Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Resnet”

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.

87 records · Page 5
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Red Wine Fermentation Alters Grape Seed Morphology and Internal Porosity

During wine fermentation, grape berry components, including seeds, undergo extensive physical and chemical changes that result in the release of flavonoids, such as tannins, from seeds into wine. Understanding changes in seed morphology during fermentation is crucial for aiding the development of accurate prediction models for flavonoid extraction during winemaking, which enhances fermentation management and ensures consistency in wines from year to year. High-resolution x-ray microcomputed tomography (x-ray μCT) was used to investigate the effect of red wine fermentation on changes in grape seed morphology. Using a PyTorch-based implementation of a fully convolutional network with a Resnet-101 backbone for semantic segmentation of x-ray μCT images, we quantified extensive alteration to grape seed structure during fermentation. Image analyses revealed the development of a pore network breaking apart the seed endosperm by the end of fermentation, leading to an increase in surface area. Fermentation significantly altered grape seed morphology. Such alterations could enable transport of seed flavonoids from inside the endosperm and integument to outside the seed. Further research on the physical processes occurring in seeds during wine fermentation is necessary to build better physiochemical models.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRS2020

NeMO-Net – The Neural Multi-Modal Observation & Training Network for Global Coral Reef Assessment

We present NeMO-Net, the Srst open-source deep convolutional neural network (CNN) and interactive learning and training software aimed at assessing the present and past dynamics of coral reef ecosystems through habitat mapping into 10 biological and physical classes. Shallow marine systems, particularly coral reefs, are under significant pressures due to climate change, ocean acidification, and other anthropogenic pressures, leading to rapid, often devastating changes, in these fragile and diverse ecosystems. Historically, remote sensing of shallow marine habitats has been limited to meter-scale imagery due to the optical effects of ocean wave distortion, refraction, and optical attenuation. NeMO-Net combines 3D cm-scale distortion-free imagery captured using NASA FluidCam and Fluid lensing remote sensing technology with low resolution airborne and spaceborne datasets of varying spatial resolutions, spectral spaces, calibrations, and temporal cadence in a supercomputer-based machine learning framework. NeMO-Net augments and improves the benthic habitat classification accuracy of low-resolution datasets across large geographic ad temporal scales using high-resolution training data from FluidCam.NeMO-Net uses fully convolutional networks based upon ResNet and ReSneNet to perform semantic segmentation of remote sensing imagery of shallow marine systems captured by drones, aircraft, and satellites, including WorldView and Sentinel. Deep Laplacian Pyramid Super-Resolution Networks (LapSRN) alongside Domain Adversarial Neural Networks (DANNs) are used to reconstruct high resolution information from low resolution imagery, and to recognize domain-invariant features across datasets from multiple platforms to achieve high classification accuracies, overcoming inter-sensor spatial, spectral and temporal variations.Finally, we share our online active learning and citizen science platform, which allows users to provide interactive training data for NeMO-Net in 2D and 3D, integrated within a deep learning framework. We present results from the PaciSc Islands including Fiji, Guam and Peros Banhos 1 1 2 1 3 1 where 24-class classification accuracy exceeds 91%.

Chirayath, Ved↗
NASA NTRS2018

Deep Neural Network for Precision Multi-band Infrared Image Segmentation

Image segmentation is one of the fundamental steps in computer vision. Separating targets from background clutter with high precision is a challenging operation for both humans and computers. Currently, segmenting objects from IR images is done by tedious manual work. The implementation of a Deep Neural Network (DNN) to perform precision segmentation of multi-band IR video images is presented. A customized pix2pix DNN with multiple layers of generative encoder/decoder and discriminator architecture is used in the IR image segmentation process. Real and synthetic images and ground truths are employed to train the DNN. Iterative training is performed to achieve optimum accuracy of segmentation using a minimal number of training data. Special training images are created to enhance the missing features and to increase the segmentation accuracy of the objects. Retraining strategies are developed to minimize the DNN training time. Single pixel accuracy has been achieved in IR target boundary segmentation using DNNs. The segmentation accuracy between the customized pix2pix DNN and simple thresholding, GraphCut, simple neural network and ResNet models are compared.

Torres, Gil↗
NASA NTRS2022

Microstructure Segmentation with Deep Learning Encoders Pre-Trained on a Large Microscopy Dataset

This study examined the improvement of microscopy segmentation accuracy by transfer learning from a large dataset of microscopy images called MicroNet. Many neural network encoder architectures, including VGG, Inception, and ResNet, were trained on over 100,000 labelled microscopy images from 54 classes. These pre-trained encoders were then embedded into multiple segmentation architectures including U-Net and DeepLabV3+ to evaluate segmentation performance on newly created benchmark microscopy datasets. Compared to ImageNet pre-training, models pre-trained on MicroNet generalized better to out-of-distribution micrographs taken under different imaging and sample conditions and were more accurate with less training data. When training with only a single Ni-superalloy image, pre-training on MicroNet produced a 72.2 percent reduction in relative segmentation error. These results suggest that transfer learning from large in-domain datasets generate models with learned feature representations that are more useful for downstream tasks and will likely improve any microscopy image analysis technique that can leverage pre-trained encoders.

machine learning↗
NASA NTRSDate not supplied

Concept-based Analysis of Neural Networks via Vision-Language Models

The analysis of vision-based deep neural networks (DNNs) is highly desirable but it is very challenging due to the difficulty of expressing formal specifications for vision tasks and the lack of efficient verification procedures. In this paper, we propose to leverage emerging multimodal, vision-language, foundation models (VLMs) as a lens through which we can reason about vision models. VLMs have been trained on a large body of images accompanied by their textual description, and are thus implicitly aware of high-level, human-understandable concepts describing the images. We describe a logical specification language Con spec designed to facilitate writing specifications in terms of these concepts. To define and formally check Con spec specifications, we build a map between the internal representations of a given vision model and a VLM, leading to an efficient verification procedure of natural-language properties for vision models. We demonstrate our techniques on a ResNet-based classifier trained on the RIVAL-10 dataset using CLIP as the multimodal model.

Large Vision Language Models↗
NASA NTRS2024

EdgeCortix SAKURA-I Machine-Learning, PCIe Accelerator SEE Heavy Ion Test Report

To enable autonomy in space, machine-learning and computer vision applications become invaluable for sensor processing. However, these algorithms are computationally complex and unfeasible for many embedded central processing units (CPUs) and usually require external coprocessors, such as graphics processing units (GPUs) or accelerators specific to the application, including application specific integrated circuits (ASICs). In power-constrained systems, GPUs tend to consume more power than is acceptable (>40W), so lower-power accelerators have shown promise to provide the performance needed under spacecraft constraints. For radiation engineers, developing methodologies that can properly test CPUs, GPUs, and accelerators, and enable comparisons between them remains a necessary complication to solve as the devices become more complex. The methodology in this test aims to be a start in developing a baseline single-event effect (SEE) test for client-device machine learning accelerators. This category of devices do not host their own operating system. This testing campaign is a continuation of a previous 200 MeV proton test performed in January 2024. This report covers two heavy ion tests of the SAKURA-I card: one in April 2024, and one in June 2024. Additional data was needed after the April test due to ion-range issues experienced at higher linear-energy transfers (LETs). These range issues are described in more detail in Section 8. This experiment characterizes SEEs and data error susceptibility of the EdgeCortix SAKURA-I machine-learning accelerator under heavy ions. The device was monitored for single event upsets (SEUs) and single event functional interrupts (SEFIs) at the Lawrence Berkeley National Laboratory’s 88-inch cyclotron. The SAKURA-I board accelerates machine-learning inference applications on a host computer through a PCIex16 connection. For the purposes of devising an end to end automated analysis workflow for this experiment, the YOLO-V5 and SSD300 objection-detection models, and the ResNet-50, EfficientNet, and MobileNetV2 image classification models were used as a representative suite of analytical machine-learning models.

Seth S Roffe↗
DOE OSTI2020

Systematic Evaluation of Backdoor Data Poisoning Attacks on Image Classifiers

Backdoor data poisoning attacks have recently been demonstrated in computer vision research as a potential safety risk for machine learning (ML) systems. Traditional data poisoning attacks manipulate training data to induce unreliability of an ML model, whereas backdoor data poisoning attacks maintain system performance unless the MLmodel is presented with an input containing an embedded“trigger” that provides a predetermined response advantageous to the adversary. Our work builds upon prior back-door data-poisoning research for ML image classifiers and systematically assesses different experimental conditions including types of trigger patterns, persistence of trigger patterns during retraining, poisoning strategies, architectures (ResNet-50, NasNet, NasNet-Mobile), datasets (Flowers, CIFAR-10), and potential defensive regularization techniques (Contrastive Loss, Logit Squeezing, Manifold Mixup,Soft-Nearest-Neighbors Loss). Experiments yield four key findings. First, the success rate of backdoor poisoning at-tacks varies widely, depending on several factors, including model architecture, trigger pattern and regularization technique. Second, we find that poisoned models are hard to detect through performance inspection alone. Third, regularization typically reduces backdoor success rate, although it can have no effect or even slightly increase it, depending on the form of regularization. Finally, backdoors inserted through data poisoning can be rendered ineffective after just a few epochs of additional training on a small set of clean data without affecting the model’s performance.

Truong, Loc T.↗
DOE OSTI2021

Binary Complex Neural Network Acceleration on FPGA

Being able to learn from complex data with phase information is imperative for many signal processing applications. Today’s real-valued deep neural networks (DNNs) have shown efficiency in latent information analysis but fall short when applied to the complex domain. Deep complex networks (DCN) , in contrast, can learn from complex data, but have high computational costs; therefore, they cannot satisfy the instant decision making requirements of many deployable systems dealing with short observations or short signal bursts. Recent, Binarized Complex Neural Network (BCNN), which integrates DCNs with binarized neural networks (BNN), shows great potential in classifying complex data in real-time. In this paper, we propose a structural pruning based accelerator of BCNN, which is able to provide more than 5000 frames/s inference throughput on edge devices. The high performance comes from both the algorithm and hardware sides. On the algorithm side, we conduct structural pruning to the original BCNN models and obtain 20 × pruning rates with negligible accuracy loss; on the hardware side, we propose a novel 2D convolution operation accelerator for the binary complex neural network. Experimental results show that the proposed design works with over 90% utilization and is able to achieve the inference throughput of 5882 frames/s and 4938 frames/s for complex NIN-Net and ResNet-18 using CIFAR-10 dataset and Alveo U280 Board.

Peng, Hongwu↗
DOE OSTI2022

Panel-Segmentation: A Python Package for Automated Solar Array Metadata Extraction Using Satellite Imagery

The NREL Python Panel-Segmentation package is a toolkit that automates the process of extracting accurate and valuable metadata related to solar array installations, using publicly available Google Maps satellite imagery. Previously published work includes automated azimuth estimation for individual solar installations in satellite images. Our continued research focuses on automated detection and classification of solar installation mounting configuration (tracking or fixed-tilt; rooftop, ground, or carport). Specifically, a Faster-RCNN Resnet-50 feature pyramid network (FPN) model was trained and validated on 862 manually labeled satellite images. This model was used to perform object detection on satellite imagery, locating and classifying individual solar installations' mounting configuration and type. Model results showed a mean average precision score (mAP) of 77.79%, with the model strongest at detecting fixed-tilt ground mount and fixed-tilt carport installations. The object detection model and its outputs have been incorporated into the Panel-Segmentation package's automated metadata extraction pipeline, which returns the mounting configuration and azimuth for individual solar arrays in satellite imagery. The complete image data set with labels has been released on the U.S. Department of Energy (DOE) DuraMAT DataHub, to encourage further research in this area.

deep learning↗
DOE OSTI2022

Modeling Combustion Reaction ODEs with Neural Networks

The chemistry of combustion reactions is complex as it involves many species and reactions. In practice, such a system is often modeled computationally using an empirically derived chemical mechanism. Given the large range of reaction rates, the system is then time evolved using a stiff ODE solver. However, even when reduced chemical mechanisms are employed, the system can become computationally expensive for two-dimensional or three-dimensional systems. Such systems can also face problems with instability. As such, it is desirable to find a cheaper, stable alternative to solving the reaction system. Neural networks offer the potential to learn these reaction ODEs and time evolve a combustion reaction in a more cost-efficient manner than stiff ODE solvers. In this study, a variety of neural networks are trained on zero-dimensional Cantera simulations of methane combustion with varying initial conditions. Several predictive approaches as well as several neural network architectures (artificial neural network with dropout, ResNet, and Neural ODE) are compared in their ability to predict combustion trajectories. Promising models are then identified.

combustion kinetics↗
DOE OSTI2024

Harnessing ML Privacy by Design Through Crossbar Array Non-idealities

Deep Neural Networks (DNNs), handling computeand data-intensive tasks, often utilize accelerators like Resistiveswitching Random-access Memory (RRAM) crossbar for energyefficient in-memory computation. Despite RRAM’s inherent nonidealities causing deviations in DNN output, this study transforms the weakness into strength. By leveraging RRAM non-idealities, the research enhances privacy protection against Membership Inference Attacks (MIAs), which reveal private information from training data. RRAM non-idealities disrupt MIA features, increasing model robustness and revealing a privacy-accuracy tradeoff. Empirical results with four MIAs and DNNs trained on different datasets demonstrate significant privacy leakage reduction with a minor accuracy drop (e.g., up to 2.8% for ResNet-18 with CIFAR-100).

artificial intelligence↗
DOE OSTI2025

DNN-based Signal Processing for Liquid Argon Time Projection Chambers

We investigate a deep learning-based signal processing for liquid argon time projection chambers (LArTPCs), a leading detector technology in neutrino physics. Identifying regions of interest (ROIs) in LArTPCs is challenging due to signal cancellation from bipolar responses and various detector effects observed in real data. We approach ROI identification as an image segmentation task, and employ a U-ResNet architecture. The network is trained on samples that incorporate detector geometry information and include a range of detector variations. Our approach significantly outperforms traditional methods while maintaining robustness across diverse detector conditions. This method has been adopted for signal processing in the Short-Baseline Neutrino program and provides a valuable foundation for future experiments such as the Deep Underground Neutrino Experiment.

Bhat, Avinay [Chicago U.]↗
DOE OSTI2025

Enhancing Cluster Identification in Atom Probe Tomography Data Using Transfer Learning

Atom Probe Tomography (APT) is a powerful technique for visualizing the atomic-scale distribution of solutes in materials, but quantitative cluster analysis of APT datasets remains a challenge due to the need for subjective parameter selection in clustering algorithms. While distance-based and density-based methods such as HDBSCAN are widely used, their performance is highly sensitive to user-defined parameters, which undermines reproducibility and accuracy. This study proposes an image-based, deep learning-aided workflow for automating parameter selection and cluster detection in APT data analysis. By projecting 3D APT point clouds onto 2D planes, we leverage pretrained convolutional neural networks (ConvNeXt-Tiny and ResNet-50) through transfer learning to predict the number of clusters present in synthetic datasets. The output is used to guide K-means clustering and estimate HDBSCAN parameters, specifically minimum cluster size and minimum sample points. This approach reduces reliance on manual parameter tuning, improving consistency and scalability. The methodology demonstrates the feasibility of using image-based deep learning for interpreting complex spatial patterns in APT data, enabling faster and more objective analysis. The complete workflow and code are made publicly available to support reproducibility and future research.

Density-based clustering↗
DOE OSTI2020

End-to-End Physics Event Classification with CMS Open Data: Applying Image-Based Deep Learning to Detector Data for the Direct Classification of Collision Events at the LHC

This paper describes the construction of novel end-to-end image-based classifiers that directly leverage low-level simulated detector data to discriminate signal and background processes in pp collision events at the Large Hadron Collider at CERN. To better understand what end-to-end classifiers are capable of learning from the data and to address a number of associated challenges, we distinguish the decay of the standard model Higgs boson into two photons from its leading background sources using high-fidelity simulated CMS Open Data. We demonstrate the ability of end-to-end classifiers to learn from the angular distribution of the photons recorded as electromagnetic showers, their intrinsic shapes, and the energy of their constituent hits, even when the underlying particles are not fully resolved, delivering a clear advantage in such cases over purely kinematics-based classifiers.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2022

Building surrogate models of nuclear density functional theory with Gaussian processes and autoencoders

From the lightest Hydrogen isotopes up to the recently synthesized Oganesson (Z = 118), it is estimated that as many as about 8,000 atomic nuclei could exist in nature. Most of these nuclei are too short-lived to be occurring on Earth, but they play an essential role in astrophysical events such as supernova explosions or neutron star mergers that are presumed to be at the origin of most heavy elements in the Universe. Understanding the structure, reactions, and decays of nuclei across the entire chart of nuclides is an enormous challenge because of the experimental difficulties in measuring properties of interest in such fleeting objects and the theoretical and computational issues of simulating strongly-interacting quantum many-body systems. Nuclear density functional theory (DFT) is a fully microscopic theoretical framework which has the potential of providing such a quantitatively accurate description of nuclear properties for every nucleus in the chart of nuclides. Thanks to high-performance computing facilities, it has already been successfully applied to predict nuclear masses, global patterns of radioactive decay like β or γ decay, and several aspects of the nuclear fission process such as, e.g., spontaneous fission half-lives. Yet, predictive simulations of nuclear spectroscopy—the low-lying excited states and transitions between them—or of nuclear fission, or the quantification of theoretical uncertainties and their propagation to basic or applied nuclear science applications, would require several orders of magnitude more calculations than currently possible. However, most of this computational effort would be spent into generating a suitable basis of DFT wavefunctions. Such a task could potentially be considerably accelerated by borrowing tools from the field of machine learning and artificial intelligence. In this paper, we review different approaches to applying supervised and unsupervised learning techniques to nuclear DFT.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗