Abstract:Noise attenuation is a crucial phase in seismic signal processing. Enhancing the signal-to-noise ratio (SNR) of registered seismic signals improves subsequent processing and, eventually, data analysis and interpretation. In this work, a novel noise reduction framework based on an intelligent deep convolutional neural network is proposed that works on segments of the time-frequency domain and, hence named as DeepSeg. The proposed network is efficient in learning sparse representation of the data simultaneously in the time-frequency domain and adaptively capturing seismic signals corrupted with noise. DeepSeg is able to achieve impressive denoising performance even when seismic signal shares common frequency band with noise. The proposed approach properly tackles a variety of correlated (color) and uncorrelated noise, and other nonseismic signals. DeepSeg can boost the SNR considerably even in extremely noisy environments with minimal changes to the signal of interest. The effectiveness of the proposed methodology is demonstrated in enhancing passive seismic event detection/denoising. However, there are other obvious applications of the DeepSeg in active and passive seismic fields, e.g., seismic imaging, preprocessing of ambient noise data, and microseismic event monitoring. It is worth pointing out here that the deep neural network is trained exclusively using synthetic seismic data, negating the need for real data during the training phase. Furthermore, the proposed setup is general and its potential applications are not confined to passive event denoising or even seismic. The method proposed is also adaptable to other diverse signals in different settings, like medical images/signals [magnetic resonance imaging (MRI), electroencephalogram (EEG) signals, electrocardiograms (ECG) signals, and retinal images, to name a few], radar signals, speech signals, fault detection in electrical/mechanical systems, daily life images, etc. Experiments on synthetic and real seismic data reveal the efficacy and supremacy of the proposed method in terms of SNR improvement and required training data when compared to the state-of-the-art deep neural network-based denoising technique.

Across-domains transferability of Deep-RED in de-noising and compressive sensing recovery of seismic data

Intelligent Reconstruction for Spatially Irregular Seismic Data by Combining Compressed Sensing with Deep Learning

Domain Adaptation in Automatic Picking of Phase Velocity Dispersions Based on Deep Learning

Residual Learning of Deep Convolutional Neural Network for Seismic Random Noise Attenuation

Seismic data recovery using deep targeted denoising priors in an alternating optimization framework

DeepSeg: Deep Segmental Denoising Neural Network for Seismic Data

Domain Expansion via Network Adaptation for Solving Inverse Problems

Deep learning for high-resolution seismic imaging

Transfer learning in large-scale ocean bottom seismic wavefield reconstruction

Seismic Signal Denoising and Decomposition Using Deep Neural Networks

Seismic data denoising and deblending using deep learning

Reconstructing missing seismic data using Deep Learning

Deep-Learning Inversion of Seismic Data

Applications of Deep Neural Networks in Exploration Seismology: a technical survey

Rapid seismic domain transfer: Seismic velocity inversion and modeling using deep generative neural networks

A Convolutional Autoencoder Method for Simultaneous Seismic Data Reconstruction and Denoising

Adapting the Residual Dense Network for Seismic Data Denoising and Upscaling

Deep Nonlocal Regularizer: A Self-Supervised Learning Method for 3-D Seismic Denoising

ROBUST DEEP LEARNING-BASED SEISMIC INVERSION WORKFLOW USING TEMPORAL CONVOLUTIONAL NETWORKS

Seismic Coherent Noise Removal With Residual Network and Synthetic Seismic Samples

Deep CNN for Coherent Seismic Noise Removal: A Perspective