3D Reconstruction of embedded object using ground penetrating radar

Ground Penetrating Radar (GPR) is a non-destructive device widely used to locate and map underground utilities such as pipes, cables, etc. Its principle is based on the reflection signal of a transmitter-receiver antenna that strikes underground objects by means of the propagation of a short pulse of electromagnetic waves into the ground. The GPR will produce a hyperbolic curve as a result of the object’s presence. Accurate interpretation of hyperbola curves is greatly important and highly depends on user expertise; thus, it is considered a challenge. To address this issue, this study aims to develop 3D reconstructions of embedded objects. In this study, C-scan images were acquired, and 3D interpolation and the Synthetic Aperture Focusing Technique (SAFT) were introduced. In this framework, the acquired data is subjected to pre-processing techniques via time-zero correction, background removal using average background subtraction, and Kirchoff’s migration method. The software Reflex 3D Scan has been used to analyse and preprocess the 3D reconstruction of embedded objects. The obtained results show that 3D interpolation and SAFT methods are not only able to reconstruct 3D models but are also able to reveal information on the dimension and location of the buried object represented by voxel points in the 3D space cube.


Introduction
The idea of detecting buried objects that could render the ground and its contents clear has been the subject of ongoing research, given that considerable scientific and engineering effort has been put into developing suitable methods for the exploration.Ground-penetrating radar (GPR) has been found to be an especially attractive option.The term 'ground penetrating radar' refers to a range of electromagnetic techniques designed primarily for the location of objects buried underground [1].The GPR technology is used by means of moving the antenna along the surface where the transmitted electromagnetic (EM) waves travel through the medium to detect and locate objects in soil or concrete reflecting upon another medium with sufficient dielectric contrast.The receiver then processes the reflected wave to create a time series, known as an A-scan.Consecutive patterns in A-scan will create an image called B-scan.Bscan exhibits characteristics like the hyperbola pattern, where the vertical axis indicates how long it takes for a signal to travel from the transmitter to the receiver while the horizontal axis indicates the travelling distance of the antenna [2].Even though the recognition of GPR images has achieved a certain level of success, interpreting the hyperbolic pattern information is still a challenging and ongoing research project [3].Hyperbola patterns of A-scan shaping B-scan and C-scan built from successive slices of B-scans are as shown in Figure 1.In the last few years, the GPR technology has been improved to meet the requirements needed for radar investigations on embedded elements beneath the earth's surface, within concrete walls, etc.The limitation of this technology, however, comes from the analysis of the seismic data, which is complex and can only be understood by experts in the field.In addition, the wide approach to 2D GPR profile investigations does not wholly determine the embedded object's true geometry and location [5].Therefore, this paper proposes the Synthetic Aperture Focusing Technique and 3D interpolation by threshold to reconstruct the 3D model of a buried metal sphere.A C-scan of the buried metal sphere, which is built from consecutive B-scans in the x and y directions, will be used for reconstruction.

Material and Method
Figure 2 shows the flow chart for developing the 3D visualisation of the embedded object.There are two main parts to this project, which are the flow and details of the data collection and the processing of the collected data.The 3D reconstruction is divided into three main steps: image acquisition, preprocessing, and processing algorithms for the reconstruction of the 3D model.

GPR data acquisition
In this work, the GPR data has been collected at Agency Nuclear.The experimental setup is shown in Figure 3.It consist of grid paper of 60 cm x 60 cm with line spacing of 5 cm for scanning guidance, position of the sample and the GPR system with antenna of 1.2 GHz.Metal ball with sphere shape having diameter of 16.5 cm was wrapped with aluminum foil used as sample.The selected depth is 30 cm from the ground surface.Dry sand is selected as medium for the test bed since it considered as the 'ideal' condition to obtain clear hyperbolic signature of the buried object.

GPR preprocessing
Pre-processing plays an important role prior to the reconstruction method as it is used to enhance the quality of the image and get rid of any unnecessary or unwanted details or noise.There are few steps that are implemented in the pre-processing of the data such as time-zero correction, background removal and migration.

Time-zero correction and background removal.
The raw B-scan profiles must go through timezero correction where adjustments will be made to move zero time to the surface location.The background removal then eliminates any unwanted noise from the hyperbolic image to bring focus on the hyperbola lines which gives clear image for observation or processing of the data later on [2].

Migration.
Migration refers to pre-stack migration that must be performed prior to the major processing of seismic data.It gives better resolution on the observed data as it helps in moving dipping events to the correct position, increases spatial resolution and disperses diffraction.Kirchoff's migration takes a geometrical approach in which it corrects the amplitudes and phase before the summing [6].

3D reconstruction method
In this work, interpolation and SAFT technique have been used to achieved 3D model of metal ball.

Interpolation. Interpolation of 2D
images involve obtaining the missing information on the amplitudes of the voxels lying in the spatial region between B-scans.Linear interpolation has been used using the Eq. 1 [7].
2.3.2.Synthetic Aperture Focusing Technique.The algorithm will have to determine transmitter and receiver locations based on calibrated velocity of GPR signals.Two-way travel time between transmitter and receiver when it detects the object will have a calculation in 2D image Eq. 2 [7]: In this work, the SAFT using an overlay technique will stack all of the B-scans images (X and Y direction) and being summed to reconstructed 3D image using Eq. 3 [7]: (3)

Results and Discussion
High amplitude signals are visible where known buried object occur.Figure 4 shows the hyperbolic signature based on the slicing view in x-axis.Based on Figure 4, it shows that the horizontal line refers to the distance (m) while the vertical refers to the time (ns).This data shows the 2D image namely the x-cut slices with the interval of 3cm.When it sliced through (a), the contrast of hyperbolic is very low as it represents the beginning of the buried object.At (b), (c), (d) and (e) they show the contrast of hyperbolic curve being very high compared to (a) because they cut through the buried object in the middle.The highest contrast of the hyperbolic signature can be seen at (c) and (d).Next, at (f) it shows slow contrast of the hyperbolic shape being similar with the neighborhood pixel indicating the end of the sphere.Figure 5 shows the 2D image slices in the direction of y-cut with slicing interval of 3 cm.Based on Figure 5 it is observed that the hyperbolic signature of buried metal sphere does not appear in the same hyperbolic curve as when compared with the slices in the x-plane.The y-cut slices in (a) and (f) show contrast of hyperbolic is gradually diminishing indicating the beginning and the end of the sphere respectively.The highest contrast of the hyperbolic signature can be seen at (c) and (d) which means the hyperbolic is situated at the middle of the metal sphere.The hyperbolic curve at (b) and (e) are situated right before the edge of the sphere as the contrast is higher than (a) and (f) but not higher than (c) and (d).The result from C-scanning shows the depth of the sphere is at 28 cm with and an increment of 1 cm instead of the proposed 30 cm.This is a close approximate to the proposed depth of 30 cm as at depth 30 cm the cross section of the sphere may not be in its full shape and would cover a very small area from the top of the sphere that is contacted by the EM waves.The C-scan images are captured with 2 cm differences between cross sections.Hence, cross sections at 28 cm, 30 cm, 32 cm, and so on depth are acquired until 44 cm which is at the end of the hyperbolic signal of the object ergo, no visible or defined shape of the sphere can be made pass 44 cm depth.Figure 6 shows the cross section of metal ball with respective depth.As shown in Figure 6, as the cross sections of the C-scan is going down, the diameter of the sphere seems much more dilated.This is because the EM wave that reflects on the surface of the sphere to create the signature hyperbola has only accurately enveloped certain area of the metal sphere at the surface but dilates beyond a certain point which pose for limitation of the system to create effective EM curve that represents the exact diameter and thickness of underground objects.Figure 7 shows the 3D model presented by linear interpolation of a single slice in each of the three model planes (x, y and z).Despite the lack of information on the full volume of the underground object, this representation of the target object still gives an initial idea on the object position and dimension.The slice at z-plane at depth 36cm tells us the observed object is close to circle-like shape while at xyplane the slices are displayed as semi circles.The slice at z-plane at depth 36cm tells us the observed object is close to circle-like shape while at xy-plane the slices are displayed as semi circles Before 3D interpolation, a threshold is chosen.The signals in the data having amplitude higher than the chosen threshold and the smallest distance to the point of observation will be plotted.This gives a possibility to define the shape of the target object by looking pass other unnecessary parts of the 3Ddata volume straight to the plotted points.As seen in Figure 8, by thresholding and interpolation, a shape is made out of the plotted points and gaps are filled to make a full volume in the 3D cube display.the shape does not necessarily follow the spherical shape and volume of the metal sphere sample buried but it is more curved and cylindrical as the volume follows the hyperbolic shape from the attained data.

Conclusion
This paper has presented 3D reconstruction of embedded object using GPR images.The reconstruction methods presented shows promising result as they are able to simulate and mimic the metal sphere in 3D model.While simple 3D interpolation after threshold gives full volume of the 3D model simulation, SAFT gives the best resolution on the shape of the sphere as seen from the more defined contours of the sphere and ability to follow the surface hemisphere of the metal sphere.The main difference between the interpolation method by thresholding and SAFT for 3D reconstruction can be seen by their purpose and operation.Interpolation by thresholding primarily gives focus on filling in the missing data points to create the full volume of the concentrated data points at the focus region whereas SAFT aims to improve the focusing of the GPR data by synthesizing a larger effective .

Figure 2 .
Figure 2. The flowchart of the proposed method.

Figure 3 .
Figure 3. Experimental setup: (a) grid paper for scanning line, (b) position of buried object and (c) GPR system

Figure 4 .
Figure 4. Hyperbolic signature based on the slicing view in x-axis.

Figure 5 .
Figure 5. Hyperbolic signature based on the slicing view in y-axis.

Figure 6 .
Figure 6.Cross section of the object at depth of (a) 28 cm, (b)32 cm and, (c) 44 cm.

Figure 7 .
Figure 7. Interpolation of single slice at x, y and z plane

Figure 8 .
Figure 8. 3D model of embedded object by 3D interpolation SAFT method enhance the resolution and image quality of 3D representation of embedded objects by applying focusing algorithm on radar signals providing clearer image of the subsurface.While SAFT is responsible for aggregating amplitudes from all A-scan before the A-scans are superimposed into 3D space, different amplitude threshold was chosen to show where and to what extent the focused points are superimposed and to which capacity the model is able to simulate the exact volume of the metal sphere.The amplitude thresholds were chosen at 2000 MHz and 4000 MHz as seen from Figure9.For the amplitude threshold taken at 2000 MHz, the volume is seen more dispersed whereas the amplitude threshold taken at 3000 MHz and 4000 MHz, the voxels are not as dispersed and are more focused.The contours of the models are more defined and visible giving the round shape of the sphere especially from the "top" view.

Figure 9 .
Figure 9. 3D model by SAFT with threshold amplitude at (a) 2000 MHz; (b) 4000 MHz 8th International Conference on Man Machine Systems 2023 Journal of Physics: Conference Series 2641 (2023) 012022 IOP Publishing doi:10.1088/1742-6596/2641/1/0120227 aperture.Both techniques are necessary for a more accurate and effective analysis of the embedded object.