Expand Dimensional of Seismic Data and Random Noise Attenuation Using Low-Rank Estimation

Abstract

Random noise attenuation in seismic data requires employing leading-edge methods to attain reliable denoised data. Efficient noise removal, effective signal preservation and recovery, reasonable processing time with a minimum signal distortion and seismic event deterioration are properties of a desired noise suppression algorithm. There are various noise attenuation methods available that more or less have these properties. We aim to obtain more effective denoised seismic data by assuming 3-D seismic data as a tensor in order three and increasing its dimension to 4-D seismic data by employing continuous wavelet transform (CWT). First, we map 3-D block seismic data to smaller blocks to estimate the low-rank component. The CWT of the tensor is calculated along the third dimension to extract the singular values and their related left/right vectors in the wavelet domain. Afterward, the effective low-rank component is extracted using optimized coefficients for each singular value. Thresholding is applied in the wavelet domain along the third dimension to calculate effective coefficients. Two synthetic and field data examples are considered for performance evaluation of the proposed method, and the results were compared with the competitive random noise suppression methods, such as the tensor optimum shrinkage singular value decomposition, the iterative block tensor singular value thresholding, and the block matching 4-D algorithms. Qualitative and quantitative comparison of the proposed method with other methods indicates that the proposed method efficiently eliminates random noise from seismic data.

Effect of transfer learning on the performance of VGGNet-16 and ResNet-50 for the classification of organic and residual waste

“…Recently, great attention has been caught to applying deep learning for the waste classification related to computer version (CV) with the development of computer hardware (Nasri et al, 2020). Compared with traditional CV algorithms like scale-invariant feature transform (SIFT), supporting vector machine (SVM), and principal component (PCA) (Soleimani, 2016a,b;Lu and Chen, 2022), deep learning has the ability to automatically extract the representation and equips with more applicability, robustness, generalization, and scability (Lin et al, 2022;Mafakheri et al, 2022;Saad and Chen, 2022).…”

Reconstruction of seismic data based on SFISTA and curvelet transform

“…The performance of these methods could be investigated more by comparing the average amplitude spectrum (Mafakheri et al, 2022). The average amplitude spectrum presents the original data and those obtained by FISTA and SFISTA methods, as shown in Figure 8 by the black, pink, and blue lines, respectively.…”

Machine learning-assisted processing workflow for multi-fiber DAS microseismic data

“…Frontiers in Earth Science frontiersin.org cased borehole using a wireline system (e.g., Eaton et al, 2022;Wang et al, 2022), which usually is less sensitive and yields noisier data compared with cemented fiber due to inferior coupling between the cable and the surrounding medium. To handle noisy data processing, our workflow could benefit from further new samples with diverse noise types and advanced noise attenuation tools (e.g., Mahdavi et al, 2021;Mafakheri et al, 2022) as part of the data pre-conditioning step. An advantage of CNN machine-learning techniques is that the network can be retrained as more data become available, which makes it possible to use transfer learning, i.e., to train a network using DAS data from one well and apply it to DAS data from other wells.…”