A concealed disease-oriented ground penetrating radar spatiotemporal registration and detection method

Abstract

The present application relates to ground penetrating radar data processing technical field, especially to a kind of ground penetrating radar space-time registration and detection method for hidden disease.It includes the following steps:S1: establish the continuous mapping relationship of radar signal and geographic space;S2: construct local trajectory index mechanism based on anisotropy measurement;S3: construct space-time joint correction space based on error prior distribution;S4: parameter inversion based on maximum a posteriori probability estimation constraint;S5: establish registration quality evaluation and signal feature decoupling mechanism;S6: construct double physical property index and adaptive discrimination model driven by dielectric priori.This application overcomes the influence of GPS positioning error and trajectory swing, realizes centimeter-level physical space alignment, significantly reduces false alarm caused by registration deviation;By introducing the physical constraint of frequency dimension, the signal-to-clutter ratio in complex urban background is improved, and the false alarm rate is significantly reduced.

Description

Technical Field

[0001] This invention relates to the field of ground-penetrating radar data processing technology, and in particular to a ground-penetrating radar spatiotemporal registration and detection method for concealed defects. Background Technology

[0002] With the rapid advancement of global urbanization, urban underground spaces have become complex networks carrying pipelines for transportation, energy, communication, and drainage. However, this vast underground system is facing multiple challenges, including aging infrastructure, changing geological environments, and high-intensity surface loads, leading to the proliferation and spread of various hidden underground defects. Problems such as road collapses, underground cavities, loose soil, and pipeline leaks not only cause huge economic losses but also pose a direct threat to public safety. Unlike visible defects such as surface cracks or ruts, hidden underground defects are often highly concealed in their early stages, developing slowly and not easily detected by traditional surface inspection methods, until they develop into catastrophic ground collapses. Therefore, how to achieve early detection, accurate location, and dynamic monitoring of hidden underground defects has become a common concern in the fields of civil engineering and urban disaster prevention and mitigation.

[0003] Among numerous non-destructive testing technologies, ground-penetrating radar (GPR) is widely recognized as one of the most effective means of detecting underground road defects due to its excellent penetration capability in non-metallic media, high-resolution shallow imaging, and sensitivity to differences in dielectric constant. In particular, vehicle-mounted array GPR systems, by integrating multi-channel antenna arrays, high-precision positioning and attitude determination systems, and high-speed data acquisition units, can perform full-coverage tomographic scanning of urban roads at normal driving speeds without disrupting traffic, greatly improving detection efficiency and data coverage.

[0004] Although vehicle-mounted ground-penetrating radar technology is relatively mature, it still faces significant technical bottlenecks in the practical detection of specific defects such as early-stage, minute leaks in underground pipelines. Because the moisture diffusion range and physical property differences caused by early leaks are small, a single static radar scan often fails to distinguish it from the background clutter of the surrounding complex underground medium. Therefore, time-series comparative analysis based on data from multiple periods in the same survey area becomes crucial for identifying these evolving defects. However, this technical approach is limited by the following issues in engineering implementation:

[0005] First, the non-repeatability of vehicle-mounted acquisition trajectories leads to spatiotemporal misalignment. Although the vehicle-mounted system is equipped with a satellite positioning module, due to the positioning errors of civilian navigation systems and the influence of road traffic flow, it is difficult for vehicles to strictly maintain a driving line consistent with the baseline trajectory during re-measurement, often resulting in nonlinear lateral swaying and deviation. This physical spatial misalignment means that the radar images acquired in two consecutive measurements are not on the same cross-section, rendering direct image comparison and analysis ineffective.

[0006] Secondly, traditional registration and detection algorithms have poor adaptability. Existing ground-penetrating radar image registration techniques mostly rely on significant feature points, such as the apex of a pipeline hyperbola or inflection points of a layer, but they are difficult to apply to uniform road sections lacking features. Meanwhile, the commonly used time-domain amplitude difference method is extremely sensitive to registration residuals; even a small spatial misalignment can generate high-amplitude false artifacts at strong reflective interfaces such as road structure layers, thus masking genuine weak leakage signals. Furthermore, existing methods often neglect the absorption characteristics of water media for high-frequency components of electromagnetic waves, i.e., the frequency redshift effect, and lack a discrimination mechanism based on multi-dimensional physical properties.

[0007] In summary, existing technologies have significant limitations in achieving accurate spatial alignment of multi-period vehicle radar data and extraction of weak water body signals. Therefore, there is an urgent need for a spatiotemporal joint registration and detection method that can adapt to lateral vehicle trajectory deviation, possess coarse-to-fine hierarchical alignment capabilities, and integrate multi-dimensional physical property characteristics such as amplitude and frequency. Summary of the Invention

[0008] The purpose of this invention is to overcome the defects of the prior art and provide a ground-penetrating radar spatiotemporal joint hierarchical registration and detection method for detecting hidden diseases, which has the ability to improve the identification accuracy of ground-penetrating radar for early leakage diseases.

[0009] The objective of this invention can be achieved through the following technical solutions:

[0010] A ground-penetrating radar spatiotemporal registration and detection method for concealed defects includes the following steps:

[0011] S1: Establish a continuous mapping relationship between radar signals and geographic space.

[0012] S2: Construct a local trajectory indexing mechanism based on anisotropic measurement.

[0013] S3: Construct a spatiotemporal joint correction space based on the prior distribution of errors.

[0014] S4: Parameter inversion based on maximum a posteriori probability estimation constraints.

[0015] S5: Establish a registration quality assessment and signal feature decoupling mechanism.

[0016] S6: Construct a dielectric prior-driven dual physical property index and adaptive discrimination model.

[0017] Beneficial effects

[0018] Compared with the prior art, the present invention has the following advantages:

[0019] (1) This invention achieves centimeter-level precise resetting in physical space, significantly eliminating false alarms caused by trajectory misalignment. For nonlinear lateral offsets caused by the vehicle's "snake-like movement" during re-measurement by vehicle-mounted ground-penetrating radar, existing technologies typically rely solely on GPS coordinate interpolation, which is insufficient to correct meter-level physical misalignments. This invention creatively constructs a hierarchical registration architecture based on "physical geometric constraints." First, it utilizes the BallTree high-dimensional spatial index combined with the physical width constraint of the array radar to lock the search range of massive trajectory points within the physically feasible domain. Then, it uses the maximum a posteriori probability estimation under the prior positioning error to transform the complex trajectory correction into a constrained parameter inversion problem. Comparative experimental data shows that compared with the commonly used GPS linear interpolation method, this invention reduces the average spatial registration error between the two data periods from 45.2 cm to 4.8 cm, and increases the average waveform correlation coefficient from 0.64 to 0.89. This method fundamentally solves the problem of high-amplitude "artifacts" generated during image differencing due to spatial alignment deviations in traditional methods, effectively avoiding false alarms.

[0020] (2) A dual-criteria model based on multi-dimensional physical properties was established, which significantly improved the identification accuracy of hidden and minor seepage defects. Addressing the challenges of early-stage seepage defects being characterized by low moisture content, weak signals, and susceptibility to being masked by uneven roadbed media, this invention overcomes the limitations of traditional methods that rely solely on a single indicator of "amplitude intensity." This invention innovatively introduces the instantaneous frequency attribute characterizing spectral redshift, utilizing the strong absorption characteristics of water media for high-frequency electromagnetic wave components, and constructs a dual-temporal difference criterion model of "abnormal energy enhancement + abnormal frequency attenuation." Comparative experimental data shows that by introducing physical constraints in the frequency dimension, this invention improves the signal-to-noise ratio by 12.8 dB compared to the traditional single-threshold method in complex urban environments; simultaneously, it effectively eliminates non-water body interference that, while exhibiting strong reflection, lacks frequency shift, significantly reducing the false alarm rate from 42.5% to 6.2%.

[0021] (3) It possesses full-process automated processing capabilities and strong engineering adaptability. The algorithm proposed in this invention can automatically complete the entire process from massive trajectory cleaning, coarse and fine graded registration to automatic defect extraction without manual selection of feature points or intervention in the registration process. In particular, the optimal offset solution method based on correlation maximization and physical prior constraints has strong robustness and can adapt to different road conditions, different acquisition speeds, and different array specifications of vehicle-mounted radar systems. Experiments have verified that its registration success rate reaches 86.8%, which is significantly better than the 65.4% of the traditional method, improving the processing efficiency and engineering practical value of massive urban road radar data. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method of the present invention;

[0023] Figure 2 This is a schematic diagram of the original B-Scan image of the ground-penetrating radar in an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of the B-Scan map of the monitoring ground-penetrating radar after spatiotemporal joint registration according to an embodiment of the present invention;

[0025] Figure 4 This is a schematic diagram of the hidden disease identification results based on the dual criterion model in an embodiment of the present invention ((a) - original B-scan radar profile, (b) - identification result). Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 some embodiments of the present invention, but not all embodiments.

[0027] Example 1

[0028] like Figure 1 As shown, this embodiment provides a ground-penetrating radar spatiotemporal registration and detection method for concealed defects, including the following steps:

[0029] S1: Establish a continuous mapping relationship between radar signals and geographic space.

[0030] This method acquires baseline and monitoring ground-penetrating radar (GPR) data for the same target area collected at different times. The original discrete trajectory data undergoes planar projection processing, and the projected trajectories are then reconstructed into a continuous sequence, establishing a one-to-one correspondence between radar scan channels and geospatial coordinates. The output is a continuous spatial coordinate sequence using a unified spatiotemporal reference, providing standardized input data for subsequent spatial registration.

[0031] S2: Construct a local trajectory indexing mechanism based on anisotropic measurement.

[0032] A local moving coordinate system is established along the tangent and normal directions of the radar trajectory. The spatiotemporal corridor constraint is constructed using the vehicle kinematic boundary and the physical width of the array radar. Within this feasible region, an anisotropic distance metric function that integrates longitudinal velocity and lateral deviation weights is defined to retrieve the best matching neighborhood of the monitoring trajectory relative to the reference trajectory, thereby overcoming the registration error caused by non-uniform sampling and non-rigid deformation of the trajectory.

[0033] S3: Construct a spatiotemporal joint correction space based on the prior distribution of errors. For the random jitter of radar signal acquisition, perform subspace alignment based on surface reflection phase; at the same time, based on the statistical distribution law of sensor positioning error and array geometric parameters, derive the probabilistic feasible region of lateral offset parameter, and strictly restrict the solution space of subsequent optimization problems within the physical confidence interval.

[0034] S4: Parameter inversion based on maximum a posteriori probability estimation constraints.

[0035] The distribution characteristics of GPS positioning error are modeled as a prior probability regularization term, and the similarity of radar waveforms is modeled as a likelihood function. A global objective function that integrates physical prior and data observation is constructed. The objective function is solved by an optimization algorithm to obtain the optimal physical offset in the sense of maximum a posteriori probability, thereby achieving the optimal balance between waveform coherence and positioning statistics.

[0036] S5: Establish a registration quality assessment and signal feature decoupling mechanism.

[0037] The validity of the registration results is determined by using correlation evaluation index. Only when there is a valid physical overlap between the benchmark data and the monitoring data is the radar echo signal that has completed spatial registration extracted. The echo signal is then processed by signal transformation to achieve feature decoupling, and the instantaneous amplitude attribute characterizing the reflected energy intensity and the instantaneous frequency attribute characterizing the medium attenuation characteristics are obtained.

[0038] S6: Construct a dielectric prior-driven dual physical property index and adaptive discrimination model.

[0039] Based on the dielectric response mechanism of water-containing media, an energy focusing index characterizing echo energy and a redshift index characterizing spectral shift are constructed respectively. An adaptive floating threshold is established using the statistical distribution characteristics of baseline data, and an anomaly discrimination model with dual constraints is constructed to achieve quantitative identification and spatial location of hidden micro-leakage signals.

[0040] Specifically,

[0041] In step S1, the reference ground-penetrating radar data refers to the radar echo spectrum and its synchronously collected GPS trajectory data obtained during the first general survey of the target pipeline area using a vehicle-mounted array ground-penetrating radar system at a historical time (e.g., January 2023); the monitoring ground-penetrating radar data refers to the data obtained during the resurvey of the same area using the same or similar vehicle-mounted ground-penetrating radar system at the current time (e.g., June 2023), both of which include radar waveform data and GPS positioning data.

[0042] The planar projection processing involves converting the latitude and longitude coordinates of the original GPS data. ( Latitude Convert longitude to universal transverse Mercator projection coordinates ( The x-axis is... (Using the vertical axis) to eliminate the effects of Earth's curvature.

[0043] The continuous reconstruction is based on radar channel numbers. Using the independent variable, linear interpolation is performed on the projected coordinates. Due to the time-interval scanning of ground-penetrating radar and other sensors, and the non-uniform time sampling of GPS, there is a mismatch in sampling times between the two. Through interpolation calculation, precise two-dimensional planar coordinates are assigned to each radar scan signal, ultimately outputting a continuous spatial coordinate sequence corresponding one-to-one with each radar data point. This provides a unified coordinate reference for subsequent spatial registration, among which The first The horizontal and vertical coordinates of the radar.

[0044] In step S2, the specific implementation process of the local trajectory indexing mechanism is as follows:

[0045] The construction of the local coordinate system involves selecting trajectory points as reference points from the continuous spatial coordinate sequence of the ground-penetrating radar output in step S1. And select corresponding candidate points from the continuous spatial coordinate sequence of the ground-penetrating radar as monitoring points. , based on the baseline trajectory point With the origin as the starting point, the longitudinal direction along the tangent of the trajectory is defined as... The normal direction is defined as the transverse direction. , will monitor points Coordinate mapping to this local system ;

[0046] The spatiotemporal corridor constraint defines a physically feasible region and retains only candidate points that satisfy the following dual constraints:

[0047] Lateral constraints are ,in This is the lateral deviation. This refers to the physical width of the array radar. For positioning tolerance;

[0048] Vertical constraints are ,in For longitudinal deviation, The maximum speed of the vehicle. The time interval between two consecutive scans of the ground-penetrating radar is used to limit the maximum longitudinal drift of the vehicle.

[0049] Anisotropic nearest neighbor retrieval employs the BallTree algorithm to construct a high-dimensional spatial index for benchmark points, and calculates monitoring points within the spatiotemporal corridor constraints. With reference point Anisotropic distance :

[0050]

[0051] in These are the normalized weighting factors for the vertical and horizontal directions, respectively. (Selection) The smallest point is selected as the best matching point.

[0052] Simultaneously, extract the lateral deviation corresponding to the neighborhood of the best matching point. The initial lateral offset serves as the coarse spatial registration of the reference ground-penetrating radar and the monitoring ground-penetrating radar data. Pass it on to the next step.

[0053] In step S3, the specific implementation process of the spatiotemporal joint correction space is as follows:

[0054] Subspace alignment based on surface reflection phase, i.e., vertical phase alignment, addresses radar antenna jitter caused by road surface undulations. It extracts the initial wave jump point of each signal, i.e., the maximum amplitude phase point, and uses a cross-correlation alignment algorithm to uniformly correct the surface reflection times of all scan channels to zero time. Eliminate vertical non-rigid deformation caused by jitter in the acquisition system;

[0055] The probabilistic feasible region is based on the statistical characteristics of GPS positioning errors. Physical width of array radar Derive the confidence boundaries for parameter inversion. Define the lateral offset. search range for:

[0056]

[0057] in, This refers to the initial lateral offset calculated at the optimal matching point in step S2. Here is the confidence coefficient. This represents the standard deviation of GPS positioning error.

[0058] In step S4, the specific implementation process of parameter inversion is as follows:

[0059] Prior modeling will accurately measure the lateral offset. Treat it as a random variable, following a Gaussian distribution ,in This is the coarse registration offset. This represents the standard deviation of GPS positioning error.

[0060] Construct the objective function and define the global cost function for fine registration. The weighted sum of the relevance term and the regularization term:

[0061]

[0062] in, To monitor the lateral offset of the baseline data The Pearson correlation coefficient between radar waveforms in the overlapping region is given. For physical penalty terms based on GPS error, Hyperparameters for adjusting weights;

[0063] The optimization algorithm solves for the optimal physical offset within the feasible region. .

[0064] In step S5, the specific implementation process of the registration quality assessment and signal feature decoupling mechanism is as follows:

[0065] The registration quality assessment is based on the optimal physical offset obtained in step S4. Lateral positional correction is performed on the monitoring ground-penetrating radar data to achieve fine-grained registration in physical space. A correlation threshold is set. If the calculated global maximum correlation coefficient If the signal is positive, it is determined that there is valid overlap, and the registered single-channel radar waveform within the overlapping area is extracted as the real signal. ;

[0066] When it is determined that there is valid overlap, the real signal Constructing analytic signals using Hilbert transform The corresponding expression is:

[0067]

[0068] in Represents the Hilbert transform operator. The unit is imaginary. The instantaneous amplitude... and instantaneous frequency The calculation expressions are as follows:

[0069]

[0070]

[0071] In step S6, the specific implementation process of indicator construction and discrimination model is as follows:

[0072] Energy Focus Index :

[0073]

[0074] in For the energy of the current moment, The background reference energy characterizes the energy abrupt change caused by dielectric contrast.

[0075] Redshift index :

[0076]

[0077] in The frequency at the current moment. Background reference frequency, characterizing the amount of the center frequency drifting to lower frequencies relative to the background spectrum;

[0078] Adaptive threshold setting: Calculate the mean of the index series of defect-free road sections in the baseline period. (Energy Mean) and frequency mean ) and standard deviation (Energy Standard Deviation) and frequency standard deviation ), set a dynamic threshold. The energy enhancement threshold is The frequency attenuation threshold is ,in These are the confidence coefficients for energy and frequency, respectively.

[0079] The dual criterion expression is:

[0080]

[0081] In the formula, The result of the binary determination of hidden diseases (1 represents the presence of abnormality, 0 represents no abnormality). and These are the energy focus index and redshift index at the current moment, respectively. and These are the corresponding adaptive thresholds. Leakage is determined only when the energy index is significantly higher than the upper limit and the redshift index is significantly lower than the lower limit.

[0082] The specific data processing procedure of this embodiment will be described in detail below:

[0083] This embodiment selects an in-service main road (asphalt concrete pavement) in a city in northern China as the test site. A diameter [missing information] was buried under the pavement in the test area. The rainwater pipeline is buried at a depth of approximately 0.5m. This section of road has historically experienced ground subsidence due to leaks at pipeline joints. The method described in this invention was used for full-process detection and verification.

[0084] The system employs a vehicle-mounted three-dimensional array ground-penetrating radar system. The antenna array contains 20 channels with an equivalent physical width of 1.5m. The center frequency is 400 MHz, the time window range is 60 ns, the sampling point count is 512 points / channel, the triggering method is distance wheel triggering, and the channel spacing is set to 0.05m.

[0085] In step S1, baseline ground-penetrating radar data (collected in January 2023) and monitoring ground-penetrating radar data (collected in June 2023) for an underground pipeline area along a municipal road are acquired. The sampling frequency of the original GPS positioning data is 1Hz, and the center frequency of the radar antenna is 400MHz. After coordinate projection and interpolation, a continuous spatial coordinate sequence corresponding one-to-one with the radar channel number is generated.

[0086] In step S2, the BallTree spatial indexing algorithm is used to perform a neighborhood search on the monitoring trajectory. Vertical velocity weights are set. Lateral deviation weight The calculation results show that at a distance of approximately 240m from the starting point, the initial lateral offset of the monitoring trajectory relative to the baseline trajectory was calculated. exist The fluctuations indicate that the vehicle deviated significantly to the right in this section of the road.

[0087] In steps S3 and S4, the standard deviation of GPS positioning error is set. Confidence coefficient Constructing a dynamic search interval And it is limited to the physical coverage area of ​​the radar array. Set the lateral offset grid search step size. And calculate the objective function. The regularization parameter The algorithm converges to the optimal offset. Through registration, the correlation coefficient was increased from 0.32 to 0.89, successfully eliminating the spatial incoherence caused by trajectory misalignment.

[0088] In step S5, Hilbert transform and property analysis are performed on the aforementioned suspected abnormal regions. The dielectric constant is changed from the background value. (Dry soil) rises to (Moist soil), while displaying the base period signal frequency as During the monitoring period, the main frequency of the signal dropped to Frequency shift .

[0089] In step S6, background statistical parameters, including the mean energy background value, are calculated using data from defect-free road sections during the baseline period. Standard deviation Set threshold Energy index of anomaly points in actual measurements The frequency background mean was determined to be an energy anomaly. Standard deviation Set threshold Instantaneous frequency of abnormal points in measured values The frequency was determined to be abnormal. Based on the comprehensive assessment: this area simultaneously meets the conditions of both energy enhancement and frequency attenuation, and the algorithm automatically marks it as a "high-confidence leakage problem".

[0090] Drilling and sampling confirmed that there was a disconnect at the rainwater pipe interface at a depth of 0.58m underground. The surrounding soil was in a fluid state, and the measured volumetric water content was 27.3%, which was consistent with the radar inversion results.

[0091] In this embodiment, the original B-Scan map of the ground penetrating radar is as follows: Figure 2 As shown.

[0092] In this embodiment, the B-Scan map of the ground-penetrating radar after spatiotemporal joint registration is used as follows: Figure 3 As shown.

[0093] In this embodiment, a schematic diagram of the hidden disease identification results based on the dual-criteria model is shown below. Figure 4 As shown in Figure 4. Figure 4(a) shows the original B-scan radar profile during the monitoring period. The red dashed box in the figure represents the "criteria search area" set by the algorithm. This search area constructs a local calculation window centered on the target pipeline (red dot). The algorithm extracts instantaneous amplitude and instantaneous frequency attributes only within this defined area, thereby eliminating interference from shallow roadbed structures and other irrelevant clutter, improving detection efficiency and accuracy. Figure 4(b) shows the final binarized identification result output by the algorithm after integrating both energy enhancement and frequency redshift indicators within this criteria search area. The white pixel connected area in the figure represents the spatial distribution range of successfully extracted minor underground pipeline leaks.

[0094] Comparative experiments and verification of beneficial effects

[0095] To objectively verify the effectiveness of the "spatiotemporal joint hierarchical registration and dual criterion detection method" proposed in this invention in complex engineering environments, this embodiment designs a comparative experiment with commonly used methods in the prior art.

[0096] This experiment selected five segments of vehicle-mounted ground-penetrating radar (GPR) inspection data from the aforementioned municipal roads and surrounding areas with similar road conditions as the test set, with a total survey line length of 2.5 km. Data acquisition was performed using a 400 MHz center frequency vehicle-mounted array antenna with a lane spacing of 0.05 m. The true values ​​of underground defects for all test road segments were determined jointly by pipeline endoscopic inspection reports and excavation verification records for some locations to ensure the reliability of the evaluation benchmark. The method of this invention adopts an adaptive threshold strategy, while the comparison method uses an empirically fixed threshold. All methods were run on the same computing platform and with the same dataset.

[0097] The conventional processing flow commonly used in current engineering practice is selected as the existing method for comparison, which mainly relies on spatial distance for coarse trajectory alignment and uses a single instantaneous amplitude attribute to set a fixed threshold to extract abnormal areas.

[0098] To quantitatively evaluate algorithm performance, the following evaluation metrics are defined:

[0099] (a) Average Spatial Registration Error (MSRE): Using known pipeline centerline feature points or manually marked corresponding points as control points, calculate the average value (unit: cm) of the lateral physical distance deviation between the registered monitoring image and the reference image.

[0100] (b) Average correlation coefficient of waveform: Within the effective physical overlap area after registration, the Pearson correlation coefficient of the corresponding channel A-scan waveform is calculated and the mean value is taken to characterize the degree of waveform recovery.

[0101] (c) Signal-to-noise ratio (SNR): defined as... .in, This represents the average energy within the leakage anomaly region, verified by truth values. The average energy is the energy of the disease-free background area at the same depth layer and 1.0m away from the anomaly area.

[0102] (d) False Alarm Rate (FAR): Defined as the proportion of areas that are judged as abnormal by the algorithm but are actually free of disease to the total number of all detected abnormal areas.

[0103] (e) Composite metric (F1-Score): The harmonic mean calculated based on precision and recall, using the formula: The results of the comparative experiments are shown in Table 1.

[0104] Table 1 Comparison of Experimental Results

[0105]

[0106] As shown in the table, compared with existing methods, the method of this invention significantly reduces the average spatial registration error (MSRE) from 45.2 cm to 4.8 cm, while increasing the average waveform correlation coefficient from 0.64 to 0.89 and the registration success rate from 65.4% to 86.8%, effectively overcoming the misalignment distortion caused by the nonlinear oscillation of the vehicle-mounted acquisition trajectory. Furthermore, this invention employs a dual physical attribute criterion model based on energy and frequency, effectively suppressing interference from strong reflective background clutter, increasing the signal-to-noise ratio (SNR) from 8.5 dB to 21.3 dB, and reducing the false alarm rate (FAR) from 42.5% to 6.2%. In summary, the method of this invention significantly reduces false alarms caused by registration deviation and background clutter, greatly improving the accuracy and robustness of automatic identification of early-stage, hidden, and minor leaks.

[0107] Table 2 Symbol Annotation Table

[0108]

Claims

1. A ground-penetrating radar spatiotemporal registration and detection method for concealed defects, characterized in that, Includes the following steps: S1: Establish a continuous mapping relationship between radar signals and geographic space; S2: Construct a local trajectory indexing mechanism based on anisotropic measurement; S3: Construct a spatiotemporal joint correction space based on the prior distribution of errors; S4: Parameter inversion based on maximum a posteriori probability estimation constraints; S5: Establish a registration quality assessment and signal feature decoupling mechanism; S6: Construct a dielectric prior-driven dual physical property index and adaptive discrimination model; Step S2 specifically involves: A local moving coordinate system is established along the tangent and normal directions of the radar trajectory. The spatiotemporal corridor constraint is constructed using the vehicle kinematic boundary and the physical width of the array radar. Within this feasible region, an anisotropic distance metric function that integrates longitudinal velocity and lateral deviation weights is defined to retrieve the best matching neighborhood of the monitoring trajectory relative to the reference trajectory, thereby overcoming the registration error caused by non-uniform sampling and non-rigid deformation of the trajectory. The specific implementation process of the local trajectory indexing mechanism is as follows: Constructing a local coordinate system: Select trajectory points as reference points from the continuous spatial coordinate sequence of the reference ground-penetrating radar output in step S1. And select corresponding candidate points from the continuous spatial coordinate sequence of the ground-penetrating radar as monitoring points. , based on the baseline trajectory point With the origin as the starting point, the longitudinal direction along the tangent of the trajectory is defined as... The normal direction is defined as the transverse direction. , will monitor points Coordinates mapped to this local coordinate system ; The spatiotemporal corridor constraint defines a physically feasible region and retains only candidate points that satisfy the following dual constraints: Lateral constraints are ,in This is the lateral deviation. This refers to the physical width of the array radar. For positioning tolerance; Vertical constraints are ,in For longitudinal deviation, The maximum speed of the vehicle. The time interval between two consecutive scans by the ground-penetrating radar is used to limit the maximum longitudinal drift of the vehicle. Anisotropic nearest neighbor retrieval employs the BallTree algorithm to construct a high-dimensional spatial index for benchmark points, and calculates monitoring points within the spatiotemporal corridor constraints. With reference point Anisotropic distance : in These are the normalized weighting factors for the vertical and horizontal directions, respectively; Select The point with the smallest value is selected as the optimal matching point; simultaneously, the lateral deviation corresponding to the neighborhood of this optimal matching point is extracted. The initial lateral offset serves as the coarse spatial registration of the reference ground-penetrating radar and the monitoring ground-penetrating radar data. Pass it on to the next step.

2. The ground-penetrating radar spatiotemporal registration and detection method for concealed defects according to claim 1, characterized in that, Step S1 specifically involves: The system acquires benchmark ground-penetrating radar (GPR) data and monitoring GPR data collected at different times for the same target area. It performs planar projection processing on the original discrete trajectory data and reconstructs the projected trajectory into a continuous system, establishing a one-to-one correspondence between radar scan channels and geospatial coordinates. The output is a continuous spatial coordinate sequence using a unified spatiotemporal reference, providing standardized input data for subsequent spatial registration. The reference ground-penetrating radar data refers to the radar echo spectrum and GPS trajectory data simultaneously collected when the target pipeline area is first surveyed using a vehicle-mounted array ground-penetrating radar system at a historical time; the monitoring ground-penetrating radar data refers to the data obtained when the same or similar vehicle-mounted ground-penetrating radar system is used to re-survey the same area at the current time, and both include radar waveform data and GPS positioning data. The planar projection processing involves converting the latitude and longitude coordinates of the original GPS data. Convert to universal transverse Mercator projection coordinates To eliminate the effects of Earth's curvature; The continuous reconstruction is based on radar channel numbers. Using the independent variable, linear interpolation is performed on the projected coordinates, ultimately outputting a continuous spatial coordinate sequence corresponding one-to-one with each radar data point. This provides a unified coordinate reference for subsequent spatial registration.

3. The ground-penetrating radar spatiotemporal registration and detection method for concealed defects according to claim 1, characterized in that, Step S3 specifically involves performing subspace alignment based on surface reflection phase to address the random jitter in radar signal acquisition; simultaneously, based on the statistical distribution of sensor positioning errors and array geometric parameters, deriving the probabilistic feasible region of the lateral offset parameters, and strictly limiting the solution space of the subsequent optimization problem to the physical confidence interval. Among them, the subspace alignment based on surface reflection phase, namely vertical phase alignment, is designed to address the radar antenna jitter caused by road surface undulations. It extracts the first wave jump point of each signal, i.e. the maximum amplitude phase point, and uses the cross-correlation alignment algorithm to uniformly correct the surface reflection time of all scan channels to the zero time point, thereby eliminating the vertical non-rigid deformation caused by the jitter of the acquisition system. The probabilistic feasible region is based on the statistical characteristics of GPS positioning errors. Physical width of array radar Derive the confidence boundaries for parameter inversion; define the lateral offset. search range for: in, This refers to the initial lateral offset calculated at the optimal matching point in step S2. , is the confidence coefficient This represents the standard deviation of GPS positioning error.

4. The ground-penetrating radar spatiotemporal registration and detection method for concealed defects according to claim 3, characterized in that, Step S4 specifically involves, The distribution characteristics of GPS positioning error are modeled as a prior probability regularization term, and the similarity of radar waveforms is modeled as a likelihood function. A global objective function that integrates physical prior and data observation is constructed. The objective function is solved by an optimization algorithm to obtain the optimal physical offset in the sense of maximum a posteriori probability, thereby achieving the optimal balance between waveform coherence and positioning statistics. Prior modeling will accurately measure the lateral offset. Treat it as a random variable, following a Gaussian distribution ,in This is the coarse registration offset. This represents the standard deviation of GPS positioning error. Construct the objective function and define the global cost function for fine registration. The weighted sum of the relevance term and the regularization term: in, To monitor the lateral offset of the baseline data The Pearson correlation coefficient between radar waveforms in the overlapping region is given. For physical penalty terms based on GPS error, Hyperparameters for adjusting weights; The optimization algorithm solves for the optimal physical offset within the feasible region. .

5. The ground-penetrating radar spatiotemporal registration and detection method for concealed defects according to claim 1, characterized in that, Step S5 specifically involves: The validity of the registration results is determined by using correlation evaluation indicators. Radar echo signals that have completed spatial registration are extracted only when there is a valid physical overlap between the benchmark data and the monitoring data. The echo signal is processed by signal transformation to achieve feature decoupling, and the instantaneous amplitude attribute characterizing the reflected energy intensity and the instantaneous frequency attribute characterizing the medium attenuation characteristics are obtained. The specific implementation process of the registration quality assessment and signal feature decoupling mechanism is as follows: The registration quality assessment is based on the optimal physical offset obtained in step S4. Lateral positional correction is performed on the monitoring ground-penetrating radar data to achieve fine-grained registration in physical space; a correlation threshold is set. If the calculated global maximum correlation coefficient If the signal is positive, it is determined that there is valid overlap, and the registered single-channel radar waveform within the overlapping area is extracted as the real signal. ; When it is determined that there is valid overlap, the real signal Constructing analytic signals using Hilbert transform The corresponding expression is: in Represents the Hilbert transform operator. The unit is the imaginary number; the instantaneous amplitude and instantaneous frequency The calculation expressions are as follows: 。 6. The ground-penetrating radar spatiotemporal registration and detection method for concealed defects according to claim 5, characterized in that, Step S6 specifically involves: Based on the dielectric response mechanism of water-containing media, an energy focusing index characterizing echo energy and a redshift index characterizing spectral shift are constructed respectively. An adaptive floating threshold is established using the statistical distribution characteristics of baseline data, and an anomaly discrimination model with dual constraints is constructed to achieve quantitative identification and spatial location of hidden micro-leakage signals. The specific implementation process of indicator construction and discrimination model is as follows: Energy Focus Index : in For the energy of the current moment, The background reference energy characterizes the energy abrupt change caused by dielectric contrast. Redshift index : in The frequency at the current moment. Background reference frequency, characterizing the amount of the center frequency drifting to lower frequencies relative to the background spectrum; Adaptive threshold setting: Calculate the mean of the index series of defect-free road sections in the baseline period. Standard deviation The mean Including average energy and frequency mean The standard deviation Including energy standard deviation and frequency standard deviation Set dynamic thresholds; Energy enhancement threshold is The frequency attenuation threshold is ,in These are the confidence coefficients for energy and frequency, respectively. The dual criterion expression is: In the formula, The result is a binary representation of the hidden disease, where 1 represents the presence of an anomaly and 0 represents the absence of an anomaly. and These are the energy focus index and redshift index at the current moment, respectively. and These are the corresponding adaptive thresholds. Leakage is determined only when the energy index is significantly higher than the upper limit and the redshift index is significantly lower than the lower limit.

Images