Ground-based vs UAV-based GPR

The original article is published on the SPH Engineering web site.

This report presents the results of a test survey comparing the capabilities and effectiveness of ground- and UAV (Unmanned Aerial Vehicle)-based ground penetrating radar (GPR) systems.

Introduction

During the test, a single RadSys Zond Aero 500 NG GPR was used in both ground-based and UAV-based setups. By using the same GPR in both configurations, the impact that different antennas may have on the results and interpretations was reduced.The survey was conducted over an asphalt strip, beneath which there are several soil layers and numerous compact local targets.

Data was collected by Sergey Kucenko¹ and Sergey Zelenkov². Data processing and reporting were done by Sergey Zelenkov², Alexey Dobrovolskiy¹, and Matiss Brants¹.

¹SPH Engineering, Riga, Latvia

²Radar Systems, Inc., Riga, Latvia

Data Samples to access >

Methods

The UAV test range

The test was conducted on the 18th of September 2024 in the SPH Engineering UAV test range in Baloži, Latvia (N 56.8631°, E 24.1119°): a 4-hectare large, open and flat field surrounded by forests (figure 1). The site hosts numerous buried infrastructure objects like steel pipes, water barrels, reinforced pipes, etc., for geophysical testing purposes. In the present case, however, the GPR was tested over an asphalt strip surrounding the field. Beneath the asphalt are layers of disturbed sand with some rock inclusions, which rest on top of natural peaty soil.

Figure 1. Aerial view of SPH Engineering’s UAV test site in Balozi, Latvia.

Survey execution

The test was done on a sunny, warm day, with no significant rainfall during the previous three days. The 170 m long survey line (Figure 2) followed a straight patch of rather damaged asphalt with potholes. While the UAV-based system was flown in a perfectly straight line following a pre-set course, the ground-based cart system had to be guided around patches of potholes, and thus, the line was less straight.

Figure 2. The survey paths of the UAV-based GPR setup (black, solid line) and the ground-based GPR setup (red, dotted line).

Used device - Radar Systems Inc. Zond Aero 500 NG GPR

When comparing UAV-based and ground-based GPRs, many parameters have to be considered, such as the antenna type, central frequency, gain, digitizing technique, dynamic range, filtering, etc. To avoid misunderstandings and reduce unfair advantages, these parameters have to be as similar as possible. The best way to achieve this similarity is to use the same antenna in all tested setups, as in the present test.

The Zond Aero 500 NG is a universal GPR system for terrestrial and airborne surveys. It uses real-time sampling (RTS) technology and advanced built-in digital signal processing to provide high-quality data with a high signal-to-noise ratio. The central frequency is 500 MHz. The system can be mounted on a UAV or cart or simply dragged with a tow rope.

In the UAV-based test, the GPR was mounted on the rails of a DJI Matrice 350 RTK UAV (figure 3a). The UAV was equipped with a laser-guided True-Terrain Following system to keep the GPR antenna at a steady altitude of 0.5 meters above the ground. The velocity of the setup was set to 1 m/s. The raw data were collected on SPH Engineering’s SkyHub onboard computer attached to the UAV.

Mounting the GPR antenna on a cart (figure 3b) allowed to keep a constant distance from the ground with a good coupling. The cart was guided by an operator and had to be maneuvered around any low obstacles, which a drone setup could simply fly over undisturbed. For the ground-based setup, the raw data was recorded on a laptop attached to the handle of the cart.

Figure 3.a. The Zond Aero 500 NG GPR mounted on DJI Matrice 350 RTK drone in UAV-mode.

Figure 3.a. The Zond Aero 500 NG GPR mounted on a cart in ground-based mode.

Data processing

Data processing was done using Prism v2.70 software, developed by Radar Systems Inc. The first step in data processing was to clip the ends of the profile where only static signal was recorded and correct for the slightly longer path of the ground-based setup due to potholes in the road, as mentioned previously. The X-interpolation by GPS coordinates helped to rebuild the drone-based GPR profile equidistantly to align it with ground-based data. Only background removal was used for data filtering, and the same signal gain was applied to both data sets. The dielectric permittivity coefficient was set to 9.

Results

The GPR data sets from both setups are remarkably similar (figure 4). Most of the major horizontal features from the ground-based setup dataset are clearly visible also in the UAV-based dataset. The main difference is the increased signal strength and higher resolution of the boundaries in the ground-based data. The ground-based data also provided deeper signal penetration as there are more pronounced and resolved boundaries at a depth of 2 to 4 meters. The noise in the UAV-based data is more prominent, especially at the deeper levels.

Figure 4. The filtered data from a) UAV-based and b) ground-based GPR setups.

The resolution of local targets (i.e., rocks or other compact objects) is lower for the UAV-based data than ground-based data. These local targets have a hyperbolic signal shape (roughly an upside-down “U”); in the present case, they are very prominent at the beginning of the profiles (figure 5), which coincide with the road embankment. In the UAV-based data, the hyperbolas have shorter “branches” and thus are less visible. This is due to the airborne antenna’s elevation above the ground and the antenna's narrower radiation pattern refraction on the ground surface. The ground-based antenna has a broader radiation pattern due to the coupling with the ground, which can capture the reflected waves from the local targets earlier and therefore generate more prominent hyperbolas.

Figure 5. A zoomed-in view on a portion of the a) UAV-based, and b) ground-based GPR profiles, where many hyperbolic reflections from local targets are visible.

When analyzing the unfiltered data and its frequency spectrum (figure 6), the different nature of the signal noise is revealed. For the UAV-based data, the most prominent is the low-frequency “ringing” noise (visible as strong horizontal lines before background signal removal), while the ground-based data has a “WoW” noise, which is characterized by the decaying saturation of the signal. The high-frequency noise is also different between both datasets. The central frequency of the spectrum is shifted to the left for the ground-based data, and they look predictably better due to the antenna being closer to the ground (the difference is about 6 dB). For the airborne data, the high-frequency noise is much stronger due to the influence of air-reflected signals from the surface and shallow objects. These signals travel much faster as the wave velocity depends on the medium’s permittivity where it is propagating (for air, it is close to the speed of light). However, such noise can be filtered with a simple band-pass filter.

Figure 6. Acquired data average frequency spectrum for a) UAV-based GPR and b) ground-based GPR setups. Note the different dB scales.

Conclusions

The GPR on the UAV platform provided results comparable to those of the ground-based GPR at the particular test site.
Data acquired with ground-based GPR has a lower noise level, higher signal-to-noise ratio, and, as a result, more pronounced accurate signals and a better depth penetration;
The mobility of a UAV allows an airborne GPR to be flown in a more rugged terrain and follow a more straight path than a ground-based setup;
Ground-based GPR captured better hyperbola signatures from local targets, resulting in a slightly higher resolution for small objects, but diffraction hyperbolas were present in both data sources.