Effect of ground undulation and mounted vehicle velocity variation on stepped frequency continuous wave GPR data Smitha Nage Gowda 1,2 Deverakonda R. Ullas Bharadwaj 1 Sridhara Abilash 1 Saligrama N. Sridhara 1 Vipula Singh 1 1 Department of Electronics and Communication Engineering, Rama Nagappa Shetty Institute of Technology, Bangalore 560098, India. 2 Corresponding author. Email: smithanesara81@gmail.com Abstract. This paper describes a method to generate dataset on stepped frequency continuous wave (SFCW) ground penetrating radar (GPR) for land mine detection. Probability of target detection as well as accuracy of GPR is hindered by ground undulations and GPR mounted vehicle velocity variation. This paper proposes a novel method of filtering out ground undulation effect by ground bounce removal filter and also mitigating GPR mounted vehicle velocity variations. This work also focuses on migration of simulated B-scan and C-scan data using Kirchhoff and F-K migration algorithms. The irregular surface condition of the ground or ground undulation is modelled and a ground bounce removal filter is developed to eliminate the effects of ground undulation. Non-uniform sampling of B-scan replicates the scenario of variation in velocity of the GPR mounted vehicle. The Kirchhoff and F-K migration algorithms applied to the outcome of ground bounce removal filter dataset results in no/less error with respect to true depth and position of the landmine in all possible scenarios. An interactive graphical user interface (GUI) for generating and testing the SFCW GPR data is also discussed in this paper. Key words: B-scan, C-scan, GPR, ground bounce, migration, SFCW. Received 2 February 2016, accepted 27 October 2016, published online 15 December 2016 Introduction Ground penetrating radar (GPR) is a non-destructive method that uses electromagnetic radiation to image the subsurface structure. In military applications, GPRs are used for detection of landmines and tunnels. GPRs can be operated either in time domain or in frequency domain. Time domain GPRs work with single small pulse resulting in higher range resolution, whereas frequency domain GPRs consider a continuous signal wherein the carrier wave frequency varies continuously or with a fixed step. The latter is called the stepped frequency continuous wave (SFCW) GPR (Nicolaescu et al., 2003). The SFCW has the advantage of having a wider dynamic range, high mean power and low noise figure. It gives high depth of penetration and using this radar, 3D imaging can also be done (Nicolaescu et al., 2003; Sharma et al., 2013). B-scan GPR image, appears as a hyperbola, due to various trip times of electromagnetic wave (or cross range of the target) as the antenna is moving in the scanning direction. Long trails of the hyperbola in the B-scan results in low resolution of target position due to the change in medium. Using the migration (transform) technique resolves this problem; by transforming an unfocussed GPR image to a focused image giving rise to true depth as well as size of target. This transforming technique is referred to as migration or focusing (Ozdemir et al., 2014). In case of a vehicle-mounted GPR, radar can look down into the soil or it can look ahead in forward-mode. The effective range of vehicle-mounted GPR depends on the speed of the vehicle and the relative dielectric constant of the ground. This work focuses on removing ground undulation both in linear and non-linear surfaces. Migration algorithms focus the image giving target depth with minimal error. GPR mounted vehicle velocity is also considered, wherein effects of variation in vehicle velocity will not deviate migration results. This makes our work on migration and ground undulation removal effective in all scenarios considered in this part of work. This paper is organised as follows. The next section covers difficulties of landmine detection. In the third section, data generation for a SFCW GPR has been covered. Modelling of ground undulation and ground bounce removal filter is covered in the fourth section. The fifth section includes Kirchhoff and F-K migration techniques. In the sixth section, effects of variation of GPR mounted vehicle velocity is shown. The final section includes results and discussions. Difficulties in landmine detection The probability of land mine detection in a GPR has to be very high, as good as 100%, with 100% accuracy and 0% false alarm rate in view of the consequences of non-detection. Problems in landmine detection as well as localisation are: (a) ground surface topography can vary from smooth to rough soil; and (b) the depth and angle of buried landmines can vary. We have focused on these two problems in this research work. We have modelled soil undulation and designed a filter to remove the ground bounce due to undulation. In order to get the true location of the landmine, migration algorithms have been implemented. Realistically, ground topography does not follow a single statistical structure due to undulation. Ground undulation may be more at one location and less at another. In order to understand the received signal of GPR, first consider a transmitter and receiver over a flat surface with a buried object as the target. In this case, the signal received by GPR is characterised by CSIRO PUBLISHING Exploration Geophysics http://dx.doi.org/10.1071/EG16011 Journal compilation Ó ASEG 2016 www.publish.csiro.au/journals/eg