≥120, V≥70 is set. The morphological closing operation (3×3 rectangular kernel, dilation 2 times and erosion 1 time) is applied to fill the contour breaks and connect the discrete curve segments. Continuous contours with an area of ≥100 pixels are screened, arranged in ascending order by x coordinate, and the key points of the curve are extracted. The pixel coordinates are mapped to the actual physical quantity through bilinear interpolation, realizing the accurate conversion from image pixel coordinates to actual physical quantity coordinates. The calculation formula is: ; In this calculation formula, , They are the actual physical horizontal and vertical coordinate values after conversion (the unit is determined by the scale unit parsed by OCR); , are the horizontal and vertical pixel coordinates of the curve points in the image (the origin is the upper left corner of the image, , ); , The width and height of the image (in pixels, such as the width and height values obtained by img_shape[::-1][:2]); , Represents the x-axis scale range (i.e., the maximum / minimum value of the actual physical quantity); and represents the y-axis scale range (i.e., the maximum / minimum value of the actual physical quantity).

[0046] 4. Data analysis and model calculation: Perform numerical integration on the liquid water signal in the T2 spectrum, dynamically calculate the unfrozen water content, and conduct freeze-thaw coupling analysis. The specific calculation and analysis method is as follows: 1) Calculation of unfrozen water content In this embodiment, numerical integration is performed on the T2 spectrum of each temperature point (using the Simpson's rule with a step size of 0.01 ms), and the integration range is set from 10 ms to 1000 ms. Since the ice signal with T2 < 10 μs will be filtered by the hardware, this integration range can effectively obtain the liquid water signal.

[0047] The integral value of the liquid water signal is: ; Perform integral calculation on the liquid water signal in the T2 spectrum, and dynamically calculate the content of unfrozen water at each temperature based on the following nuclear magnetic signal intensity formula; , where is the nuclear magnetic signal intensity at temperature T; is the nuclear magnetic signal intensity corresponding to the initial liquid water; w0 is the initial total water content.

[0048] According to the unfrozen water content at each temperature point, the system automatically generates a temperature-unfrozen water content curve, and accurately marks key inflection points such as the freezing point and the stable section, intuitively presenting the trend of the unfrozen water content changing with temperature.

[0049] 2) Freeze-thaw coupling analysis In this embodiment, by establishing a linear mapping relationship between the gray value (0 - 255) of the nuclear magnetic image and the pore radius (such as gray = 200 corresponding to R = 0.1 μm), a pore-fracture co-evolution matrix is constructed to achieve accurate analysis of the dynamics of unfrozen water and pore evolution during the freeze-thaw process. The specific method is as follows: 2.1: Extract pore distribution characteristics and establish a mapping relationship (1) Inversion of pore radius from T2 spectrum: Using the linear relationship between the transverse relaxation time and the pore radius : ; where the surface relaxation rate takes the empirical value , and substituting The distribution curve is converted into the pore radius distribution.

[0050] (2) Pore classification and quantification The pores are divided into ultra - micropores ( ), micropores ( ), small pores ( ), mesopores ( ), and macropores ( ). By integrating the spectral peak area, the proportion of the pore volume in each pore size interval is calculated: ; ; where, is the pore volume in a certain pore size interval, is the total pore volume.

[0051] When the proportion of the micropore volume exceeds 60%, the change in the micropore radius ( ) dominates the evolution, reflecting the mesoscopic frost heaving mechanism; when the proportion of the macropore volume exceeds 10%, the change in the crack width ( ) becomes the main variable, corresponding to the macroscopic failure process.

[0052] The mapping of the nuclear magnetic resonance (NMR) image gray value - pore radius establishes a linear relationship between the gray value (0 - 255) of the NMR image and the pore radius: ; By calibrating the parameters and through the mercury intrusion porosimetry (MIP), the quantitative conversion from the image gray value to the pore radius is realized.

[0053] 2.2: Analyze the pore evolution law Construct the pore - crack co - evolution matrix, and define the evolution matrix to describe the coupling relationship between the pore structure, unfrozen water, and temperature during the freeze - thaw process: ; where, ΔR is the change in the pore radius (μm / ℃), ∇w u is the unfrozen water gradient (% / mm), and ∇T is the temperature gradient (℃ / mm).

[0054] is the change in the radius of the micro - pores with a pore size less than 0.2 μm due to the ice crystal growth stress during the freeze - thaw process, and its physical meaning is to quantify the frost heaving characteristics of the micro - pores at the mesoscopic scale represents the change in the width of the cracks with a width greater than 1 μm due to the ice wedge effect during the freeze - thaw process, and its physical meaning is to characterize the crack propagation rate at the macroscopic scale.

[0055] Through the differences in the pore structure evolution at the microscopic and macroscopic scales, they jointly provide key parameters for the analysis of the pore-water-thermal coupling mechanism during the multi-scale freeze-thaw process.

[0056] 5. Verification and Optimization: Through the accuracy verification experiment, verify the measurement accuracy of the unfrozen water identification method and the pore radius measurement accuracy in this embodiment; through the robustness test, verify the anti-salt ion interference ability of the unfrozen water identification method in this embodiment.

[0057] 1) Accuracy Verification Experiment Based on the theoretical model and simulation analysis of the method in this embodiment, predict the relevant accuracy and capabilities of the Qinghai-Tibet frozen soil samples: through the nuclear magnetic signal integration model and the temperature-water coupling algorithm, the average error of the unfrozen water content can be controlled within 5%, which is significantly lower than the 15%-20% error of the traditional differential scanning calorimetry (DSC). The improvement in accuracy stems from the non-destructive quantification ability of nuclear magnetic technology for liquid water signals, avoiding the interference of the sample structure in traditional destructive testing; Based on the theoretical mapping relationship between the T2 spectrum and the pore radius, combined with the calibration of the surface relaxation rate, when the pore radius R = 0.5μm, the measurement error ≤ 0.05μm, and the resolution can reach 0.1μm. This accuracy can effectively capture the microscopic characteristics of the soil pore structure and provide a quantitative basis for freeze-thaw damage analysis; through the pore-fracture co-evolution matrix and the freeze-thaw coupling model, the error in the classification of the frost heave risk level is better than that of the traditional empirical model (the error of the traditional model ≥ 30%), and the prediction accuracy is improved by more than 50%. It can realize the dynamic classification of low risk (|λ1| < 0.3), medium risk (0.3 ≤ |λ1| < 0.7), and high risk (|λ1| ≥ 0.7), providing risk warning support based on eigenvalue thresholds for projects such as the Qinghai-Tibet Railway and cold region tunnels.

[0058] 2) Robustness Test In the range of salt content from 0 to 5%, the system signal-to-noise ratio is stable ≥ 90dB, and the salt ion interference suppression rate > 90%, indicating that this technology has a strong resistance to salt ion interference. The multi-thread parallel computing architecture supports processing 10 frames of images per second, meeting the high-efficiency requirements of real-time monitoring.

[0059] In addition, during the implementation of the identification method in this embodiment, the following method is used to optimize the parameters: The optimal parameters of the Hough transform (ρ = 0.5 - 2, θ = π / 360 - π / 90) were determined through grid search (ρ = 1, θ = π / 180), improving the accuracy of axis detection. The OCR regular expression was extended to support formats such as "3.2×10⁻³" and "5.6×10²", enhancing the recognition rate. A deep learning model (CNN) was introduced to optimize curve segmentation, further optimizing the effects of image processing and feature extraction.

[0060] Example 2 Example 2 proposes a nuclear magnetic resonance and image intelligent dual-modal unfrozen water dynamic recognition system to implement the unfrozen water dynamic precise recognition method of the above Example 1. The system includes a multi-modal data acquisition subsystem, an intelligent image processing unit, and a data analysis module.

[0061] In this Example 2, the multi-modal data acquisition subsystem is used to obtain the transverse relaxation time T2 spectrum and nuclear magnetic resonance images of frozen soil samples at different temperature nodes. The acquisition subsystem includes: 1) Nuclear magnetic resonance acquisition module: A low-field nuclear magnetic resonance instrument equipped with a 0.5T permanent magnet, with a minimum echo time TEmin = 0.1ms, supporting CPMG sequence acquisition. The low-field nuclear magnetic resonance instrument is equipped with an anti-interference double-layer shielded probe coil, with a magnetic field uniformity ≤ 10ppm. Through dynamic shimming technology and frequency adaptive filtering algorithm, the signal-to-noise ratio ≥ 90dB, ensuring the reliability of signals under extreme low temperature (-30°C) and complex pore structure conditions, and is suitable for frozen soil detection.

[0062] 2) Temperature control module: Integrating a liquid nitrogen circulation pipeline and a PID temperature controller, with a working temperature range of -30°C to 20°C and a temperature control accuracy of ±0.01°C. The temperature closed-loop control is realized through the PID temperature controller, and the temperature sensor used is a PT100 platinum resistance, with a temperature measurement accuracy of ±0.01°C; through a gradient cooling design (rate 1°C / min) and a 9-temperature node scanning strategy, the entire process of soil sample freezing and thawing is comprehensively covered.

[0063] In this Example 2, the intelligent image processing unit is equipped with the OpenCV and Tesseract OCR libraries, and is used for axis detection, curve fitting extraction, scale parsing, and data conversion of nuclear magnetic resonance images; The intelligent image processing unit adopts a multi-threaded parallel architecture, supporting multi-threaded parallel computing. The single-frame image processing time < 0.5 seconds, supporting continuous monitoring at 10 frames per second, meeting the real-time requirements of the project.

[0064] The intelligent image processing unit includes an adaptive color segmentation module, supporting the recognition of red and blue double-color curves. The HSV threshold ranges are as follows: Red: lower = [0, 120, 70], upper = [10, 255, 255], Blue: lower = [100, 120, 70], upper = [130, 255, 255]; In this Embodiment 2, the data analysis module is based on the Python platform and internally built with a special analysis model for Qinghai-Tibet soil, a freeze-thaw coupling analysis module, and a data verification function.

[0065] 1) Special analysis model for Qinghai-Tibet soil: It includes a dynamic calculation model for unfrozen water content and a co-evolution model for pores and fractures.

[0066] 1.1) The dynamic calculation model for unfrozen water content dynamically calculates the unfrozen water content ω by integrating the peak area of the liquid water in the T2 spectrum and combining it with the initial total water content w0.

[0067] 1.2) The co-evolution model for pores and fractures introduces the pore characteristic factor Fs and the surface relaxation rate ρ for calibration, and combines the verified data of the mercury intrusion method to achieve a pore radius resolution of 0.1 μm and can identify the expansion of microfractures (accuracy ±0.05 μm). The specific method is as follows: The co-evolution model for pores and fractures introduces the pore characteristic factor (micropores <0.2 μm, fractures > 1 μm): ; In the formula, is the volume of micropores with a pore diameter less than 0.2 μm, which is calculated by integrating the peak area in the corresponding relaxation time interval in the T2 spectrum and reflects the mesoscopic frost heaving mechanism. is the volume of fractures with a width greater than 1 μm, which is determined by mapping the gray level of the nuclear magnetic resonance image and the pore radius calibrated by the mercury intrusion method, and characterizes the macroscopic fracture expansion.

[0068] is calibrated with the surface relaxation rate to establish a multi-scale evolution equation set: ; Among them, the evolution coefficients , are obtained by measuring the mercury intrusion radii of micropores and fractures , by the mercury intrusion method and fitting.

[0069] Within the temperature range ΔT, the gradients , ∇T are approximately constant, and integrating with respect to time gives the state change amount: ; After normalizing the time variable ( = 1), the co-evolution model of pores and fractures has the following relationship with the co-evolution matrix of pores and fractures: ; The specific calibration process is as follows: First, data collection is carried out. The pore radius is measured by mercury intrusion porosimetry (MIP), and the micropore volume is calculated by measuring the T2 spectrum of nuclear magnetic resonance. Fracture volume And obtain ; Then, parameter fitting is carried out. Substitute the data into the evolution equation and adjust k1 and k2 to make ; Finally, verification is carried out. In the freeze-thaw experiment, ensure that the fracture expansion amount calculated by the model and the scanning deviation ≤ 0.05 μm.

[0070] Freeze-thaw coupling analysis module: Used to construct a pore-fracture co-evolution matrix (such as matrix ), and through the matrix eigenvalues ( , , sorted by the modulus length ), analyze and divide the frost heaving risk levels (low / medium / high risk). The specific risk division is as follows: Low risk: , corresponding to the stable section of permafrost areas, with a frost heaving deformation rate < 0.5% / year, unfrozen water gradient , temperature gradient , weak water-thermal-mechanical coupling, and high soil stability.

[0071] Medium risk: , corresponding to the transition section of seasonal frozen soil areas, with a frost heaving deformation rate of 0.5% - 1.5% / year, And Medium coupling ( , ), and significant but controllable pore-fracture evolution.

[0072] High risk: , corresponding to the water-rich fracture development area, such as the subgrade thaw settlement section of the Qinghai-Tibet Railway, with a frost heaving deformation rate > 1.5% / year, , , rapid fracture expansion ( ), and high disaster risk.

[0073] Data verification function: Automatically correct calculation errors by comparing experimental data with the theoretical model.

[0074] Data analysis module: The integrated cubic spline interpolation algorithm, as a piecewise cubic polynomial interpolation method, constructs a cubic polynomial between adjacent data points to ensure the continuity of function values, first-order derivatives, and second-order derivatives, thus forming a smooth interpolation curve. This algorithm can fit discrete data such as pore distribution with high precision, meet the requirement of interpolation error ≤ 2%, and can generate a continuous pore structure curve. On the one hand, it provides support for the visual analysis of frost heaving risk, including establishing the correlation model between pore size distribution and frost heaving potential energy and classifying the frost heaving risk level. On the other hand, it provides continuous medium parameters such as pore connectivity interpolation quantification for the simulation of unfrozen water migration, thereby improving the accuracy of freeze-thaw coupling analysis.

[0075] System expansion and technology upgrade: In terms of system expansion, the unfrozen water dynamic precise identification system of this Embodiment 2 also has the following expansion directions: Technology upgrade direction: In the future, a nuclear magnetic resonance instrument with a higher magnetic field strength (1.0T) can be introduced, which is expected to further improve the T2 spectrum resolution to ±0.005 ms and obtain more refined sample information. Develop a multi-physical field coupling model, integrate data in multiple aspects such as temperature, seepage, and stress, and realize the full-dimensional analysis of the frost heaving mechanism to provide more comprehensive theoretical support for engineering applications. Combine edge computing technology to achieve real-time analysis and early warning of millisecond-level data and meet the requirements of rapid decision-making at the engineering site.

[0076] Application expansion scenarios: In the field of ecological restoration in frozen soil areas, by monitoring the coordinated changes of soil moisture-salt-pores, it provides a scientific basis for formulating reasonable vegetation restoration strategies. In underground energy storage projects, it evaluates the thermal stability of frozen soil, optimizes the design of ground source heat pump systems, improves energy utilization efficiency, and contributes to the development and utilization of clean energy.

[0077] In addition, the unfrozen water dynamic precise identification method and system of the present invention have the implementation closed-loop property, which is specifically manifested as: Data flow: Starting from nuclear magnetic signal acquisition, through image processing, feature extraction, and model calculation, it finally provides a basis for engineering decision-making, and the engineering practice feedback further optimizes the entire technical process.

[0078] Verification closed-loop: First, perform parameter calibration in the laboratory, then go to the field actual environment for verification, perform parameter iteration according to the verification results, and finally achieve large-scale application, forming a complete technical implementation closed-loop, continuously improving the reliability and practicability of the technology.

[0079] Embodiment 3 Embodiment 3 of the present invention proposes a computer-readable storage medium, which stores program codes for executing the above recognition method. The codes include an image preprocessing unit, a Hough transform detection unit, an OCR parsing unit, and an unfrozen water content calculation unit. Among them, the image preprocessing unit supports batch processing and can process multiple nuclear magnetic resonance images simultaneously, significantly improving the processing speed; the unfrozen water content calculation unit supports multi-core parallel computing, and the computing speed is significantly improved compared with single-core computing.

[0080] The above are only the preferred embodiments of the present invention and do not impose any limitations on the present invention. Any person skilled in the art, within the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification and other changes to the technical solution and technical content disclosed by the present invention, which are all within the content of the technical solution of the present invention and still fall within the protection scope of the present invention.

Claims

1. A method for dynamically identifying unfrozen water with intelligent dual modalities of nuclear magnetic resonance and images, characterized in that, It includes the following steps: S1: Seal the frozen soil sample in a nuclear magnetic resonance sample tube, and realize gradient cooling of the sample within a preset low temperature range through temperature control; S2: Use nuclear magnetic resonance technology to scan and obtain the transverse relaxation time T2 spectrum and nuclear magnetic resonance image of the frozen soil sample at different temperature nodes; S3: After preprocessing the nuclear magnetic resonance image, use image recognition technology to automatically extract the temperature-signal intensity curve in the nuclear magnetic resonance image; S4: Integrate the liquid water signal in the T2 spectrum, and dynamically calculate the content of unfrozen water at each temperature based on the following nuclear magnetic signal intensity formula; ， Wherein, is the nuclear magnetic signal intensity at temperature T; is the nuclear magnetic signal intensity corresponding to the initial liquid water; w0 is the initial total water content.

2. The method for dynamically identifying unfrozen water by nuclear magnetic resonance and image intelligent dual - modality according to claim 1, wherein, It also includes steps S5 and S6: S5: Invert the pore radius using the transverse relaxation time T2 spectrum, extract the pore distribution characteristics, calculate the proportion of pore volume in each pore size interval, and establish a linear relationship between the gray value of the nuclear magnetic resonance image and the pore radius; S6: Construct a pore-fracture co-evolution matrix, and define the evolution matrix M to describe the coupling relationship between the pore structure, unfrozen water, and temperature during the freeze-thaw process; ； Among them, is the radius change amount generated by the growth stress of ice crystals during the freeze-thaw process in micro-pores with a pore diameter less than 0.2 μm. represents the width change amount caused by ice wedging during the freeze-thaw process in cracks with a width greater than 1 μm, ∇w u is the unfrozen water gradient, and ∇T is the temperature gradient.

3. The nuclear magnetic resonance and image intelligent dual-modal unfrozen water dynamic recognition method according to claim 2, wherein Step S5 is specifically: S5.1: Invert the pore radius using the linear relationship between the transverse relaxation time T2 spectrum and the pore radius R, and convert the T2 distribution curve into a pore radius distribution; ； Among them, the surface relaxation rate takes an empirical value ; S5.2: Pore classification and quantification: According to the pore size, the pores are divided into ultra-micropores, micropores, small pores, medium pores, and large pores. By integrating the peak area of the T2 spectrum, calculate the proportion of pore volume in each pore size interval; ； Among them, is the pore volume in a certain pore size range, is the total pore volume; When the proportion of micropore volume exceeds 60%, the change in micropore radius plays a leading role in evolution and reflects the mesoscopic frost heaving mechanism; when the proportion of macropore volume exceeds 10%, the change in crack width becomes the main variable, corresponding to the macroscopic failure process; S5.3: Measure the pore radius distribution of the soil sample by mercury intrusion method, extract the gray value of the pore region of the nuclear magnetic resonance image of the same soil sample, establish a mapping relationship between the gray value of the nuclear magnetic resonance image and the pore radius, and thus construct a linear relationship between the gray value of the nuclear magnetic resonance image and the pore radius.

4. According to the nuclear magnetic resonance and image intelligent dual-modal unfrozen water dynamic recognition method described in claim 1, characterized in that In step S1, the preset low temperature range is -30°C to 20°C; In step S2, a 0.5T permanent magnet low-field nuclear magnetic resonance instrument is used to perform nuclear magnetic resonance scanning on the frozen sample, and nuclear magnetic resonance scanning is triggered every 5°C, so as to obtain the transverse relaxation time T2 spectrum and nuclear magnetic resonance image of the frozen soil sample at a total of 9 temperature nodes of 20°C, 10°C, 0°C, -2°C, -5°C, -7°C, -10°C, -15°C, and -30°C, comprehensively covering the whole process of soil sample freeze-thaw; During the nuclear magnetic data acquisition process, the CPMG sequence is used, the echo time TE = 0.2ms, and the number of accumulations is 32 times. The system can ensure that the signal-to-noise ratio ≥ 90dB within the temperature range of -30°C to 20°C.

5. The method for dynamically identifying unfrozen water by nuclear magnetic resonance and image intelligent dual modality according to claim 1, characterized in that, In step S3, The process of the image preprocessing includes: Gray-scale processing: Convert the nuclear magnetic resonance image from the RGB color space to a gray-scale image; Binarization processing: Use the adaptive threshold method, set the threshold to 120, and convert the gray-scale image to a binary image; Median filtering processing: Use a 3×3 median filter to remove the salt-and-pepper noise in the image; The image recognition technology includes: Use the improved Hough transform to identify the temperature axis and the signal intensity axis, and combine the Canny edge detection algorithm and the least squares method for fitting to realize the correction of the coordinate axis inclination; Integrate the Tesseract OCR engine with regular expressions to extract and parse scale values, automatically calibrate scale offset errors, and accurately convert the scale information in the image into actual physical quantity values; Based on HSV color space segmentation and morphological closing operation, realize pixel-level extraction of the temperature-signal curve.

6. The intelligent dual-modal unfrozen water dynamic recognition method combining nuclear magnetic resonance and image according to claim 5, characterized in that The parameter settings of the improved Hough transform are: ρ = 1, θ = π / 180, threshold = 100, minimum line segment length = 100 pixels; The specific method for realizing pixel-level extraction of the temperature-signal curve based on HSV color space segmentation and morphological closing operation is as follows: First, perform color segmentation on the red curve through the HSV color space, and set the HSV threshold range as lower = [0, 120, 70], upper = [10, 255, 255]; Second, use the OpenCV library to extract the curve contour and filter out noise points with an area less than 100 pixels; Finally, convert the pixel coordinates into actual values, and use the linear interpolation method to calculate the temperature and signal intensity values corresponding to the curve points.

7. A nuclear magnetic resonance and image intelligent dual-modal unfrozen water dynamic recognition system for implementing the recognition method described in any one of claims 1-6, characterized in that, Including: Multi-modal data acquisition subsystem: including a nuclear magnetic resonance acquisition module and a temperature control module, used to obtain the transverse relaxation time T2 spectrum and nuclear magnetic resonance image of the frozen soil sample at different temperature nodes; Intelligent image processing unit: equipped with the OpenCV and Tesseract OCR libraries, used for axis detection, curve fitting extraction and scale parsing of nuclear magnetic resonance images; Data analysis module: based on the Python platform, built-in unfrozen water content calculation algorithm and pore evolution model.

8. The intelligent dual-modal unfrozen water dynamic recognition system combining nuclear magnetic resonance and image according to claim 7, characterized in that In the multi-modal data acquisition subsystem, Nuclear magnetic resonance acquisition module: use a low-field nuclear magnetic resonance instrument equipped with a 0.5T permanent magnet, the minimum echo time TEmin = 0.1ms, support CPMG sequence acquisition, equipped with an anti-interference double-layer shielded probe coil, magnetic field uniformity ≤ 10ppm, signal-to-noise ratio ≥ 90dB; Temperature control module: integrate a liquid nitrogen circulation pipeline and a PID temperature controller, the working temperature range is -30°C to 20°C, the temperature control accuracy is ±0.01°C, realize temperature closed-loop control through the PID temperature controller, the temperature sensor is a PT100 platinum resistance, and the temperature measurement accuracy is ±0.01°C; The intelligent image processing unit supports multi-threaded parallel computing, and the intelligent image processing unit includes an adaptive color segmentation module, which supports the recognition of red and blue double-color curves. The HSV threshold ranges are as follows: Red: lower = [0, 120, 70], upper = [10, 255, 255], Blue: lower = [100, 120, 70], upper = [130, 255, 255]; The data analysis module built-in data verification function automatically corrects calculation errors by comparing experimental data with theoretical models.

9. The nuclear magnetic resonance and image intelligent dual-modal unfrozen water dynamic recognition system according to claim 7, characterized in that The data analysis module integrates a cubic spline interpolation algorithm with an interpolation error ≤ 2%, and supports visual output of pore distribution and classification of frost heaving risk levels.

10. A computer-readable storage medium stores program code for executing the recognition method according to any one of claims 1-6, characterized in that The code includes an image preprocessing unit, a Hough transform detection unit, an OCR parsing unit, and an unfrozen water content calculation unit. Among them, the image preprocessing unit supports batch processing, and the unfrozen water content calculation unit supports multi-core parallel computing.

Images