Parallel implementation of vertex component analysis for hyperspectral endmember extraction

International Conference with Peer Review 2012 IEEE International Conference in Geoscience and Remote Sensing Symposium (IGARSS), 22-27 July 2012, Munich, Germany

Vertex component analysis GPU-based implementation for hyperspectral unmixing

Endmember extraction (EE) is a fundamental and crucial task in hyperspectral unmixing. Among other methods vertex component analysis ( VCA) has become a very popular and useful tool to unmix hyperspectral data. VCA is a geometrical based method that extracts endmember signatures from large hyperspectral datasets without the use of any a priori knowledge about the constituent spectra. Many Hyperspectral imagery applications require a response in real time or near-real time. Thus, to met this requirement this paper proposes a parallel implementation of VCA developed for graphics processing units. The impact on the complexity and on the accuracy of the proposed parallel implementation of VCA is examined using both simulated and real hyperspectral datasets.

Parallel hyperspectral unmixing method on GPU

Many Hyperspectral imagery applications require a response in real time or near-real time. To meet this requirement this paper proposes a parallel unmixing method developed for graphics processing units (GPU). This method is based on the vertex component analysis (VCA), which is a geometrical based method highly parallelizable. VCA is a very fast and accurate method that extracts endmember signatures from large hyperspectral datasets without the use of any a priori knowledge about the constituent spectra. Experimental results obtained for simulated and real hyperspectral datasets reveal considerable acceleration factors, up to 24 times.

Large area image sensing structures based on a-SiC : H: a dynamic characterization

In recent works large area hydrogenated amorphous silicon p-i-n structures with low conductivity doped layers were proposed as single element image sensors. The working principle of this type of sensor is based on the modulation, by the local illumination conditions, of the photocurrent generated by a light beam scanning the active area of the device. In order to evaluate the sensor capabilities is necessary to perform a response time characterization. This work focuses on the transient response of such sensor and on the influence of the carbon contents of the doped layers. In order to evaluate the response time a set of devices with different percentage of carbon incorporation in the doped layers is analyzed by measuring the scanner-induced photocurrent under different bias conditions.

Novel structure for large area image sensing

This work presents preliminary results in the study of a novel structure for a laser scanned photodiode (LSP) type of image sensor. In order to increase the signal output, a stacked p-i-n-p-i-n structure with an intermediate light-blocking layer is used. The image and the scanning beam are incident through opposite sides of the sensor and their absorption is kept in separate junctions by an intermediate light-blocking layer. As in the usual LSP structure the scanning beam-induced photocurrent is dependent on the local illumination conditions of the image. The main difference between the two structures arises from the fact that in this new structure the image and the scanner have different optical paths leading to an increase in the photocurrent when the scanning beam is incident on a region illuminated on the image side of the sensor, while a decreasing in the photocurrent was observed in the single junction LSP. The results show that the structure can be successfully used as an image sensor even though some optimization is needed to enhance the performance of the device.

Determining the number of clusters in categorical data

Cluster analysis for categorical data has been an active area of research. A well-known problem in this area is the determination of the number of clusters, which is unknown and must be inferred from the data. In order to estimate the number of clusters, one often resorts to information criteria, such as BIC (Bayesian information criterion), MML (minimum message length, proposed by Wallace and Boulton, 1968), and ICL (integrated classification likelihood). In this work, we adopt the approach developed by Figueiredo and Jain (2002) for clustering continuous data. They use an MML criterion to select the number of clusters and a variant of the EM algorithm to estimate the model parameters. This EM variant seamlessly integrates model estimation and selection in a single algorithm. For clustering categorical data, we assume a finite mixture of multinomial distributions and implement a new EM algorithm, following a previous version (Silvestre et al., 2008). Results obtained with synthetic datasets are encouraging. The main advantage of the proposed approach, when compared to the above referred criteria, is the speed of execution, which is especially relevant when dealing with large data sets.

Implementação em hardware reconfigurável de método de separação de dados hiperespetrais

Relatório do Trabalho Final de Mestrado para obtenção do grau de Mestre em Engenharia de Electrónica e Telecomunicações

An efficient scalable RNS architecture for large dynamic ranges

This paper proposes an efficient scalable Residue Number System (RNS) architecture supporting moduli sets with an arbitrary number of channels, allowing to achieve larger dynamic range and a higher level of parallelism. The proposed architecture allows the forward and reverse RNS conversion, by reusing the arithmetic channel units. The arithmetic operations supported at the channel level include addition, subtraction, and multiplication with accumulation capability. For the reverse conversion two algorithms are considered, one based on the Chinese Remainder Theorem and the other one on Mixed-Radix-Conversion, leading to implementations optimized for delay and required circuit area. With the proposed architecture a complete and compact RNS platform is achieved. Experimental results suggest gains of 17 % in the delay in the arithmetic operations, with an area reduction of 23 % regarding the RNS state of the art. When compared with a binary system the proposed architecture allows to perform the same computation 20 times faster alongside with only 10 % of the circuit area resources.

Efficient feature selection filters for high-dimensional data

Feature selection is a central problem in machine learning and pattern recognition. On large datasets (in terms of dimension and/or number of instances), using search-based or wrapper techniques can be cornputationally prohibitive. Moreover, many filter methods based on relevance/redundancy assessment also take a prohibitively long time on high-dimensional. datasets. In this paper, we propose efficient unsupervised and supervised feature selection/ranking filters for high-dimensional datasets. These methods use low-complexity relevance and redundancy criteria, applicable to supervised, semi-supervised, and unsupervised learning, being able to act as pre-processors for computationally intensive methods to focus their attention on smaller subsets of promising features. The experimental results, with up to 10(5) features, show the time efficiency of our methods, with lower generalization error than state-of-the-art techniques, while being dramatically simpler and faster.

Large pseudoscalar Yukawa couplings in the complex 2HDM

We start by presenting the current status of a complex flavour conserving two-Higgs doublet model. We will focus on some very interesting scenarios where unexpectedly the light Higgs couplings to leptons and to b-quarks can have a large pseudoscalar component with a vanishing scalar component. Predictions for the allowed parameter space at end of the next run with a total collected luminosity of 300 fb(-1) and 3000 fb(-1) are also discussed. These scenarios are not excluded by present data and most probably will survive the next LHC run. However, a measurement of the mixing angle phi(tau), between the scalar and pseudoscalar component of the 125 GeV Higgs, in the decay h -> tau(+)tau(-) will be able to probe many of these scenarios, even with low luminosity. Similarly, a measurement of phi(t) in the vertex (t) over bar th could help to constrain the low tan beta region in the Type I model.

Large-area homogeneous periodic surface structures generated on the surface sputtered boron carbide thin films by femtosecond laser processing

Amorphous and crystalline sputtered boron carbide thin films have a very high hardness even surpassing that of bulk crystalline boron carbide (≈41 GPa). However, magnetron sputtered B-C films have high friction coefficients (C.o.F) which limit their industrial application. Nanopatterning of materials surfaces has been proposed as a solution to decrease the C.o.F. The contact area of the nanopatterned surfaces is decreased due to the nanometre size of the asperities which results in a significant reduction of adhesion and friction. In the present work, the surface of amorphous and polycrystalline B-C thin films deposited by magnetron sputtering was nanopatterned using infrared femtosecond laser radiation. Successive parallel laser tracks 10 μm apart were overlapped in order to obtain a processed area of about 3 mm2. Sinusoidal-like undulations with the same spatial period as the laser tracks were formed on the surface of the amorphous boron carbide films after laser processing. The undulations amplitude increases with increasing laser fluence. The formation of undulations with a 10 μm period was also observed on the surface of the crystalline boron carbide film processed with a pulse energy of 72 μJ. The amplitude of the undulations is about 10 times higher than in the amorphous films processed at the same pulse energy due to the higher roughness of the films and consequent increase in laser radiation absorption. LIPSS formation on the surface of the films was achieved for the three B-C films under study. However, LIPSS are formed under different circumstances. Processing of the amorphous films at low fluence (72 μJ) results in LIPSS formation only on localized spots on the film surface. LIPSS formation was also observed on the top of the undulations formed after laser processing with 78 μJ of the amorphous film deposited at 800 °C. Finally, large-area homogeneous LIPSS coverage of the boron carbide crystalline films surface was achieved within a large range of laser fluences although holes are also formed at higher laser fluences.

Rheo-optical characterization of liquid crystalline acetoxypropylcellulose melt undergoing large shear flow and relaxation after flow cessation

The rheological and structural characteristics of acetoxypropylcellulose (APC) nematic melt are studied at shear rates ranging from 10 s(-1) to 1000 s(-1) which are relevant to extrusion based processes. APC shows a monotonic shear thinning behavior over the range of shear rates tested. The negative extrudate-swell shows a minimum when a critical shear rate (gamma) over dot(c) is reached. For shear rates smaller than (gamma) over dot(c), the flow-induced texture consists of two set of bands aligned parallel and normal to the flow direction. At shear rates larger than (gamma) over dot(c), the flow induced texture is reminiscent of a 2 fluids structure. Close to the shearing walls, domains elongated along the flow direction and stacked along the vorticity are imaged with POM, whereas SALS patterns indicate that the bulk of the sheared APC is made of elliptical domains oriented along the vorticity. No full nematic alignment is achieved at the largest shear rate tested. Below (gamma) over dot(c), the stress relaxation is described by a stretched exponential. Above (gamma) over dot(c), the stress relaxation is described by a fast and a slow process. The latter coincides with the growth of normal bands thicknesses, as the APC texture after flow cessation consists of two types of bands with parallel and normal orientations relative to the flow direction. Both bands thicknesses do not depend on the applied shear rate, in contrast to their orientation. (C) 2015 Elsevier Ltd. All rights reserved.

Optimizing memory transactions for large-scale programs

Even though Software Transactional Memory (STM) is one of the most promising approaches to simplify concurrent programming, current STM implementations incur significant overheads that render them impractical for many real-sized programs. The key insight of this work is that we do not need to use the same costly barriers for all the memory managed by a real-sized application, if only a small fraction of the memory is under contention lightweight barriers may be used in this case. In this work, we propose a new solution based on an approach of adaptive object metadata (AOM) to promote the use of a fast path to access objects that are not under contention. We show that this approach is able to make the performance of an STM competitive with the best fine-grained lock-based approaches in some of the more challenging benchmarks. (C) 2015 Elsevier Inc. All rights reserved.

Parallel hyperspectral compressive sensing method on GPU

Remote hyperspectral sensors collect large amounts of data per flight usually with low spatial resolution. It is known that the bandwidth connection between the satellite/airborne platform and the ground station is reduced, thus a compression onboard method is desirable to reduce the amount of data to be transmitted. This paper presents a parallel implementation of an compressive sensing method, called parallel hyperspectral coded aperture (P-HYCA), for graphics processing units (GPU) using the compute unified device architecture (CUDA). This method takes into account two main properties of hyperspectral dataset, namely the high correlation existing among the spectral bands and the generally low number of endmembers needed to explain the data, which largely reduces the number of measurements necessary to correctly reconstruct the original data. Experimental results conducted using synthetic and real hyperspectral datasets on two different GPU architectures by NVIDIA: GeForce GTX 590 and GeForce GTX TITAN, reveal that the use of GPUs can provide real-time compressive sensing performance. The achieved speedup is up to 20 times when compared with the processing time of HYCA running on one core of the Intel i7-2600 CPU (3.4GHz), with 16 Gbyte memory.

Recurring contact opportunities within groups of devices

The capability to anticipate a contact with another device can greatly improve the performance and user satisfaction not only of mobile social network applications but of any other relying on some form of data harvesting or hoarding. One of the most promising approaches for contact prediction is to extrapolate from past experiences. This paper investigates the recurring contact patterns observed between groups of devices using an 8-year dataset of wireless access logs produced by more than 70000 devices. This effort permitted to model the probabilities of occurrence of a contact at a predefined date between groups of devices using a power law distribution that varies according to neighbourhood size and recurrence period. In the general case, the model can be used by applications that need to disseminate large datasets by groups of devices. As an example, the paper presents and evaluates an algorithm that provides daily contact predictions, based on the history of past pairwise contacts and their duration. Copyright © 2015 ICST.