GNGTS 2021 - Atti del 39° Convegno Nazionale

GNGTS 2021 S essione 3.3 502 Fig. 2 - Portion of a seismic line of the WS10 exploration project used for the NN performance assessment. The prediction results are shown in red. A key benefit of the proposed prediction procedure is that it can be applied to any kind of dataset to automatically extract the reflectors, once the training phase is concluded on a set of randomly generated data containing various types and levels of noises. Therefore, in order to verify this capability and evaluate the performances, we tested the NN on a totally different da- taset, namely a Ground Penetrating Radar (GPR) survey collected on a Glacier in the Eastern Alps (further details about the data in Colucci et al., 2014). Also in this case (Fig. 3), the NN is able to extract all the relevant reflectors, which are related, from top to bottom to completely different glaciological and geological units, often characterized by high-amplitude scattering. Figure 3: Example of 2-D GPR profile used to test the NN. The results of the prediction are shown in red.