Search NASA⌕ Search

SEARCH · Search NASA

Results for “Model Inference”

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.

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

At least 1,765 records · Page 98
All sourcesNASA NTRSDOE OSTI
NASA NTRS1996

Numerical modelling of the circulation of superrotating atmospheres: Venus and Titan

Atmospheric superrotation is known to exist on Venus where the cloud cover, at 70 km above the surface, rotates about 60 times faster than the solid planet. Superrotation is strongly suspected on Titan on the basis of numerical studies and observations. The numerical simulation of superrotation is described. It is shown how the general circulation models, originally developed for terrestrial meteorology and climatology, can be used to infer the circulation in such atmospheres. The emphasis is on the case of Titan for which a specific general circulation model was developed.

Hourding, F.↗
DOE OSTI2015

ON THE MAGNETIC FIELD OF PULSARS WITH REALISTIC NEUTRON STAR CONFIGURATIONS

We have recently developed a neutron star model fulfilling global and not local charge neutrality, both in the static and in the uniformly rotating cases. The model is described by the coupled Einstein-Maxwell-Thomas-Fermi equations, in which all fundamental interactions are accounted for in the framework of general relativity and relativistic mean field theory. Uniform rotation is introduced following Hartle's formalism. We show that the use of realistic parameters of rotating neutron stars, obtained from numerical integration of the self-consistent axisymmetric general relativistic equations of equilibrium, leads to values of the magnetic field and radiation efficiency of pulsars that are very different from estimates based on fiducial parameters that assume a neutron star mass M = 1.4 M {sub ☉}, radius R = 10 km, and moment of inertia I = 10{sup 45} g cm{sup 2}. In addition, we compare and contrast the magnetic field inferred from the traditional Newtonian rotating magnetic dipole model with respect to the one obtained from its general relativistic analog, which takes into account the effect of the finite size of the source. We apply these considerations to the specific high-magnetic field pulsar class and show that, indeed, all of these sources can be described as canonical pulsars driven by the rotational energy of the neutron star, and have magnetic fields lower than the quantum critical field for any value of the neutron star mass.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2026

Probabilistic Inference of Low-Surface-Brightness Galaxy Morphological Parameters Using Simulation-Based Inference

Low-surface-brightness galaxies (LSBGs) are diffuse, often dark-matter-dominated systems whose faintness makes their structural parameters difficult to measure reliably in wide-field imaging surveys. Robust parameter inference, including uncertainty quantification, is important for population studies and for comparisons with models of galaxy formation, as future surveys are expected to produce increasingly large samples of diffuse galaxies. In practice, LSBG profile modeling is sensitive to sky- background errors, masking choices, contaminating background sources, and the computational cost of obtaining posterior-level uncertainties for large samples. Motivated by these questions, we develop a simulation-based inference (SBI) framework for estimating posterior distributions of LSBG morphological parameters from simulated galaxy images. Using PyImfit, we generate DES-like single-Sersic profile LSBG images with known position angle, ellipticity, Sersic index, effective surface brightness, and effective radius. We then train a normalizing-flow-based neural posterior estimator using the sbi package to infer these parameters from the simulated images. For isolated simulated galaxies, the SBI posterior recovers the true input parameters, produces posterior predictive residuals consistent with the assumed noise model, and shows good empirical calibration in a DES-motivated test regime. We also compare SBI with PyImfit-based MCMC inference and find broadly comparable posterior constraints, while SBI enables substantially faster posterior sampling after training. Finally, we test robustness to compact background contaminants. A model trained only on isolated galaxies produces undercovered posteriors on contaminated images, whereas training on simulations with variable contaminant positions and fluxes improves calibration across contaminated test sets. These results demonstrate the promise of SBI for scalable, uncertainty-aware LSBG morphology inference, while emphasizing that posterior reliability strongly depends on whether training simulations include relevant observational complications.

Batbayar, Bilguun [U. Chicago (main)]↗
NASA NTRSDate not supplied

Development of an Optical Model for the MEDLI2 Radiometer

We present an optical model of the MEDLI2 radiometer, which was flown onboard the Mars 2020 vehicle during atmospheric entry on February 18, 2021. The proposed model takes into account the radiometer geometry and optical phenomena such as reflection, refraction, and absorption to estimate the signal loss inherent in the design of the radiometer. The model serves as a filter, which can be used to convert predicted total radiative heat fluxes on the radiometer window into predicted measured heat fluxes, enabling more accurate comparisons of the radiometer flight data with predicted radiometer measurements. Optical properties of the radiometer’s sapphire window are inferred from spectral transmissivity measurements. A modified Lorentz Oscillator Model is used to represent the dielectric function of the window and is fit to the transmission data. The resulting optical model of the radiometer is validated against experimental data used to measure the dependence of the radiometer output voltage on the incident angle of the light impinging on it. Comparison of predicted radiometer measurements with the MEDLI2 flight data show good agreement between the predicted and measured peak heat fluxes, though offset at different times. A linear increase in the signal loss due to radiometer optics alone is also demonstrated, suggesting that the changing shock angle over the course of the trajectory impacts the signal loss.

MEDLI2↗
DOE OSTI2022

Physics Informed Neural Networks as Computational Physics Emulators

This report is a brief overview and evaluation of Physics Informed Neural Networks (PINNs). Karniadakis and co-workers, e.g., Karniadakis et al. (2021) assert that the PINNs approach integrates seamlessly both data and mathematical physics models, even in partially understood, uncertain and high-dimensional contexts. They further claim that PINNs are effective and efficient for ill-posed and inverse problems, and when combined with domain decomposition, are scalable to large problems and a tool to discover hidden physics. While they demonstrate the capabilities in specific academic instances, their overarching claims about PINNs seem to be an overstatement, at least at the current time. We have briefly considered a few of the limitations of PINNs in this investigation. It is not clear to us if the PINNs approach can ever be competitive with approaches that use specialized algorithms to achieve high-accuracy solutions of governing equations and other techniques that can combine observational data with such solutions. As an example of the latter, consistent with the principles of Bayesian inference, data assimilation, or more generally data-model fusion, is a process that fuses observational data typically with a computational model that respects certain constraints such as conservation laws. For example, improvements in observational network combined with data assimilation have been key in improving weather predictions over the past four decades Kalnay (2003).

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2021

AeroCom Phase III Multi-Model Evaluation of the Aerosol Life Cycle and Optical Properties Using Ground and Space-Based Remote Sensing as Well as Surface In Situ Observations

Within the framework of the AeroCom (Aerosol Comparisons between Observations and Models) initiative, the state-of-the-art modelling of aerosol optical properties is assessed from 14 global models participating in the phase III control experiment (AP3). The models are similar to CMIP6/AerChemMIP Earth System Models (ESMs) and provide a robust multi-model ensemble. Inter-model spread of aerosol species lifetimes and emissions appears to be similar to that of mass extinction coefficients (MECs), suggesting that aerosol optical depth (AOD) uncertainties are associated with a broad spectrum of parameterised aerosol processes. Total AOD is approximately the same as in AeroCom phase I (AP1) simulations. However, we find a 50 % decrease in the optical depth (OD) of black carbon (BC), attributable to a combination of decreased emissions and lifetimes. Relative contributions from sea salt (SS) and dust (DU) have shifted from being approximately equal in AP1 to SS contributing about 2∕3 of the natural AOD in AP3. This shift is linked with a decrease in DU mass burden, a lower DU MEC, and a slight decrease in DU lifetime, suggesting coarser DU particle sizes in AP3 compared to AP1. Relative to observations, the AP3 ensemble median and most of the participating models underestimate all aerosol optical properties investigated, that is, total AOD as well as fine and coarse AOD (AODf, AODc), Ångström exponent (AE), dry surface scattering (SCdry), and absorption (ACdry) coefficients. Compared to AERONET, the models underestimate total AOD by ca. 21 % ± 20 % (as inferred from the ensemble median and interquartile range). Against satellite data, the ensemble AOD biases range from −37 % (MODIS-Terra) to −16 % (MERGED-FMI, a multi-satellite AOD product), which we explain by differences between individual satellites and AERONET measurements themselves. Correlation coefficients (R) between model and observation AOD records are generally high (R>0.75), suggesting that the models are capable of capturing spatio-temporal variations in AOD. We find a much larger underestimate in coarse AODc (∼ −45 % ± 25 %) than in fine AODf (∼ −15 % ± 25 %) with slightly increased inter-model spread compared to total AOD. These results indicate problems in the modelling of DU and SS. The AODc bias is likely due to missing DU over continental land masses (particularly over the United States, SE Asia, and S. America), while marine AERONET sites and the AATSR SU satellite data suggest more moderate oceanic biases in AODc. Column AEs are underestimated by about 10 % ± 16 %. For situations in which measurements show AE > 2, models underestimate AERONET AE by ca. 35 %. In contrast, all models (but one) exhibit large overestimates in AE when coarse aerosol dominates (bias ca. +140 % if observed AE < 0.5). Simulated AE does not span the observed AE variability. These results indicate that models overestimate particle size (or underestimate the fine-mode fraction) for fine-dominated aerosol and underestimate size (or overestimate the fine-mode fraction) for coarse-dominated aerosol. This must have implications for lifetime, water uptake, scattering enhancement, and the aerosol radiative effect, which we can not quantify at this moment. Comparison against Global Atmosphere Watch (GAW) in situ data results in mean bias and inter-model variations of −35 % ± 25 % and −20 % ± 18 % for SCdry and ACdry, respectively. The larger underestimate of SCdry than ACdry suggests the models will simulate an aerosol single scattering albedo that is too low. The larger underestimate of SCdry than ambient air AOD is consistent with recent findings that models overestimate scattering enhancement due to hygroscopic growth. The broadly consistent negative bias in AOD and surface scattering suggests an underestimate of aerosol radiative effects in current global aerosol models. Considerable inter-model diversity in the simulated optical properties is often found in regions that are, unfortunately, not or only sparsely covered by ground-based observations. This includes, for instance, the Sahara, Amazonia, central Australia, and the South Pacific. This highlights the need for a better site coverage in the observations, which would enable us to better assess the models, but also the performance of satellite products in these regions. Using fine-mode AOD as a proxy for present-day aerosol forcing estimates, our results suggest that models underestimate aerosol forcing by ca. −15 %, however, with a considerably large interquartile range, suggesting a spread between −35 % and +10 %.

space-based remote sensing↗
DOE OSTI2023

Exploring unsupervised top tagging using Bayesian inference

Recognizing hadronically decaying top-quark jets in a sample of jets, or even its total fraction in the sample, is an important step in many LHC searches for Standard Model and Beyond Standard Model physics as well. Although there exists outstanding top-tagger algorithms, their construction and their expected performance rely on Montecarlo simulations, which may induce potential biases. For these reasons we develop two simple unsupervised top-tagger algorithms based on performing Bayesian inference on a mixture model. In one of them we use as the observed variable a new geometrically-based observable \tilde{A}_{3} A ̃ 3 , and in the other we consider the more traditional \tau_{3}/\tau_{2} τ 3 / τ 2 N N -subjettiness ratio, which yields a better performance. As expected, we find that the unsupervised tagger performance is below existing supervised taggers, reaching expected Area Under Curve AUC \sim 0.80-0.81 ∼ 0.80 − 0.81 and accuracies of about 69% - − 75% in a full range of sample purity. However, these performances are more robust to possible biases in the Montecarlo that their supervised counterparts. Our findings are a step towards exploring and considering simpler and unbiased taggers.

Alvarez, Ezequiel↗
DOE OSTI2015

X-ray continuum as a measure of pressure and fuel–shell mix in compressed isobaric hydrogen implosion cores

Pressure, by definition, characterizes the conditions within an isobaric implosion core at peak compression [Gus'kov et al., Nucl. Fusion 16, 957 (1976); Betti et al., Phys. Plasmas 8, 5257 (2001)] and is a key parameter in quantifying its near-ignition performance [Lawson, Proc. Phys. Soc. London, B 70, 6 (1957); Betti et al., Phys. Plasmas 17, 058102 (2010); Goncharov et al., Phys. Plasmas 21, 056315 (2014); and Glenzer et al., Phys. Plasmas 19, 056318 (2012)]. At high spectral energy, where the x-ray emission from an imploded hydrogen core is optically thin, the emissivity profile can be inferred from the spatially resolved core emission. This emissivity, which can be modeled accurately under hot-core conditions, is dependent almost entirely on the pressure when measured within a restricted spectral range matched to the temperature range anticipated for the emitting volume. In this way, the hot core pressure at the time of peak emission can be inferred from the measured free-free emissivity profile. The pressure and temperature dependences of the x-ray emissivity and the neutron-production rate explain a simple scaling of the total filtered x-ray emission as a constant power of the total neutron yield for implosions of targets of similar design over a broad range of shell implosion isentropes. This scaling behavior has been seen in implosion simulations and is confirmed by measurements of high-isentrope implosions [Sangster et al., Phys. Plasmas 20, 056317 (2013)] on the OMEGA laser system [Boehly et al., Opt. Commun. 133, 495 (1997)]. Attributing the excess emission from less-stable, low-isentrope implosions, above the level expected from this neutron-yield scaling, to the higher emissivity of shell carbon mixed into the implosion's central hot spot, the hot-spot “fuel–shell” mix mass can be inferred.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
NASA NTRS1992

Derivation of surface properties from Magellan altimetry data

The fit of the Hagfors model to the Magellan altimetry data provides a means to characterize the surface properties of Venus. However, the derived surface properties are only meaningful if the model provides a good representation of the data. The Hagfors model provides a good representation of the data. The Hagfors model is generally a realistic fit to surface scattering properties of a nadir-directed antenna such as the Magellan altimeter; however, some regions of the surface of Venus are poorly described by the existing model, according to the goodness of fit parameter provided on the ARCDR CD-ROMs. Poorly characterized regions need to be identified and fit to new models in order to derive more accurate surface properties for use in inferring the geological processes that affect the surface in those regions. We have compared the goodness of fit of the Hagfors model to the distribution of features across the planet, and preliminary results show a correlation between steep topographic slopes and poor fits to the standard model, as has been noticed by others. In this paper, we investigate possible relations between many classes of features and the ability of the Hagfors model to fit the observed echo profiles. In the regions that are not well characterized by existing models, we calculate new models that compensate for topographic relief in order to derive improved estimates of surface properties. Areas investigated to date span from longitude 315 through 45, at all latitudes covered by Magellan. A survey of those areas yields preliminary results that suggest that topographically high regions are well suited to the current implementation of the Hagfors model. Striking examples of such large-scale good fits are Alpha Regio, the northern edges of Lada Terra, and the southern edge of Ishtar Terra. Other features that are typically well fit are the rims of coronae such as Heng-O and the peaks of volcanos such as Gula Mons. Surprisingly, topographically low regions, such as the ubiquitous plains areas, are modeled poorly in comparison. However, this generalization has has exceptions: Lakshmi Planum is an elevated region that is not well fit compared to the rest of neighboring Ishtar, while the southern parts of topographically low Guinevere Planitia are characterized quite well by the Hagfors model. Features that are candidates for improved models are impact craters, coronae, ridges of significant scale, complex ridged terrains, moderate-sized mountains, and sharp terrain boundaries. These features are chosen because the goodness of fit is likely to be most affected either by departures from normal incidence angles or by sharp changes in terrain type within a single footprint. Most large features that are elevated with respect to their surroundings will suffer from steep slope effects, and smaller coronae and impact craters will probably suffer due to rapid changes in their appearance within a single footprint (10-20 km).

Lovell, Amy J.↗
NASA NTRS2005

PlanWorks: A Debugging Environment for Constraint Based Planning Systems

Numerous planning and scheduling systems employ underlying constraint reasoning systems. Debugging such systems involves the search for errors in model rules, constraint reasoning algorithms, search heuristics, and the problem instance (initial state and goals). In order to effectively find such problems, users must see why each state or action is in a plan by tracking causal chains back to part of the initial problem instance. They must be able to visualize complex relationships among many different entities and distinguish between those entities easily. For example, a variable can be in the scope of several constraints, as well as part of a state or activity in a plan; the activity can arise as a consequence of another activity and a model rule. Finally, they must be able to track each logical inference made during planning. We have developed PlanWorks, a comprehensive system for debugging constraint-based planning and scheduling systems. PlanWorks assumes a strong transaction model of the entire planning process, including adding and removing parts of the constraint network, variable assignment, and constraint propagation. A planner logs all transactions to a relational database that is tailored to support queries for of specialized views to display different forms of data (e.g. constraints, activities, resources, and causal links). PlanWorks was specifically developed for the Extensible Universal Remote Operations Planning Architecture (EUROPA(sub 2)) developed at NASA, but the underlying principles behind PlanWorks make it useful for many constraint-based planning systems. The paper is organized as follows. We first describe some fundamentals of EUROPA(sub 2). We then describe PlanWorks' principal components. We then discuss each component in detail, and then describe inter-component navigation features. We close with a discussion of how PlanWorks is used to find model flaws.

Daley, Patrick↗
DOE OSTI2023

Bayesian inference of in-medium baryon-baryon scattering cross sections from HADES proton flow data

Within a Bayesian statistical framework using a Gaussian Process emulator for an isospin-dependent Boltzmann-Uehling-Uhlenbeck (IBUU) transport model simulator of heavy-ion reactions at intermediate energies, we infer from the HADES proton flow data the posterior probability distribution functions of in-medium baryon-baryon scattering cross section modification factor X with respect to free-space and the corresponding incompressibility K of nuclear matter as well as their correlation function. In conclusion, the mean value of X is found to be $X$ = $1.32_ {+0.28} ^ {-0.40} $ at 68% confidence level assuming the nuclear incompressibility K will not exceed 400 MeV, providing circumstantial evidence for enhanced baryon-braryon scattering cross sections in hot and dense nuclear matter.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2019

Automatic extraction of cancer registry reportable information from free-text pathology reports using multitask convolutional neural networks

Objective: In this work we implement 2 different multitask learning (MTL) techniques, hard parameter sharing and cross-stitch, to train a word-level convolutional neural network (CNN) specifically designed for automatic extraction of cancer data from unstructured text in pathology reports. We show the importance of learning related information extraction (IE) tasks leveraging shared representations across the tasks to achieve state-of-the-art performance in classification accuracy and computational efficiency. Materials and Methods: Multitask CNN (MTCNN) attempts to tackle document information extraction by learning to extract multiple key cancer characteristics simultaneously. We trained our MTCNN to perform 5 information extraction tasks: (1) primary cancer site (65 classes), (2) laterality (4 classes), (3) behavior (3 classes), (4) histological type (63 classes), and (5) histological grade (5 classes). We evaluated the performance on a corpus of 95 231 pathology documents (71 223 unique tumors) obtained from the Louisiana Tumor Registry. We compared the performance of the MTCNN models against single-task CNN models and 2 traditional machine learning approaches, namely support vector machine (SVM) and random forest classifier (RFC). Results: MTCNNs offered superior performance across all 5 tasks in terms of classification accuracy as compared with the other machine learning models. Based on retrospective evaluation, the hard parameter sharing and cross-stitch MTCNN models correctly classified 59.04% and 57.93% of the pathology reports respectively across all 5 tasks. The baseline models achieved 53.68% (CNN), 46.37% (RFC), and 36.75% (SVM). Based on prospective evaluation, the percentages of correctly classified cases across the 5 tasks were 60.11% (hard parameter sharing), 58.13% (cross-stitch), 51.30% (single-task CNN), 42.07% (RFC), and 35.16% (SVM). Moreover, hard parameter sharing MTCNNs outperformed the other models in computational efficiency by using about the same number of trainable parameters as a single-task CNN. Conclusions: The hard parameter sharing MTCNN offers superior classification accuracy for automated coding support of pathology documents across a wide range of cancers and multiple information extraction tasks while maintaining similar training and inference time as those of a single task–specific model.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗
NASA NTRS1979

The grabens of Canyonlands National Park, Utah - Geometry, mechanics, and kinematics

Results are reported for three years of field and analytical study of a system of young and very well exposed grabens in southeastern Utah. A kinematic and mechanical model is developed for these specific grabens with the hope that this will aid in interpreting graben systems elsewhere on the earth and other planets. Because studies on other planets must be based entirely on photo analysis, an attempt is made to determine to what extent three-dimensional geometry could be inferred from photographs alone. An equation based on the model of graben geometry and kinematics relates thickness of faulted plates to graben parameters measurable on photographs. The initial width of a nascent graben is a simple function of the dip of the faults bounding the graben and of the thickness of the faulted brittle plate. The uniform spacing of the grabens is a natural result of the geometry of the system.

Mcgill, G. E.↗
DOE OSTI2024

Tokamak divertor plasma emulation with machine learning

Abstract Future tokamak devices that aim to create conditions relevant to power plant operations must consider strategies for mitigating damage to plasma facing components in the divertor. One of the goals of MAST-U tokamak operations is to inform these considerations by researching advanced divertor configurations that aid stable plasma detachment. Machine design, scenario planning and detachment control would all greatly benefit from tools that enable rapid calculation of scenario-relevant quantities given some input parameters. This paper presents a method for generating large, simulated scrape-off layer data sets, which was applied to generate a data set of steady-state Hermes-3 simulations of the MAST-U tokamak. A machine learning model was constructed using a Bayesian approach to hyperparameter optimisation to predict diagnosable output quantities given control-relevant input features. The resulting best-performing model, which is based on a feedforward neural network, achieves high accuracy when predicting electron temperature at the divertor target and carbon impurity radiation front position and runs in around 1 ms in inference mode. Techniques for interpreting the predictions made by the model were applied, and a high-resolution parameter scan of upstream conditions was performed to demonstrate the utility of rapidly generating accurate predictions using the emulator. This work represents a step forward in the design of machine learning-driven emulators of tokamak exhaust simulation codes in operational modes relevant to divertor detachment control and plasma scenario design.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2023

Machine Learning 5G Attack Detection in Programmable Logic

Machine learning-assisted network security may significantly contribute to securing 5G components. However, machine learning network security inference speeds generally require tens to hundreds of milliseconds thereby introducing significant latency in 5G operations. The inference latency can be reduced by deploying the machine learning model to programmable logic in a field programmable gate array (FPGA) at the cost of a small loss in accuracy. In order to quantify this loss, as well as to establish baseline performance inference speeds for programmable logic implementations, this work explores an autoencoder and a ß-variational autoencoder deployed on two different FPGA evaluation boards and compares accuracy and performance against an NVIDIA A100 GPU implementation. A publicly available 5G dataset containing 10 types of attacks along with normal traffic is introduced as part of the evaluation.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Pushing the frontiers in climate modelling and analysis with machine learning

Climate modelling and analysis are facing new demands to enhance projections and climate information. Here, in this study, we argue that now is the time to push the frontiers of machine learning beyond state-of-the-art approaches, not only by developing machine-learning-based Earth system models with greater fidelity, but also by providing new capabilities through emulators for extreme event projections with large ensembles, enhanced detection and attribution methods for extreme events, and advanced climate model analysis and benchmarking. Utilizing this potential requires key machine learning challenges to be addressed, in particular generalization, uncertainty quantification, explainable artificial intelligence and causality. This interdisciplinary effort requires bringing together machine learning and climate scientists, while also leveraging the private sector, to accelerate progress towards actionable climate science.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2018

3-D Semantic Information Inference from Airborne Video. Final Report

We developed a full 3D modeling pipeline focusing on an important application scenario: city-scale 3D reconstruction from aerial imagery. We propose an approach to solve camera pose estimation and dense reconstruction from Wide Area Aerial Surveillance (WAAS) videos captured by an airborne platform. Our approach solves them in an online fashion: it incrementally updates a sparse 3D map and estimates the camera pose as each new frame arrives; depth maps of selected key frames are computed using a variational method and integrated to produce a full 3D model via volumetric reconstruction. In practice, aerial imagery is usually captured using a multicamera system. We propose an approach for camera pose estimation of multi-camera aerial imagery which is parallelized on multiple GPUs for efficiency. The approach is also extended for progressive 3D model acquisition with a hand-held camera. We also developed an offline approach since online approach is not a necessity and accuracy has higher priority over efficiency in many scenarios. We presented MeshRecon, a mesh-based offline system composed of three modules: a dense point cloud is generated using multi-resolution plane sweep method; an initial mesh model is extracted from the point cloud via global optimization considering visibility information of all images; the mesh model is then iteratively refined to capture structural details by optimizing the photometric consistency and spatial regularization. The major processes are parallelized on GPU for efficiency. For the aerial imagery case, we evaluate our system on several real-world multicamera aerial imagery datasets, each covering an urban scenario of several square kilometers. Quantitative result shows that the reconstructed geometric 3D model is highly accurate with error smaller than 1 meter over the entire city. Besides aerial imagery, we also evaluate its performance on general geometric 3D model acquisition of real-world objects. Result shows that the system is robust and flexible for various types of objects at different scales in both indoor and outdoor environments. Based on city 3D models reconstructed at different times, we present a system for city-scale geometry change detection by performing comparisons at the 3D geometry level. Our system is able to detect geometry changes at different scales, ranging from a building cluster to small-scale vegetation changes, with high accuracy.

97 MATHEMATICS AND COMPUTING↗