A hybrid approach to seismic deblending: when physics meets self-supervision

Abstract

To limit the time, cost, and environmental impact associated with the acquisition of seismic data, in recent decades considerable effort has been put into so-called simultaneous shooting acquisitions, where seismic sources are fired at short time intervals between each other. As a consequence, waves originating from consecutive shots are entangled within the seismic recordings, yielding so-called blended data. For processing and imaging purposes, the data generated by each individual shot must be retrieved. This process, called deblending, is achieved by solving an inverse problem which is heavily underdetermined. Conventional approaches rely on transformations that render the blending noise into burst-like noise, whilst preserving the signal of interest. Compressed sensing type regularization is then applied, where sparsity in some domain is assumed for the signal of interest. The domain of choice depends on the geometry of the acquisition and the properties of seismic data within the chosen domain. In this work, we introduce a new concept that consists of embedding a self-supervised denoising network into the Plug-and-Play (PnP) framework. A novel network is introduced whose design extends the blind-spot network architecture of [28] for partially coherent noise (i.e., correlated in time). The network is then trained directly on the noisy input data at each step of the PnP algorithm. By leveraging both the underlying physics of the problem and the great denoising capabilities of our blind-spot network, the proposed algorithm is shown to outperform an industry-standard method whilst being comparable in terms of computational cost. Moreover, being independent on the acquisition geometry, our method can be easily applied to both marine and land data without any significant modification.Preprint. Under review.

Transfer learning for self-supervised, blind-spot seismic denoising

“…Whilst N2V's successor, StructN2V, can suppress coherent noise, it requires a consistent noise pattern for which a specific noise mask is built, e.g., masking the noise along a specific direction. This has shown great promise for trace-wise noise suppression, such as dead sensors (Liu et al, 2022a) or blended data (Luiken et al, 2022), however it is not practical for the suppression of the general seismic noise field which is continuously evolving. In this work, we have proposed to initially train a network on simplistic synthetic datasets and then fine-tune the model in a selfsupervised manner on the noisy field data.…”

“…Building on this, Liu et al (2022a,b) proposed to extend the blind-spot property to become a blind-trace, adapting the previously proposed selfsupervised denoiser for the suppression of trace-wise noise. Similarly, both Luiken et al (2022) and Wang et al (2022) proposed blind-trace networks implemented at the architecture level, as opposed to the likes of Krull et al (2019); Birnie et al (2021); Liu et al (2022a) which were implemented as processing steps. Both Liu et al (2022a) and Luiken et al (2022) illustrated successful suppression of tracewise noise, specifically poorly coupled receivers in common shot gathers and blending noise in common channel gathers, respectively.…”

“…Similarly, both Luiken et al (2022) and Wang et al (2022) proposed blind-trace networks implemented at the architecture level, as opposed to the likes of Krull et al (2019); Birnie et al (2021); Liu et al (2022a) which were implemented as processing steps. Both Liu et al (2022a) and Luiken et al (2022) illustrated successful suppression of tracewise noise, specifically poorly coupled receivers in common shot gathers and blending noise in common channel gathers, respectively. Similar to Birnie et al (2021), both these studies highlighted the potential of self-supervised learning to outperform conventional methods, when applied optimally.…”

“…Earlier implementations of selfsupervised, blind-spot networks have all focussed on suppressing a single component of the noise field as opposed to tackling the noise field as a whole. In order to tackle the continually evolving noise field throughout a field seismic recording, it is not possible to use a rigid scheme involving a strict noise mask-such as the trace-wise noise suppression of Liu et al (2022a) and Luiken et al (2022)-instead a more flexible solution is needed.…”

Deep preconditioners and their application to seismic wavefield processing

“…The success of deep learning in various scientific disciplines has recently gained the attention of the geophysical community. Encouraging applications of deep learning in various seismic processing and interpretation tasks have been reported, ranging from denoising (Saad and Chen, 2020;Birnie and Alkhalifah, 2022), deblending (Richardson and Feller, 2019;Sun et al, 2020;Luiken et al, 2022), velocity analysis (Yang and Ma, 2019;Sun et al, 2020;Kazei et al, 2021), fault and geobodies interpretation (Waldeland et al, 2018;Shi et al, 2019;Wu et al, 2020), to reservoir characterization (Zhao, 2018;Alfarraj and AlRegib, 2019;Das et al, 2019). We refer the reader to for an exhaustive literature review on the topic.…”