SEARCH · Search NASA

Results for “2D Image”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 73 records · Page 4

All sources NASA NTRS DOE OSTI

DOE OSTI2023

Wavelength domain spatial frequency modulation imaging: enabling fiber optic delivery and detection

Spatial frequency modulation imaging (SPIFI) provides a simple architecture for modulating an extended illumination source that is compatible with single pixel imaging. We demonstrate wavelength domain SPIFI (WD-SPIFI) by encoding time-varying spatial frequencies in the spectral domain that can produce enhanced resolution images, like its spatial domain counterpart, spatial domain (SD) SPIFI. However, contrary to SD-SPIFI, WD-SPIFI enables remote delivery by single mode fiber, which can be attractive for applications where free-space imaging is not practical. Finally, we demonstrate a cascaded system incorporating WD-SPIFI in-line with SD-SPIFI enabling single pixel 2D imaging without any beam or sample scanning.

47 OTHER INSTRUMENTATION↗

DOE OSTI2021

Molecular imaging of plant-microbe interactions on the Brachypodium seed surface

Plant growth-promoting rhizobacteria (PGPR) play a crucial role in biological control and pathogenic defense on and with-in plant tissues, however the mechanism(s) by which plants associate with PGPR to elicit such beneficial effects needs further study. Here, we present high mass resolution time-of-flight secondary ion mass spectrometry (ToF-SIMS) imaging of Brachypodium distachyon (Brachypodium) seeds with and without exposure to two model PGPR, i.e., Gram-negative Pseudomonas fluorescens SBW25::GFP (P.) and Gram-positive Arthrobacter chlorophenolicus A6::GFP (A.). Delayed image extraction was used to image PGPB-treated seed sections to reveal morphological changes. ToF-SIMS spectral compari-son, principal component analysis (PCA), and two-dimensional (2D) imaging show that the selected PGPR have different effects on the host seed surface, resulting in changes in chemical composition and morphology. Metabolite prod-ucts and biomarkers, such as flavonoids, phenolic com-pounds, fatty acids, and indole-3-acetic acid (IAA), were ob-served on the PGPR-treated seed surfaces. These compounds had different distributions on the Brachypodium seed surface for the two PGPR, indicating that the different bacteria elicited distinct responses from the host. Our results illustrate that ToF-SIMS is an effective tool to study plant-microbe interac-tions and to provide insightful information with submicrome-ter lateral resolution of the chemical distributions associated with morphological features, potentially offering a new way to study the mechanisms underlying beneficial roles of PGPR.

PGPR, Brachypodium distachyon, ToF-SIMS, metabolit↗

NASA NTRS2021

Unsteady Pressure-Sensitive-Paint Shot Noise Reduction

In the study of unsteady Pressure-Sensitive Paint (uPSP) image data sets, pixel intensity values collected by Phantom cameras from multiple perspectives are used to calculate the surface pressure of an experimental model. This paper focuses on mitigation of shot noise which is introduced into the image data set by the cameras. Shot noise impacts the quality of data collected, increasing the system error in its pressure value calculations. This paper characterises statistical methods to reduce shot noise within experimental data by taking an average or weighted average across an area of pixels of their respective counts. Each combined-pixel has a higher total effective intensity value and therefore exhibits reduced noise throughout its time history. This is demonstrated by calculating the standard deviations of the combined pixel values through time, which decrease as the combined pixel area increases. Finally, these methods are compared across experimental wind tunnel conditions to examine their effectiveness at different Mach numbers. Ultimately, the results are used to recommend a blur filter type and size which will be applied to the raw image data prior to processing, as well as a minimum camera resolution for future experiments. Downstream of the filter application, the software to convert the 2D image data sets into surface pressure readings on a 3D grid use a ratio of decimation that assigns multiple pixels to each 3D grid node. This decimation ratio will also be aligned with the size of the blur filter, resulting in a more manageable data set size and compromised spatial resolution. Combined with the effect of the blur filter, the resulting pixel intensity readings will exhibit significantly reduced shot noise, reducing the magnitude of uncertainty and error in the final calculated pressure. For future tests, the camera spatial resolution must be sufficient to capture small flow and model features even after the recommended blur filter is applied. The image data set is taken from a uPSP experiment of the Space Launch System Block 1 cargo vehicle model in September 2019.

Lucy Tang↗

NASA NTRS2021

Unsteady Pressure-Sensitive Paint (UPSP) Shot Noise Reduction

Lucy Tang↗

DOE OSTI2024

High-Density Polyethylene Custom Focusing Lenses for High-Resolution Transient Terahertz Biomedical Imaging Sensors

Transient terahertz time-domain spectroscopy (THz-TDS) imaging has emerged as a novel non-ionizing and noninvasive biomedical imaging modality, designed for the detection and characterization of a variety of tissue malignancies due to their high signal-to-noise ratio and submillimeter resolution. We report our design of a pair of aspheric focusing lenses using a commercially available lens-design software that resulted in about 200 × 200-μm2 focal spot size corresponding to the 1-THz frequency. The lenses are made of high-density polyethylene (HDPE) obtained using a lathe fabrication and are integrated into a THz-TDS system that includes low-temperature GaAs photoconductive antennae as both a THz emitter and detector. The system is used to generate high-resolution, two-dimensional (2D) images of formalin-fixed, paraffin-embedded murine pancreas tissue blocks. The performance of these focusing lenses is compared to the older system based on a pair of short-focal-length, hemispherical polytetrafluoroethylene (TeflonTM) lenses and is characterized using THz-domain measurements, resulting in 2D maps of the tissue refractive index and absorption coefficient as imaging markers. For a quantitative evaluation of the lens effect on the image resolution, we formulated a lateral resolution parameter, R2080, defined as the distance required for a 20–80% transition of the imaging marker from the bare paraffin region to the tissue region in the same image frame. The R2080 parameter clearly demonstrates the advantage of the HDPE lenses over TeflonTM lenses. The lens-design approach presented here can be successfully implemented in other THz-TDS setups with known THz emitter and detector specifications.

47 OTHER INSTRUMENTATION↗

NASA NTRS1992

PCIPS 2.0: Powerful multiprofile image processing implemented on PCs

Over the years, the processing power of personal computers has steadily increased. Now, 386- and 486-based PC's are fast enough for many image processing applications, and inexpensive enough even for amateur astronomers. PCIPS is an image processing system based on these platforms that was designed to satisfy a broad range of data analysis needs, while requiring minimum hardware and providing maximum expandability. It will run (albeit at a slow pace) even on a 80286 with 640K memory, but will take full advantage of bigger memory and faster CPU's. Because the actual image processing is performed by external modules, the system can be easily upgraded by the user for all sorts of scientific data analysis. PCIPS supports large format lD and 2D images in any numeric type from 8-bit integer to 64-bit floating point. The images can be displayed, overlaid, printed and any part of the data examined via an intuitive graphical user interface that employs buttons, pop-up menus, and a mouse. PCIPS automatically converts images between different types and sizes to satisfy the requirements of various applications. PCIPS features an API that lets users develop custom applications in C or FORTRAN. While doing so, a programmer can concentrate on the actual data processing, because PCIPS assumes responsibility for accessing images and interacting with the user. This also ensures that all applications, even custom ones, have a consistent and user-friendly interface. The API is compatible with factory programming, a metaphor for constructing image processing procedures that will be implemented in future versions of the system. Several application packages were created under PCIPS. The basic package includes elementary arithmetics and statistics, geometric transformations and import/export in various formats (FITS, binary, ASCII, and GIF). The CCD processing package and the spectral analysis package were successfully used to reduce spectra from the Nordic Telescope at La Palma. A photometry package is also available, and other packages are being developed. A multitasking version of PCIPS that utilizes the factory programming concept is currently under development. This version will remain compatible (on the source code level) with existing application packages and custom applications.

Smirnov, O. M.↗

DOE OSTI2026

3D reconstruction and neural rendering for adversarial machine learning

While evasion attacks on computer vision systems have been widely studied, creating attacks that remain effective under significant changes in viewpoint continues to be challenging. Traditional approaches often rely on affine transformations of images, but these approaches degrade at larger perspective shifts and often produce unrealistic or ineffective perturbations. Recent methods use differentiable renderers to improve viewpoint robustness, but they typically depend on manually constructed 3D models. We introduce a semi-automated pipeline that generates physically printable and perspective-invariant adversarial patches using only a small set of 2D images. Our method integrates 3D reconstruction, neural rendering, adversarial patch optimization, and an object detection victim model into a unified workflow. We use 2D Gaussian Splatting for high fidelity mesh reconstruction and FlexPara for surface parameterization that produces texture maps suitable for patch editing. Together, these components form a fully differentiable pipeline in PyTorch3D that links texture modification to model outputs, enabling efficient optimization of patches that remain effective across many viewpoints. The complete process, from image capture to patch printing and physical evaluation, can be completed within a few hours. We demonstrate the effectiveness of the resulting patches through attacks on the YOLOv8 object detection model and discuss remaining challenges and opportunities for improving robustness and scalability.

Singhvi, Vivaan [ORNL] (ORCID:0009000586288221)↗

DOE OSTI2025

Sparse measurement medical CT reconstruction using multi-fused block matching denoising priors

A major challenge for medical X-ray CT imaging is reducing the number of X-ray projections to lower radiation dosage and reduce scan times without compromising image quality. However these under-determined inverse imaging problems rely on the formulation of an expressive prior model to constrain the solution space while remaining computationally tractable. Traditional analytical reconstruction methods like Filtered Back Projection (FBP) often fail with sparse measurements, producing artifacts due to their reliance on the Shannon-Nyquist Sampling Theorem. Consensus Equilibrium, which is a generalization of Plug and Play, is a recent advancement in Model-Based Iterative Reconstruction (MBIR), has facilitated the use of multiple denoisers are prior models in an optimization free framework to capture complex, non-linear prior information. However, 3D prior modelling in a Plug and Play approach for volumetric image reconstruction requires long processing time due to high computing requirement. Instead of directly using a 3D prior, this work proposes a BM3D Multi Slice Fusion (BM3D-MSF) prior that uses multiple 2D image denoisers fused to act as a fully 3D prior model in Plug and Play reconstruction approach. Our approach does not require training and are thus able to circumvent ethical issues related with patient training data and are readily deployable in varying noise and measurement sparsity levels. In addition, reconstruction with the BM3D-MSF prior achieves similar reconstruction image quality as fully 3D image priors, but with significantly reduced computational complexity. We test our method on clinical CT data and demonstrate that our approach improves reconstructed image quality.

Hossain, Maliha [ORNL]↗

DOE OSTI2025

Hyper Spectral Anomaly Detection

Anomaly detection is a common machine learning (ML) task with growing importance in the fields of imaging, quality assurance, and multiple security related disciplines. Anomaly detection is more difficult than traditional machine learning methods due to the inherent unlabeled nature of the datasets. Existing anomaly detection architectures commonly face challenges with explainability, retaining information related to the relational structure of the data, and false positive rates. Hyperspectral Imaging Anomaly Detection (HSI) is a statistical model that employs vertex and edge weighted graphs to preserve the data’s relationships on different topographical scales. The model is able to generalize from anomaly detection in 2D images to novel datasets related to cyber-security. Furthermore, the use of multi-spectral and other filtering methods results in fewer false positives and increases the explainability of model predictions. When applying HSI to cyber-security datasets, we are able to successfully detect malicious activity with a relatively high degree of accuracy.

97 - MATHEMATICS AND COMPUTING↗

DOE OSTI2022

Gas puff imaging on the TCV tokamak

We present the design and operation of a suite of Gas Puff Imaging (GPI) diagnostic systems installed on the Tokamak à Configuration Variable (TCV) for the study of turbulence in the plasma edge and Scrape-Off-Layer (SOL). These systems provide the unique ability to simultaneously collect poloidal 2D images of plasma dynamics at the outboard midplane, around the X-point, in both the High-Field Side (HFS) and Low-Field Side (LFS) SOL, and in the divertor region. We describe and characterize an innovative control system for deuterium and helium gas injection, which is becoming the default standard for the other gas injections at TCV. Extensive pre-design studies and the different detection systems are presented, including an array of avalanche photodiodes and a high-speed CMOS camera. First results with spatial and time resolutions of up to ≈2 mm and 0.5 µs, respectively, are described, and future upgrades of the GPI diagnostics for TCV are discussed.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

NASA NTRSDate not supplied

Celestial Navigation in Cislunar Space with autoNGC

Celestial navigation (CelNav) is a source of navigation observables where images of known solar system bodies are used to locate a spacecraft, beneficial within the solar system for both cislunar and deep space missions. CelNav provides a variety of design benefits to support and enable current and new autonomous space operations- using only a camera and a processor to produce in-situ measurements for navigation. This technology reduces subscription to ground-based tracking during all phases of a mission, freeing up resources for other operational needs. This also supports secure navigation since it eliminates the need for ground contact. CelNav enables missions where the light time delay between Earth and the spacecraft is too long (or the Earth to spacecraft line of sight is obscured) to support critical operations. It also enables smaller mission classes, where Deep Space Network (DSN)time is cost prohibitive, to reduce its cost by focusing primarily on data downlink. Finally, it enables the NASA Artemis program and other cislunar human space flight by providing redundant navigation to traditional radiometric tracking. In this presentation, we discuss the implementation of a CelNav app in autonomous Navigation, Guidance, and Control (autoNGC), a comprehensive flight software suite for onboard autonomy that is built on the core Flight System (cFS). The presentation also summarizes the results of flight software-in-the-loop (SIL) and processor-in-the-loop (PIL) demonstrations. Both are high-fidelity simulations with the use of a camera emulator hosted on a GPU server that simulates images that would be captured by the camera. The CelNav app leverages the use of cGIANT (cFS Goddard Image Analysis and Navigation Tool).Previously developed for the autoNGC software suite, cGIANT is an onboard autonomous image processing and optical navigation (OpNav) tool that performs limb-based OpNav and Terrain Relative Navigation. The added CelNav capability of cGIANT generates bearing measurements to multiple known celestial bodies (planets, moons, asteroids, comets, etc.) in monocular (2D) images. These observables are then fed to the Goddard Enhanced Onboard Navigation System (GEONS)navigation filter app, enabling us to navigate the spacecraft autonomously. In early 2025, the autoNGC CelNav capability is planned to be flight tested as part of the onboard autonomy experiment on the Cislunar Autonomous Positioning System Technology Operations and Navigation Experiment(CAPSTONE) spacecraft that is currently in a Lunar Near Rectilinear Halo Orbit(NRHO).

celestial navigation↗

NASA NTRS2011

Machine Vision for Relative Spacecraft Navigation During Approach to Docking

This paper describes a machine vision system for relative spacecraft navigation during the terminal phase of approach to docking that: 1) matches high contrast image features of the target vehicle, as seen by a camera that is bore-sighted to the docking adapter on the chase vehicle, to the corresponding features in a 3d model of the docking adapter on the target vehicle and 2) is robust to on-orbit lighting. An implementation is provided for the case of the Space Shuttle Orbiter docking to the International Space Station (ISS) with quantitative test results using a full scale, medium fidelity mock-up of the ISS docking adapter mounted on a 6-DOF motion platform at the NASA Marshall Spaceflight Center Flight Robotics Laboratory and qualitative test results using recorded video from the Orbiter Docking System Camera (ODSC) during multiple orbiter to ISS docking missions. The Natural Feature Image Registration (NFIR) system consists of two modules: 1) Tracking which tracks the target object from image to image and estimates the position and orientation (pose) of the docking camera relative to the target object and 2) Acquisition which recognizes the target object if it is in the docking camera Field-of-View and provides an approximate pose that is used to initialize tracking. Detected image edges are matched to the 3d model edges whose predicted location, based on the pose estimate and its first time derivative from the previous frame, is closest to the detected edge1 . Mismatches are eliminated using a rigid motion constraint. The remaining 2d image to 3d model matches are used to make a least squares estimate of the change in relative pose from the previous image to the current image. The changes in position and in attitude are used as data for two Kalman filters whose outputs are smoothed estimate of position and velocity plus attitude and attitude rate that are then used to predict the location of the 3d model features in the next image.

Chien, Chiun-Hong↗

NASA NTRS1993

Remote sensing estimates of cirrus particle size for tropical and midlatitude cirrus: Hexagonal crystals and ice spheres

A large discrepancy exists in current estimates of a mean cirrus particle size appropriate for calculations of the effects of these ice clouds on solar and thermal infrared radiative fluxes. For spheres with large size parameter (x = (2 pi r / lambda) is greater than 30, where r is particle radius), and moderate absorption (n(sup i) x less than 1, where n(sup i) is imaginary index of refraction for ice), the optimal effective particle radius is given by: r(sub e) = integral of r(exp 3)n(r)dr / integral of r(exp 2)n(r)dr. For the remote sensing of cirrus particle size at wavelengths of 0.83, 1.65, and 2.21 mu m, a 50 mu m ice sphere would have a size parameter of about 200, and values of n(sup i) x of 0, 0.045, and 0.06, satisfying the above conditions. However, while r(sub e) is a well-defined parameter for spheres, this cross-section area-weighted particle radius can only be extended to non-spherical particles by defining some equivalent sphere, typically an equivalent volume or equivalent cross-section area sphere. Using equivalent volume spheres, values of r(sub e) obtained over Lake Michigan on October 28, 1986, during FIRE phase I varied from 200 mu m (King Air 2D Imaging probes) to 60 mu m (Landsat reflectances at 0.83, 1.65, and 2.2 mu m), to 25 mu m (HIS spectrometer thermal emission between 8 and 12 mu m). Three major uncertainties were identified in this comparison: small ice particles missed by the 2D-C aircraft probes, uncertain ice refractive index, and uncertainties in the single scatter albedos and scattering phase functions used in the radiative calculations. Since the first FIRE cirrus results, advances have been made in all three areas. The present paper reports on improvements in the radiative modeling of ice particles at 0.83, 1.65, and 2.21 mu m wavelengths appropriate for comparisons to Landsat Thematic Mapper data. The paper also includes new results for Landsat observations of ice clouds in the eastern and western tropical Pacific.

Wielicki, Bruce A.↗

DOE OSTI2023

Classifying metal‐binding sites with neural networks

Abstract To advance our ability to predict impacts of the protein scaffold on catalysis, robust classification schemes to define features of proteins that will influence reactivity are needed. One of these features is a protein's metal‐binding ability, as metals are critical to catalytic conversion by metalloenzymes. As a step toward realizing this goal, we used convolutional neural networks (CNNs) to enable the classification of a metal cofactor binding pocket within a protein scaffold. CNNs enable images to be classified based on multiple levels of detail in the image, from edges and corners to entire objects, and can provide rapid classification. First, six CNN models were fine‐tuned to classify the 20 standard amino acids to choose a performant model for amino acid classification. This model was then trained in two parallel efforts: to classify a 2D image of the environment within a given radius of the central metal binding site, either an Fe ion or a [2Fe‐2S] cofactor, with the metal visible (effort 1) or the metal hidden (effort 2). We further used two sub‐classifications of the [2Fe‐2S] cofactor: (1) a standard [2Fe‐2S] cofactor and (2) a Rieske [2Fe‐2S] cofactor. The accuracy for the model correctly identifying all three defined features was >95%, despite our perception of the increased challenge of the metalloenzyme identification. This demonstrates that machine learning methodology to classify and distinguish similar metal‐binding sites, even in the absence of a visible cofactor, is indeed possible and offers an additional tool for metal‐binding site identification in proteins.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2022

Seeking regularity from irregularity: unveiling the synthesis–nanomorphology relationships of heterogeneous nanomaterials using unsupervised machine learning

Nanoscale morphology of functional materials determines their chemical and physical properties. However, despite increasing use of transmission electron microscopy (TEM) to directly image nanomorphology, it remains challenging to quantify the information embedded in TEM data sets, and to use nanomorphology to link synthesis and processing conditions to properties. We develop an automated, descriptor-free analysis workflow for TEM data that utilizes convolutional neural networks and unsupervised learning to quantify and classify nanomorphology, and thereby reveal synthesis–nanomorphology relationships in three different systems. While TEM records nanomorphology readily in two-dimensional (2D) images or three-dimensional (3D) tomograms, we advance the analysis of these images by identifying and applying a universal shape fingerprint function to characterize nanomorphology. After dimensionality reduction through principal component analysis, this function then serves as the input for morphology grouping through unsupervised learning. We demonstrate the wide applicability of our workflow to both 2D and 3D TEM data sets, and to both inorganic and organic nanomaterials, including tetrahedral gold nanoparticles mixed with irregularly shaped impurities, hybrid polymer-patched gold nanoprisms, and polyamide membranes with irregular and heterogeneous 3D crumple structures. In each of these systems, unsupervised nanomorphology grouping identifies both the diversity and the similarity of the nanomaterial across different synthesis conditions, revealing how synthetic parameters guide nanomorphology development. Our work opens possibilities for enhancing synthesis of nanomaterials through artificial intelligence and for understanding and controlling complex nanomorphology, both for 2D systems and in the far less explored case of 3D structures, such as those with embedded voids or hidden interfaces.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2025

Nanotomography for Quantitative 3D Particle Reconstruction

Particulates are ubiquitous across fuel cycle operations and carry critical information about particle formation, processing, and potential proliferation-related activities. Traditional analytical techniques, including micro-Raman spectroscopy and standard electron microscopy, are often limited in spatial resolution or dimensionality, particularly when used to examine metallic or submicron-scale features. Understanding particle morphology, phase distribution, and internal porosity is essential for constraining formation conditions, thermodynamic environments, and material transport behavior. In this report, we demonstrate the application of plasma focused ion beam nanotomography to reconstruct micron-scale particulates at nanoscale resolution. Using high-resolution backscattered electron imaging and Avizo software, we obtained 3D reconstructions that enabled quantitative analysis of particle morphology, phase composition, and internal voids. Representative examples include a Ta particle with a large central void and a composite particle with embedded tetrahedral crystalline structures. These reconstructions reveal structural and compositional details that are inaccessible through conventional 2D imaging. The results demonstrate that nanotomography provides both qualitative and quantitative insights into particle formation and behavior. Using nanotomography, porosity and phase distributions can be quantified to inform models of particle density, transport, and solidification conditions. Beyond technical insights, the workflow developed here establishes a transferable capability for analyzing heterogeneous particles and has potential applications in bulk materials studies via x-ray computed tomography or other volumetric imaging modalities. Ongoing efforts are focused on optimizing the workflow to process multiple particles simultaneously, increasing throughput and statistical robustness. Overall, this work illustrates the power of nanotomography as a tool for connecting particulate morphology to formation mechanisms, composition, and transport, thereby strengthening analytical capabilities for nuclear forensics, fuel cycle analysis, and related scientific investigations.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

DOE OSTI2024

Hyperspectral acquisition with ScanImage at the single pixel level: application to time domain coherent Raman imaging

We present a comprehensive strategy and its practical implementation using the commercial ScanImage software platform to perform hyperspectral point scanning microscopy when a fast time-dependent signal varies at each pixel level. In the proposed acquisition scheme, the scan along the X-axis is slowed down while the data acquisition is maintained at a high pace to enable the rapid acquisition of the time-dependent signal at each pixel level. The ScanImage generated raw 2D images have a very asymmetric aspect ratio between X and Y, the X axis encoding both for space and time acquisition. The results are X-axis macro-pixel where the associated time-dependent signal is sampled to provide hyperspectral information. We exemplified the proposed hyperspectral scheme in the context of time-domain coherent Raman imaging, where a pump pulse impulsively excites molecular vibrations that are subsequently probed by a time-delayed probe pulse. In this case, the time-dependent signal is a fast acousto-optics delay line that can scan a delay of 4.5ps in 25 μ s at each pixel level. With this acquisition scheme, we demonstrate ultra-fast hyperspectral vibrational imaging in the low frequency range [10 cm −1 , 150 cm −1 ] over a 500 μm field of view (64 x 64 pixels) in 130ms (∼ 7.5 frames/s). The proposed acquisition scheme can be readily extended to other applications requiring the acquisition of a fast-evolving signal at each pixel level.

Metais, Samuel↗

DOE OSTI2025

Toward more-robust, AI-enabled subsurface seismic imaging for geotechnical applications

Non-invasive seismic imaging has the potential to cost-effectively evaluate large volumes of subsurface material to inform geotechnical site investigation. However, seismic imaging using full waveform inversion (FWI) requires significant computational time and is dependent on an initial starting model. As a result, FWI has not yet been widely adopted into geotechnical practice. Previous efforts, on relatively simple two-layered models, indicate that data-driven artificial intelligence (AI) models may be as effective as FWI at predicting 2D images of shear wave velocity (V s ). Furthermore, the AI model predictions can be made almost instantaneously after data acquisition and do not require an initial starting model. We examine the generality of these findings by developing a new AI model for subsurface seismic imaging, whereby we make several notable contributions. First, we architect a multimodal AI model that combines time- and frequency-domain representations of the seismic wavefield to predict a 50 m by 20 m subsurface image of V s . Second, we developed a new diverse dataset of 100,000 images with their corresponding seismic wavefields to train the AI model. Third, we propose four physics-informed data augmentations for data-driven seismic imaging. Fourth, we develop two prediction consistency tests to evaluate the model’s performance when the true subsurface is unknown. Our final model, which has been made publicly available, is capable of predicting a subsurface V s image from a single seismic wavefield with an average, mean absolute percent error (MAPE) of 24 %. The predictive model is applied to a field dataset and shown to be consistent with local geology and shear-wave refraction measurements from the same location.

Artificial intelligence↗