Publications

2022

2022

1. 

   Equivariant imaging for self-supervised regularly undersampled seismic data interpolation

   Weiwei Xu , Vincenzo Lipari, Paolo Bestagini , and 3 more authors

   In SEG International Exposition and Annual Meeting , 2022

   Abs

   Because of the restriction of complex field conditions and economic circumstance, seismic data is usually undersampled in the spatial domain, which needs to be interpolated to meet the requirements of following seismic data processing such as seismic imaging. We present a seismic data interpolation method via an end-to-end self-supervised deep learning framework. Specifically, a CNN is trained only using the observed undersampled seismic data itself. Furthermore, based on the equivariance of seismic data with respect to shift and undersampling, a training strategy that enforces both the measurement consistency and the equivalence is utilized. Experiments on regularly undersampled synthetic and field data interpolation show the effectiveness of our presented method in comparison with deep image prior (DIP) based interpolation method.

2. DIPPAS: A deep image prior PRNU anonymization scheme

   Francesco Picetti , Sara Mandelli , Paolo Bestagini , and 2 more authors

   EURASIP Journal on Information Security, 2022

   Abs

   Source device identification is an important topic in image forensics since it allows to trace back the origin of an image. Its forensics counterpart is source device anonymization, that is, to mask any trace on the image that can be useful for identifying the source device. A typical trace exploited for source device identification is the photo response non-uniformity (PRNU), a noise pattern left by the device on the acquired images. In this paper, we devise a methodology for suppressing such a trace from natural images without a significant impact on image quality. Expressly, we turn PRNU anonymization into the combination of a global optimization problem in a deep image prior (DIP) framework followed by local post-processing operations. In a nutshell, a convolutional neural network (CNN) acts as a generator and iteratively returns several images with attenuated PRNU traces. By exploiting straightforward local post-processing and assembly on these images, we produce a final image that is anonymized with respect to the source PRNU, still maintaining high visual quality. With respect to widely adopted deep learning paradigms, the used CNN is not trained on a set of input-target pairs of images. Instead, it is optimized to reconstruct output images from the original image under analysis itself. This makes the approach particularly suitable in scenarios where large heterogeneous databases are analyzed. Moreover, it prevents any problem due to the lack of generalization. Through numerical examples on publicly available datasets, we prove our methodology to be effective compared to state-of-the-art techniques.

3. Interpolation of missing shots via plug and play method with CSGs trained deep denoiser

   Weiwei Xu , Vincenzo Lipari, Paolo Bestagini , and 2 more authors

   In 83rd EAGE Annual Conference & Exhibition , 2022

   Abs

   Due to the restriction of complex field conditions, the trace interval in common receiver gathers (CRGs) is often larger than that in common shot gathers (CSGs). This impacts on the stability and precision of the following seismic data processing steps. To solve this issue, we present a Plug and Play method CSGs-trained deep denoiser for the interpolation of missing shots. Specifically, based on the spatial reciprocity theorem, instead of collecting or constructing training datasets, CSGs are used as the training dataset to train a deep convolutional neural network (CNN) Gaussian denoiser. This trained denoiser is then plugged into the alternating direction method of multiplier (ADMM) framework to solve the interpolation inverse problem. A numerical example on field data shows the effectiveness of the presented method in comparison to CNN-POCS method.

4. Intelligent seismic deblending through deep preconditioner

   Weiwei Xu , Vincenzo Lipari, Paolo Bestagini , and 3 more authors

   IEEE Geoscience and Remote Sensing Letters, 2022

   Abs

   Seismic deblending is an ill-posed inverse problem that involves counteracting the effect of a blending matrix derived from the shots’ position and firing time. In this letter, we propose a seismic deblending method based on so-called deep preconditioners. A convolutional autoencoder (AE) is first trained in a patch-wise fashion to learn an effective sparse representation of the common receiver gathers (CRGs) we aim to reconstruct. Then, the decoder branch of the trained AE is used as a nonlinear preconditioner for the deblending problem. Particularly, to avoid the explicit creation of a training dataset, we suggest to use the common shot gathers (CSGs) of the blended dataset itself to train the AE network, as they are not affected by incoherent blending noise. Numerical examples on synthetic and field datasets demonstrate the effectiveness of the proposed method.

2021

2021

1. Anti-aliasing add-on for deep prior seismic data interpolation

   Francesco Picetti , Vincenzo Lipari, Paolo Bestagini , and 1 more author

   In 2021 IEEE International Conference on Image Processing (ICIP) , 2021

   Abs

   Data interpolation is a fundamental step in any seismic processing workflow. Among machine learning techniques recently proposed to solve data interpolation as an inverse problem, Deep Prior paradigm aims at employing a convolutional neural network to capture priors on the data in order to regularize the inversion. However, this technique lacks of reconstruction precision when interpolating highly decimated data due to the presence of aliasing. In this work, we propose to improve Deep Prior inversion by adding a directional Laplacian as regularization term to the problem. This regularizer drives the optimization towards solutions that honor the slopes estimated from the interpolated data low frequencies. We provide some numerical examples to showcase the methodology devised in this manuscript, showing that our results are less prone to aliasing also in presence of noisy and corrupted data.

2. Post-Stack Inversion with Uncertainty Estimation through Bayesian Deep Image Prior

   Vicenzo Lipari , Francesco Picetti , P Bestagini , and 1 more author

   In 82nd EAGE Annual Conference & Exhibition , 2021

   Abs

   We propose a Bayesian framework for post-stack inversion and uncertainty estimation based on deep priors. A Convolutional Neural Network acts like a nonlinear preconditioner to the inversion problem, capturing the priors from the data in its inner layers. At the same time, it also provides an estimate of the aleatoric uncertainty; this is achieved by minimizing a joint objective function in the CNN parameters space. Then, in a Bayesian framework, Montecarlo dropout is leveraged in order to sample the posterior and characterize the inherent uncertainty. Through synthetic examples we prove our methodology to be effective.

2020

2020

1. A deep prior convolutional autoencoder for seismic data interpolation

   Fantong Kong , Vincenzo Lipari, Francesco Picetti , and 2 more authors

   In EAGE 2020 Annual Conference & Exhibition Online , 2020

   Abs

   A properly designed skip-connection convolutional autoencoder deep generator is able to capture the inner structure of shot gathers from subsampled seismic data without any pre-training procedure. The complete interpolated data can be reconstructed by feeding the autoencoder with multidimensional random noise and minimizing the mean squared error between generated and measured data. The performances achieved on synthetic and field data show the effectiveness of the proposed method.

2. Deep prior-based seismic data interpolation via multi-res u-net

   Fantong Kong , Francesco Picetti , Vincenzo Lipari, and 2 more authors

   In SEG International Exposition and Annual Meeting , 2020

   Abs

   Interpolation of seismic data is an important pre-processing step in most seismic processing workflows. Through the deep image prior paradigm, it is possible to use Convolutional Neural Networks for seismic data interpolation without the costly and prone-to-overfitting training stage. The proposed method makes use of the multi-res U-net architecture as a deep prior to perform interpolation of time slices in order to reconstruct 3D shot gathers. Numerical examples on different corrupted synthetic datasets demonstrate the validity and effectiveness of the proposed approach.

3. A tool for processing seismic images: Study on CycleGAN

   Francesco Picetti , Federico Lombardi , Vincenzo Lipari, and 3 more authors

   In 3rd Asia Pacific Meeting on Near Surface Geoscience & Engineering , 2020

   Abs

   The advent of deep learning techniques had a huge impact in the geophysical community. Convolutional Neural Networks have been investigated for interpretation tasks and, lately, for seismic imaging problems traditionally approached with analytical methods. In this manuscript we employ a state-of-the-art CycleGAN for processing migrated images. We show an application scenario in which this post-processing machine transforms migrated images into corresponding reflectivity model of the subsurface. The proposed methodology achieves promising results in both synthetic and field data. The reconstructed reflectivity model exhibits a meaningful topology. However, the results suggest that computer-vision techniques need to be adapted to tackle the complexity of seismic imaging problems. For this purpose, we believe that geophysical knowledge can be embedded into more accurate network design.

4. Deep prior-based unsupervised reconstruction of irregularly sampled seismic data

   Fantong Kong , Francesco Picetti , Vincenzo Lipari, and 3 more authors

   IEEE Geoscience and Remote Sensing Letters, 2020

   Abs

   Irregularity and coarse spatial sampling of seismic data strongly affect the performances of processing and imaging algorithms. Therefore, interpolation is a usual preprocessing step in most of the processing workflows. In this work, we propose a seismic data interpolation method based on the deep prior paradigm: an ad hoc convolutional neural network is used as a prior to solve the interpolation inverse problem, avoiding any costly and prone-to-overfitting training stage. In particular, the proposed method leverages a multiresolution U-Net with 3-D convolution kernels exploiting correlations in cubes of seismic data, at different scales in all directions. Numerical examples on different corrupted synthetic and field data sets show the effectiveness and promising features of the proposed approach.

2019