Ground penetrating radar image quality evaluation method and system for asphalt pavement cavity disease identification

Abstract

The invention discloses a ground penetrating radar image quality evaluation method and system for asphalt pavement cavity disease identification, and belongs to the technical field of ground penetrating radar images. In order to solve the problem that no method for evaluating the quality of the ground penetrating radar image of the asphalt pavement cavity disease exists at present, the method comprises the following steps of: acquiring a detection B-scan image of a detection road aiming at the detection road, and establishing a corresponding control group forward model according to a design file of the detection road; according to the control group, a multi-layer asphalt pavement structure in a design file of a detected road is changed into a single-layer pavement structure; recording an image simulated by the forward modeling model of the control group as a B-scan image of the control group; and then image quality evaluation of the B-scan image is detected according to the SSIM-S index. The method is used for quality evaluation of the ground penetrating radar image.

Description

Technical Field

[0001] The invention belongs to the technical field of ground penetrating radar images, and relates to a method and system for evaluating the quality of ground penetrating radar images of asphalt pavement cavity diseases. Background Art

[0002] Ground Penetrating Radar (GPR) is a non-destructive and rapid detection technology that uses electromagnetic wave reflection to determine the distribution of underground media. It has become one of the preferred methods for detecting hidden road defects. The key to using GPR to accurately detect hidden road defects is GPR image recognition, but the quality of current ground penetrating radar images is uneven. The quality of ground penetrating radar images will seriously affect the identification of hidden road defects. The higher the quality of the ground penetrating radar image, the more obvious the features of the hidden defects on the image, and the higher the detection accuracy. On the contrary, it will lead to a higher probability of misjudgment and missed judgment, and the potential of GPR technology cannot be fully utilized. Therefore, ground penetrating radar image quality assessment is a bottleneck that restricts this technology from supporting the development of smart cities and transportation construction.

[0003] At present, there is no unified standard and method for the quality assessment of ground penetrating radar images. Therefore, when the image quality is poor, if there is no way to evaluate its quality, and all ground penetrating radar images are directly regarded as undifferentiated uniform quality images for road cavity disease identification, the identification effect of some road diseases will not be ideal. Summary of the invention

[0004] The invention aims to solve the problem that there is no quality assessment method for ground penetrating radar images of asphalt pavement cavity diseases at present.

[0005] A ground penetrating radar image quality assessment method for asphalt pavement cavity disease identification includes the following steps:

[0006] For the inspection road, a ground penetrating radar image of the inspection road, i.e., the inspection B-scan image, is obtained. At the same time, a corresponding control group forward model is established according to the design file of the inspection road. The control group is a control group that changes the multi-layer asphalt pavement structure in the design file of the inspection road into a single-layer pavement structure; the image simulated by the control group forward model is recorded as the control group B-scan image; then the image quality of the inspection B-scan image is evaluated according to the SSIM-S index;

[0007]

[0008] Where SSIM(x,y) represents the structural similarity index between image x and image y, SROCC(x,y) represents the Spearman rank correlation coefficient index between image x and image y, and image x and image y are the detection B-scan image and the control group B-scan image, respectively.

[0009] Furthermore, the forward model uses a 400 MHz Ricker wavelet as an excitation source signal.

[0010] Furthermore, in the process of acquiring the ground penetrating radar image of the detected road, the antenna transceiver spacing is set to be less than 25 cm. Preferably, the antenna transceiver spacing is set to [10 cm, 20 cm].

[0011] Furthermore, the antenna transceiver spacing in the forward model is equal to the antenna transceiver spacing set in the process of acquiring the ground penetrating radar image of the detection road.

[0012] A ground penetrating radar image quality assessment system for asphalt pavement cavity disease identification, comprising:

[0013] Ground penetrating radar image acquisition unit: for the detection road, acquire the ground penetrating radar image of the detection road, that is, detect the B-scan image;

[0014] Control group forward model building unit: a corresponding control group forward model is built according to the design documents of the test road. The control group is to change the multi-layer asphalt pavement structure in the design documents of the test road into a single-layer pavement structure;

[0015] Control group image acquisition unit: based on the established control group forward model, simulates the control group B-scan image;

[0016] Quality assessment unit: performs image quality assessment of B-scan images based on the SSIM-S index;

[0017]

[0018] Where SSIM(x,y) represents the structural similarity index between image x and image y, SROCC(x,y) represents the Spearman rank correlation coefficient index between image x and image y, and image x and image y are the detection B-scan image and the control group B-scan image, respectively.

[0019] Furthermore, the forward model uses a 400 MHz Ricker wavelet as an excitation source signal.

[0020] Furthermore, in the process of acquiring the ground penetrating radar image of the detected road, the antenna transceiver spacing is set to be less than 25 cm. Preferably, the antenna transceiver spacing is set to [10 cm, 20 cm].

[0021] Furthermore, the antenna transceiver spacing in the forward model is equal to the antenna transceiver spacing set in the process of acquiring the ground penetrating radar image of the detection road.

[0022] Beneficial effects:

[0023] The present invention calculates the SSIM of the B-scan images of the experimental group and the control group, and uses the SROCC between the waveforms of the experimental group and the control group as an aid to comprehensively evaluate the characteristics of the asphalt pavement cavity disease. The present invention uses SSIM-S for evaluation, and the SSIM-S range is [0,1]. The larger the value, the higher the similarity of the image, and the better the result of the forward simulation of the experimental group. The SSIM-S of the present invention not only takes into account the particularity of the B-scan image, but also solves the problem that when the dielectric constant of the target body and the background differ greatly during radar propagation, the medium absorbs the electromagnetic wave energy strongly, and the simulated B-scan image features are not obvious, resulting in inaccurate evaluation of the quality of the B-scan image of the experimental group. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is the Blackman-Harris Window time domain waveform (400MHz and 800MHz);

[0025] Figure 2 Blackman-Harris Window spectrum diagram (400MHz and 800MHz);

[0026] Figure 3 Ricker time domain waveform (400MHz and 800MHz);

[0027] Figure 4 Ricker spectrum diagram (400MHz and 800MHz);

[0028] Figure 5 Schematic diagram of parameter distribution of UPML boundary conditions;

[0029] Figure 6 Schematic diagram of the forward model of asphalt pavement void disease with changes in depth direction;

[0030] Figure 7 Schematic diagram of the forward model of asphalt pavement void disease corresponding to the change of the sending and receiving spacing;

[0031] Figure 8 Schematic diagram of the forward model of asphalt pavement cavity disease corresponding to the change of antenna transmission and reception source spacing and cavity depth;

[0032] Fig. 9 This is a schematic diagram of the forward model for the control group;

[0033] Fig.10 Schematic diagram of B-scan data for the experimental and control groups;

[0034] Fig.11 This is a display image where the B-scan features are not obvious;

[0035] Fig.12 There is a reflection signal display diagram for the wave field snapshot;

[0036] Fig.13 The waveform diagrams of the experimental group and the control group are shown;

[0037] Fig.14 The process and evaluation flow chart for the SSIM-S indicator construction;

[0038] Fig.15 According to the data in Table 3, it is a three-dimensional folded surface diagram;

[0039] Fig.16 This is the B-scan image of the cavity disease with a depth of 0 cm from the pavement base;

[0040] Fig.17 The SSIM-S curve of the depth of the void disease from the pavement base;

[0041] Fig.18 It is the SSIM-S curve of TA and RA spacing;

[0042] Fig.19 It is a snapshot of the wave field at different cavity damage depths when the antenna transmission and reception spacing is 5 cm;

[0043] Fig. 20 The wave field snapshots of different antenna receiving and transmitting spacing when the cavity damage depth is 25cm. DETAILED DESCRIPTION

[0044] In order to solve the problems in the background technology, the present invention considers the B-scan data of ground penetrating radar, combines the waveform diagram, integrates the image quality assessment method in the field of computer vision, and proposes an index suitable for ground penetrating radar image quality assessment. The present invention is described in detail below in conjunction with specific embodiments. Specific implementation method one:

[0046] This embodiment is a ground penetrating radar image quality assessment method for asphalt pavement cavity disease identification.

[0047] Maxwell's equations are the theoretical basis for quantitatively analyzing the detection performance of ground penetrating radar. They describe the relationship between the electric field and the magnetic field, and explain the propagation law of high-frequency electromagnetic waves in the medium when the ground penetrating radar is working. FDTD is used as a numerical simulation method for electromagnetic fields to decompose the target into Yee units. This form is similar to the resolution unit of a computer, gradually approaching the final target. When expressing the free current density, when solving electromagnetic problems, Maxwell's equations are not enough to explain the size of each electromagnetic field. Therefore, it is also necessary to consider the relationship between the various field quantities, and combine the constitutive relationship to decompose the current density into conductor current density and applied current density.

[0048] The equations obtained after decomposition consist of two vector equations. Each of their vector equations in three-dimensional space can be decomposed into three scalar equations, equivalent to six scalar equations. The equation form is as follows:

[0049]

[0050] In the formula, E represents the electric field intensity (V / m), H represents the magnetic field intensity (A / m), J represents the free current density (A / m2), M represents the magnetic current density (V / m2), μ and ε represent the magnetic permeability and permittivity of the medium, and the subscripts x, y, and z represent the components of the corresponding physical quantities in the three directions of three-dimensional space, respectively.

[0051] In the Cartesian coordinate system, E x =E y =H z = 0. Maxwell's equations form a two-dimensional finite difference equation after sampling the spatial components in the two-dimensional finite difference time domain method:

[0052]

[0053]

[0054] in, Indicates that at time step and spatial grid points ; Δt is the time step; Δy is the space step (in the y direction); represents the electric field component at time step n and spatial grid point (i,j+1).

[0055] In FDTD numerical calculation, the stability condition will affect the calculation accuracy. Under two-dimensional FDTD conditions, the stability condition for ground penetrating radar electromagnetic wave fluctuation is expressed as formula (10):

[0056]

[0057] Where c is the speed of light, Δx and Δy are the steps of the two-dimensional Yee unit in the X and Y directions, and Δt is the time step of the wave.

[0058] Excitation source and absorbing boundary conditions:

[0059] In forward modeling, the excitation source is an important parameter. It affects the behavior of electromagnetic waves in a specific medium. Ground penetrating radar is a device for detecting hidden road defects, and the main excitation source function is the pulse function. Blackman-Harris Window and Ricker are one of the commonly used excitation source functions in the current FDTD method, and are most widely used in simulated seismic exploration and geological detection. The present invention selects the above two excitation source functions for comparison and drawing. On this basis, the time domain waveforms of the two frequencies of 400MHz and 800MHz are compared at the same time.

[0060] The function expression of Blackman-Harris Window is as follows:

[0061]

[0062] In the formula, a n is the center frequency of the excitation source, and is a fixed parameter. t represents time, and T is the duration of the excitation source function. The T value is calculated by formula (19):

[0063]

[0064] In the formula, f c is the center frequency of the excitation source function.

[0065] Through programming calculation, the time domain waveform and spectrum diagram of the main frequency of 400Mhz and 800Mhz are obtained, as shown in Figure 1-Figure 2 shown.

[0066] Ricker wavelet is a zero-phase wavelet in radar signal simulation, and its expression is as follows:

[0067]

[0068] In the formula, f c is the center frequency of the excitation source function. e is the natural logarithm. π is the circumference of a circle.

[0069] Based on the above content, through programming calculation, the time domain waveform and spectrum diagram of the main frequency of 400Mhz and 800Mhz under the Ricker excitation source are obtained, as shown in Figure 3-Figure 4 shown.

[0070] analyze Figure 1-Figure 2 and Figure 3-Figure 4The time domain waveform and spectrum of the excitation source function. The Ricker wavelet is a spike pulse shaped waveform that is similar in shape to a bell curve. This shape makes the Ricker wavelet excellent in target detection and imaging. Because it can produce high-resolution echoes and provide better target positioning capabilities. There are two different side lobes in the waveform, and the center frequency of the signal source is more suitable for simulating complex propagation medium models, such as asphalt road structure layers. In contrast, the Blackman-Harris pulse waveform is smoother and is not suitable for the detection of sudden target bodies such as road pothole diseases. The Ricker wavelet has a wider spectrum width than the Blackman-Harris pulse, which means that it can provide better frequency resolution. Obviously, the Ricker wavelet is more representative in simulating the detection of asphalt road pothole diseases.

[0071] By comparing the time domain waveforms and spectra of 400MHz and 800MHz, it is found that no matter whether the excitation source is Ricker wavelet or Blackman-Harris pulse, 400MHz is more suitable for detecting hidden pavement defects than 800MHz, especially when the depth of the defect is below the base layer. Figure 1-Figure 2 In the case of Blackman-Harris excitation, the oscillation speed of 800MHz is twice as fast as that of 400MHz. This will result in a faster propagation speed of 800MHz in the same medium, but the energy loss will also increase, and the detection depth will be greatly attenuated. Similarly, Figure 3-Figure 4 In the case of Ricker excitation, the time domain waveform of 800MHz is also narrower than that of 400MHz, which means that the period is shorter and the oscillation speed is faster. From the perspective of the spectrum diagram, the spectrum bandwidth of 800MHz is much larger than that of 400MHz. The wider the spectrum bandwidth, the shorter the duration. Therefore, the present invention takes into account the location of the cavity disease in the asphalt pavement and the multi-layer structure of the asphalt pavement, and adopts a 400MHz Ricker wavelet excitation source as the forward pulse model.

[0072] In the asphalt pavement detection process based on ground penetrating radar, it cannot be guaranteed that the detected target body is of regular shape or has broken boundaries. This type of problem will cause the spatial waveform to be truncated and abnormal electromagnetic wave reflection to occur when FDTD is used to simulate the electromagnetic field. In order to make the forward modeling results more accurate, a truncation method is set at the boundary grid of the model to achieve the propagation effect of simulating infinite space in a limited grid space. The present invention adopts a uniaxial anisotropic ideal matching layer (UPML) absorbing boundary condition. This method is suitable for uniaxial anisotropic media and performs wave field absorption at the boundary grid. UPML calculation is short in time, and the calculation formula based on Maxwell's equations is as follows:

[0073]

[0074] In the formula, S x , S y and S z They represent the uniaxial anisotropy parameter tensor, j is the imaginary unit, ω is the angular frequency, and ε is the dielectric constant.

[0075] The UPML of the two-dimensional GPR wave has four plane regions and four edge regions. z , H x , H y Components, then three parameters of UPML are expressed as equations (16)-(18). The parameter distribution diagram of UPML boundary conditions is as follows: Figure 5 shown.

[0076] S x =κ x +σ x / jωε 0 (16)

[0077] S y =κ y +σ y / jωε 0 (17)

[0078] S z =1 (18)

[0079] In the formula, the parameter κ x and κ y It is used to absorb the dropout wave reaching the UPML layer; the parameter σ x and σ y is the attenuation factor in the UPML region. In order to make the waves fully absorbed and achieve better simulation results, the parameters σ and κ should have appropriate spatial distribution in the UPML region.

[0080] σ i (i) = σ max (i / d i ) m (19)

[0081]

[0082] Among them, σ i (i) Conductivity at position i, κ i (i) Dielectric loss factor at position i, σ max The maximum value of conductivity, κ max The maximum value of dielectric loss factor; d i The length scale at spatial position i (PML medium thickness) is generally d i=(8~10)Δ, Δ space step; m is used to adjust the index of conductivity and dielectric loss factor change, generally m=4; ε r Dielectric constant.

[0083] A forward model will be needed to verify the SSIM-S index in the future. The forward model needs to be established. The above content is the theoretical basis of the forward model.

[0084] Forward modeling of asphalt pavement void disease:

[0085] The forward model here is a simulation method. First, the asphalt pavement structure and cavity disease are simulated by computer simulation, and then the simulated ground penetrating radar image is obtained by calculation, and the SSIM-S index is verified by the forward model. The new evaluation index is also inspired by the forward model, so the SSIM-S index of the present invention is determined based on the relationship between the forward model and the image quality evaluation index. At the same time, because the cavity disease characteristics of the experimental group in the forward model are affected by the multi-layer structure of the road, its characteristics are definitely not standard and obvious, and the road structure in the control group model has only one layer of structure, which greatly reduces other influencing factors, and the simulated cavity disease characteristics are relatively more standard and obvious. Therefore, by calculating the SSIM-S index of the two groups of models, the image quality of the experimental group model can be obtained.

[0086] The present invention constructs three different forward models for ground penetrating radar detection of asphalt pavement cavity damage. Figure 6-Figure 8 As shown. The first layer is the surface layer, and the material is asphalt mixture. The second layer is the base layer, and the material is cement-stabilized gravel. The third layer is the roadbed, and the main materials are compacted soil and gravel. TA is the ground penetrating radar transmitting antenna, and RA is the radar receiving antenna. When electromagnetic waves propagate between a variety of different media, the spacing between the antenna transceiver source will affect the final signal processing result, and the burial depth of the target body will affect the detection accuracy. Therefore, the present invention analyzes the impact of cavity diseases at different burial depths and different antenna transceiver spacings on the detection effect of asphalt pavement cavity diseases.

[0087] The first step is to simulate the asphalt pavement cavity disease at different depths. The pavement structure is divided into the most common three-layer model. The Yee unit size is selected as 1cm, and the size in the z direction is negligible. The center frequency of the radar excitation source is set to 400Mhz, which is the best frequency for asphalt pavement cavity disease in actual detection. The cavity disease is Figure 6-Figure 8 The cube in the figure has a boundary length of 0.3m. Generally speaking, void damage refers to the voids formed by the material falling off at the base and below the base. Therefore, in order to fully simulate the location of void damage in the asphalt pavement, the present invention places it in three different layers. Figure 6As shown in the figure, in the first model, the distance x between the cavity disease and the top of the base layer is 0m. Each time the depth steps down by 0.05m, and so on, the distance between the cavity disease and the top of the base layer in the final model is 0.4m. The excitation source is Ricker, and the transmitter position starts at 0.1m from the left end of the surface layer. After each signal is emitted by the transmitter, the transmitter and receiver step to the right by 0.01m. The boundary condition is UPML.

[0088] Secondly, the forward simulation of different antenna transmission and reception spacing is carried out. Figure 7 As shown, the distance d between the transmitting and receiving sources of the first model antenna is 0.05m. After the forward modeling is completed, the distance between the transmitting and receiving sources is changed, and the step distance of the transmitting and receiving distance d is 0.05m, and finally d is 1m.

[0089] Figure 8 The relationship between the changing antenna-transmitter-receiver source spacing and the cavity depth.

[0090] A total of 180 asphalt road cavity disease forward models are established in the present invention. The parameters such as antenna spacing and cavity depth of each model are shown in Table 1.

[0091] Table 1 Forward model of asphalt pavement void disease

[0092]

[0093]

[0094] The electromagnetic parameters of the materials of each structural layer of the asphalt pavement are set as shown in Table 2. The interior of the cavity disease is mainly composed of air, so the electromagnetic parameter values ​​are consistent with those of air.

[0095] Table 2 Electromagnetic parameters of each structural layer and cavity in the forward model of asphalt pavement

[0096]

[0097] In order to compare the distance between the best antenna and the receiving source of the ground penetrating radar, the present invention also constructs 180 control group models. The control group model corresponds one-to-one to the forward model of the asphalt pavement cavity disease. The schematic diagram of the control group model is as follows: Fig. 9 As shown. It differs from the forward model of cavity disease in the asphalt pavement structure. The present invention changes the three-layer asphalt pavement structure of the experimental group model into a single-layer pavement structure and applies it to the control group model. The location of the cavity disease is the same as that of the experimental group model. This method can reduce the loss of signal propagation in different media and obtain the best cavity disease characteristics. The present invention uses the control group model characteristics as a benchmark to compare the experimental group model characteristics and study the relationship between antenna spacing and cavity disease.

[0098] Forward model results:

[0099] The present invention simulates radar scanning data based on the established asphalt pavement cavity disease experimental group and control group models. This data is A-scan. The present invention uses programming to integrate multiple A-scans to obtain B-scan data of the asphalt pavement cavity disease experimental group and control group, a part of which is as follows Fig.10 As shown, Fig.10 The images on the left are from the experimental group, and the images on the right are from the control group.

[0100] B-scan data is a form of time domain data that displays the echo signal in the underground medium as brightness or color that changes over time. Usually, the X-axis represents the measurement position, the Y-axis represents time, and the brightness or color represents the intensity or amplitude of the echo signal. The cavity disease of the experimental group model showed non-standard hyperbolic characteristics on the B-scan, and the internal waveform oscillation was relatively messy. With the change of time domain, the diffraction characteristics on both sides of the cavity disease of the experimental group model did not tend to converge. The hyperbolic characteristics of the cavity disease of the control group model are standard, and the internal waveform oscillates beautifully. The diffraction characteristics on both sides do not exceed the top amplitude.

[0101] The evaluation indicators of forward modeling data are as follows:

[0102] Generally, B-scan data can reflect the signal characteristics of hidden diseases of asphalt pavement and the surrounding environmental noise. The stronger the signal received by the radar receiving antenna, the more obvious the characteristics are on the B-scan image. In the actual detection process, professionals also use the characteristics on the B-scan image as the basis for on-site exploration. The pavement structure layer number of the forward model of the control group is one layer, and the radar signal propagation loss inside it will be greatly reduced compared to the experimental group model. The B-scan data obtained by the control group model is also cleaner and more accurate, and can be used as a benchmark for measuring the quality of B-scan data. The present invention uses SSIM to calculate the correlation between the B-scan data of the experimental group and the control group models.

[0103] SSIM is an indicator used to measure the similarity between two images. Compared with the traditional image quality measurement indicators, it is more consistent with the human eye's judgment of image quality. When the input images are of equal size, the SSIM operation is shown in formula (23).

[0104]

[0105] In the formula, μ x and μ y is the mean value of image x and the mean value of image y, σ x and σ y is the standard deviation of the two images, σ xy is the covariance between the two images, C 1 and C 2is a constant. Image x represents the detection B-scan image (experimental group), and image y represents the B-scan image obtained by the forward model simulation of the control group;

[0106] It should be noted that: in the actual detection process, the actual ground penetrating radar image will be obtained through the above content, and then the corresponding forward model can be established according to the design file of the detection road. The average value of the actual B-scan image of the ground penetrating radar is μ x ; The average value of the image obtained by the corresponding forward model simulation is μ y ; Thus, the SSIM index can be calculated; similarly, the subsequent SSIM-S index can also be calculated.

[0107] The SSIM range is [0,1]. The larger the value, the higher the image similarity and the better the quality. However, for some special images, such as medical CT and ground penetrating radar B-scan data, SSIM is not flexible enough and is not sensitive enough to image changes. At the same time, SSIM is not accurate enough in evaluating images with large grayscale differences. There is a large difference in grayscale contrast between strong amplitude and weak amplitude in B-scan images. During the research process, it was also found that in a few forward model simulations, the hole signal was observed in the wave field snapshot to have been reflected to the receiving antenna, but the hole feature was not obvious on the B-scan image, such as Figure 11 to Figure 12 As shown, Fig.11 It can be seen that the B-scan features are not obvious. Fig.12 It can be seen that the wave field snapshot has a reflection signal.

[0108] Therefore, the present invention takes the above problems into consideration and introduces the Spearman rank correlation coefficient based on SSIM to comprehensively evaluate the forward simulation results. SROCC is a non-parametric indicator for measuring the correlation between two variables. It is also the Pearson correlation coefficient between rank variables. Its calculation formula is shown in formula (24) (25).

[0109] r s =ρ R(X)R(Y) (twenty four)

[0110]

[0111] Where n is the number of observed samples; R(x i ) represents the i-th observation level of image x, represents the average rank of the corresponding observations; r(y i ) represents the rank of the i-th observation of image y, Represents the average value of the corresponding observations; the observations are the feature vectors of the image.

[0112] The SROCC value is between -1 and 1. Coefficients close to 1 and -1 indicate a higher similarity. The present invention draws on the waveform of seismic waves during propagation and establishes a forward simulation waveform of asphalt pavement cavity disease based on forward simulation data. The present invention generates all waveforms of the experimental group and the control group by reading the forward simulation data and interpreting them through programming. Some of the waveforms are as follows: Fig.13 As shown, Fig.13 The images on the left are from the experimental group, and the images on the right are from the control group.

[0113] Due to the existence of gain and the influence of the multi-layer position of the experimental group model, the amplitude of the cavity disease waveform of the experimental group is stronger than that of the control group. However, this strong amplitude cannot change the standard characteristics of the cavity disease. Therefore, the present invention calculates the SSIM of the B-scan images of the experimental group and the control group, and uses the SROCC between the waveforms of the experimental group and the control group as an aid to comprehensively evaluate the characteristics of the cavity disease in the asphalt pavement. The present invention proposes a new evaluation index SSIM-S based on the ground penetrating radar image of the cavity disease in the asphalt pavement. The calculation formula of SSIM-S is shown in formula (26).

[0114]

[0115] The SSIM-S range is [0,1]. The larger the value, the higher the image similarity and the better the forward simulation results of the experimental group. SSIM-S not only takes into account the particularity of B-scan images, but also solves the problem that when the dielectric constant of the target and the background differ greatly during radar propagation, the medium absorbs the electromagnetic wave energy strongly, and the simulated B-scan image features are not obvious, resulting in inaccurate evaluation of the quality of the experimental group B-scan images. The SSIM-S evaluation index is shown below. Fig.14 As shown. First, SSIM is calculated by B-scan images of the forward model of the experimental group and the control group. Secondly, the Spearman rank correlation coefficient is calculated based on the waveform diagrams of the two groups. Finally, SSIM-S is obtained by substituting it into formula (26).

[0116] Evaluation of Ground Penetrating Radar Forward Simulation of Cavity Defects in Asphalt Pavement Based on SSIM-S:

[0117] The present invention uses a total of 360 forward models in the experimental group and the control group, and calculates the SSIM-S index according to the B-scan graph and waveform graph corresponding to each model. The SSIM-S index of each model is shown in Table 3.

[0118] Table 3 SSIM-S index of each model

[0119]

[0120]

[0121] In order to facilitate the display of the data in Table 3, the present invention makes a three-dimensional folded surface diagram, such as Fig.15 shown. Fig.15 SSIM-S, antenna transceiver spacing and the depth of void defects from the pavement base are the Z axis, X axis and Y axis respectively.

[0122] from Fig.15 It can be observed that the entire three-dimensional folded image is similar to a ridge. As the distance between TA and RA increases and the depth of the cavity disease becomes shallower, the overall SSIM-S value becomes lower. This shows that the larger the antenna transmission and reception distance, the less suitable it is for detecting cavity diseases close to the road surface. When such detection results are reflected on the B-scan image, the characteristics of the cavity disease and the characteristics of the asphalt pavement layer will be mixed, and the diffraction wave characteristics at both ends of the cavity disease will almost disappear. Fig.16 Therefore, in actual detection, the relevant parameters of the ground penetrating radar antenna can be better set according to this characteristic.

[0123] In the three-dimensional folded surface diagram, the present invention projects the diagram vertically on the YZ plane, and obtains the SSIM-S index curve of the antenna receiving and transmitting spacing under different cavity disease depth ranges according to the interpolation method. Fig.17 shown.

[0124] Depend on Fig.17 It can be observed that among the 20 groups of different antenna transceiver spacings, the SSIM-S value of the antenna transceiver spacing of 5 cm is the highest, and the effect of detecting the target image is the best. As the antenna transceiver spacing increases, the SSIM-S value of the cavity disease of the same depth decreases, and the B-scan image quality deteriorates. When the antenna transceiver spacing is [5cm, 25cm], the SSIM-S value of the forward model increases with the increase of the depth of the cavity disease. When the antenna transceiver spacing is [30cm, 100cm], the SSIM-S value of the forward model fluctuates repeatedly with the increase of the depth of the cavity disease. The shorter the antenna transceiver spacing, the longer the forward simulation takes and the higher the accuracy. Therefore, considering the accuracy and time of the model simulation, the present invention believes that the antenna transceiver spacing can be selected as [10cm, 20cm].

[0125] In the three-dimensional folded surface diagram, the present invention projects the diagram vertically on the XZ plane, and obtains the SSIM-S index curve diagram of the cavity disease at different depths under different antenna transceiver spacing ranges according to the interpolation method. Fig.18 shown.

[0126] Depend on Fig.18It can be observed that among the 9 groups of different cavity disease depths, the SSIM-S value of the cavity disease 40cm away from the pavement base is relatively the highest, and the characteristics of the detected target image are the best. As the depth of the cavity disease increases, the SSIM-S value of the same antenna transceiver spacing becomes larger, which means that its B-scan image quality improves. When the antenna transceiver spacing is less than 25cm, SSIM-S is positively correlated with the cavity disease depth and negatively correlated with the antenna transceiver spacing. When the antenna transceiver spacing is greater than 25cm, the above rule changes. Therefore, the present invention believes that in actual detection or forward simulation, when the antenna transceiver spacing cannot be changed, deeper cavity diseases should be detected as much as possible. This method can maximize the accuracy of detection.

[0127] In view of the above reasons, the present invention believes that the increase in the antenna receiving and transmitting distance first leads to a longer time for the direct wave to reach the antenna receiving end. The direct wave, the road surface reflected wave and the cavity disease interface reflected wave arrive at the antenna receiving end together. In the B-scan image, the characteristics of the three waveforms are mixed together, making it difficult to distinguish the relationship between the asphalt pavement layers. Secondly, the deeper the cavity disease, the road surface medium will affect the propagation speed of the electromagnetic wave, resulting in a longer time for the disease signal to reach the antenna receiving end. The arrival time of the disease signal is generally longer than the arrival time of the direct wave and the road surface layer reflected wave. The signal interference between the three will be reduced. Fig.19 Wavefield snapshots of different cavity damage depths are shown.

[0128] Therefore, in Fig.18 It is shown in the figure that under different antenna receiving and transmitting distances, for most antenna receiving and transmitting distances, the deeper the cavity defect is, the higher the SSIM-S is. When the receiving and transmitting distance of a few antennas is between [60cm, 85cm], the arrival time of the defect signal and the direct wave signal is similar. Fig. 20 As shown, when the antenna transmission and reception distances are 60cm and 85cm, the time it takes for multiple radar signals to arrive at the receiving end is similar. Specific implementation method 2:

[0130] This embodiment is a ground penetrating radar image quality assessment system for asphalt pavement cavity disease identification, which is a computer program corresponding to the ground penetrating radar image quality assessment method for asphalt pavement cavity disease identification described in the first embodiment. More specifically, the ground penetrating radar image quality assessment system for asphalt pavement cavity disease identification described in this embodiment includes:

[0131] Ground penetrating radar image acquisition unit: for the detection road, obtain the ground penetrating radar image of the detection road, that is, detect the B-scan image; in the process of obtaining the ground penetrating radar image of the detection road, set the antenna transceiver spacing to be less than 25cm, preferably, set the antenna transceiver spacing to [10cm, 20cm].

[0132] Control group forward model establishment unit: establish a corresponding control group forward model according to the design file of the detection road, the control group is to change the multi-layer asphalt pavement structure in the design file of the detection road into a single-layer pavement structure; the forward model uses a 400MHz Ricker wavelet as the excitation source signal. The antenna transceiver spacing in the forward model is equal to the antenna transceiver spacing set in the process of obtaining the ground penetrating radar image of the detection road.

[0133] Control group image acquisition unit: based on the established control group forward model, simulates the control group B-scan image;

[0134] Quality assessment unit: performs image quality assessment of B-scan images based on the SSIM-S index;

[0135]

[0136] Where SSIM(x,y) represents the structural similarity index between image x and image y, SROCC(x,y) represents the Spearman rank correlation coefficient index between image x and image y, and image x and image y are the detection B-scan image and the control group B-scan image, respectively.

[0137] The present invention may also have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. A ground penetrating radar image quality assessment method for asphalt pavement cavity disease identification, characterized in that: The following steps are involved: For the inspection road, a ground penetrating radar image of the inspection road, i.e., the inspection B-scan image, is obtained. At the same time, a corresponding control group forward model is established according to the design file of the inspection road. The control group is a control group that changes the multi-layer asphalt pavement structure in the design file of the inspection road into a single-layer pavement structure; the image simulated by the control group forward model is recorded as the control group B-scan image; then the image quality of the inspection B-scan image is evaluated according to the SSIM-S index; Where SSIM(x,y) represents the structural similarity index between image x and image y, SROCC(x,y) represents the Spearman rank correlation coefficient index between image x and image y, and image x and image y are the detection B-scan image and the control group B-scan image, respectively.

2. The method for evaluating the quality of ground penetrating radar images for identifying asphalt pavement voids according to claim 1, characterized in that: The forward model uses a 400 MHz Ricker wavelet as an excitation source signal.

3. The method for evaluating the quality of ground penetrating radar images for identifying asphalt pavement voids according to claim 2, characterized in that: In the process of acquiring the ground penetrating radar image of the detection road, the antenna transmission and reception distance is set to be less than 25 cm.

4. The method for evaluating the quality of ground penetrating radar images for identifying asphalt pavement voids according to claim 3 is characterized in that: When acquiring the ground penetrating radar image of the detected road, the antenna transmit and receive spacing is set to [10cm, 20cm].

5. A method for evaluating ground penetrating radar image quality for identifying asphalt pavement voids according to claim 3 or 4, characterized in that: The antenna transceiver spacing in the forward model is equal to the antenna transceiver spacing set in the process of acquiring the ground penetrating radar image of the detection road.

6. A ground penetrating radar image quality assessment system for asphalt pavement cavity disease identification, characterized in that: include: Ground penetrating radar image acquisition unit: for the detection road, acquire the ground penetrating radar image of the detection road, that is, detect the B-scan image; Control group forward model building unit: a corresponding control group forward model is built according to the design documents of the test road. The control group is to change the multi-layer asphalt pavement structure in the design documents of the test road into a single-layer pavement structure; Control group image acquisition unit: based on the established control group forward model, simulate the control group B-scan image; Quality assessment unit: performs image quality assessment of B-scan images based on the SSIM-S index; Where SSIM(x,y) represents the structural similarity index between image x and image y, SROCC(x,y) represents the Spearman rank correlation coefficient index between image x and image y, and image x and image y are the detection B-scan image and the control group B-scan image, respectively.

7. A ground penetrating radar image quality assessment system for asphalt pavement cavity disease identification according to claim 6, characterized in that: The forward model uses a 400 MHz Ricker wavelet as an excitation source signal.

8. The ground penetrating radar image quality assessment system for asphalt pavement cavity disease identification according to claim 7, characterized in that: In the process of acquiring the ground penetrating radar image of the detection road, the antenna transmission and reception distance is set to be less than 25 cm.

9. A ground penetrating radar image quality assessment system for asphalt pavement cavity disease identification according to claim 8, characterized in that: When acquiring the ground penetrating radar image of the detected road, the antenna transmit and receive spacing is set to [10cm, 20cm].

10. A ground penetrating radar image quality assessment system for asphalt pavement cavity disease identification according to claim 8 or 9, characterized in that: The antenna transceiver spacing in the forward model is equal to the antenna transceiver spacing set in the process of acquiring the ground penetrating radar image of the detection road.

Images