The main objective of the ExoMars 2020 mission is to search for signs of past and/or present life on Mars. Toward this goal, a rover was designed to investigate the shallow subsurface which is the most likely place where signs of life may be preserved, beneath the hostile surface of Mars. The rover of the ExoMars 2020 mission has on board a polarimetric ground penetrating radar called WISDOM (Water Ice Subsurface Deposits Observation on Mars). Thanks to its large frequency bandwidth of 2.5 GHz, WISDOM is able to probe down to a depth of approximately 3 m on sedimentary rock with a vertical resolution of a few centimeters.The main scientific objectives of WISDOM are to characterize the shallow subsurface of Mars, to help understand the local geological context and to identify the most promising location for drilling. The WISDOM team is currently working on the preparation of the scientific return of the ExoMars 2020 mission. In particular, tools are developed to interpret WISDOM experimental data and, more specifically, to extract information from the radar signatures of expected buried reflectors. Insights into the composition of the ground (through the retrieval of its permittivity) and the geological context of the site can be inferred from the radar signature of buried rocks since the shape and the density of rocks in the subsurface is related to the geological processes that have shaped and placed them there (impact, fluvial processes, volcanism). This paper presents results obtained by automatic detection of structures of interest on a radargram, especially radar signature of buried rocks. The algorithm we developed uses a neural network to identify the position of buried rocks/blocs and then a Hough transform to characterize each signature and to estimate the local permittivity of the medium. Firstly, we will test the performances of the algorithm on simulated data constructed with a 3D FDTD code. This code allows us to simulate radar operation in realistic environments. Secondly, we will test our algorithm on experimental data acquired in a semi-controlled environment. Lastly, we will present experimental data acquired during a recent field campaign (July 2017) in the south of France and we will validate our method and illustrate the ability of WISDOM to provide clues about the geological context of a site.