Intelligent evaluation method for asphalt pavement structure damage based on ground penetrating radar signal processing

Abstract

The present application relates to the technical field of radar detection, and particularly relates to a kind of intelligent evaluation method of asphalt pavement structure damage based on ground penetrating radar signal processing, comprising: according to the difference between the road surface reflection center frequency of each single-channel radar data section of asphalt pavement and the local instantaneous center frequency of each time, obtain the center frequency attenuation distribution matrix, determine the path average frequency attenuation value of the minimum frequency difference path of the cumulative situation of frequency attenuation difference in center frequency attenuation distribution matrix;Noise data is applied to center frequency attenuation distribution matrix, to obtain the noise frequency attenuation matrix;Search the optimal path in noise frequency attenuation matrix, based on the difference between the minimum frequency difference path and the optimal path, obtain the path vertical deviation index;According to path average frequency attenuation value and path vertical deviation index, realize the reliable and accurate evaluation of the damage state of asphalt pavement structure.

Description

Technical Field

[0001] This invention relates to the field of radar detection technology, and specifically to an intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing. Background Technology

[0002] The service life of asphalt pavement is significantly limited by the bonding state between the asphalt surface layer and the semi-rigid base layer. Under uneven construction or repeated traffic loads, micron-sized voids easily form at the interlayer interface. These voids absorb moisture through capillary action, forming a hidden micro-moistened void layer, which is an early cause of pavement shoving, rutting, potholes, and other defects. Currently, ground-penetrating radar (GPR) is the mainstream non-destructive testing method, which detects the internal structure by emitting electromagnetic waves and receiving reflected signals. Existing technologies mostly infer the interface state by analyzing the local characteristics of the reflected waves (such as the amplitude, instantaneous frequency, or spectrum at a specific location) and attempt to establish a single mapping relationship between these local characteristics and damage. However, the asphalt pavement structure is a complex, non-homogeneous medium. The propagation of radar signals within it is affected by multiple scattering, attenuation, and interference, making the signal characteristics at a single location or at a single moment extremely unstable and random. This analysis method based on isolated features ignores the consistency information of the overall attenuation law and propagation path of radar signals in the time and space dimensions during the process of radar signals passing through the medium. As a result, the evaluation of the overall bonding state and damage degree of the interlayer interface has poor anti-interference ability and insufficient reliability, and it is difficult to stably and accurately reflect the true health status of the structure. Summary of the Invention

[0003] The technical problem that this invention aims to solve is: how to overcome the problem of unstable and inaccurate evaluation results caused by existing ground-penetrating radar detection methods relying on local and isolated signal features.

[0004] The purpose of this invention is to provide an intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing. The specific technical solution adopted is as follows:

[0005] This invention provides an intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing, comprising:

[0006] The center frequency attenuation distribution matrix is ​​obtained by comparing the surface reflection center frequency of each single radar data segment of the asphalt pavement with the local instantaneous center frequency at each moment.

[0007] Determine the path-average frequency attenuation value of the path with the minimum frequency difference in the cumulative difference matrix; the cumulative difference matrix represents the accumulation of frequency attenuation differences in the center frequency attenuation distribution matrix;

[0008] Noise data is applied to the center frequency attenuation distribution matrix to obtain a noisy frequency attenuation matrix; the noise data is obtained from the center frequency attenuation distribution matrix and the minimum frequency difference path.

[0009] Search for the optimal path in the noise frequency attenuation matrix, and obtain the path vertical deviation index based on the difference between the minimum frequency difference path and the optimal path;

[0010] The structural damage of asphalt pavement is evaluated based on the path average frequency attenuation value and the path vertical deviation index.

[0011] In an exemplary embodiment, the process of obtaining the center frequency of the road surface reflection includes:

[0012] Determine the target time window in the single-channel radar data segment;

[0013] A fast Fourier transform is performed on the radar data of the target time window to obtain the power spectral density corresponding to each frequency.

[0014] The road surface reflection center frequency is obtained by weighting the power spectral density corresponding to each frequency.

[0015] In an exemplary embodiment, the process of obtaining the local instantaneous center frequency includes:

[0016] For any moment in the single-channel radar data segment, a short-time Fourier transform is used to obtain the power spectral density corresponding to each frequency at that moment.

[0017] Using the power spectral density corresponding to each frequency at that moment as weights, a weighted average is performed on each frequency at that moment to obtain the local instantaneous center frequency at that moment.

[0018] In an exemplary embodiment, the process of obtaining the center frequency attenuation distribution matrix includes:

[0019] Calculate the difference between the road surface reflection center frequency of any single-channel radar data segment and the local instantaneous center frequency at various times within that single-channel radar data segment, and determine the maximum value between the difference and 0, which is used as the element value at the corresponding position in the center frequency attenuation distribution matrix.

[0020] In an exemplary embodiment, the process of obtaining the cumulative difference matrix includes:

[0021] For the first channel in the cumulative difference matrix, the cumulative difference value at each time step of the first channel in the cumulative difference matrix is ​​obtained from the center frequency attenuation distribution matrix;

[0022] For the associated time of the target time in the cumulative difference matrix, a first feature, a second feature, and a third feature of the associated time are determined; the first feature is the cumulative difference value of the associated time in the previous single channel; the second feature is the element difference value between the element of the target time and the element of the associated time in the previous single channel in the center frequency attenuation distribution matrix; the third feature is the layer continuity constraint term obtained from the index difference between the target time and the associated time; the target time is any time in any other single channel in the cumulative difference matrix, and the associated time is any time within the local neighborhood of the target time;

[0023] The first feature, second feature, and third feature of the associated time are fused to obtain the fused feature of the associated time.

[0024] The minimum value is determined from the fusion features of each associated time of the target time to obtain the cumulative difference value of the target time.

[0025] In an exemplary embodiment, the process of obtaining the minimum frequency difference path includes:

[0026] Determine the minimum cumulative difference value in each channel of the cumulative difference matrix;

[0027] The minimum frequency difference path is formed by the time corresponding to the minimum cumulative difference value in each channel.

[0028] In an exemplary embodiment, the process of obtaining the path average frequency attenuation value includes:

[0029] The average frequency attenuation value of the path is obtained by integrating the element values ​​of each time point in the center frequency attenuation distribution matrix of the path with the minimum frequency difference.

[0030] In an exemplary embodiment, the process of acquiring the noise data includes:

[0031] A binary mask matrix is ​​determined, wherein the foreground region of the binary mask matrix is ​​obtained by the minimum frequency difference path;

[0032] The background region in the binary mask matrix is ​​mapped onto the center frequency attenuation distribution matrix to obtain the data standard deviation of the background region in the center frequency attenuation distribution matrix;

[0033] A simulated background fluctuation matrix is ​​generated from the standard deviation of the data, and the simulated background fluctuation matrix is ​​the noise data.

[0034] In an exemplary embodiment, the process of obtaining the path vertical deviation index includes:

[0035] Determine the time difference between the path with the minimum frequency difference and the path with the optimal path that belong to the same single channel;

[0036] By integrating the time differences of each individual track, the path vertical deviation index is obtained.

[0037] In an exemplary embodiment, the evaluation of asphalt pavement structural damage based on the path average frequency attenuation value and the path vertical deviation index includes:

[0038] Based on the preset frequency attenuation threshold and path deviation threshold, the magnitudes of the average frequency attenuation value of the path and the frequency attenuation threshold, as well as the magnitudes of the vertical deviation index of the path and the path deviation threshold, are determined respectively.

[0039] The assessment results determine the type of structural damage to the asphalt pavement.

[0040] This invention has the following beneficial effects: By extracting individual radar data segments from asphalt pavement, a center frequency attenuation distribution matrix relative to the road surface reference is constructed. This comprehensively depicts the continuous dynamic process of electromagnetic wave attenuation with depth / time as the center frequency propagates in the pavement structure from a spatiotemporal perspective, providing the core data foundation for subsequent analysis and upgrading the analysis object from a point to a surface (spatiotemporal distribution). The minimum difference path can be understood as the main or optimal path of systematic and consistent attenuation of the signal when it propagates in the medium. The average frequency attenuation value is a stable characteristic quantity that characterizes the overall attenuation intensity of the structure. It filters out random interference and reflects the basic absorption characteristics of the structure itself for electromagnetic wave energy. By introducing specific noise based on the original data characteristics, the scenario of signal interference is simulated, and the optimal path is re-searched in the new noise matrix and compared with the original optimal path. By comparing paths with small differences, the path vertical deviation index is obtained. This index quantifies the stability of the main attenuation path under disturbance. If the structure is uniform and the interface is good, the main propagation path should be relatively stable, and the deviation index should be small. If the structure is non-uniform, loose, or damaged, the propagation path is easily disturbed and changes, and the deviation index should be large. Therefore, this index is a sensitive indicator reflecting the uniformity of the structure and the stability of the interface. Finally, by comprehensively analyzing the path average frequency attenuation value and the path vertical deviation index, this method is more comprehensive and has stronger anti-interference ability than methods that rely on single or local features. It can more stably and accurately reveal the overall health status of the asphalt pavement structure (especially the interlayer interface). From the perspective of the spatiotemporal integrity of signal propagation and path robustness, a new feature extraction and evaluation framework is proposed, which can more accurately and reliably assess the damage status of the asphalt pavement structure. Attached Figure Description

[0041] Figure 1This is a flowchart of an intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing, provided in one embodiment of the present invention.

[0042] Figure 2 This is a flowchart illustrating the process of obtaining the cumulative difference matrix according to an embodiment of the present invention;

[0043] Figure 3 This is a flowchart of noise data acquisition provided in one embodiment of the present invention. Detailed Implementation

[0044] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. All data and information collected in this application have been obtained with full consent.

[0046] This embodiment provides an intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing, applicable to the evaluation of asphalt pavement structural damage. In an exemplary embodiment, the radar equipment is mounted on a detection vehicle, vertically facing the road surface. To simultaneously acquire multiple radar data, the radar equipment can be a multi-channel radar system. The ground-penetrating radar transmitting antennas can be designed in a parallel array, arranging antennas of different frequencies laterally on the detection vehicle, with each antenna transmitting and receiving independently and acquiring data synchronously; or a coaxial multi-frequency antenna design, where a single antenna element is designed with a multi-band response, simultaneously transmitting multiple frequency band signals through a frequency synthesizer. The frequency range is set according to the actual detection needs.

[0047] To improve the reliability of asphalt pavement structural damage assessment and meet real-time or near-real-time detection requirements, this embodiment extracts real-time acquired radar data into multiple non-overlapping radar data segments. The length of each radar data segment, i.e., the corresponding duration, is set according to actual needs. Each radar data segment is then analyzed separately to obtain the asphalt pavement structural damage assessment result for each segment.

[0048] For any given radar data segment, taking the current segment as an example, it includes several single-channel radar data segments. A single-channel radar data segment is essentially a time-series A-scan (single-channel waveform). The number of single-channel radar data segments included in a radar data segment, i.e., the number of lateral detection channels, is set according to actual needs. In this embodiment, the number of lateral detection channels is 50, corresponding to an actual road surface length of approximately 0.5 meters to 2 meters. The specific value depends on the sampling rate. The number of lateral detection channels corresponds to the number of radar frequencies.

[0049] For each single-channel radar data segment in the current radar data segment, time alignment is required to eliminate the longitudinal time deviation of the road surface reflection position between different channels caused by vehicle vibration. In an exemplary embodiment, the peak time of the single-channel radar data segment is picked using an amplitude threshold detection method or a cross-correlation algorithm, and the peak time is shifted and aligned to the zero point of the time axis.

[0050] Establish a radar data coordinate grid based on the current radar data segment. Where z is the depth sampling point index (since different depths correspond to different times, the depth sampling point index can also be understood as a time-series index, i.e., a time index), and its value range is [value range missing]. H represents the height of the interlayer interface search zone defined based on the design thickness of the pavement structure (e.g., covering a range of 10cm above and below the bottom of the surface layer; H can also be understood as the number of moments included in the duration of the current radar data segment); x represents the lateral detection track index within the current radar data segment, with a value range of... W represents the number of lateral detection channels, i.e., the number of single-channel radar data segments. Therefore, the physical consistency of the depth sampling point index z is ensured through the aforementioned time alignment operation.

[0051] like Figure 1 As shown in the figure, the intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing provided in this embodiment includes the following steps:

[0052] Step S1: Based on the difference between the surface reflection center frequency of each single radar data segment of the asphalt pavement and its local instantaneous center frequency at each moment, obtain the center frequency attenuation distribution matrix.

[0053] Step S2: Determine the path average frequency attenuation value of the path with the smallest frequency difference in the cumulative difference matrix;

[0054] Step S3: Apply noise data to the center frequency attenuation distribution matrix to obtain the noise-added frequency attenuation matrix;

[0055] Step S4: Search for the optimal path in the noise frequency attenuation matrix, and obtain the path vertical deviation index based on the difference between the path with the minimum frequency difference and the optimal path;

[0056] Step S5: Evaluate the structural damage of asphalt pavement based on the path average frequency attenuation value and the path vertical deviation index.

[0057] The following detailed explanation of each step, in conjunction with the accompanying drawings, is provided.

[0058] Step S1: Based on the difference between the surface reflection center frequency of each single radar data segment of the asphalt pavement and its local instantaneous center frequency at each moment, obtain the center frequency attenuation distribution matrix.

[0059] Because the center frequency of a ground-penetrating radar (GPR) transmitting antenna can drift due to changes in ambient temperature during long-term operation, and because differences in road surface roughness across different road sections can cause varying degrees of scattering of incident wave energy, it is necessary to extract the incident wave characteristics of each single-channel radar data segment as a dynamic benchmark for relative comparison in order to eliminate interference from system and environmental factors on absolute frequency values.

[0060] For any single-channel radar data segment, the road surface reflection center frequency of the single-channel radar data segment is first determined. The road surface reflection center frequency characterizes the center frequency of the single-channel radar data segment during road surface reflection and is an overall characteristic of the single-channel radar data segment. In an exemplary embodiment, a target time window is determined in the single-channel radar data segment. The selection of the target time window can characterize the overall characteristics of the radar data wave of the single-channel radar data segment. For example, the target time window can cover the main energy areas of the road surface direct wave and reflected wave of the single-channel radar data segment. In an exemplary embodiment, the maximum peak time in the single-channel radar data segment is located, and a data segment with a preset length centered on the maximum peak time is determined as the target time window. The preset length can be obtained by determining the antenna center frequency corresponding to the single-channel radar data segment, taking its reciprocal to obtain the period, and usually taking 2 to 3 times the period as the total length of the target time window to cover the complete Ricker wavelet oscillation. For example, the total length is set to 1.5 nanoseconds. It should be understood that if the peak time is the zero point of time alignment mentioned above, in order to avoid the radar data to the left of the zero point of time being truncated and lost, this embodiment can first obtain the target time window and the road surface reflection center frequency, and then perform the above time alignment operation.

[0061] Then, a Fast Fourier Transform (FFT) is performed on the radar data of the target time window within the single-channel radar data segment to obtain the power spectral density of the road surface reflection signal at each frequency. It should be understood that each frequency falls within the effective bandwidth of the ground-penetrating radar antenna, which is set according to the actual situation; for example, for a 2GHz antenna, it can be from 0.5GHz to 3.5GHz. Finally, a weighted integral method is used to calculate the center frequency of the road surface reflection of the single-channel radar data segment. Specifically, the power spectral density corresponding to each frequency within the single-channel radar data segment is used as a weight, and a weighted average is performed on each frequency within the single-channel radar data segment to obtain the center frequency of the road surface reflection of the single-channel radar data segment. The calculation formula is as follows:

[0062] ;

[0063] in, This represents the road surface reflection center frequency of the x-th single-channel radar data segment, characterizing the initial spectral state of the radar wave when it enters the road surface structure at the x-th channel position. For frequency variables, This represents the lower limit of the effective bandwidth range of a ground-penetrating radar antenna. This indicates the upper limit of the effective bandwidth range of a ground-penetrating radar antenna. Represents frequency variable The corresponding power spectral density.

[0064] It should be understood that if radar data in a certain single-channel radar data segment is lost or extremely weak, the calculated... If the denominator approaches or equals 0, it becomes meaningless, leading to computational overflow. Therefore, when the denominator is less than a preset minimum value, the single-channel radar data segment x is marked as invalid. In the subsequent generation of the center frequency attenuation distribution matrix, the values ​​of all elements of this invalid channel x can be obtained through linear interpolation of the element values ​​at the corresponding depth z of its adjacent valid channels (e.g., channels x-1 and x+1). This yields the surface reflection center frequency of each single-channel radar data segment.

[0065] If trace amounts of moisture accumulate between asphalt pavement layers (water-bearing interlayer delamination), the polar relaxation effect of water molecules will significantly absorb the high-frequency components of electromagnetic waves penetrating the interface, causing the signal center frequency to attenuate relative to the incident wave. Dry asphalt mixtures or intact interlayer interfaces, however, have a relatively smaller impact on frequency.

[0066] For a single-channel radar data segment, the local instantaneous center frequency at each moment within the segment is determined. This local instantaneous center frequency represents a local characteristic at that moment. The local instantaneous center frequency can be obtained using known time-frequency analysis techniques such as short-time Fourier transform or wavelet transform. In an exemplary embodiment, for any moment within the single-channel radar data segment, a short-time Fourier transform is used to obtain the power spectral density corresponding to each frequency at that moment. It should be understood that each frequency is also within the effective bandwidth of the ground-penetrating radar antenna. For any given moment, a time window corresponding to that moment is constructed (e.g., a time window centered on that moment), and the type of time window function is determined (e.g., Hanning window, Hamming window, or Gaussian window). The window length is typically very short, and a certain overlap ratio (e.g., 50%) can be set between adjacent windows. Then, the window function is placed starting from the beginning of the single-channel radar data segment, and the radar signal segments within the window are windowed (to reduce spectral leakage). A Fourier transform is then performed on the windowed signal segments, and the power spectral density at each frequency at that moment is calculated. By continuously sliding the window in this way, the power spectral density at each frequency at each time point can be obtained. It should be understood that for the start and end times, a time window centered on them cannot be obtained. Therefore, zero-padding can be performed on the missing parts within the time window to ensure the integrity of the time window.

[0067] The selection of the window length for the short-time Fourier transform should take into account both time and frequency resolution, and is typically 2-4 times the period of the radar signal's dominant frequency. For example, for an antenna with a center frequency of 2 GHz, its signal period is approximately 0.5 nanoseconds, so the window length can be set to 1 to 2 nanoseconds. The overlap ratio between adjacent windows is usually set to 50% to 75% to ensure the continuity of time and frequency information.

[0068] For any given moment, the local instantaneous center frequency is obtained by weighting the power spectral density corresponding to each frequency at that moment and averaging the values ​​of all frequencies at that moment. It should be understood that the local instantaneous center frequency is calculated using the same method as the road surface reflection center frequency mentioned above.

[0069] Then, based on the difference between the surface reflection center frequency of each single radar data segment of the asphalt pavement and its local instantaneous center frequency at each moment, the center frequency attenuation distribution matrix is ​​obtained.

[0070] Taking the z-th moment of the x-th single-channel radar data segment as an example, calculate the difference between the road surface reflection center frequency of the x-th single-channel radar data segment and the local instantaneous center frequency of the x-th single-channel radar data segment at the z-th moment. Then, take the maximum value between this difference and 0 as the element value at the corresponding position of the z-th moment of the x-th single-channel radar data segment in the center frequency attenuation distribution matrix. The calculation formula is as follows:

[0071] ;

[0072] in, Represents the center frequency attenuation distribution matrix The element value at the z-th time position of the x-th single-channel radar data segment. Let represent the local instantaneous center frequency at time z of the x-th single-channel radar data segment, and max denote the maximum value function. This yields the center frequency attenuation distribution matrix. Center frequency attenuation distribution matrix It is A matrix of dimensionality.

[0073] The above formula calculates the frequency loss of the local instantaneous center frequency relative to the incident wave on the same track. The max function is used to remove the negative attenuation value caused by calculation noise (i.e., the case of abnormal frequency increase). The numerical unit is GHz. The larger the value, the stronger the absorption of high-frequency components of radar electromagnetic waves by the medium at that location, indicating that there may be abnormal dielectric loss.

[0074] In this embodiment, to facilitate subsequent data processing, the center frequency attenuation distribution matrix is... Perform a normalization operation, specifically: obtain the center frequency attenuation distribution matrix. Calculate the center frequency attenuation distribution matrix by finding the maximum value in the matrix. The ratio of each element to the maximum value is used as the normalized result for each element. Based on the normalized results of each element, a normalized center frequency attenuation distribution matrix is ​​generated. The center frequency attenuation distribution matrix will be discussed later. All matrices are normalized, which not only facilitates data processing but also eliminates the influence of dimensions. It is worth noting that if the maximum value in the center frequency attenuation distribution matrix is ​​0, then no normalization is required, and the normalized matrix is ​​the original matrix.

[0075] Step S2: Determine the path average frequency attenuation value of the path with the smallest frequency difference in the cumulative difference matrix.

[0076] In the non-homogeneous medium of asphalt pavement, the random scattering signals generated by the discrete distribution of dry coarse aggregates often resemble the redshift signals caused by trace amounts of moisture in terms of spectral characteristics, making it difficult to distinguish between "true" and "false" signals. The core of this step lies in introducing simulated fluctuations equivalent to the current pavement background texture features to conduct a structural stability "stress test" on the extracted interlayer interface path. This method is based on the principle that real water-bearing interlayer peeling is a laterally connected physical entity with topological stability under noise interference; while the false signals caused by discrete aggregates are random high-value permutations that will cause significant path jumps under interference. Through this mechanism, this step decouples the medium's loss properties (frequency attenuation) from its structural properties (path stability), thereby overcoming the limitations of relying solely on signal strength for discrimination.

[0077] In order to achieve the attenuation distribution matrix at the center frequency The most likely trajectory representing the continuous direction of the interlayer interface is precisely located. This step uses a dynamic programming algorithm to search for the connected path with the minimum cumulative difference, which represents the layer that is most significant and continuous in terms of energy loss.

[0078] First, the center frequency attenuation distribution matrix... Determine the cumulative difference matrix Cumulative difference matrix Characterizing the center frequency attenuation distribution matrix The cumulative frequency attenuation difference. Cumulative difference matrix. Obtained by a step-by-step recursive approach, such as Figure 2 As shown, the following is a specific process for obtaining the cumulative difference matrix:

[0079] Step S21: For the first channel in the cumulative difference matrix, obtain the cumulative difference value of the first channel at each time step from the center frequency attenuation distribution matrix.

[0080] First, initialize the cumulative difference matrix. The cumulative difference values ​​in the first single channel, since there is no prior constraint on the path starting point in the depth direction, will have a center frequency attenuation distribution matrix. The element values ​​at each time step in the first single track are used as the cumulative difference matrix. The cumulative difference value at the corresponding time in the first single track, i.e. .

[0081] Step S22: For the associated time of the target time in the cumulative difference matrix, determine the first feature, second feature and third feature of the associated time.

[0082] Then determine the cumulative difference matrix. The cumulative difference values ​​at each time point from the second track to the last track in the cumulative difference matrix are used. For ease of explanation, any time point in any track from the second track to the last track in the cumulative difference matrix is ​​set as the target time. The local neighborhood range of the target time is determined. In this embodiment, the local neighborhood range includes the time point before the target time, the target time, and the time point after the target time. The associated time is any time point within the local neighborhood range of the target time. It should be understood that if the target time is the start time or the end time, the obtained local neighborhood range only includes the time point on one side. For example, the local neighborhood range of the start time only includes the start time and the time point after the start time, so these two times are taken as the local neighborhood range of the start time; similarly, the local neighborhood range of the end time only includes the time point before the end time and the end time, so these two times are taken as the local neighborhood range of the end time.

[0083] The first characteristic of the associated time is determined as the cumulative difference value of that time in the previous single track. If the target time is the z-th time in the x-th single track of the cumulative difference matrix, the cumulative difference value of the target time is... The associated time of the target time is the k-th time. If the previous channel of the x-th channel is the (x-1)-th channel, then the first feature, i.e., the cumulative difference at the associated time in the previous channel, is... . The settings ensure that the depth jump between adjacent paths and between paths within the same path does not exceed one instant, thereby guaranteeing the physical continuity of the interface and avoiding non-physical vertical abrupt changes.

[0084] The second characteristic of the associated time is determined; the second characteristic is the center frequency attenuation distribution matrix. The element difference between the element at the target time and the element at the associated time in the previous single channel, and the attenuation distribution matrix of the target time at the center frequency. The elements in are Center frequency attenuation distribution matrix In the context, the element that is associated with the previous single track is... The element difference between the two is The element that is associated with the previous single-path element at that moment. As with The frequency attenuation value at the previous single-channel potential connection point. It characterizes the difference in dielectric loss properties between adjacent passes; the smaller the value, the more similar the dielectric properties.

[0085] A third feature is determined for the associated time, which is a hierarchical continuity constraint term obtained from the index difference between the target time and the associated time. In an exemplary embodiment, the hierarchical continuity constraint term is... ,in, This represents the index difference between the z-th time step and the k-th time step. Since the z-th time step and the k-th time step differ by one time step, or they overlap, therefore... The calculation result is 1 or 0. For the preset balance coefficient (in this embodiment, The recommended value range is 0.5 to 1.0, depending on the depth sampling rate (used to balance the magnitude of spectral numerical differences with spatial geometric distances). It should be understood that... Dimensions The reciprocal of the dimensions is used to ensure that the dimensions are eliminated after multiplying the two.

[0086] Step S23: Fuse the first, second, and third features at the associated time to obtain the fused features at the associated time.

[0087] The average values ​​of the first, second, and third features at the correlation time are calculated to obtain the fused features at the correlation time. The calculation formula is as follows:

[0088] ;

[0089] in, This represents the fusion feature at time k in the x-th channel. This represents the cumulative difference value at time k in the (x-1)th channel. This represents the attenuation distribution matrix at the center frequency at time z in the x-th channel. The elements in This represents the attenuation distribution matrix at the center frequency at time k in the (x-1)th channel. The elements in This represents the number of time intervals between the z-th time and the k-th time. This is the preset balance coefficient.

[0090] Step S24: Determine the minimum value from the fusion features of each associated time at the target time to obtain the cumulative difference value at the target time.

[0091] Since the associated times at time z in track x are time z-1, time z, and time z+1, we obtain the fusion features at time z-1, time z, and time z+1 in track x. The minimum value among these three fusion features is selected as the cumulative difference value at time z in track x. .

[0092] The cumulative difference matrix is ​​obtained through the above method. The cumulative difference values ​​at each time point from the second track to the last track are used to construct the cumulative difference matrix. .

[0093] Then, determine the cumulative difference matrix. The path with the minimum frequency difference. Wherein, the cumulative difference matrix is ​​determined. The minimum cumulative difference value in each channel is used to obtain the index of the minimum cumulative difference value in each channel, i.e., the time corresponding to the minimum cumulative difference value in each channel. The minimum frequency difference path is constructed from the times corresponding to the minimum cumulative difference values ​​in each channel. Simultaneously, the time corresponding to the minimum cumulative difference value in each track is recorded. Taking the x-th track as an example, the time corresponding to its minimum cumulative difference value is... .

[0094] Then, based on the cumulative difference matrix Minimum frequency difference path The path with the minimum frequency difference is obtained. The path average frequency attenuation value. Among them, the path with the minimum frequency difference is obtained. Attenuation distribution matrix at the center frequency at each time point The element values ​​in the matrix, i.e., the time corresponding to the minimum cumulative difference value in each channel, are the attenuation distribution matrix at the center frequency. The element values ​​in the matrix are then used to calculate the average value of these element values. The result is used as the path average frequency attenuation value, and the calculation formula is as follows:

[0095] ;

[0096] in, This represents the path average frequency attenuation value, which serves as a characteristic quantity to characterize the degree of loss in the stratum medium. This represents the time corresponding to the minimum cumulative difference value of the x-th track. Attenuation distribution matrix at center frequency The element values ​​in.

[0097] Step S3: Apply noise data to the center frequency attenuation distribution matrix to obtain the noise-added frequency attenuation matrix.

[0098] Due to differences in asphalt mixture gradation and compaction degree across different road sections, the background noise level of radar echoes is spatially non-stationary. To enable subsequent detection to adapt to current road conditions, it is necessary to quantify the signal fluctuation characteristics in non-interface regions.

[0099] Noise data is applied to the center frequency attenuation distribution matrix to obtain the noisy frequency attenuation matrix. The noise data represents interference equivalent to the current environment. This is achieved by applying noise data to the original signal (i.e., the center frequency attenuation distribution matrix). Injecting disturbances equivalent to the current environment into the algorithm simulates the robustness of the path extraction under more severe aggregate distribution conditions.

[0100] First, determine the noise data, which is derived from the center frequency attenuation distribution matrix. and minimum frequency difference path Obtained. In an exemplary embodiment, such as Figure 3 As shown below, a specific process for obtaining noise data is given:

[0101] Step S31: Determine the binary mask matrix.

[0102] Using the minimum frequency difference path As a spatial reference, construct an attenuation distribution matrix with respect to the center frequency. A binary mask matrix of the same dimension. Calculated by the path of minimum frequency difference. The foreground region of the binary mask matrix is ​​obtained. Specifically, only the path with the minimum frequency difference can be selected. The time positions of each included single channel are used as the foreground region of the binary mask matrix; to ensure reliability, for the path with the minimum frequency difference... At any given moment, expand to the left and right of that moment. At any given time (in this embodiment) By taking values ​​of 2 to 3 time points to cover the Fresnel zone influence area of ​​the interlayer interface, the time range of that time point is obtained, thus yielding the path of minimum frequency difference. The time range of each included single channel will be the path with the smallest frequency difference. The positions of the time ranges of each included single channel are used as the foreground regions of the binary mask matrix. Regions other than the foreground regions in the binary mask matrix are marked as background regions.

[0103] Step S32: Map the background region in the binary mask matrix to the center frequency attenuation distribution matrix to obtain the data standard deviation of the background region in the center frequency attenuation distribution matrix.

[0104] Map the background region in the binary mask matrix to the center frequency attenuation distribution matrix. In the process, obtain the attenuation distribution matrix of the background region at the center frequency in the binary mask matrix. For each element in the binary mask matrix, calculate the attenuation distribution matrix of the background region at the center frequency. The standard deviation of each element in the data is denoted as the background frequency fluctuation standard deviation. The standard deviation of the background frequency fluctuation This directly reflects the random fluctuation amplitude of the frequency attenuation value within the current radar data segment, caused by the uneven distribution of dry aggregates and voids. The standard deviation of this background frequency fluctuation is... This will serve as the benchmark for generating the simulated background fluctuation matrix.

[0105] Step S33: Generate the simulated background fluctuation matrix from the data standard deviation.

[0106] Based on the standard deviation of background frequency fluctuation Generate a simulated background fluctuation matrix, which represents the desired noise data. The simulated background fluctuation matrix has the following dimensions: In one exemplary embodiment, each element in the simulated background fluctuation matrix follows a mean of 0 and a standard deviation equal to the standard deviation of the background frequency fluctuation. The Gaussian distribution (i.e., normal distribution) is used to obtain the simulated background wave matrix. It should be understood that, in principle, an infinite number of simulated background wave matrices can be obtained, from which any one can be selected.

[0107] If the standard deviation of background frequency fluctuation If the standard deviation is too small, the desired noise data may not meet the requirements, for example, negative noise may be superimposed later. Therefore, this embodiment presets a lower limit value for the standard deviation. If the background frequency fluctuation standard deviation is too small, the lower limit value will be set. If it is less than the lower limit of this standard deviation, then the standard deviation of the background frequency fluctuation will be... The standard deviation is fixed at this lower limit to avoid background frequency fluctuations in the standard deviation. Too small. The lower limit of the standard deviation is determined based on the instrument noise level of the radar system itself, to ensure that the injected simulated noise intensity is not lower than the system noise floor. Specifically, a segment of unloaded signal can be collected under electromagnetic shielding conditions, the standard deviation of its equivalent frequency fluctuation can be calculated, and this can be used as the lower limit of the standard deviation.

[0108] Simulate the background fluctuation matrix Superimposed on the center frequency attenuation distribution matrix Above, the background fluctuation matrix will be simulated. and center frequency attenuation distribution matrix The values ​​of elements at the same position are added together to generate a noise attenuation matrix. The calculation method is as follows:

[0109] ;

[0110] in, Represents the noise attenuation matrix The element at time z in the x-th single track. Represents the simulated background fluctuation matrix The element at time z in the x-th single track.

[0111] It should be understood that, in order to avoid Less than 0, in this embodiment, if If it is less than 0, then Set to 0.

[0112] Step S4: Search for the optimal path in the noise frequency attenuation matrix, and obtain the path vertical deviation index based on the difference between the path with the minimum frequency difference and the optimal path.

[0113] Obtain the noise attenuation matrix Then, according to step S2, based on the attenuation distribution matrix of the center frequency... The method for obtaining the cumulative difference matrix and the method for obtaining the minimum frequency difference path in the cumulative difference matrix are obtained, and the noise attenuation matrix is ​​searched to obtain the noise addition frequency attenuation matrix. The path with the minimum frequency difference in the cumulative difference matrix is ​​defined as the optimal path in the noise frequency attenuation matrix. The optimal path in the noise frequency attenuation matrix is ​​defined as the noise response path. Simultaneously, record the noise-added response path. The time corresponding to the minimum cumulative difference value in each track, taking the x-th track as an example, is the time corresponding to the minimum cumulative difference value. .

[0114] By comparing the spatial geometric position differences of the optimal path before and after interference, the topological stability of the current radar data segment is quantified, thereby outputting binary features for final classification.

[0115] If the asphalt pavement corresponding to the current radar data segment is a continuous interface formed by the accumulation of trace amounts of moisture, its energy characteristics have significant lateral connectivity. Even with background noise superimposed, the optimal path location searched by the algorithm will still remain near the real interface, i.e., the path with the minimum frequency difference. With noise-adding response path The difference is small, even approaching 0; conversely, if the asphalt pavement corresponding to the current radar data segment is an artifact formed by random scattering of discrete aggregates, its connectivity is fragile. After superimposed noise, the optimal path will undergo significant random jumps, i.e., the path with the smallest frequency difference. With noise-adding response path The differences are significant. Therefore, based on the minimum frequency difference path obtained above... With noise-adding response path The difference is used to obtain the path vertical deviation index. Specifically: determine the path with the minimum frequency difference. With noise-adding response path The time differences within the same track are then combined to obtain the path vertical deviation index, calculated using the following formula:

[0116] ;

[0117] in, This represents the path vertical deviation index. Represents the path of minimum frequency difference With noise-adding response path The time difference in the x-th channel. Since the number of times in each channel is H, H is added to the denominator in the calculation formula to normalize the time difference and eliminate dimensions, which facilitates subsequent data processing.

[0118] If the asphalt pavement corresponding to the current radar data segment is a continuous interface formed by the accumulation of trace amounts of moisture, and its energy characteristics exhibit significant lateral connectivity, then the path vertical deviation index... The vertical deviation index is relatively small; conversely, if the asphalt pavement corresponding to the current radar data segment is an artifact formed by random scattering of discrete aggregates, its connectivity is fragile, and after the noise is superimposed, the optimal path will undergo significant random jumps, then the path vertical deviation index will be relatively large. Relatively large.

[0119] Step S5: Evaluate the structural damage of asphalt pavement based on the path average frequency attenuation value and the path vertical deviation index.

[0120] By following the steps above, the path-average frequency attenuation value corresponding to the current radar data segment is obtained. Vertical deviation index of the path Based on the path average frequency attenuation value corresponding to the current radar data segment Vertical deviation index of the path This allows for the evaluation of asphalt pavement structural damage corresponding to the current radar data segment. Among these factors is the path-average frequency attenuation value. The path vertical deviation index characterizes the medium loss properties of asphalt pavement. To characterize the structural topological properties of asphalt pavement, a two-dimensional physical state space is constructed based on the orthogonality between the two. Within this space, by defining logical boundaries, often confused "true and false anomalies" are separated into independent engineering defect types, thereby enabling the evaluation of structural damage to asphalt pavement.

[0121] In one exemplary embodiment, a frequency attenuation threshold and a path deviation threshold are preset. These thresholds can be set based on experience and judgment. To avoid the uncertainty caused by manually setting fixed thresholds, this step utilizes the inherent statistical distribution characteristics of asphalt pavement materials to determine the frequency attenuation threshold and the path deviation threshold.

[0122] For the frequency attenuation threshold, to ensure the reliability of the statistical model, at least 3-5 reference road segments at different locations with a cumulative total length of not less than 100 meters should be selected. The health status of the reference road segments can be verified through core sampling or confirmed by existing maintenance records. The path deviation threshold characterizes the maximum average vertical drift that a real continuous interface can tolerate under noise interference. Its value should be slightly larger than the maximum jump in a single step of the dynamic programming algorithm (i.e., one depth sampling point), while being significantly smaller than the random path jump amplitude that may be caused by large discrete aggregates. For example, if typical coarse aggregates span an average of 3-4 depth sampling points on radar images, the path deviation threshold can be set between 1.5 and 2.0 to effectively distinguish between structural jumps and physical interface drift.

[0123] The frequency attenuation threshold is used to distinguish whether the medium has abnormal dielectric loss (i.e., whether moisture is present). Several reference asphalt road sections known to be healthy, dry, and uniformly graded are selected. Following the steps above, the path-average frequency attenuation value of each reference asphalt road section is obtained. The mean and standard deviation of the path-average frequency attenuation value for each reference asphalt road section are calculated. Then, the frequency attenuation threshold is obtained using the following method:

[0124] ;

[0125] in, Indicates the frequency attenuation threshold. This represents the mean of the path-average frequency attenuation values ​​for each reference asphalt road segment. This represents the standard deviation of the path-average frequency attenuation value for each reference asphalt road segment. This represents the preset statistical significance coefficient (e.g., taking...). (This represents the 95% confidence interval).

[0126] Since the path drift of a real continuous interface should be mainly limited by the search step size constraint of the algorithm, this embodiment sets the path deviation threshold to a constant related to the depth sampling resolution (e.g., taking...). The path deviation threshold is used to distinguish whether the asphalt pavement corresponding to the current radar data segment has structural connectivity that resists random interference.

[0127] Based on the frequency attenuation threshold and the path deviation threshold, the magnitudes of the path average frequency attenuation value and the frequency attenuation threshold, as well as the magnitudes of the path vertical deviation index and the path deviation threshold, are determined respectively. Specifically: if the path average frequency attenuation value is greater than or equal to the frequency attenuation threshold, and the path vertical deviation index is less than the path deviation threshold, it indicates that there is water-bearing interlayer stripping in the asphalt pavement corresponding to the current radar data segment, because: high frequency attenuation (path average frequency attenuation value greater than or equal to the frequency attenuation threshold) indicates the presence of a high-loss medium (moisture) between layers; low vertical deviation (path vertical deviation index less than the path deviation threshold) indicates that the high-loss area is continuous and stable in the lateral direction, consistent with the physical characteristics of water-bearing layers; if the path average frequency attenuation value is less than the frequency attenuation threshold, and the path vertical deviation index is less than the path deviation threshold, it indicates that there is dry interlayer bonding in the asphalt pavement corresponding to the current radar data segment, because: low frequency attenuation (path average frequency attenuation value greater than or equal to the frequency attenuation threshold) indicates the presence of a high-loss medium (moisture) between layers. A mean frequency attenuation value less than the frequency attenuation threshold indicates dryness of the medium; a low vertical deviation (path vertical deviation index less than the path deviation threshold) indicates a smooth and continuous interlayer interface, belonging to a normal construction joint or interlayer interface, and the asphalt pavement corresponding to the current radar data segment is healthy / dry; if the path vertical deviation index is greater than or equal to the path deviation threshold, it indicates that the asphalt pavement corresponding to the current radar data segment has a discrete medium distribution, because: regardless of whether the frequency attenuation value is large or small, as long as the vertical deviation is significant (path vertical deviation index greater than or equal to the path deviation threshold), it indicates that the extracted path has undergone a positional jump under noise interference, which corresponds to a random distribution area of ​​coarse aggregate or voids, without physically continuous layers, belonging to non-interlayer disease signals. This allows for the determination of the asphalt pavement structural damage category from the judgment results, effectively eliminating the misjudgment of high-frequency attenuation caused by aggregate scattering in traditional methods, and ensuring the specificity of water damage disease identification.

[0128] Using the above process, this embodiment can perform real-time damage assessment on the asphalt pavement corresponding to each radar data segment. In subsequent applications, the damage assessment results of the asphalt pavement corresponding to each radar data segment are correlated with each asphalt pavement segment and visualized. Simultaneously, according to the mileage order of the asphalt pavement, the damage assessment results of each asphalt pavement segment are spatially stitched together to generate a continuously distributed strip-shaped pavement structure damage distribution map. Furthermore, in the map, road segments with structural damage type categorized as water-bearing interlayer stripping can be highlighted (e.g., red) to prompt maintenance departments to conduct borehole verification or grouting repair; road segments with structural damage type categorized as discrete medium distribution can be secondary-marked (e.g., yellow) to indicate the need to monitor their loosening trend; and road segments with structural damage type categorized as dry interlayer bonding can be marked as safe (e.g., green), requiring no intervention.

[0129] This embodiment also provides an intelligent evaluation system for asphalt pavement structural damage based on ground-penetrating radar signal processing, including: a memory and a processor; the memory is connected to the processor, and the memory is used to store program instructions; the processor is used to implement the steps in the above-described embodiment of the intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing when the program instructions are executed.

[0130] In one exemplary embodiment, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the embodiment of the intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing.

[0131] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0132] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A smart evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing, characterized in that, include: The center frequency attenuation distribution matrix is ​​obtained by comparing the surface reflection center frequency of each single radar data segment of the asphalt pavement with the local instantaneous center frequency at each moment. Determine the path-average frequency attenuation value of the path with the minimum frequency difference in the cumulative difference matrix; the cumulative difference matrix represents the accumulation of frequency attenuation differences in the center frequency attenuation distribution matrix; Noise data is applied to the center frequency attenuation distribution matrix to obtain a noisy frequency attenuation matrix; the noise data is obtained from the center frequency attenuation distribution matrix and the minimum frequency difference path. Search for the optimal path in the noise frequency attenuation matrix, and obtain the path vertical deviation index based on the difference between the minimum frequency difference path and the optimal path; The structural damage of asphalt pavement is evaluated based on the path average frequency attenuation value and the path vertical deviation index.

2. The intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing as described in claim 1, characterized in that, The process of obtaining the center frequency of the road surface reflection includes: Determine the target time window in the single-channel radar data segment; A fast Fourier transform is performed on the radar data of the target time window to obtain the power spectral density corresponding to each frequency. The road surface reflection center frequency is obtained by weighting the power spectral density corresponding to each frequency.

3. The intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing as described in claim 1, characterized in that, The process of obtaining the local instantaneous center frequency includes: For any moment in the single-channel radar data segment, a short-time Fourier transform is used to obtain the power spectral density corresponding to each frequency at that moment. Using the power spectral density corresponding to each frequency at that moment as weights, a weighted average is performed on each frequency at that moment to obtain the local instantaneous center frequency at that moment.

4. The intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing as described in claim 1, characterized in that, The process of obtaining the center frequency attenuation distribution matrix includes: Calculate the difference between the road surface reflection center frequency of any single-channel radar data segment and the local instantaneous center frequency at various times within that single-channel radar data segment, and determine the maximum value between the difference and 0, which is used as the element value at the corresponding position in the center frequency attenuation distribution matrix.

5. The intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing as described in claim 1, characterized in that, The process of obtaining the cumulative difference matrix includes: For the first channel in the cumulative difference matrix, the cumulative difference value at each time step of the first channel in the cumulative difference matrix is ​​obtained from the center frequency attenuation distribution matrix; For the associated time of the target time in the cumulative difference matrix, a first feature, a second feature, and a third feature of the associated time are determined; the first feature is the cumulative difference value of the associated time in the previous single channel; the second feature is the element difference value between the element of the target time and the element of the associated time in the previous single channel in the center frequency attenuation distribution matrix; the third feature is the layer continuity constraint term obtained from the index difference between the target time and the associated time; the target time is any time in any other single channel in the cumulative difference matrix, and the associated time is any time within the local neighborhood of the target time; The first feature, second feature, and third feature of the associated time are fused to obtain the fused feature of the associated time. The minimum value is determined from the fusion features of each associated time of the target time to obtain the cumulative difference value of the target time.

6. The intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing as described in claim 1, characterized in that, The process of obtaining the minimum frequency difference path includes: Determine the minimum cumulative difference value in each channel of the cumulative difference matrix; The minimum frequency difference path is formed by the time corresponding to the minimum cumulative difference value in each channel.

7. The intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing as described in claim 6, characterized in that, The process of obtaining the path average frequency attenuation value includes: The average frequency attenuation value of the path is obtained by integrating the element values ​​of each time point in the center frequency attenuation distribution matrix of the path with the minimum frequency difference.

8. The intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing as described in claim 1, characterized in that, The process of acquiring the noise data includes: A binary mask matrix is ​​determined, wherein the foreground region of the binary mask matrix is ​​obtained by the minimum frequency difference path; The background region in the binary mask matrix is ​​mapped onto the center frequency attenuation distribution matrix to obtain the data standard deviation of the background region in the center frequency attenuation distribution matrix; A simulated background fluctuation matrix is ​​generated from the standard deviation of the data, and the simulated background fluctuation matrix is ​​the noise data.

9. The intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing as described in claim 1, characterized in that, The process of obtaining the path vertical deviation index includes: Determine the time difference between the path with the minimum frequency difference and the path with the optimal path that belong to the same single channel; By integrating the time differences of each individual track, the path vertical deviation index is obtained.

10. The intelligent evaluation method for asphalt pavement structural damage based on ground-penetrating radar signal processing as described in claim 1, characterized in that, The evaluation of asphalt pavement structural damage based on the path average frequency attenuation value and the path vertical deviation index includes: Based on the preset frequency attenuation threshold and path deviation threshold, the magnitudes of the average frequency attenuation value of the path and the frequency attenuation threshold, as well as the magnitudes of the vertical deviation index of the path and the path deviation threshold, are determined respectively. The assessment results determine the type of structural damage to the asphalt pavement.

Images