Tensor Super-resolution for Seismic Data

Liao, Songjie; Liu, Xiaoyang; Qian, Feng; Yin, Miao; Hu, Guangmin

doi:10.1109/icassp.2019.8683419

Cited by 7 publications

(3 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Following this key idea, the SC method vectorizes the image, making a sparse approximation to estimate the useful structure where the clean image Y obeys the criteria [41]. However, since seismic data has high-dimensional properties, the vectorization operation of the traditional SC approach pulls down the original multidimensional structure of the volume [33]. To overcome this defect, we introduce the t-product to the SC method [29], [30], whereby equation ( 6) can be rewritten as a soft or hard threshold based on experience to control the fidelity term [42]…”

Section: Problem Statement and Formulation A Problem Statementmentioning

confidence: 99%

“…Recent research has suggested that the tensor-tensor product (t-product) can be introduced into SC models, which enables the direct data to process on high dimensional structures [29], [30], such as K-tensor singular value decomposition (K-TSVD) [31], and two-dimensional sparse coding (2DSC) [32]. However, the existing approaches rely on the preset fixed threshold strategy, leading to an inability to effectively extract useful features from seismic data [33]- [35]. In the absence of noise variance as a priori knowledge, the application of predetermined threshold strategies will cause part of the valuable information to be removed together with the seismic noise.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

3-D Seismic Noise Attenuation via Tensor Sparse Coding With Spatially Adaptive Coherence Constraint

Qian

Fan

et al. 2021

IEEE Access

Self Cite

View full text Add to dashboard Cite

Tensor sparse coding (TSC) is a method used to excavate 3D volume structures extended by sparse coding (SC), which is increasingly applied in data noise attenuation. Existing TSC approaches control the intensity of noise attenuation by using a predetermined soft or hard threshold that relies on the noise variance. However, the noise variance in seismic data is unknown and varies with time and space, leading to the conventional TSC method not being able to track this change. To address this issue, we proposed a tensor sparse coding model with spatially adaptive coherent constraint (TSC-SAC) to find an optimal adjustable threshold without the demand for prior knowledge of the noise variance. The threshold is determined by the coherence of the residual with respect to the dictionary. Moreover, a tensor spatial coherence orthogonal matching pursuit algorithm (TSC-OMP) is developed for solving sparse representation. Unlike the existing threshold strategy in traditional TSC methods, TSC-OMP utilizes an ideal spatially adaptive coherence threshold to regulate the sparsity, which can effectively preserve the valuable information in processing for noise suppression. By comparing with four state-of-the-art denoising algorithms, we then demonstrated the superior performance of TSC-SAC on both a synthetic and two field data sets.INDEX TERMS Tensor sparse coding, spatially adaptive coherence threshold, tensor spatial coherence orthogonal matching pursuit (TSC-OMP).

show abstract

Section: Problem Statement and Formulation A Problem Statementmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

3-D Seismic Noise Attenuation via Tensor Sparse Coding With Spatially Adaptive Coherence Constraint

Qian

Fan

et al. 2021

IEEE Access

Self Cite

View full text Add to dashboard Cite

show abstract

“…For instance, the UC Berkeley INTEL project [7], [8] reported 40% data missings and 8% data errors. Many scientific researches such as geoexploration [9] depend on complete data to draw accurate conclusion, since data analysis on incomplete data leads to inaccurate and even wrong conclusions. Therefore, various data completion methods [8], [10], [11] have been designed to recover the incomplete data.…”

Section: Introductionmentioning

confidence: 99%

High-Performance Homomorphic Matrix Completion on Multiple GPUs

2020

Self Cite

View full text Add to dashboard Cite

In various applications such as trajectory tracking in mobile social networks and online recommendation systems, the massive raw data are often incomplete due to various unpredictable or unavoidable reasons. Matrix completion algorithms are effective for reconstructing two-dimensional data, but sending raw data containing personal, sensitive information to cloud computing nodes for matrix completion may lead to privacy exposure issue. The homomorphic matrix completion is a promising approach to perform matrix completion while preserving privacy. However, CPU-based homomorphic matrix completion has low performance, making it impractical to process multiple or large-scale data completion tasks in realtime. In this paper, we propose a high-performance homomorphic matrix completion scheme by exploiting commodity GPUs (Graphics Processing Units) that are widely available in HPC servers and cloud computing nodes. First, we design and implement a baseline GPU-based homomorphic matrix completion, and propose techniques to optimize memory accesses, GPU utilization, and communications. Second, we propose a shard mode for large-scale matrices exceeding GPU memory capacity. Third, we propose a multi-GPU mode to fully utilize multiple GPUs in computing nodes. Experiment results show that the proposed scheme is both fast and accurate. On matrices of varying sizes, the proposed scheme running on a single Tesla V100 GPU achieves up to 116.23× speedups over the CPU MATLAB implementation running on dual Xeon CPUs. The multi-GPU mode achieves up to 1.84× speedups on two GPUs versus on a single GPU. For large-scale matrices, the shard mode achieves up to 174.92× speedups on a single GPU over the CPU MATLAB implementation on two CPUs, and further achieves up to 1.35× speedups when running on two GPUs using the multi-GPU mode.INDEX TERMS GPU, homomorphic matrix completion, least squares minimization.

show abstract

Seismic image super-resolution reconstruction through deep feature mining network

Zeng

Pan

et al. 2023

Appl Intell

View full text Add to dashboard Cite

Tensor Super-resolution for Seismic Data

Cited by 7 publications

References 14 publications

3-D Seismic Noise Attenuation via Tensor Sparse Coding With Spatially Adaptive Coherence Constraint

3-D Seismic Noise Attenuation via Tensor Sparse Coding With Spatially Adaptive Coherence Constraint

High-Performance Homomorphic Matrix Completion on Multiple GPUs

Seismic image super-resolution reconstruction through deep feature mining network

Contact Info

Product

Resources

About