OpenFWI is a collection of large-scale, multi-structural benchmark datasets for machine learning driven seismic FWI. We release twelve datasets synthesized from different priors, including one 3D dataset. We also provide baseline experimental results with four deep learning methods: InversionNet, VelocityGAN, UPFWI and InversionNet3D.
This review paper delves into how contemporary scientific machine-learning (ML) techniques, and deep neural networks in particular, have been harnessed to tackle CWI problems. We present a structured framework that consolidates existing research spanning multiple domains, including computational imaging, wave physics, and data science.
In this paper, we revisit FWI from a new perspective: generating both modalities simultaneously. We found that both modalities can be jointly generated from a shared latent space using a diffusion process. Remarkably, our jointly generated seismic-velocity pairs inherently satisfy the governing PDE without requiring additional constraints. This reveals an interesting insight: the diffusion process inherently learns a scoring mechanism in the latent space, quantifying the deviation from the governing PDE.
This review will survey methods for incorporating physics knowledge with machine learning (primarily deep neural networks) to solve computational seismic inversion problems. We will provide a structured framework of the existing research in the seismic inversion community, and will identify technical challenges, insights, and trends.
This paper demonstrates that only linear mapping necessitates paired data, while both the encoder and decoder can be learned from their respective domains through self-supervised learning. This unveils an intriguing phenomenon (named Auto-Linear) where the self-learned features of two separate domains are automatically linearly correlated.
This study explores the use of a convolutional neural network (CNN) to enhance prostate imaging under these constraints. Our approach leverages supervised learning to manage the extremely narrow data acquisition apertures that hinder traditional full-waveform inversion (FWI) methods. We validated our method using synthetic prostate phantoms and finite difference ultrasound data simulations.
In this paper, we introduce EFWI, a comprehensive benchmark dataset that is specifically designed for elastic FWI. EFWI encompasses 8 distinct datasets that cover diverse subsurface geologic structures (flat, curve, faults, etc). The benchmark results produced by three different deep-learning methods are provided.
The research reported here investigates the use of a convolutional neural network (CNN) to learn a mapping from USCT waveform data to speed of sound estimates. The CNN was trained using a supervised approach with a task-informed loss function aiming at preserving features of the image that are relevant to the detection of lesions.
we propose the use of smaller probes capable of generating limited ultrasound brain image fragments from various angles. Subsequently, we have developed a new brain imaging method, termed BrainPuzzle, to restore brain images from these limited fragments. Unlike traditional transformer-based image generation models, BrainPuzzle not only uses a transformer to recognize and rearrange the fragments into their correct positions but also integrates a graph convolutional network (GCN) to automatically capture the spatial relationships among the fragments, thereby enhancing the model’s capabilities.
FULL-waveform inversion (FWI) plays an important role in various applications. However, the existing computational methods for solving full-waveform inversion are not only computationally expensive, but also yields low-resolution results. To resolve those issues, we employ data-driven techniques to solve the full-waveform inversion.
VelocityGAN interprets data-driven inversion method as an image-to-image-translation problem using generative adversarial networks (GANs) structure. It learns an effective regularization that is customized to the inversion problem.
We developed a new hybrid computational approach to solve full-waveform inversion that combines physics-driven models with data-driven methodologies. Specifically, we develop a data augmentation method that not only improves the representativity of training set, but also incorporates important governing physics into training and therefore improve the inversion accuracy.
We developed a detection method using cascaded region-based convolutional neural networks. Our method would capture events in different sizes, while incorporating contextual information to enrich features for each individual proposal.
We developed an end-to-end framework which can automatically learn the hyper-parameter in the denoising algorithm so that we do not need to manually set the hyper-parameter. Specifically, our network structure consists of two modules, an adaptive filtering module for signal denoising, and a classification module for signal classification.
The accuracy of machine-learning based detection methods rely on sufficient amount of high-quality training data, which itself can be expensive to obtain due to the requirement of domain knowledge and subject matter expertise. We resolve this dilemma by answering two questions: (1) provided with limited number of reliable labels, can we use them to generate more synthetic labels; (2) Can we use those synthetic lables to improve the detectability?