Geophysical Data Processing

Understanding our planet is one of the most exciting and challenging tasks that researchers all around the world are facing.

We at ISPL are working on imaging the Earth’s subsurface applying the most recent advances in the signal processing community.

Seismic Imaging

Since the easily accessible reservoirs have already been exploited, oil and gas companies are looking for energy reservoirs in increasingly complex environments: new exploration scenarios require higher resolution and fidelity, posing more and more difficult challenges to the scientific community.

Read more

In order to infer a quantitative description of the rock layers, geologists need images of the subsurface, obtained by very expensive exploration campaigns:

1. The area of interest is investigated by means of artificial seismic waves (i.e. small artificial earthquakes).
   A controlled source emits signals that propagate through the media; these waves are reflected and transmitted depending on the subsurface topology and lithology, so that an array of sensors records the wave-field. Acquisition files can occupy up to 10 terabytes of disk space.
2. The acquired data need to be processed in order to attenuate noise and remove distortions. These operations are very complex and thus require a great computational effort. Moreover, they strongly influence the final outcomes of the exploration process.
3. The recorded data are compared with the known source signals in order to infer the position of reflectors, along with their coefficients. This process of relocating seismic events from the sensor positions is called migration and it requires several weeks of computations on modern supercomputing infrastructures (capable of 10¹⁵ operations per second).

ISPL is actively involved in two main research topics:

• Least-Squares Reverse Time Migration, an iterative technique that has proven excellent performance in recovering the amplitude of reflectivity coefficients (very important in order to characterize the rock materials), providing high resolution images with low artefacts.
• Full Waveform Inversion exploits the amplitude and phase information of the seismic waveforms in order to reconstruct the velocity models of the subsurfaces (i.e. to infer the rock layer properties).

Deep Learning for Geophysics

Inverse problems in the field of seismic imaging have at least three distinctive features: their size, both in terms of data and model space; their strongly ill-posed nature; and the absence of ground truth.
In the light of these considerations, it is possible to notice how many geophysical applications could benefit from the joint use of signal processing and machine learning paradigms. Indeed, the availability of huge datasets enables the use of data-driven approaches and not only model-based analysis, which helps scientists relax strong assumptions on the physical phenomena under evaluation.

We can use CNNs to solve minimization problems, turning their overfitting downside into an asset. This means that, rather than using the amount of available data for generalizing the application scenarios, we specialize the architecture to the analysis of a single image.
We are also focusing on how to embed physical constraints into network design. Indeed, geophysical and computer vision applications do not share the same exact problems, and thus the architectures coming from image processing applications must be accurately handled. As an example, the loss functions can be modified in order to satisfy some prior knowledge (e.g. to enforce sparsity in the solution), and the convolution layers can be turned into 3D operators.

PoliMine – Humanitarian Demining System

Buried landmines and unexploded remnants of war are a constant threat for the population of many countries that have been hit by wars in the past years. The huge amount of casualties has been a strong motivation for the research community toward the development of safe and robust techniques designed for landmine clearance.

Nonetheless, being able to detect and localize buried landmines with high precision in an automatic fashion is still considered a challenging task due to the many different boundary conditions that characterize this problem (e.g. several kinds of objects to detect, different soils and meteorological conditions, etc.).

We propose a novel technique for buried object detection tailored to unexploded landmine discovery. The proposed solution exploits a specific kind of convolutional neural network (CNN) known as autoencoder to analyze volumetric data acquired with ground penetrating radar (GPR) using different polarizations. This method works in an anomaly detection framework: we only train the autoencoder on GPR data acquired on landmine-free areas. The system then recognizes landmines as objects that are dissimilar to the soil used during the training step. Experiments conducted on real data show that the proposed technique requires little training and no ad-hoc data pre-processing to achieve accuracy higher than 93% on challenging datasets.