Method and system for detecting depth of freezing face of seasonally frozen soil subgrade based on ground penetrating radar signal characteristics

Abstract

The application discloses a method and system for detecting the depth of a frozen surface of a seasonally frozen soil subgrade based on ground penetrating radar signal characteristics. Compared with traditional methods that rely on empirical formulas or single signal analysis, the application significantly improves the automation and anti-interference ability of the detection, and can effectively adapt to complex and variable seasonally frozen soil environments. The multi-source data set constructed by combining indoor scale test, numerical simulation and field measurement enhances the generalization and robustness of the model. In practical application, the application realizes rapid and non-destructive detection of the depth of the frozen surface, greatly reduces the use of destructive detection methods such as drilling cores, reduces the detection cost and period, and provides reliable technical support for construction quality control and operation period safety evaluation of the seasonally frozen soil subgrade.

Description

Technical Field

[0001] This invention relates to the field of roadbed engineering technology, specifically to a method and system for detecting the depth of the frozen surface of seasonally frozen soil roadbeds based on the characteristics of ground-penetrating radar signals. Background Technology

[0002] The freezing depth of soil in seasonally frozen zones refers to the maximum vertical depth from the ground surface that freezes due to the drop in temperature during winter. Unlike permafrost regions, the freezing depth in seasonally frozen zones is not a fixed value but varies significantly with the seasons. In the design of highways, railways, and other engineering projects, this depth is a critical parameter that directly affects the stability and durability of the overall structure.

[0003] There are currently four commonly used methods for detecting the depth of frozen surfaces: (1) Measure the ground temperature and calculate it based on certain engineering experience. The drawback of this method is that it is a point measurement method and has poor ability to detect areas. At the same time, because the ground temperature is affected by factors such as groundwater and soil properties, it cannot completely correspond to the depth of the frozen surface, resulting in low detection accuracy.

[0004] (2) Obtain and calculate the surface temperature through large-scale remote sensing. This method can detect the depth of the frozen surface over a large area, but it is affected by vegetation, atmosphere, etc., and its accuracy is not high.

[0005] (3) Sampling is carried out by core drilling, excavation and other methods. Since it is a direct observation, this method has high detection accuracy, but it cannot be carried out on a large scale. Moreover, the detection work requires a lot of manpower and material resources, the detection efficiency is low, and it will also damage the roadbed.

[0006] (4) The depth of the frozen surface can be calculated by measuring the resistivity of the soil using electrical methods. This method is theoretically feasible, but in actual engineering, it requires the installation of electrodes, resulting in high overall testing costs, a large amount of preliminary work, and low efficiency.

[0007] It can be seen that all four detection methods mentioned above have shortcomings, therefore a more efficient and non-destructive method is urgently needed to detect the freezing depth of the subgrade in seasonally frozen soil. At present, research on ground-penetrating radar for subgrade detection focuses on the identification of hidden defects and soil physical parameters, such as the invention patent CN119148095A "A method, equipment and medium for identifying hidden road defects using three-dimensional ground-penetrating radar", the invention patent CN118642062A "A method for identifying underground road defects based on target detection and ground-penetrating radar", and the invention patent CN118519112A "A method, equipment and storage medium for estimating the resilient modulus of the roadbed based on ground-penetrating radar".

[0008] In some studies involving ground-penetrating radar (GPR) detection of permafrost, the invention patent CN119086237A, which discloses "a method for identifying the upper limit of permafrost using multi-frequency composite array GPR," targets the identification of the upper limit of permafrost and uses the method of inversely calculating the thickness of the thawed soil layer using electromagnetic waves, without utilizing the rich information of GPR signals in the time-frequency domain and energy domain. The invention patent CN107272073B, which discloses "a method for calculating the relative water content of permafrost using GPR," targets the detection of water content within permafrost and is unrelated to the detection of the depth of the frozen surface.

[0009] In summary, no relevant technologies for detecting the depth of the frozen surface of permafrost roadbed using ground-penetrating radar signal characteristics have been published at this stage. Summary of the Invention

[0010] To address the technical problems mentioned above, this invention aims to construct a method and system for rapidly identifying the freezing depth of seasonally frozen soil subgrades with high efficiency and over a wide range, thereby providing a basis for the design and construction of roads in seasonally frozen areas.

[0011] To achieve the above objectives, this invention provides a method for detecting the depth of the frozen surface of seasonally frozen soil subgrade based on ground-penetrating radar signal characteristics, comprising: S1. Divide the subgrade into testing units based on the frozen soil section to be tested; S2. Based on the detection units, construct a ground-penetrating radar model for predicting the depth of frozen surfaces on roadbeds; S3. Based on the detection section of the seasonally frozen soil subgrade to be detected, obtain the feature values ​​of the dataset to be detected, and input the dataset to be detected into the ground penetrating radar identification model for predicting the depth of the frozen surface of the subgrade to obtain the depth of the frozen surface of the seasonally frozen soil subgrade.

[0012] Preferably, S2 includes: constructing a ground-penetrating radar (GPR) database for identifying the depth of frozen roadbed surfaces based on the detection unit, and using a sparrow search algorithm to optimize the hyperparameters of the convolutional neural network, long short-term memory network, and multi-head attention mechanism, and using the optimized hyperparameters to construct a network for training to obtain a GPR prediction model for identifying the depth of frozen roadbed surfaces.

[0013] Preferably, in step S2, the method for constructing a ground-penetrating radar (GPR) database for identifying the depth of frozen surfaces on roadbeds includes: Based on the detection unit, ground-penetrating radar is used to collect echo signal data, and the echo signal data is preprocessed to obtain the ground-penetrating radar time-domain basic dataset. Based on the ground-penetrating radar time-domain basic dataset, Hippotransform and spectral estimation analysis are performed to obtain the feature values ​​in the time domain, frequency domain, and time-frequency domain. Combined with the freezing surface depth of the seasonally frozen soil subgrade corresponding to the detection unit, a ground-penetrating radar identification subgrade freezing surface depth database is constructed.

[0014] Preferably, the methods for obtaining feature values ​​in the time domain, frequency domain, and time-frequency domain include: Based on the ground-penetrating radar time-domain basic dataset, the peak amplitude, average amplitude, average energy, amplitude squared difference, and amplitude kurtosis are calculated for each A-Scan signal as time-domain indicators. Perform a Hibbert transform on each A-Scan signal and combine it with the original signal to construct an analytic signal. Calculate the instantaneous amplitude, instantaneous phase, and instantaneous frequency as time-frequency domain indicators. Power spectrum estimation is performed on each A-Scan signal, and the total broadband energy, main frequency band energy, low frequency band energy, peak frequency, and center frequency are calculated as frequency domain indicators.

[0015] Preferably, in step S2, the method for obtaining the ground-penetrating radar-based prediction model for the depth of the frozen surface of the roadbed includes: Based on the ground-penetrating radar database of roadbed frozen surface depth, a random segmentation method was used to divide the training set and the test set. A sparrow search algorithm is established to optimize the number of convolutional kernels, convolutional layer size, number of units in long short-term memory network layers, number of heads in multi-head attention layers, attention dimension, learning rate, and batch size of convolutional neural networks. The optimal hyperparameters are output through iterative updates using the sparrow search algorithm. Based on the optimal hyperparameters, a convolutional neural network, a long short-term memory network, and a multi-head attention mechanism network are constructed. The network weights are updated using the mean squared error as the loss function and the network is trained by updating the network weights based on the Adam optimizer to obtain a ground-penetrating radar model for predicting the depth of frozen surfaces on roadbeds.

[0016] Preferably, in step S3, the method for obtaining the feature values ​​of the dataset to be detected includes: Ground penetrating radar (GPR) is used to collect echo signal data, and the echo signal data is preprocessed to obtain a basic GPR time-domain dataset. Hilbert transform and spectral estimation analysis are performed on the basic GPR time-domain dataset to obtain eigenvalues ​​in the time domain, frequency domain, and time-frequency domain.

[0017] This invention also provides a system for detecting the freezing surface depth of seasonally frozen soil subgrade based on ground-penetrating radar signal characteristics. The system is used to implement the above method and includes: The segmentation module is used to divide the inspection unit according to the inspection section of the frozen soil subgrade to be inspected; The module is used to build a prediction model for the depth of the frozen surface of the roadbed identified by ground penetrating radar based on the detection unit; The detection module is used to obtain the feature values ​​of the detection dataset based on the detection section of the seasonally frozen soil subgrade, and input the detection dataset into the ground penetrating radar to identify the depth prediction model of the frozen surface of the subgrade to obtain the depth of the frozen surface of the seasonally frozen soil subgrade.

[0018] Compared with the prior art, the beneficial effects provided by the present invention are as follows: (1) The method of the present invention can use vehicle-mounted or hand-push ground-penetrating radar, and the maximum efficiency of detecting the freezing surface depth of the seasonally frozen soil subgrade can reach 60km / h, which is more than 3 times higher than the detection efficiency of the existing methods.

[0019] (2) The method of the present invention has strong adaptability and can be used on all types of two-dimensional or three-dimensional ground penetrating radars.

[0020] (3) Compared with traditional methods such as ground temperature measurement, excavation, and core drilling, the method of the present invention belongs to the area non-destructive measurement, which does not damage the original structure of the roadbed and can realize large-scale continuous detection of the freezing surface depth of the seasonally frozen soil roadbed.

[0021] (4) The method of the present invention makes full use of the time domain, frequency domain and time-frequency domain information in the ground penetrating radar data, and has higher accuracy than existing radar detection and large-scale remote sensing detection.

[0022] (5) The method of the present invention uses the sparrow search algorithm to optimize the hyperparameters of CNN and LSTM models. CNN is used to extract local features from the original time series data, and LSTM is used to model the long-term dependencies of the data. The results fully reflect the multi-level features of the ground penetrating radar data time series and have better generalization performance. Attached Figure Description

[0023] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram illustrating the implementation steps of an embodiment of the present invention; Figure 2 This is a schematic diagram of the numerical simulation model of the ground-penetrating radar according to an embodiment of the present invention; Figure 3 This is a schematic diagram of an indoor test of ground-penetrating radar according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the field testing area for permafrost subgrade according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the A-Scan signal of the ground-penetrating radar according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the SSA-CNN-LSTM-multi-head attention network process according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the sparrow search algorithm in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Example 1 This embodiment provides a method for detecting the depth of the frozen surface of a seasonally frozen soil subgrade based on the characteristics of ground-penetrating radar signals. The steps are as follows: Figure 1 As shown. Specifically includes: S1. Divide the subgrade into testing units based on the frozen soil section to be tested.

[0028] S2. Based on the detection unit, construct a ground-penetrating radar model for predicting the depth of the frozen surface of the roadbed.

[0029] S201. In this embodiment, the width of the ground-penetrating radar device is 1.5m. In combination with the actual engineering detection accuracy requirements, the size of the detection unit is set to 1.5m×1.5m.

[0030] S202. Based on the detection units, construct a database of ground-penetrating radar identification of roadbed freezing surface depth.

[0031] (1) A scaled-down model box measuring 1.5m × 1.5m × 2m is installed indoors. The model box has a temperature control function with a temperature adjustment range of -10℃ to 10℃. Figure 3 As shown. A 1.6m thick water-bearing soil subgrade was filled inside the model box. The temperature of the model box was adjusted to -5℃ and maintained for 24 hours. Then, ground-penetrating radar was used to collect echo signals, and the depth of the frozen surface was determined by excavation. This experiment was repeated 30 times to obtain the ground-penetrating radar dataset for the indoor scaled-down experiment.

[0032] (2) A simulation model of the seasonally frozen soil subgrade was built using the finite difference software GPRMax. The model size was 1.5m × 7.5m × 2m. Figure 2 As shown in the figure. Ground-penetrating radar (GPR) echoes were extracted at different freezing depths to obtain a numerical simulation GPR dataset containing 50 samples. Table 1 shows the electromagnetic parameters of the roadbed material used in the numerical simulation. The antenna excitation source was a Ricker wavelet at a frequency of 450 MHz, the channel spacing was 0.02 m, and the transmit / receive antenna spacing was 0.05 m. The electromagnetic parameter settings for the material in the GPRMax numerical simulation are shown in Table 1.

[0033] Table 1 .

[0034] (3) Ground-penetrating radar echo data were obtained through field measurements in actual frozen soil subgrade engineering. The freezing surface depth of the frozen soil subgrade was determined at the center of each detection unit using methods such as excavation and core drilling, resulting in a measured ground-penetrating radar dataset containing 450 samples, such as... Figure 4 As shown, data alignment between radar echoes and the depth of the frozen surface is achieved through coordinate positioning.

[0035] S203. Preprocess the ground-penetrating radar echo data obtained in step S202, including time gain, zero-time correction, background removal, and filtering, to obtain the ground-penetrating radar time-domain basic dataset.

[0036] S204. Perform Hippotransform and spectral estimation analysis on the ground-penetrating radar data obtained in S203 to obtain 13 eigenvalues ​​in the time domain, frequency domain, and time-frequency domain. Combined with the freezing surface depth of the seasonally frozen soil subgrade corresponding to S202, construct a ground-penetrating radar database for identifying the freezing surface depth of the subgrade.

[0037] (1) For each A-Scan signal of the time-domain basic dataset obtained in step S203 x ( n )(like Figure 5 The calculations (as shown) yield five time-domain metrics: peak amplitude, average amplitude, average energy, amplitude squared difference, and amplitude kurtosis. The time-domain metrics within a detection unit are the average of the calculated values ​​of all A-Scan signals within that unit.

[0038] The time-domain index is calculated as follows: Peak amplitude: ; Average amplitude: ; Total Energy: ; Amplitude squared difference: ; Amplitude kurtosis: ; In the formula, N Let n be the total number of A-Scan channels within the detection unit, n = 0, 1, 2, …, N .

[0039] (2) For each A-Scan signal of the ground-penetrating radar time-domain basic dataset obtained in step S203 x ( n Performing the Hibert transform yields the corresponding Hibert transform function. ,Will x (n )and Combined construction of analytical signals f ( n The system calculates three time-frequency domain metrics: instantaneous amplitude, instantaneous phase, and instantaneous frequency. The time-frequency domain metric for each detection unit is the average of the calculated values ​​of all A-Scan signals within it.

[0040] The Hilbert transform method is as follows: To carry out x ( n Fourier transform: Constructing a Hiffel filter: Hilbert transform function: Analyzing the signal: The time-frequency domain index is calculated as follows: Instantaneous amplitude: Instantaneous phase: Instantaneous frequency: (3) A-Scan signals of the ground-penetrating radar time-domain basic dataset obtained in step S203 x ( n Power spectrum estimation is performed, and five frequency domain indices are calculated: total broadband energy, main frequency band energy, low frequency band energy, peak frequency, and center frequency.

[0041] Estimated power spectral density: ; Energy compensation factor of window function: ; Total bandwidth energy: ; Main frequency band energy: ; Low-frequency energy: ; Peak frequency: ; Average frequency: .

[0042] In the formula, x (n) represents a single A-Scan signal; fs The sampling frequency is determined by the Nyquist theorem, and is set to 2GHz in this embodiment; w ( n ) is a window function; L This represents the total number of sub-intervals into which the entire A-Scan signal is divided; MThe number of sampling points contained in each sub-interval; fk The actual physical frequency, In this embodiment f min is 0. f max fs / 2; k This is the integer index number in the discrete Fourier transform result; fc The center frequency of the ground-penetrating radar is 450MHz in this embodiment; fcutoff The low-frequency cutoff frequency is selected in this embodiment. fc / 3; To distinguish frequencies, BW is the antenna bandwidth, which is -10dB in this embodiment.

[0043] S205. A random segmentation method is used to divide the ground-penetrating radar (GPR) database for identifying the depth of frozen surfaces on roadbeds. Each detection area is a data sample, with characteristic values ​​including peak amplitude, average amplitude, total energy, squared difference of amplitude, amplitude kurtosis, broadband total energy, main band energy, low band energy, peak frequency, center frequency, instantaneous amplitude, instantaneous phase, and instantaneous frequency. The target value is the depth of frozen surfaces. 80% of the samples in the database are used as the training set, and 20% are used as the test set.

[0044] S206. A Sparrow Search + Convolutional Neural Network + Long Short-Term Memory Network + Multi-Head Attention Mechanism (SSA-CNN-LSTM-MATT) algorithm is used to train the training set to obtain a ground-penetrating radar model for predicting the depth of frozen roadbed surfaces. Figure 6 As shown.

[0045] (1) Establish a sparrow search algorithm to optimize the hyperparameters of the subsequent network. In this embodiment, the number of convolutional kernels in the CNN layer is included. Cf Convolutional layer size Ck Number of LSTM layer units Lh Multi-head attention layer count Ah Attention dimension Ad Learning rate η Batch size B, such as Figure 7 As shown.

[0046] The position of a sparrow is represented as: Fitness function: Sparrow finder location update: Mahjong joiner position update: Sparrow Watcher Location Update: Where Xbest is the current global optimal position; β K is a random number that controls the step size; fi , fg , fw These are the current individual fitness, the global optimum, and the worst fitness, respectively.

[0047] Through iterative updates, the hyperparameters Xbest of the optimal solution are finally output.

[0048] (2) Establish a CNN-LSTM multi-head attention network The network is constructed using the optimal hyperparameters obtained from SSA. The input samples are... , where B is the size of the batch.

[0049] Extracting local patterns using one-dimensional convolution: In the formula, Wconv These are the convolution kernel weights, with a size of Ck The quantity is Cf ; bconv For kernel bias; Xreshape There are B samples, each of which is a sequence of length 13 and has 1 channel. Zconv The output shape is Where L' consists of the number of features and hyperparameters Cf Sure.

[0050] The output of CNN Zconv Input LSTM, assuming the number of hidden units in LSTM is... Lh : The final hidden state sequence of LSTM As input, where dmo = Lh .

[0051] The multi-head attention mechanism is as follows: In the formula, Q=HWQ, K=HWK, V=HWV, and the projection matrix is... , Ah The number of heads in a multi-head attention layer.

[0052] Global average pooling is applied to the sequence dimension of the attention output to obtain a vector representation: The predicted value is output through the fully connected layer: Finally, the network weights are updated based on the Adam optimizer, using MSE as the loss function.

[0053] S3. Based on the detection section of the seasonally frozen soil subgrade to be detected, obtain the feature values ​​of the dataset to be detected, and input the dataset to be detected into the ground penetrating radar identification model for predicting the depth of the frozen surface of the subgrade to obtain the depth of the frozen surface of the seasonally frozen soil subgrade.

[0054] S301. Divide the detection unit on the frozen soil subgrade into detection units of the same size as in step S201. Use three-dimensional ground penetrating radar to collect echo signal data. Through time gain, zero-time correction, background removal and filtering as in step S202, obtain the basic time domain dataset of ground penetrating radar.

[0055] S302. Perform row Hilbert transform and spectral estimation analysis on the data to be detected obtained in S301 to obtain the same 13 feature values ​​as in S204, forming the dataset to be detected.

[0056] S303. Input the dataset to be detected from step S302 into the SSA-CNN-LSTM-multi-head attention network model obtained in step S205 to achieve rapid and non-destructive detection of the freezing surface depth of the permafrost subgrade. Table 2 shows a partial comparison between the estimated value and the actual core drilling freezing depth.

[0057] Table 2 .

[0058] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for detecting the depth of the frozen surface of seasonally frozen soil subgrade based on ground-penetrating radar signal characteristics, characterized in that, include: S1. Divide the subgrade into testing units based on the frozen soil section to be tested; S2. Based on the detection units, construct a ground-penetrating radar model for predicting the depth of frozen surfaces on roadbeds; S3. Based on the detection section of the seasonally frozen soil subgrade to be detected, obtain the feature values ​​of the dataset to be detected, and input the dataset to be detected into the ground penetrating radar identification model for predicting the depth of the frozen surface of the subgrade to obtain the depth of the frozen surface of the seasonally frozen soil subgrade.

2. The method for detecting the freezing surface depth of seasonally frozen soil subgrade based on ground-penetrating radar signal characteristics according to claim 1, characterized in that, S2 includes: constructing a ground-penetrating radar (GPR) database for identifying the depth of frozen roadbed surfaces based on the detection unit; optimizing the hyperparameters of the convolutional neural network, long short-term memory network, and multi-head attention mechanism using the sparrow search algorithm; training the network using the optimized hyperparameters to obtain a GPR prediction model for identifying the depth of frozen roadbed surfaces.

3. The method for detecting the freezing surface depth of seasonally frozen soil subgrade based on ground-penetrating radar signal characteristics according to claim 2, characterized in that, In step S2, the method for constructing a ground-penetrating radar database for identifying the depth of frozen surfaces on roadbeds includes: Based on the detection unit, ground-penetrating radar is used to collect echo signal data, and the echo signal data is preprocessed to obtain the ground-penetrating radar time-domain basic dataset. Based on the ground-penetrating radar time-domain basic dataset, Hippotransform and spectral estimation analysis are performed to obtain the feature values ​​in the time domain, frequency domain, and time-frequency domain. Combined with the freezing surface depth of the seasonally frozen soil subgrade corresponding to the detection unit, a ground-penetrating radar identification subgrade freezing surface depth database is constructed.

4. The method for detecting the freezing surface depth of seasonally frozen soil subgrade based on ground-penetrating radar signal characteristics according to claim 3, characterized in that, Methods for obtaining eigenvalues ​​in the time domain, frequency domain, and time-frequency domain include: Based on the ground-penetrating radar time-domain basic dataset, the peak amplitude, average amplitude, average energy, amplitude squared difference, and amplitude kurtosis are calculated for each A-Scan signal as time-domain indicators. Perform a Hibbert transform on each A-Scan signal and combine it with the original signal to construct an analytic signal. Calculate the instantaneous amplitude, instantaneous phase, and instantaneous frequency as time-frequency domain indicators. Power spectrum estimation is performed on each A-Scan signal, and the total broadband energy, main frequency band energy, low frequency band energy, peak frequency, and center frequency are calculated as frequency domain indicators.

5. The method for detecting the freezing surface depth of seasonally frozen soil subgrade based on ground-penetrating radar signal characteristics according to claim 1, characterized in that, In step S2, the method for obtaining the ground-penetrating radar (GPR) prediction model for the depth of the frozen surface of the roadbed includes: Based on the ground-penetrating radar database of roadbed frozen surface depth, a random segmentation method was used to divide the training set and the test set. A sparrow search algorithm is established to optimize the number of convolutional kernels, convolutional layer size, number of units in long short-term memory network layers, number of heads in multi-head attention layers, attention dimension, learning rate, and batch size of convolutional neural networks. The optimal hyperparameters are output through iterative updates using the sparrow search algorithm. Based on the optimal hyperparameters, a convolutional neural network, a long short-term memory network, and a multi-head attention mechanism network are constructed. The network weights are updated using the mean squared error as the loss function and the network is trained by updating the network weights based on the Adam optimizer to obtain a ground-penetrating radar model for predicting the depth of frozen surfaces on roadbeds.

6. The method for detecting the freezing surface depth of seasonally frozen soil subgrade based on ground-penetrating radar signal characteristics according to claim 1, characterized in that, In step S3, the method for obtaining the feature values ​​of the dataset to be detected includes: Ground penetrating radar (GPR) is used to collect echo signal data, and the echo signal data is preprocessed to obtain a basic GPR time-domain dataset. Hilbert transform and spectral estimation analysis are performed on the basic GPR time-domain dataset to obtain eigenvalues ​​in the time domain, frequency domain, and time-frequency domain.

7. A system for detecting the depth of frozen surface of seasonally frozen soil subgrade based on ground-penetrating radar signal characteristics, the system being used to implement the method described in any one of claims 1-6, characterized in that, include: The segmentation module is used to divide the inspection unit according to the inspection section of the frozen soil subgrade to be inspected; The module is used to build a prediction model for the depth of the frozen surface of the roadbed identified by ground penetrating radar based on the detection unit; The detection module is used to obtain the feature values ​​of the detection dataset based on the detection section of the seasonally frozen soil subgrade, and input the detection dataset into the ground penetrating radar to identify the depth prediction model of the frozen surface of the subgrade to obtain the depth of the frozen surface of the seasonally frozen soil subgrade.

Images