Road hidden disease identification, positioning and segmentation method based on three-dimensional ground penetrating radar

Abstract

The invention discloses a road hidden disease identification, positioning and segmentation method based on a three-dimensional ground penetrating radar, and belongs to the technical field of road hidden disease detection. The problems that the traditional method is low in road hidden disease detection efficiency and poor in accuracy and cannot perform automatic three-dimensional reconstruction on road diseases are solved. According to the invention, by establishing a YOLO-nnUNet three-dimensional ground penetrating radar disease detection-segmentation joint algorithm, a specific method for rapid positioning, identification and three-dimensional segmentation of hidden road diseases in three-dimensional ground penetrating radar data is provided. According to the method, a complete disease identification process including three-dimensional ground penetrating radar data acquisition, two-dimensional disease defect detection, three-dimensional disease clustering positioning and three-dimensional disease boundary semantic segmentation is realized, the efficiency and accuracy of road hidden disease detection are improved, meanwhile, road hidden disease segmentation data of the three-dimensional ground penetrating radar can be obtained, and the road hidden disease detection accuracy is improved. And outputting a three-dimensional reconstruction result of the road disease. The method can be applied to road hidden disease detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of road hidden disease detection, and in particular relates to a method for identifying, locating and segmenting road hidden diseases based on three-dimensional ground penetrating radar. Background Art

[0002] With the rapid development of transportation infrastructure and the continued growth in the number of motor vehicles, road usage and load intensity are increasing. However, hidden road defects (such as voids, cracks, spalling, and cavities) are buried beneath the pavement, making them difficult to detect directly through traditional visual inspection or surface monitoring methods. Once these defects spread, they can seriously impact the structural safety and durability of roads, and may even lead to safety hazards such as road collapse and traffic accidents. Currently, methods for detecting hidden road defects primarily include manual inspections, geological surveys, and radar detection. Three-dimensional ground-penetrating radar (GPR), as a non-destructive detection technology, enables high-precision scanning of road structures. However, traditional methods rely on manual analysis of radar echo images, making automated and intelligent defect detection difficult. Therefore, to address the low efficiency and poor accuracy of traditional methods for detecting hidden road defects, a method that replaces traditional manual analysis with intelligent detection is crucial for promoting automated, intelligent, and real-time road defect detection. Furthermore, traditional methods lack automated and rapid three-dimensional reconstruction of road defects. Summary of the Invention

[0003] The purpose of this invention is to solve the problems of low efficiency and poor accuracy in detecting hidden road defects and the inability to automatically reconstruct road defects in three dimensions by traditional methods, and to propose a method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar.

[0004] The technical solution adopted by the present invention to solve the above technical problems is: a method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar, the method specifically comprising the following steps:

[0005] Step 1: Use 3D ground penetrating radar equipment to detect the 3D information of the underground space of the road to be detected, obtain 3D radar signal data, and record the radar signal data corresponding to the coordinate point (i, j, k) as D i,j,k ;

[0006] Wherein, i∈[0,N], j∈[0,L], k∈[0,M], the 3D GPR device moves in a straight line in the horizontal direction, the moving direction of the 3D GPR device is taken as the x-axis direction, N represents the x-axis coordinate corresponding to the maximum moving distance of the radar, L represents the y-axis coordinate corresponding to the maximum detection width of the radar in the direction perpendicular to the x-axis, and M represents the z-axis coordinate corresponding to the maximum depth of the radar electromagnetic wave propagating underground;

[0007] By processing the three-dimensional radar signal data, the three-dimensional radar signal data is converted into three-dimensional radar data grayscale blocks;

[0008] Step 2: Slice the grayscale block of the 3D radar data layer by layer along the z-axis to obtain a 2D grayscale image sequence of the road to be inspected. Input each grayscale image in the 2D grayscale image sequence of the road to be inspected into the trained YOLO 2D defect detection model. The YOLO 2D defect detection model outputs a defect detection result for each 2D grayscale image. The defect detection result is a bounding box of the defect location in the 2D grayscale image.

[0009] Calculate the center coordinates of the bounding box of each disease location and record the center coordinates of each bounding box And the center coordinates The z-axis coordinate of the corresponding two-dimensional grayscale image;

[0010] Step 3: Calculate the adaptive neighborhood radius ε0 of the DBSCAN algorithm according to the center coordinates of the bounding box of the location of the disease in each two-dimensional grayscale image, and cluster the diseases in each two-dimensional grayscale image according to the adaptive neighborhood radius ε0;

[0011] Merge the diseases that are clustered into one category and spatially connected into a 3D disease candidate region, and calculate the bounding box, center position and size information of each 3D disease candidate region;

[0012] Step 4: Cut out each 3D defect candidate data block from the 3D radar data block of the road to be inspected based on the bounding box, center position, and size information of the 3D defect candidate area;

[0013] Step 5: Input each 3D defect candidate data block into the trained nnUNet 3D defect segmentation model, perform semantic segmentation on each defect candidate area, and obtain the semantically segmented defect boundary;

[0014] Then, the connected domain analysis and Gaussian filtering method are used to process the disease boundary obtained by semantic segmentation to obtain the processed disease boundary.

[0015] Furthermore, the processing of the three-dimensional radar signal data is specifically as follows:

[0016] The three-dimensional radar signal data is subjected to dynamic compression, data normalization and gray value mapping processing in sequence.

[0017] Furthermore, the three-dimensional radar signal data is processed to convert the three-dimensional radar signal data into three-dimensional radar data grayscale blocks; the specific process is:

[0018] Step 11: Dynamically compress the acquired three-dimensional radar signal data through logarithmic transformation:

[0019] P i,j,k =log(1+αD i,j,k )

[0020] Among them, P i,j,k Represents the radar signal data corresponding to the coordinate point (i, j, k) after dynamic compression;

[0021] α represents the signal amplification factor;

[0022] Step 1 and 2: Calculate the mean μ of the radar signal data corresponding to each coordinate point after dynamic compression:

[0023]

[0024] Calculate the standard deviation σ of the radar signal data corresponding to each coordinate point after dynamic compression:

[0025]

[0026] The radar signal data corresponding to each coordinate point is normalized according to the mean μ and standard deviation σ:

[0027]

[0028] Among them, M i,j,k Represents the normalized data corresponding to the coordinate point (i, j, k);

[0029] Step 1-3: Perform grayscale value mapping on the normalized data corresponding to each coordinate point, and use the grayscale value mapping result of the coordinate point (i, j, k) as the grayscale value of the voxel (i, j, k):

[0030]

[0031] Among them, min(M) represents the minimum value of the normalized data corresponding to each coordinate point;

[0032] max(M) represents the maximum value of the normalized data corresponding to each coordinate point;

[0033] H i,j,kRepresents the grayscale value of voxel (i,j,k).

[0034] Furthermore, the center coordinates of the bounding box of the disease location are:

[0035]

[0036] in, The minimum horizontal coordinate of the pixel in the bounding box representing the location of the a-th disease;

[0037] The maximum horizontal coordinate of the pixel in the bounding box representing the location of the a-th disease;

[0038] The minimum vertical coordinate of the pixel in the bounding box representing the location of the a-th disease;

[0039] The maximum vertical coordinate of the pixel in the bounding box representing the location of the a-th disease;

[0040] The center coordinates of the bounding box representing the location of the a-th disease.

[0041] Furthermore, the adaptive neighborhood radius ε0 is calculated as follows:

[0042] For the center coordinates of the bounding box of the ath disease location in the two-dimensional grayscale image sequence Calculate the center coordinates of the bounding boxes of other defects in the two-dimensional grayscale image sequence and The distance is calculated, and then the calculated distances are arranged in ascending order, and the distance vector d is obtained. (k) for:

[0043]

[0044] in, Represents the distance sequence d (k) The 1st, 2nd, ..., N'th distance in ;

[0045] N' represents the total number of diseases except the a-th disease in the two-dimensional grayscale image sequence;

[0046] According to the distance vector d (k) Get N' point pairs (x n ,y n ), n=1,2,…,N':

[0047] x n =n,

[0048] Then according to N' points (x n,y n ) to obtain the cumulative sum function curve, and then obtain the adaptive neighborhood radius based on the cumulative sum function curve:

[0049]

[0050] Among them, (x'(n), y'(n)) represents the point pair (x n ,y n ) on the cumulative sum function curve, (x)(n), y)(n)) represents the point pair (x n ,y n ) on the cumulative sum function curve.

[0051] Furthermore, the specific process of the connected domain analysis is as follows:

[0052] Calculate the number of voxels in each connected diseased area respectively, and record the number of voxels in the i'th connected diseased area as V i’ , and remove the diseased areas with voxel numbers less than the threshold to obtain the remaining diseased areas.

[0053] Furthermore, the specific process of the Gaussian filtering method is:

[0054] The Gaussian filtering method is used to smooth the boundary voxels of the remaining diseased area:

[0055]

[0056] Among them, (x, y, z) represents the boundary voxel of the remaining diseased area;

[0057] H disease (x+p,y+q,z+r) represents the grayscale value of voxel (x+p,y+q,z+r) before smoothing;

[0058] Represents the Gaussian function value at the (p,q,r) position;

[0059] H smooth (x, y, z) represents the grayscale value of the boundary voxel (x, y, z) in the remaining diseased area after smoothing;

[0060] p,q,r represent the Gaussian function values The offset in ,p,q,r∈[-3σ0,3σ0], σ0 represents the standard deviation of the Gaussian distribution.

[0061] Furthermore, the Gaussian function value at the (p, q, r) position is:

[0062]

[0063] The beneficial effects of the present invention are:

[0064] This paper proposes a specific method for rapidly locating, identifying, and 3D segmenting hidden road defects in 3D GPR data by establishing a YOLO-nnUNet 3D GPR defect detection and segmentation joint algorithm. In the tasks of locating, identifying, and 3D segmenting hidden road defects, the paper optimizes the workflow of the hidden road defect segmentation task by combining the characteristics of the small number and high damage of hidden road defects with the high efficiency of 2D defect detection. This completes the defect recognition process from 3D GPR data acquisition, 2D defect defect detection, 3D defect clustering, and 3D defect boundary semantic segmentation, improving the efficiency and accuracy of hidden road defect detection. Furthermore, the method can obtain 3D GPR hidden road defect segmentation data and output 3D reconstruction results of road defects.

[0065] The present invention provides strong technical support for the automated detection and intelligent evaluation of hidden road defects by improving the effective data screening and three-dimensional extraction methods of hidden road defects, and provides a data source for road management and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a flow chart of a method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar according to the present invention;

[0067] Figure 2 It is a schematic diagram of a two-dimensional annotated image;

[0068] Figure 3 It is a schematic diagram of the 3D voxel-level annotation results;

[0069] Figure 4 It is a slice effect diagram of the three-dimensional reconstruction result of the disease;

[0070] Figure 5 It is a three-dimensional reconstruction effect diagram of the disease. DETAILED DESCRIPTION

[0071] Specific embodiment 1: This embodiment describes a method for identifying, locating, and segmenting hidden road defects based on a three-dimensional ground penetrating radar, the method specifically comprising the following steps:

[0072] Step 1: Use 3D ground penetrating radar equipment to detect the 3D information of the underground space of the road to be detected, obtain 3D radar signal data, and record the radar signal data corresponding to the coordinate point (i, j, k) as D i,j,x ;

[0073] Among them, i∈[0,N], j∈[0,L], k∈[0,M], the three-dimensional ground penetrating radar equipment moves in a straight line in the horizontal direction, the moving direction of the three-dimensional ground penetrating radar equipment is taken as the x-axis direction, N represents the x-axis coordinate corresponding to the maximum moving distance of the radar, L represents the y-axis coordinate corresponding to the maximum detection width of the radar in the direction perpendicular to the x-axis, M represents the z-axis coordinate corresponding to the maximum depth of the radar electromagnetic wave propagating underground, D i,j,k Represents the radar echo data corresponding to the coordinate point (i, j, k); it should be noted that the x-axis, y-axis, and z-axis in the present invention are the three coordinate axes of the three-dimensional rectangular coordinate system, that is, the three-dimensional ground penetrating radar device moves along the x-axis direction of the three-dimensional rectangular coordinate system;

[0074] By processing the three-dimensional radar signal data, the three-dimensional radar signal data is converted into three-dimensional radar data grayscale blocks;

[0075] Step 2: Slice the grayscale block of the 3D radar data layer by layer along the z-axis to obtain a 2D grayscale image sequence of the road to be inspected. Input each grayscale image in the 2D grayscale image sequence of the road to be inspected into the trained YOLO 2D defect detection model. The YOLO 2D defect detection model outputs a defect detection result for each 2D grayscale image. The defect detection result is a bounding box of the defect location in the 2D grayscale image.

[0076] Calculate the center coordinates of the bounding box of each disease location and record the center coordinates of each bounding box And the center coordinates The z-axis coordinate of the corresponding two-dimensional grayscale image;

[0077] Step 3: Calculate the adaptive neighborhood radius ε0 of the DBSCAN algorithm according to the center coordinates of the bounding box of the location of the disease in each two-dimensional grayscale image, and cluster the diseases in each two-dimensional grayscale image according to the adaptive neighborhood radius ε0;

[0078] Merge the defects that are clustered into one category and spatially connected (determined by the z-axis coordinate of the 2D grayscale image) into a 3D defect candidate region (i.e., extract continuous and spatially correlated 3D defect candidate regions based on the clustering results), and calculate the bounding box, center position, and size information of each 3D defect candidate region.

[0079] Step 4: Cut out each 3D defect candidate data block from the 3D radar data block of the road to be inspected based on the bounding box, center position, and size information of the 3D defect candidate area;

[0080] Step 5: Input each 3D defect candidate data block into the trained nnUNet 3D defect segmentation model, perform fine-grained semantic segmentation on each defect candidate area, and obtain the semantically segmented defect boundary;

[0081] Then, the connected domain analysis and Gaussian filtering method are used to process the disease boundary obtained by semantic segmentation to obtain the processed disease boundary, that is, to obtain the complete road hidden disease segmentation data of 3D ground penetrating radar.

[0082] like Figure 1 As shown, the training methods of the YOLO two-dimensional disease detection model and the nnUNet three-dimensional disease segmentation model in the present invention are:

[0083] A 3D ground-penetrating radar (GPR) device was used to detect 3D information of the underground space on the experimental road. The 3D radar data block (x0, y0, z0) with a 3D shape of (440, 66, 170) was obtained. The model pre-training annotation was then performed on this data block. The specific steps include:

[0084] 1) Slice the 3D radar data block layer by layer along the z-axis to generate 2D radar slices. Use the image annotation tool Labelmg to annotate the location and category labels of the disease in the 2D slices. Figure 2 The annotation results are converted into the input format of the YOLO model, and the YOLO model is pre-trained using the converted 2D images to obtain the pre-trained weights for 2D disease detection.

[0085] 2) Use 3D Slicer, a 3D voxel annotation tool, to perform voxel-level annotation on the data block to generate a 3D segmentation mask, such as Figure 3 The data is divided into two parts: training set and validation set. The training set and validation set are used to train and validate the model, and the pre-training weights for 3D disease segmentation are obtained.

[0086] The trained model is applied as follows:

[0087] Slice the 3D data layer by layer along the z-axis, read the pre-trained weight file of the YOLO model, input the 2D grayscale image sequence one by one into the YOLO 2D disease detection model for disease detection, and record the disease bounding box and slice index;

[0088] Using the calculated neighborhood radius ε0 = 10 mm and the minimum number of samples N min =2 to remove noise points, and then use the spatial center point coordinates and slice index of the two-dimensional detection frame as clustering features. The DBSCAN three-dimensional clustering algorithm is used to perform spatial clustering analysis on the detection results, and the center position and size information of each disease candidate area are calculated.

[0089] The candidate disease area is cropped from the original 3D radar data block. During the cropping process, the necessary axis conversion and flipping processing are performed on the data block of the cropped area, and each data block is converted into a standard NIfTI format file and a standard dataset structure;

[0090] The generated disease area data is input into the nnUNet model for semantic segmentation to obtain accurate disease boundaries and form a 3D segmentation file of hidden road diseases. Finally, the segmentation results are merged into complete 3D reconstruction data using the spatial positioning information of each candidate area. The slicing effect of the 3D reconstruction data is as follows: Figure 4 As shown, the three-dimensional reconstruction effect is as follows Figure 5 shown.

[0091] Specific embodiment 2: This embodiment differs from specific embodiment 1 in that the three-dimensional radar signal data is processed as follows:

[0092] The three-dimensional radar signal data is subjected to dynamic compression, data normalization and gray value mapping processing in sequence.

[0093] Other steps and parameters are the same as those in the first embodiment.

[0094] Specific embodiment three: This embodiment differs from specific embodiments one or two in that the three-dimensional radar signal data is processed to convert the three-dimensional radar signal data into three-dimensional radar data grayscale blocks; the specific process is:

[0095] Step 11: Dynamically compress the acquired three-dimensional radar signal data through logarithmic transformation:

[0096] P i,j,k =log(1+αD i,j,k )

[0097] Among them, P i,j,k Represents the radar signal data corresponding to the coordinate point (i, j, k) after dynamic compression;

[0098] α represents the signal amplification factor, the default value is 10 4 ;

[0099] Step 1 and 2: Calculate the mean μ of the radar signal data corresponding to each coordinate point after dynamic compression:

[0100]

[0101] Calculate the standard deviation σ of the radar signal data corresponding to each coordinate point after dynamic compression:

[0102]

[0103] The radar signal data corresponding to each coordinate point is normalized according to the mean μ and standard deviation σ:

[0104]

[0105] Among them, M i,j,k Represents the normalized data corresponding to the coordinate point (i, j, k);

[0106] Step 13: Perform grayscale value mapping on the normalized data corresponding to each coordinate point, and use the grayscale value mapping result of the coordinate point (i, j, k) as the grayscale value of the voxel (i, h, k):

[0107]

[0108] Among them, min(M) represents the minimum value of the normalized data corresponding to each coordinate point;

[0109] max(M) represents the maximum value of the normalized data corresponding to each coordinate point;

[0110] H i,j,k Represents the grayscale value of voxel (i,j,k).

[0111] Other steps and parameters are the same as those in the first or second embodiment.

[0112] Through the processing of this embodiment, the three-dimensional radar signal data is mapped into three-dimensional radar data grayscale blocks within the range of [0, 255].

[0113] Specific embodiment 4: This embodiment differs from any one of specific embodiments 1 to 3 in that the center coordinates of the boundary box of the location of the disease are:

[0114]

[0115] in, The minimum horizontal coordinate of the pixel in the bounding box representing the location of the a-th disease;

[0116] The maximum horizontal coordinate of the pixel in the bounding box representing the location of the a-th disease;

[0117] The minimum vertical coordinate of the pixel in the bounding box representing the location of the a-th disease;

[0118] The maximum vertical coordinate of the pixel in the bounding box representing the location of the a-th disease;

[0119] The center coordinates of the bounding box representing the location of the a-th disease.

[0120] The other steps and parameters are the same as those in the first to third embodiments.

[0121] Specific embodiment 5: This embodiment differs from any one of specific embodiments 1 to 4 in that the adaptive neighborhood radius ε0 is calculated as follows:

[0122] For the center coordinates of the bounding box of the ath disease location in the two-dimensional grayscale image sequence Calculate the center coordinates of the bounding boxes of other defects in the two-dimensional grayscale image sequence and The distance is calculated, and then the calculated distances are arranged in ascending order, and the distance vector d is obtained. (k) for:

[0123]

[0124] in, Represents the distance sequence d (k) The 1st, 2nd, ..., N'th distance in ;

[0125] N' represents the total number of diseases except the a-th disease in the two-dimensional grayscale image sequence;

[0126] According to the distance vector d (k) Get N' point pairs (x n ,y n ), n=1,2,…,N':

[0127] x n =n,

[0128] Then according to N' points (x n ,y n ) to obtain the cumulative sum function curve, and then obtain the adaptive neighborhood radius based on the cumulative sum function curve:

[0129]

[0130] Among them, (x'(n), y'(n)) represents the point pair (x n ,y n ) on the cumulative sum function curve, (x)(n), y)(n)) represents the point pair (x n ,y n ) on the cumulative sum function curve.

[0131] The other steps and parameters are the same as those in the first to fourth embodiments.

[0132] The adaptive neighborhood radius ε0 calculated according to the present invention and the set minimum number of samples N minThe DBSCAN clustering algorithm can be executed to perform spatial clustering analysis on the clustering features of the two-dimensional grayscale image sequence.

[0133] Specific embodiment 6: This embodiment differs from any one of specific embodiments 1 to 5 in that the specific process of the connected domain analysis is as follows:

[0134] Calculate the number of voxels in each connected diseased area respectively, and record the number of voxels in the i'th connected diseased area as V i’ , and remove the diseased areas with voxel numbers less than the threshold to obtain the remaining diseased areas.

[0135] The other steps and parameters are the same as those in the first to fifth embodiments.

[0136] Specific embodiment 7: This embodiment differs from any one of specific embodiments 1 to 6 in that the specific process of the Gaussian filtering method is as follows:

[0137] The Gaussian filtering method is used to smooth the boundary voxels of the remaining diseased area:

[0138]

[0139] Among them, (x, y, z) represents the boundary voxel of the remaining diseased area;

[0140] H disease (x+p,y+q,z+r) represents the grayscale value of voxel (x+p,y+q,z+r) before smoothing;

[0141] Represents the Gaussian function value at the (p,q,r) position;

[0142] H smooth (x, y, z) represents the grayscale value of the boundary voxel (x, y, z) in the remaining diseased area after smoothing;

[0143] p,q,r represent the Gaussian function values The offset in , that is, the overlapping area between each voxel of the filter and the input data, p,q,r∈[-3σ0,3σ0], σ0 represents the standard deviation of the Gaussian distribution.

[0144] The other steps and parameters are the same as those in the first to sixth embodiments.

[0145] The edge discontinuity in the 3D reconstructed data block can be reduced through processing.

[0146] Specific embodiment eight: This embodiment differs from any one of specific embodiments one to seven in that the Gaussian function value of the (p, q, r) position is:

[0147]

[0148] The other steps and parameters are the same as those in the first to seventh embodiments.

[0149] The above examples are merely illustrative of the calculation model and process of the present invention and are not intended to limit the embodiments of the present invention. Persons skilled in the art will readily appreciate that other variations or modifications based on the above description are possible. This list of embodiments is not exhaustive; however, any obvious variations or modifications derived from the technical solution of the present invention remain within the scope of protection of the present invention.

Claims

1. A method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar, characterized in that: The method specifically comprises the following steps: Step 1: Use 3D ground penetrating radar equipment to detect the 3D information of the underground space of the road to be detected, obtain 3D radar signal data, and record the radar signal data corresponding to the coordinate point (i, j, k) as D i,j,k ; Wherein, i∈[0,N], j∈[0,L], k∈[0,M], the 3D GPR device moves in a straight line in the horizontal direction, the moving direction of the 3D GPR device is taken as the x-axis direction, N represents the x-axis coordinate corresponding to the maximum moving distance of the radar, L represents the y-axis coordinate corresponding to the maximum detection width of the radar in the direction perpendicular to the x-axis, and M represents the z-axis coordinate corresponding to the maximum depth of the radar electromagnetic wave propagating underground; By processing the three-dimensional radar signal data, the three-dimensional radar signal data is converted into three-dimensional radar data grayscale blocks; Step 2: Slice the grayscale block of the 3D radar data layer by layer along the z-axis to obtain a 2D grayscale image sequence of the road to be inspected. Input each grayscale image in the 2D grayscale image sequence of the road to be inspected into the trained YOLO 2D defect detection model. The YOLO 2D defect detection model outputs a defect detection result for each 2D grayscale image. The defect detection result is a bounding box of the defect location in the 2D grayscale image. Calculate the center coordinates of the bounding box of each disease location and record the center coordinates of each bounding box And the center coordinates The z-axis coordinate of the corresponding two-dimensional grayscale image; Step 3: Calculate the adaptive neighborhood radius ε0 of the DBSCAN algorithm according to the center coordinates of the bounding box of the location of the disease in each two-dimensional grayscale image, and cluster the diseases in each two-dimensional grayscale image according to the adaptive neighborhood radius ε0; Merge the diseases that are clustered into one category and spatially connected into a 3D disease candidate region, and calculate the bounding box, center position and size information of each 3D disease candidate region; Step 4: Cut out each 3D defect candidate data block from the 3D radar data block of the road to be inspected based on the bounding box, center position, and size information of the 3D defect candidate area; Step 5: Input each 3D defect candidate data block into the trained nnUNet 3D defect segmentation model, perform semantic segmentation on each defect candidate area, and obtain the semantically segmented defect boundary; Then, the connected domain analysis and Gaussian filtering method are used to process the disease boundary obtained by semantic segmentation to obtain the processed disease boundary.

2. The method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar according to claim 1, characterized in that: The processing of the three-dimensional radar signal data is specifically as follows: The three-dimensional radar signal data is subjected to dynamic compression, data normalization and gray value mapping processing in sequence.

3. The method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar according to claim 2, characterized in that: The three-dimensional radar signal data is processed to convert the three-dimensional radar signal data into three-dimensional radar data grayscale blocks; the specific process is: Step 11: Dynamically compress the acquired three-dimensional radar signal data through logarithmic transformation: P i,j,k =log(1+αD i,j,k ) Among them, P i,j,k Represents the radar signal data corresponding to the coordinate point (i, j, k) after dynamic compression; α represents the signal amplification factor; Step 1 and 2: Calculate the mean μ of the radar signal data corresponding to each coordinate point after dynamic compression: Calculate the standard deviation σ of the radar signal data corresponding to each coordinate point after dynamic compression: The radar signal data corresponding to each coordinate point is normalized according to the mean μ and standard deviation σ: Among them, M i,j,k Represents the normalized data corresponding to the coordinate point (i, j, k); Step 1-3: Perform grayscale value mapping on the normalized data corresponding to each coordinate point, and use the grayscale value mapping result of the coordinate point (i, j, k) as the grayscale value of the voxel (i, j, k): Among them, min(M) represents the minimum value of the normalized data corresponding to each coordinate point; max(M) represents the maximum value of the normalized data corresponding to each coordinate point; H i,j,k Represents the grayscale value of voxel (i,j,k).

4. The method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar according to claim 3, characterized in that: The center coordinates of the bounding box where the disease is located are: in, The minimum horizontal coordinate of the pixel in the bounding box representing the location of the a-th disease; The maximum horizontal coordinate of the pixel in the bounding box representing the location of the a-th disease; The minimum vertical coordinate of the pixel in the bounding box representing the location of the a-th disease; The maximum vertical coordinate of the pixel in the bounding box representing the location of the a-th disease; The center coordinates of the bounding box representing the location of the a-th disease.

5. The method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar according to claim 4 is characterized in that: The calculation method of the adaptive neighborhood radius ε0 is: For the center coordinates of the bounding box of the ath disease location in the two-dimensional grayscale image sequence Calculate the center coordinates of the bounding boxes of other defects in the two-dimensional grayscale image sequence and The distance is calculated, and then the calculated distances are arranged in ascending order, and the distance vector d is obtained. (k) for: in, Represents the distance sequence d (k) The 1st, 2nd, ..., N'th distance in ; N' represents the total number of diseases except the a-th disease in the two-dimensional grayscale image sequence; According to the distance vector d (k) Get N' point pairs (x n ,y n ), n=1,2,…,N': Then according to N' points (x n ,y n ) to obtain the cumulative sum function curve, and then obtain the adaptive neighborhood radius based on the cumulative sum function curve: Among them, (x'(n), y'(n)) represents the point pair (x n ,y n ) on the cumulative sum function curve, (x)(n), y)(n)) represents the point pair (x n ,y n ) on the cumulative sum function curve.

6. The method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar according to claim 5, characterized in that: The specific process of the connected domain analysis is as follows: Calculate the number of voxels in each connected diseased area respectively, and record the number of voxels in the i'th connected diseased area as V i’ , and remove the diseased areas with voxel numbers less than the threshold to obtain the remaining diseased areas.

7. The method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar according to claim 6, characterized in that: The specific process of the Gaussian filtering method is as follows: The Gaussian filtering method is used to smooth the boundary voxels of the remaining diseased area: Among them, (x, y, z) represents the boundary voxel of the remaining diseased area; H disease (x+p,y+q,z+r) represents the grayscale value of voxel (x+p,y+q,z+r) before smoothing; Represents the Gaussian function value at the (p,q,r) position; H smooth (x, y, z) represents the grayscale value of the boundary voxel (x, y, z) in the remaining diseased area after smoothing; p,q,r represent the Gaussian function values The offset in ,p,q,r∈[-3σ0,3σ0], σ0 represents the standard deviation of the Gaussian distribution.

8. The method for identifying, locating and segmenting hidden road defects based on three-dimensional ground penetrating radar according to claim 7, characterized in that: The Gaussian function value of the (p,q,r) position is:

Images