GNGTS 2019 - Atti del 38° Convegno Nazionale

GNGTS 2019 S essione 3.1 565 • Spike-like noise model. Pre-stack seismic data can be affected by different types of random noise far more complex than simple AWGN. For instance, some of these seismic noises exhibit spike-like characteristics (Liu et al. , 2008) and are lately gaining growing interest, as they strongly affect the processing of blended data. To simulate this noise, we add spiky noise with variable percentage d of corrupted samples. Then, we convolve each noise trace with a Ricker wavelet having the same central frequency of the data. This way, we generate two corrupted datasets, corresponding to d =1 and d =3 leading to initial S/N of -34.8 dB and -39.6 dB respectively. The average results for the recovered gathers have been of 16.8 dB and 12.3 dB for increasing rates of noise. Fig. 2 shows an example of denoising for d =3. Even if the corrupted image undergoes a strong degradation, the reconstructed one presents almost all the features of the original data. Moreover, Fig. 2 reports the residual error between corrupted and reconstructed gather. Fig. 1 - Example of regular trace interpolation (top) and corresponding spectra (bottom). From left to right: original gather, decimated gather and reconstructed gather. Fig. 2 - Example of spike noise denoising. From left to right: original gather, noisy gather, reconstructed gather and residual.