Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Computational imaging”

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 1,693 records · Page 94
All sourcesNASA NTRSDOE OSTI
NASA NTRS1973

Computer enhancement of radiographs

Examination of three relevant noise processes and the image degradation associated with Marshall Space Flight Center's (MSFC) X-ray/scanning system was conducted for application to computer enhancement of radiographs using MSFC's digital filtering techniques. Graininess of type M, R single coat and R double coat X-ray films was quantified as a function of density level using root-mean-square (RMS) granularity. Quantum mottle (including film grain) was quantified as a function of the above film types, exposure level, specimen material and thickness, and film density using RMS granularity and power spectral density (PSD). For various neutral-density levels the scanning device used in digital conversion of radiographs was examined for noise characteristics which were quantified by RMS granularity and PSD. Image degradation of the entire pre-enhancement system (MG-150 X-ray device; film; and optronics scanner) was measured using edge targets to generate modulation transfer functions (MTF). The four parameters were examined as a function of scanning aperture sizes of approximately 12.5 25 and 50 microns.

Dekaney, A.↗
NASA NTRS1984

Multi-image photometric solutions for the Galilean satellites

A method for determining the limb darkening planet, acquired in the same wavelength range (lambda), within a few degrees of the same phase angle (alpha), but acquired at different illumination and viewing conditions. The two images are first geometrically registered in simple cylindrical format. A limb darkening factor is computed for each pixel pair and output as a digital image. Systematic trends with photometric coordinates across this image indicate either a calibration error or inappropriateness of the particular photometric function. This method can be used for any photometric function with a single limb darkening factor, such as the Minnaert function or the function which combines the Lommel-Seeliger and Lambert functions. By solving for many pairs of images, the limb darkening factor as a function of alpha and lambda may be determined. The two image limb darkening solutions determined for 1 Voyager image pair on Ganymede, 16 image pairs on Callisto, 44 pairs on Europa, and 19 pairs on Io are discussed.

Mcewen, A. S.↗
NASA NTRS1995

Computational strategies for tire monitoring and analysis

Computational strategies are presented for the modeling and analysis of tires in contact with pavement. A procedure is introduced for simple and accurate determination of tire cross-sectional geometric characteristics from a digitally scanned image. Three new strategies for reducing the computational effort in the finite element solution of tire-pavement contact are also presented. These strategies take advantage of the observation that footprint loads do not usually stimulate a significant tire response away from the pavement contact region. The finite element strategies differ in their level of approximation and required amount of computer resources. The effectiveness of the strategies is demonstrated by numerical examples of frictionless and frictional contact of the space shuttle Orbiter nose-gear tire. Both an in-house research code and a commercial finite element code are used in the numerical studies.

Danielson, Kent T.↗
NASA NTRS2001

Wavelet Compression of Satellite-Transmitted Digital Mammograms

Breast cancer is one of the major causes of cancer death in women in the United States. The most effective way to treat breast cancer is to detect it at an early stage by screening patients periodically. Conventional film-screening mammography uses X-ray films which are effective in detecting early abnormalities of the breast. Direct digital mammography has the potential to improve the image quality and to take advantages of convenient storage, efficient transmission, and powerful computer-aided diagnosis, etc. One effective alternative to direct digital imaging is secondary digitization of X-ray films. This technique may not provide as high an image quality as the direct digital approach, but definitely have other advantages inherent to digital images. One of them is the usage of satellite-transmission technique for transferring digital mammograms between a remote image-acquisition site and a central image-reading site. This technique can benefit a large population of women who reside in remote areas where major screening and diagnosing facilities are not available. The NASA-Lewis Research Center (LeRC), in collaboration with the Cleveland Clinic Foundation (CCF), has begun a pilot study to investigate the application of the Advanced Communications Technology Satellite (ACTS) network to telemammography. The bandwidth of the T1 transmission is limited (1.544 Mbps) while the size of a mammographic image is huge. It takes a long time to transmit a single mammogram. For example, a mammogram of 4k by 4k pixels with 16 bits per pixel needs more than 4 minutes to transmit. Four images for a typical screening exam would take more than 16 minutes. This is too long a time period for a convenient screening. Consequently, compression is necessary for making satellite-transmission of mammographic images practically possible. The Wavelet Research Group of the Department of Electrical Engineering at The Ohio State University (OSU) participated in the LeRC-CCF collaboration by providing advanced compression technology using wavelet transform. OSU developed a time-efficient software package with various wavelets to compress a serious of mammographic images. This documents reports the result of the compression activities.

Zheng, Yuan F.↗
NASA NTRS1992

DECONV-TOOL: An IDL based deconvolution software package

There are a variety of algorithms for deconvolution of blurred images, each having its own criteria or statistic to be optimized in order to estimate the original image data. Using the Interactive Data Language (IDL), we have implemented the Maximum Likelihood, Maximum Entropy, Maximum Residual Likelihood, and sigma-CLEAN algorithms in a unified environment called DeConv_Tool. Most of the algorithms have as their goal the optimization of statistics such as standard deviation and mean of residuals. Shannon entropy, log-likelihood, and chi-square of the residual auto-correlation are computed by DeConv_Tool for the purpose of determining the performance and convergence of any particular method and comparisons between methods. DeConv_Tool allows interactive monitoring of the statistics and the deconvolved image during computation. The final results, and optionally, the intermediate results, are stored in a structure convenient for comparison between methods and review of the deconvolution computation. The routines comprising DeConv_Tool are available via anonymous FTP through the IDL Astronomy User's Library.

Varosi, F.↗
DOE OSTI2021

WHISPER: Wireless Home Identification and Sensing Platform for Energy Reduction

Many regions of the world benefit from heating, ventilating, and air-conditioning (HVAC) systems to provide productive, comfortable, and healthy indoor environments, which are enabled by automatic building controls. Due to climate change, population growth, and industrialization, HVAC use is globally on the rise. Unfortunately, these systems often operate in a continuous fashion without regard to actual human presence, leading to unnecessary energy consumption. As a result, the heating, ventilation, and cooling of unoccupied building spaces makes a substantial contribution to the harmful environmental impacts associated with carbon-based electric power generation, which is important to remedy. For our modern electric power system, transitioning to low-carbon renewable energy is facilitated by integration with distributed energy resources. Automatic engagement between the grid and consumers will be necessary to enable a clean yet stable electric grid, when integrating these variable and uncertain renewable energy sources. We present the WHISPER (Wireless Home Identification and Sensing Platform for Energy Reduction) system to address the energy and power demand triggered by human presence in homes. The presented system includes a maintenance-free and privacy-preserving human occupancy detection system wherein a local wireless network of battery-free environmental, acoustic energy, and image sensors are deployed to monitor homes, record empirical data for a range of monitored modalities, and transmit it to a base station. Several machine learning algorithms are implemented at the base station to infer human presence based on the received data, harnessing a hierarchical sensor fusion algorithm. Results from the prototype system demonstrate an accuracy in human presence detection in excess of 95%; ongoing commercialization efforts suggest approximately 99% accuracy. Using machine learning, WHISPER enables various applications based on its binary occupancy prediction, allowing situation-specific controls targeted at both personalized smart home and electric grid modernization opportunities.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2025

FPGA-accelerated SpeckleNN with SNL for real-time X-ray single-particle imaging

We present the implementation of a specialized version of our previously published unified embedding model, SpeckleNN, for real-time speckle pattern classification in X-ray Single-Particle Imaging (SPI), using the SLAC Neural Network Library (SNL) on an FPGA platform. This hardware realization transitions SpeckleNN from a prototypic model into a practical edge solution, optimized for running inference near the detector in high-throughput X-ray free-electron laser (XFEL) facilities, such as those found at the Linac Coherent Light Source (LCLS). To address the resource constraints inherent in FPGAs, we developed a more specialized version of SpeckleNN. The original model, which was designed for broader classification across multiple biological samples, comprised ~5.6 million parameters. The new implementation, while reducing the parameter count to 64.6K (a 98.8% reduction), focuses on maintaining the model's essential functionality for real-time operation, achieving an accuracy of 90%. Furthermore, we compressed the latent space from 128 to 50 dimensions. This implementation was demonstrated on the KCU1500 FPGA board, utilizing 71% of available DSPs, 75% of LUTs, and 48% of FFs, with an average power consumption of 9.4W according to the Vivado post-implementation report. The FPGA performed inference on a single image with a latency of 45.015 microseconds at a 200 MHz clock rate. In comparison, running the same inference on an NVIDIA A100 GPU resulted in an average power consumption of ~73W and an image processing latency of around 400 microseconds. Our FPGA-accelerated version of SpeckleNN demonstrated significant improvements, achieving an 8.9 × speedup and a 7.8 × reduction in power consumption compared to the GPU implementation. Key advancements include model specialization and dynamic weight loading through SNL, which eliminates the need for time-consuming FPGA design re-synthesis, allowing fast and continuous deployment of models (re)trained online. These innovations enable real-time adaptive classification and efficient vetoing of speckle patterns, making SpeckleNN more suited for deployment in XFEL facilities. This implementation has the potential to significantly accelerate SPI experiments and enhance adaptability to evolving experimental conditions.

47 OTHER INSTRUMENTATION↗
NASA NTRS2001

Extending unbiased stereology of brain ultrastructure to three-dimensional volumes

OBJECTIVE: Analysis of brain ultrastructure is needed to reveal how neurons communicate with one another via synapses and how disease processes alter this communication. In the past, such analyses have usually been based on single or paired sections obtained by electron microscopy. Reconstruction from multiple serial sections provides a much needed, richer representation of the three-dimensional organization of the brain. This paper introduces a new reconstruction system and new methods for analyzing in three dimensions the location and ultrastructure of neuronal components, such as synapses, which are distributed non-randomly throughout the brain. DESIGN AND MEASUREMENTS: Volumes are reconstructed by defining transformations that align the entire area of adjacent sections. Whole-field alignment requires rotation, translation, skew, scaling, and second-order nonlinear deformations. Such transformations are implemented by a linear combination of bivariate polynomials. Computer software for generating transformations based on user input is described. Stereological techniques for assessing structural distributions in reconstructed volumes are the unbiased bricking, disector, unbiased ratio, and per-length counting techniques. A new general method, the fractional counter, is also described. This unbiased technique relies on the counting of fractions of objects contained in a test volume. A volume of brain tissue from stratum radiatum of hippocampal area CA1 is reconstructed and analyzed for synaptic density to demonstrate and compare the techniques. RESULTS AND CONCLUSIONS: Reconstruction makes practicable volume-oriented analysis of ultrastructure using such techniques as the unbiased bricking and fractional counter methods. These analysis methods are less sensitive to the section-to-section variations in counts and section thickness, factors that contribute to the inaccuracy of other stereological methods. In addition, volume reconstruction facilitates visualization and modeling of structures and analysis of three-dimensional relationships such as synaptic connectivity.

NASA Discipline Neuroscience↗
NASA NTRS1971

Spatial transform coding of color images.

The application of the transform-coding concept to the coding of color images represented by three primary color planes of data is discussed. The principles of spatial transform coding are reviewed and the merits of various methods of color-image representation are examined. A performance analysis is presented for the color-image transform-coding system. Results of a computer simulation of the coding system are also given. It is shown that, by transform coding, the chrominance content of a color image can be coded with an average of 1.0 bits per element or less without serious degradation. If luminance coding is also employed, the average rate reduces to about 2.0 bits per element or less.

Pratt, W. K.↗
NASA NTRS2000

Assessment of Internal Fabric of Particulate Materials

Particle arrangement and distribution within a soil matrix has long been recognized as having significant influence on the mechanical behavior of cohesionless soils. It is well known that two soil specimens having the same grain type (e.g., quartz, feldspar, etc.), same grain size distribution and relative density (or void ratio) can display completely different mechanical behavior. Because of the different fabric configurations in the otherwise similar specimens, they are likely to have different mechanical properties such as stress-strain response, anisotropy, dilatancy, etc. Soil Fabric is defined as the arrangement of particles, particle groups and associated pore space. In the literature, fabric analysis techniques are mainly classified as destructive (e.g., specimen stabilization, thin-sectioning, and microscopy), and nondestructive techniques (e.g., magnetic resonance imaging, ultrasonic testing, x-ray radiography, and computed tomography). Quantifying the void ratio and its distribution is the main parameter used to describe the fabric of particulate materials. This paper presents a comprehensive literature review of fabric analysis techniques applied to particulate materials. In addition, the results of a comprehensive investigation to quantify void ratio of sand specimens will be presented and discussed. The sand used in the experiments in a natural, uniform rounded to sub-rounded silica sand known as F-75 banding sand with mean particle size of 0.22 mm. Uniform specimens and specimens subjected to different axial-strain levels tested under triaxial and biaxial conditions are examined to evaluate void ratio evolution and its distribution using destructive thin-sectioning and nondestructive Computed Tomography (CT) techniques. Details of a new innovative polygon generation technique called Voronoi tessellation used to quantify void ratio of microscopic images of sand grains will be presented and compared to classical Oda's method. Finally, frequency distribution of void ratio calculated using CT analysis will be presented and discussed.

Alshibi, Khalid A.↗
DOE OSTI2023

Sharing leaky-integrate-and-fire neurons for memory-efficient spiking neural networks

Spiking Neural Networks (SNNs) have gained increasing attention as energy-efficient neural networks owing to their binary and asynchronous computation. However, their non-linear activation, that is Leaky-Integrate-and-Fire (LIF) neuron, requires additional memory to store a membrane voltage to capture the temporal dynamics of spikes. Although the required memory cost for LIF neurons significantly increases as the input dimension goes larger, a technique to reduce memory for LIF neurons has not been explored so far. To address this, we propose a simple and effective solution, EfficientLIF-Net, which shares the LIF neurons across different layers and channels. Our EfficientLIF-Net achieves comparable accuracy with the standard SNNs while bringing up to ~4.3× forward memory efficiency and ~21.9× backward memory efficiency for LIF neurons. We conduct experiments on various datasets including CIFAR10, CIFAR100, TinyImageNet, ImageNet-100, and N-Caltech101. Furthermore, we show that our approach also offers advantages on Human Activity Recognition (HAR) datasets, which heavily rely on temporal information. The code has been released at https://github.com/Intelligent-Computing-Lab-Yale/EfficientLIF-Net.

60 APPLIED LIFE SCIENCES↗
NASA NTRS2016

Thomas Leps Internship Abstract

An optical navigation system is being flown as the backup system to the primary Deep Space Network telemetry for navigation and guidance purposes on Orion. This is required to ensure Orion can recover from a loss of communication, which would simultaneously cause a loss of DSN telemetry. Images taken of the Moon and Earth are used to give range and position information to the navigation computer for trajectory calculations and maneuver execution. To get telemetry data from these images, the size and location of the moon need to be calculated with high accuracy and precision. The reentry envelope for the Orion EM-1 mission requires the centroid and radius of the moon images to be determined within 1/3 of a pixel 3 sigma. In order to ensure this accuracy and precision can be attained, I was tasked with building precise dot grid images for camera calibration as well as building a hardware in the loop test stand for flight software and hardware proofing. To calibrate the Op-Nav camera a dot grid is imaged with the camera, the error between the image dot location and the actual dot location can be used to build a distortion map of the camera and lens system so that images can be fixed to display truth locations. To build the dot grid images I used the Electro Optics Lab optical bench Bright Object Simulator System, and gimbal. The gimbal was slewed to a series of elevations and azimuths. An image of the collimated single point light source was then taken at each position. After a series of 99 images were taken at different locations the single light spots were extracted from each image and added to a composite image containing all 99 points. During the development of these grids it was noticed that an intermittent error in the artificial "star" locations occurred. Prior to the summer this error was attributed to the gimbal having glitches in it's pointing direction and was going to be replaced, however after further examining the issue I determined it to be a software issue. I have since narrowed the likely source of the error down to a Software Development Kit released by the camera supplier PixeLink. I have since developed a workaround in order to build star grids for calibration until the software bug can be isolated and fixed. I was also tasked with building a Hardware in the Loop test stand in order to test the full Op-Nav system. A 4k screen displays simulated Lunar and Terrestrial images from a possible Orion trajectory. These images are then projected through a collimator and then captured with an Op-Nav camera controlled by an Intel NUC computer running flight software. The flight software then analyzes the images to determine attitude and position, this data is then reconstructed into a trajectory and matched to the simulated trajectory in order to determine the accuracy of the attitude and position estimates. In order for the system to work it needs to be precisely and accurately aligned. I developed an alignment procedure that allows the screen, collimator and camera to be squared, centered and collinear with each other within a micron spatially and 5 arcseconds in rotation. I also designed a rigid mount for the screen that was machined on site in Building 10 by another intern. While I was working in the EOL we received a $500k Orion startracker for alignment procedure testing. Due to my prior experience in electronics development, as an ancillary duty, I was tasked with building the cables required to operate and power the startracker. If any errors are made building these cables the startracker would be destroyed, I was honored that the director of the lab entrusted such a critical component with me. This internship has cemented my view on public space exploration. I always preferred public sector to privatization because, as a scientist, the most interesting aspects of space for me are not necessarily the most profitable. I was concerned that the public sector was faltering however, and that in order to improve human space exploration I would be forced into private sector. I now know that, at least at JSC, human spaceflight is still progressing, and exciting work is still being done. I am now actively seeking employment at JSC after I complete my Ph.D and have met with my branch chiefs and mentor to discuss transitioning to a grad Co-op position.

Leps, Thomas↗
NASA NTRS1998

Comparison of MPEG digital video with super VHS tape for diagnostic echocardiographic readings

BACKGROUND: Digital recording of echocardiographic studies is on the clinical horizon. However, full digital capture of complete echocardiographic studies in traditional video format is impractical, given current storage capacity and network bandwidth. To overcome these constraints, we evaluated the diagnostic image quality of digital video by using MPEG (Motion Picture Experts Group) compression. METHODS AND RESULTS: Fifty-eight complete, consecutive studies were recorded simultaneously with the use of MPEG-1 and sVHS videotape. Each matched MPEG and sVHS study pair was reviewed by two from a total of six readers, and findings were recorded with the use of a detailed, computerized reporting tool. Intrareader and interreader discrepancies were characterized as major or minor and analyzed in total and for specific subgroups of findings (left and right ventricular parameters, valvular insufficiency, and left ventricular regional wall motion). Intrareader discrepancies were reviewed by a consensus panel for agreement with either MPEG or sVHS findings. There was an exact concordance between MPEG and sVHS readings in 83% of findings. The majority of discrepancies were minor, with major discrepancies in only 2.7% of findings. There was no difference in the rate of consensus panel agreement with MPEG or sVHS for instances of intrareader discrepancy, either in total or for any subgroup of findings. Interreader discrepancy rates were nearly identical for both MPEG and sVHS. CONCLUSIONS: MPEG-1 digital video is equivalent to sVHS videotape for diagnostic echocardiography. MPEG increases the range of practical options for digital echocardiography and offers, for the first time, the advantages of digital recording in a familiar video format.

Non-NASA Center↗
NASA NTRS1993

Data management in NOAA

The NOAA archives contain 150 terabytes of data in digital form, most of which are the high volume GOES satellite image data. There are 630 data bases containing 2,350 environmental variables. There are 375 million film records and 90 million paper records in addition to the digital data base. The current data accession rate is 10 percent per year and the number of users are increasing at a 10 percent annual rate. NOAA publishes 5,000 publications and distributes over one million copies to almost 41,000 paying customers. Each year, over six million records are key entered from manuscript documents and about 13,000 computer tapes and 40,000 satellite hardcopy images are entered into the archive. Early digital data were stored on punched cards and open reel computer tapes. In the late seventies, an advanced helical scan technology (AMPEX TBM) was implemented. Now, punched cards have disappeared, the TBM system was abandoned, most data stored on open reel tapes have been migrated to 3480 cartridges, many specialized data sets were distributed on CD ROM's, special archives are being copied to 12 inch optical WORM disks, 5 1/4 inch magneto-optical disks were employed for workstation applications, and 8 mm EXABYTE tapes are planned for major data collection programs. The rapid expansion of new data sets, some of which constitute large volumes of data, coupled with the need for vastly improved access mechanisms, portability, and improved longevity are factors which will influence NOAA's future systems approaches for data management.

Callicott, William M.↗
DOE OSTI2025

AAPM Truth‐based CT (TrueCT) reconstruction grand challenge

Background: This Special Report summarizes the 2022, AAPM grand challenge on Truth-based CT image reconstruction. Purpose: To provide an objective framework for evaluating CT reconstruction methods using virtual imaging resources consisting of a library of simulated CT projection images of a population of human models with various diseases. Methods: Two hundred unique anthropomorphic, computational models were created with varied diseases consisting of 67 emphysema, 67 lung lesions, and 66 liver lesions. The organs were modeled based on clinical CT images of real patients. The emphysematous regions were modeled using segmentations from patient CT cases in the COPDGene Phase I dataset. For the lung and liver lesion cases, 1–6 malignant lesions were created and inserted into the human models, with lesion diameters ranging from 5.6 to 21.9 mm for lung lesions and 3.9 to 14.9 mm for liver lesions. The contrast defined between the liver lesions and liver parenchyma was 82 ± 12 HU, ranging from 50 to 110 HU. Similarly, the contrast between the lung lesions and the lung parenchyma was defined as 781 ± 11 HU, ranging from 725 to 805 HU. For the emphysematous regions, the defined HU values were −950 ± 17 HU ranging from −918 to −979 HU. The developed human models were imaged with a validated CT simulator. The resulting CT sinograms were shared with the participants. The participants reconstructed CT images from the sinograms and sent back their reconstructed images. Further, the reconstructed images were then scored by comparing the results against the corresponding ground truth values. The scores included both task-generic (root mean square error [RMSE] and structural similarity matrix [SSIM]), and task-specific (detectability index [d’] and lesion volume accuracy) metrics. For the cases with multiple lesions, the measured metric was averaged across all the lesions. To combine the metrics with each other, each metric was normalized to a range of 0 to 1 per disease type, with “0” and “1” being the worst and best measured values across all cases of the disease type for all received reconstructions. Results: The True-CT challenge attracted 52 participants, out of which 5 successfully completed the challenge and submitted the requested 200 reconstructions. Across all participants and disease types, SSIM absolute values ranged from 0.22 to 0.90, RMSE from 77.6 to 490.5 HU, d’ from 0.1 to 64.6, and volume accuracy ranged from 1.2 to 753.1 mm3. The overall scores demonstrated that participant “A” had the best performance in all categories, except for the metrics of d’ for lung lesions and RMSE for liver lesions. Participant “A” had an average normalized score of 0.41 ± 0.22, 0.48 ± 0.32, and 0.42 ± 0.33 for the emphysema, lung lesion, and liver lesion cases, respectively. Conclusions: The True-CT challenge successfully enabled objective assessment of CT reconstructions with the unique advantage of access to a diverse population of diseased human models with known ground truth. This study highlights the significant potential of virtual imaging trials in objective assessment of medical imaging technologies.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2023

A combination interferometric and morphological image processing approach to rapid quality assessment of additively manufactured cellular truss core components

Advanced manufacturing (AM) processes such as laser powder bed fusion (LPBF) are increasingly capable of fabricating components with useful and unprecedented mechanical properties by incorporating complex internal bracing structures. From the standpoint of quality control and assessment, however, internally complex assemblies present significant build-verification challenges. Here we propose a hybrid approach to the inspection involving the application of computer-aided speckle interferometry (CASI) and morphological image processing as a rapid, inexpensive, and facile method for AM quality control. The described methodology has low capital equipment costs, is full-field and non-contact, can be used in an industrial setting, and has very low requirements in terms of operator training and expertise. Consisting primarily of the combination of image processing software with a simple optical system of variable sensitivity, the method is shown to be effective for inspection of a titanium honeycomb component subjected to differential pressure. Results are compared to those achieved with computed tomography (CT), immersion ultrasound testing (UT), and optical holographic interferometry. Here we propose several possible processing strategies for automated quality assessment based on this powerful hybrid approach.

36 MATERIALS SCIENCE↗
DOE OSTI2025

ExtremeMETA: High-speed Lightweight Image Segmentation Model by Remodeling Multi-channel Metamaterial Imagers

Deep neural networks (DNNs) have heavily relied on traditional computational units, such as CPUs and GPUs. However, this conventional approach brings significant computational burden, latency issues, and high power consumption, limiting their effectiveness. This has sparked the need for lightweight networks such as ExtremeC3Net. Meanwhile, there have been notable advancements in optical computational units, particularly with metamaterials, offering the exciting prospect of energy-efficient neural networks operating at the speed of light. Yet, the digital design of metamaterial neural networks (MNNs) faces precision, noise, and bandwidth challenges, limiting their application to intuitive tasks and low-resolution images. In this study, we proposed a large kernel lightweight segmentation model, ExtremeMETA. Based on ExtremeC3Net, our proposed model, ExtremeMETA maximized the ability of the first convolution layer by exploring a larger convolution kernel and multiple processing paths. With the large kernel convolution model, we extended the optic neural network application boundary to the segmentation task. To further lighten the computation burden of the digital processing part, a set of model compression methods was applied to improve model efficiency in the inference stage. The experimental results on three publicly available datasets demonstrated that the optimized efficient design improved segmentation performance from 92.45 to 95.97 on mIoU while reducing computational FLOPs from 461.07 MMacs to 166.03 MMacs. The large kernel lightweight model ExtremeMETA showcased the hybrid design’s ability on complex tasks.

large convolution kernel↗
DOE OSTI2023

NeRVI: Compressive neural representation of visualization images for communicating volume visualization results

We present NeRVI, a new deep-learning approach that compresses a large collection of visualization images generated from time-varying data for communicating volume visualization results. Based on an image-based implicit neural representation, our approach represents tens of thousands of high-resolution rendering images parametrized by different parameters via a hybrid model of multilayer perceptrons and convolutional neural networks. Here, our model predicts images and corresponding masks, and the masks are utilized for loss computation and network training to capture fine structural details and small components. In conjunction with model quantization and weight encoding, NeRVI yields highly compact compressive neural representations while preserving the image fidelity well. We demonstrate the effectiveness of NeRVI with isosurface rendering and direct volume rendering images generated from multiple data sets and compare NeRVI with other state-of-the-art deep learning-based (InSituNet, SIREN, NeRF, and NeRV) methods. Quantitative and qualitative results show that NeRVI provides an alternative solution that augments domain scientists' ability to manage, represent, and communicate scientific visualization output.

97 MATHEMATICS AND COMPUTING↗