Fully Connected U-Net and Its Application on Reconstructing Successively Sampled Seismic Data

Abstract

One of the major hot topics in seismic data processing is the reconstruction of successively sampled seismic data. There are numerous traditional methods proposed for addressing this issue; however, they still have unavoidable drawbacks, such as high computational cost and sensitive tuning parameters. In this study, we suggest a deep learning model for reconstructing successively sampled seismic data, termed fully connected U-Net (FCU-Net). FCU-Net maintains the high-resolution representations by connecting the parallel different-resolution representations and repeating multi-scale fusion. Such a structure allows FCU-Net to successfully extract multi-scale information, which is beneficial for accurate seismic data reconstruction. Additionally, the extending subnetwork of FCU-Net contains a large number of feature channels and sufficient information interaction between different resolution representations via the composite cascades, which contributes to locating successively sampled traces with big gaps and then performing the seismic interpolation. To verify the effectiveness of FCU-Net, we compare it with state-of-the-art networks, i.e., U-Net and HRNet, using synthetic and field examples. The results show that FCU-Net performs best when interpolating successively sampled seismic data, proving its superiority and availability.

Near offset reconstruction for marine seismic data using a convolutional neural network

“…The network architecture used is the U-Net (Ronneberger et al, 2015), which is a convolutional neural network (CNN) originally developed for biomedical image segmentation. This architecture has demonstrated effectiveness on various seismic imaging tasks, such as fault segmentation (Wu et al, 2019), multiple reflection removal (Bugge et al, 2020) and most relevant to this work -seismic interpolation (Fang et al, 2021;Li et al, 2023) and near offset extrapolation (Qu et al, 2021). Furthermore, it is a simple model to parameterize and train.…”

“…Recently, there has been an increase in the usage of deep learning methods for seismic interpolation and near offset reconstruction, where neural networks are trained to interpolate or generate missing traces. Examples of this have included a generative adversarial network for trace interpolation (Kaur et al, 2021), a convolutional neural network (CNN) for near offset extrapolation at shallow subsurface depths (<0.1 s of traveltime) (Qu et al, 2021), a CNN for interpolation of successively sampled seismic data (Li et al, 2023) and a multidirectional CNN for self-supervised reconstruction of gaps in seismic data (Abedi & Pardo, 2022). One of the main challenges of a deep learning approach is in creating or finding training data with rich near offset information, as source-over-cable field data is a rarity, and synthetic training data often lacks the heterogeneity and complexity of real seismic data.…”

Deep-Learning-Based Seismic-Signal P-Wave First-Arrival Picking Detection Using Spectrogram Images

“…Additionally, the connection of feature maps between the encoder and decoder allows for fine-grained information transfer during upsampling. Such U-Net algorithms have shown excellent results in the field of seismology, including the reconstruction of seismic-signal data resolution [45,46] and P-wave FAP detection studies [47][48][49].…”

“…Additionally, the connection of feature maps between the encoder and decoder allows for fine-grained information transfer during upsampling. Such U-Net algorithms have shown excellent results in the field of seismology, including the reconstruction of seismic-signal data resolution[45,46] and P-wave FAP detection studies[47][48][49].Figure7illustrates the U-Net architecture proposed in this study. The model consists mainly of an encoder and a decoder, both of which are designed with identical layers.…”