LS-SVR with variant parameters and its practical applications for seismic prospecting data denoising

Signal denoising can be considered as a function regression problem. LS-SVR (least squares-support vector regression) based on Ricker wavelet kernel function is applied to the practical seismic prospecting data denoising in this paper. To adapt LS-SVR well to the practical seismic data, the parameters including Ricker wavelet kernel parameter f and regularization parameter ? are selected automatically according to the features of data in the fixed window. The denoising experimental results for the theoretical and practical seismic data show that the performance of Ricker wavelet LS-SVR with variant parameters outperforms the one with invariant parameters in terms of the retrieved waveform in time domain and spectrum range in frequency domain.