Solving real world problems with embedded neural networks requires both training algorithms that achieve high performance and compatible hardware that runs in real time while remaining energy efficient. For the former...
详细信息
Solving real world problems with embedded neural networks requires both training algorithms that achieve high performance and compatible hardware that runs in real time while remaining energy efficient. For the former, deep learning using backpropagation has recently achieved a string of successes across many domains and datasets. For the latter, neuromorphic chips that run spiking neural networks have recently achieved unprecedented energy efficiency. To bring these two advances together, we must first resolve the incompatibility between backpropagation, which uses continuous-output neurons and synaptic weights, and neuromorphic designs, which employ spiking neurons and discrete synapses. Our approach is to treat spikes and discrete synapses as continuous probabilities, which allows training the network using standard backpropagation. The trained network naturally maps to neuromorphic hardware by sampling the probabilities to create one or more networks, which are merged using ensemble averaging. To demonstrate, we trained a sparsely connected network that runs on the TrueNorth chip using the MNIST dataset. With a high performance network (ensemble of 64), we achieve 99.42% accuracy at 108 mu J per image, and with a high efficiency network (ensemble of 1) we achieve 92.7% accuracy at 0.268 mu J per image.
Biometric systems have great importance in modern society. Today, we can find biometric systems using hardware technologies to provide an increasingly fast response. Partly this is possible due to the use of algorithm...
详细信息
In this paper a method for estimating the motion camcorder or platform used in imageprocessing tasks to determine the motion parameters is proposed. A method for determining and shaping optical flow weighted measurem...
详细信息
ISBN:
(纸本)9785919950233
In this paper a method for estimating the motion camcorder or platform used in imageprocessing tasks to determine the motion parameters is proposed. A method for determining and shaping optical flow weighted measurements for image blocks is proposed. The effectiveness of the proposed method is shown on the basis of numerical simulation data for test images. The methods and algorithms described in this paper are versatile enough and are implemented in various fields.
Diffractive optical elements (DOE) show great promise for imaging optics that are thinner and more lightweight than conventional refractive lenses while preserving their light efficiency. Unfortunately, severe spectra...
详细信息
Diffractive optical elements (DOE) show great promise for imaging optics that are thinner and more lightweight than conventional refractive lenses while preserving their light efficiency. Unfortunately, severe spectral dispersion currently limits the use of DOEs in consumer-level lens design. In this article, we jointly design lightweight diffractive-refractive optics and post-processingalgorithms to enable imaging under white light illumination. Using the Fresnel lens as a general platform, we show three phase-plate designs, including a super-thin stacked plate design, a diffractive-refractive-hybrid lens, and a phase coded-aperture lens. Combined with cross-channel deconvolution algorithm, both spherical and chromatic aberrations are corrected. Experimental results indicate that using our computational imaging approach, diffractive-refractive optics is an alternative candidate to build light efficient and thin optics for white light imaging. (C) 2015 Optical Society of America
MWSN is an emerging area of research and most of the work in the field of MWSN is done at the simulation level as there is hardly any cost effective hardware platform(node/mote) available for MWSN applications. To han...
详细信息
ISBN:
(纸本)9781479918232
MWSN is an emerging area of research and most of the work in the field of MWSN is done at the simulation level as there is hardly any cost effective hardware platform(node/mote) available for MWSN applications. To handle mobility, the MWSN node should be much more efficient than the nodes in static WSN. Moreover, a MWSN node should be capable of handling real time mobility control, path planning and navigation. The application domains of MWSN can be further expanded by incorporating swarm like intelligence in MWSN. We have developed a low cost, small form factor hardware platform which will function as a node in MWSN using custom off the shelf(COTS) products. Our mobile hardware platform, henceforth called as BSwarm robot supports self-assembly, to achieve complex tasks. The platform also support image assisted navigation and provides extensive I/O support for further feature expansion. The testbed consisting of multiple BSwarm robot can be utilized for the development and validation of algorithms/protocols related to MWSNs, distributed control of Swarm robots, real time imageprocessing etc. BSwarm robot is a multi processor based robot designed in such a way that it can be used for applications which may demand varied degree of processing, communication and input-output capabilities. This paper also highlights major factors that can be taken into consideration while choosing the hardware platform for MWSNs so that the protocol stack development for MWSNs becomes easier.
NnmfPack is a library for the nonnegative matrix factorization (NNMF) problem. Nowadays NNMF is an essential tool in many fields spanning machine learning, data analysis, image analysis or audio source separation, amo...
详细信息
NnmfPack is a library for the nonnegative matrix factorization (NNMF) problem. Nowadays NNMF is an essential tool in many fields spanning machine learning, data analysis, image analysis or audio source separation, among others. NnmfPack is an efficient numerical library conceived for shared memory heterogeneous parallel systems, and it supports, from its conception, both conventional multi-core processors and many-core coprocessors. In this article, NnmfPack is extended to handle different metrics options (-divergence), and some other parallel algorithms have been added and tested. The performance of the new functionalities of NnmfPack is tested, and some precision results of the implementations are showed using an example borrowed from the imageprocessing field.
The paper presents the developed algorithms for the reconstruction of multiple crystal lattice sites. The study was conducted with a set of crystal lattices and the developed method of modeling a three-dimensional str...
详细信息
The paper presents the developed algorithms for the reconstruction of multiple crystal lattice sites. The study was conducted with a set of crystal lattices and the developed method of modeling a three-dimensional structure of ideal crystal lattice sites. The results of the reconstruction of three-dimensional structures of lattice sites are shown using different metrics comparison. Comparative characteristics of accuracy of the algorithms are given in the paper.
Annual laminae (varves) in lake sediments are typically visually identified, measured and counted, although numerous attempts have been made to automate this process. The reason for the failure of most of these automa...
详细信息
Annual laminae (varves) in lake sediments are typically visually identified, measured and counted, although numerous attempts have been made to automate this process. The reason for the failure of most of these automated algorithms for varve counting is the complexity of the seasonal laminations, typically rich in lateral fades variations and internal heterogeneities. In the manual counting of varves, the investigator acquired and interpreted flexible numbers of complex decision criteria to understand whether a particular simple lamination is a varve or not. Fuzzy systems simulate the flexible decision making process in a computer by introducing a smooth transition between true varve and false varve. In our investigation, we use an adaptive neuro fuzzy inference system (ANFIS) to detect varves on the basis of a digital image of the sediment. The results of the application of the ANFIS to laminated sediments from the Meerfelder Maar (Eifel, Germany) and from a landslide-dammed lake in the Quebrada de Cafayate of Argentina are compared with manual varve counts and possible reasons for the differences are discussed. (C) 2015 Elsevier B.v. All rights reserved.
In Magnetic Resonance (MR) images, noise is a common issue which limits the image accuracy of any quantitative measurements. Noise elimination in MRI image pre-processing is an important step to eliminate the noise an...
详细信息
In Magnetic Resonance (MR) images, noise is a common issue which limits the image accuracy of any quantitative measurements. Noise elimination in MRI image pre-processing is an important step to eliminate the noise and to make the image fit for further steps involved in the process of analyzing. However, different types of noises produces ranges of significant impact on image quality, and thus tend to affect human interpretation and performance of computer-aided diagnosis systems. Another issue is about filtering strategies to eliminate noise and preserve high quality image depending on filter reconstruction ability and noise model. In this work three different filtering algorithms such as Median filter (MF), Adaptive filter (ADF) and Average filter (AvF) are used to remove the additive noises present in the MRI images i.e. Gaussian, Salt and pepper and speckle noise. The noise density was gradually added to MRI image up to 90% to compare performance of the filters by qualitative and quantitative evaluation. The performance of these filters are compared using the statistical parameters such as Mean Squared Error (MSE) and Peak Signal-to-Noise Ratio (PSNR). The study shows that Median filter reconstructs a high quality image than other filters in Gaussian and Salt and pepper denoising with 38.3 dB PSNR at 10% noise variance. While for speckle noise removal, Average filter is perform better than others which result of 56.2 dB PSNR at 10% noise variance. A comparison with other well-established methods, this study shows that the Median and Average filter produces better denoising results, preserving the main structures and details. (C) 2015 The Authors. Published by Elsevier B.v.
OBJECTIvES The purpose of this study was to assess the potential of iterative image reconstruction (IR) of images for radiation dose reduction in coronary computed tomography angiography (CIA). Therefore, IR in combin...
详细信息
OBJECTIvES The purpose of this study was to assess the potential of iterative image reconstruction (IR) of images for radiation dose reduction in coronary computed tomography angiography (CIA). Therefore, IR in combination with 30% tube current reduction was compared with standard scanning with filtered back projection (FBP) reconstruction. BACKGROUND Lately, new IR techniques with advanced raw data processing have been introduced by different computed tomography vendors, thus allowing for either image noise reduction at unchanged radiation dose levels or radiation dose reductions at comparable image noise levels. METHODS In this prospective, multicenter, multivendor noninferiority trial, we randomized 400 consecutive patients to 1 of 2 groups: a control group using standard FBP image reconstruction and standard tube current or an interventional group using IR technique and 30% tube current reduction. The primary endpoint was to demonstrate noninferiority in image quality (IQ) in the IR group. IQ was assessed on a 4-point scale (1, nondiagnostic IQ;4, excellent IQ). Secondary endpoints included total radiation dose estimates and the rate of downstream testing during 30-day follow-up. RESULTS Median IQ in the IR group was noninferior compared with the conventional FBP group (IR, 3.5 [interquartile range: 3.0 to 4.0];FBP, 3.4 [interquartile range: 2.8 to 4.0], p for noninferiority <0.016). The radiation exposure was significantly lower in the IR group (median dose-length-product 157 [interquartile range: 114 to 239] *** vs. 222 [interquartile range: 141 to 319] *** for IR vs. FBP, respectively, p < 0.0001). The rate of downstream testing did not differ significantly (7.7% vs. 7.9% for IR vs. FBP, respectively, p = 0.94). CONCLUSIONS Coronary CTA image quality is maintained with the combined use of a 30% reduced tube current and IR algorithms when compared with conventional FBP image reconstruction techniques and standard tube current. (C) 2015 by the American Co
暂无评论