.png){kind=icon}
.png){kind=icon}
Return
Accelerating Marchenko Multiple Elimination With Deep Parallel Convolution and Time–Frequency Network
.png){kind=icon}
PRE
AI
X
J
Z
L
X
DOI：10.1109/lgrs.2026.3710558.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The Marchenko multiple elimination (MME) method enables precise primary reconstruction in a data-driven manner, but its core computations—multidimensional convolutions and cross-correlations—are computationally intensive for large datasets. To address this, we propose DTFMarNet, a deep parallel convolution and time–frequency dual-path network that efficiently approximates key MME update operators. By organizing multishot seismic data along the trace dimension, sequential cross correlation-convolution updates are converted into parallel operations, reducing redundant computation. The deep parallel convolution structure decreases computational complexity, while the time–frequency dual-path mechanism models temporal and spectral dependencies and constrains the amplitude and phase of primaries. To overcome limited primary labels, a weakly supervised strategy leverages pseudo-References and enforces time–frequency and temporal structure constraints, thereby enhancing stability and generalization. Experiments on synthetic and real datasets demonstrate that DTFMarNet maintains effective multiple suppression and physical consistency while significantly accelerating the MME process.
Keywords:
Deep parallel convolution
Marchenko multiple elimination (MME)
multiple suppression
neural network acceleration
time–frequency dual path
Journal
I
IF:
4.4
Papers:
486
Citations:
0
Organization
Cited Papers