Several Mode Decomposition Methods In Seismic Data Denoising

Exploration of geophysical features and mysterious with the help of recorded seismic data is important to excogitate among the scientific community.Regrettably,most of the real seismic data are contaminated with noise during the recording process.The presence of noise in the data causes to cover the important information and commove for the accurate analysis.Hence,the attenuation of noise from the recorded data is a crucial task in seismic signal processing.Obtaining the high-resolution image,improving the signal-to-noise ratio(SNR),exploring the hidden information with the preservation of important features are the key points to be noted during denoising.To address these problems the number of mathematical theories and techniques have been proposed.These different algorithms have their strong properties and drawbacks.Utilizing the capability of the existing techniques and solving the shortcomings to improve the results with new idea/methods is the main task.Traditional methods like Wiener filter,Kalman filter,and Fourier strategies were mainstream first and foremost stage;however,these were touchy for only linear and stationary signal investigation.The methods like Wavelet,empirical mode decomposition(EMD),variational mode decomposition(VMD)are very popular in the field to handle the non-stationary and nonlinear signals.We proposed the new algorithm based on VMD to solve the denoising problem in a better way.In this thesis,we applied the proposed idea to attenuate the noise,to improve the signal-to-noise ratio,resolution,to observe the effect of noise in earthquake accelerogram components like peak ground acceleration,displacement,P-wave arrival detection and time estimation.We carried number of experiments on synthetic and real earthquake data,two-dimensional seismic data to show the effectiveness of the proposed model.Removing noise from seismic data is not a very complicated task but preserving the original and important feature is most sensitive during the process.EMD method can remove the noise from data but lack of strong mathematical foundation,mode mixing problem and noticeably we eliminate the first mode so there is a high possibility to lose the information that contains in the mode.The VMD method solves mode mixing problem and other defects of EMD as we do not eliminate any mode during VMD process there are very fewer chances to loss of information from the mode.The strong noise,if present in data then VMD method is less sensitives so we proposed some new techniques to improve the results.VMD disintegrates the noisy signal into number of intrinsic modes(IMF)from low to high frequency.These modes can be classified into signal dominant and noise dominant by calculating probability density function(PDF)using Kernel density estimation(KDE).High-frequency noisy modes we reconstructed with the help of some tools like continuous wavelet transform.We implemented the algorithm to observe the noise effect on peak ground acceleration,displacement in chapter 2.Additionally,we applied the proposed method for two-dimensional seismic data denoising in this chapter.Similarly we proposed the algorithm based on VMD and SG filter to observe how noise can lead to inaccurate detection of P-wave and time arrival is explained in chapter 3 with its extension for 2D.In chapter 4 we applied wavelet based geometrical mode decomposition for two-dimensional seismic data denoising.As GMD disintegrates the data into low and high-frequency IMFs we constructed hybrid algorithm based on CWT utilizing its property as a bandpass filter to attenuate the low and high frequency noise simultaneously.Numerical results from synthetic and real seismic data validate the ameliorate of the proposed method.