50175 research outputs found
Sort by
High-resolution intertidal topography from sentinel-2 multi-spectral imagery: Synergy between remote sensing and numerical modeling
No full textThe intertidal zones are well recognized for their dynamic nature and role in near-shore hydrodynamics. The intertidal topography is poorly mapped worldwide due to the high cost of associated field campaigns. Here we present a combination of remote-sensing and hydrodynamic modeling to overcome the lack of in situ measurements. We derive a digital elevation model (DEM) by linking the corresponding water level to a sample of shorelines at various stages of the tide. Our shoreline detection method is fully automatic and capable of processing high-resolution imagery from state-of-the-art satellite missions, e.g., Sentinel-2. We demonstrate the use of a tidal model to infer the corresponding water level in each shoreline pixel at the sampled timestamp. As a test case, this methodology is applied to the vast coastal region of the Bengal delta and an intertidal DEM at 10mresolution covering an area of 1134 km2 is developed from Sentinel-2 imagery. We assessed the quality of the DEM with two independent in situ datasets and conclude that the accuracy of our DEM amounts to about 1.5 m, which is commensurate with the typical error bar of the validation datasets. This DEM can be useful for high-resolution hydrodynamic and wave modeling of the near-shore area. Additionally, being automatic and numerically effective, our methodology is compliant with near-real-time monitoring constraints
Effects of Modifying the Input Features and the Loss Function on Improving Emotion Classification
No full textIn this work, we show that the discriminative power of a deep neural network can be improved at three different levels: (i) inputting discriminative features, (ii) designing an optimal network architecture and (iii) changing the loss function. This work suggests that only increasing the depth or the width of a deep network may not always be the best solution while requiring computationally efficient models. Here, we show that there is scope for improving the classifier accuracy at each of the three levels. We have carried out all our experiments on the FERplus dataset and show that the facial emotion recognition accuracy can be independently improved up to 2.5 by adding better features, 1.8 by modifying the loss function and up to 3.1 by combining the two ideas. In separate experiments, we show that the computational complexity can be reduced by a factor of 24.3, while simultaneously increasing the FER by 0.95 by modifying the architecture of the model, input features and the loss function. © 2019 IEEE
Scattering transform inspired filterbank learning from raw speech for better acoustic modeling
No full textWe propose a neural network architecture, which operates on the raw speech signal, where the first layer contains a series of 1D time-domain filters. The output of this layer is fed to the second layer, which is a bank of 2D-convolution filters that capture the spectro-temporal modulations in the speech signal. The outputs of these two layers are concatenated, normalized and then fed to a feed-forward neural network to predict the senone posteriors, which are used for ASR decoding. During the training of the neural network, we have employed different strategies, where the 1D and 2D filters are initialized with (a) Gabor filters and (b) random values and the filter coefficients are either (a) allowed to be updated along with the other affine transform parameters of the network or (b) fixed during training. ASR experiments are conducted on 160 hours of Tamil speech data and the proposed architecture gives an absolute improvement in word error rate (WER) of 1.35 and 1.21 with respect to the neural network models trained on mel-frequency cepstral coefficients and log-filterbank energy features, respectively. We have also compared the performances of various strategies for filter initialization and training and reported the WERs. © 2019 IEEE
Multi-loudspeaker Rendering of Musical Ensemble: Role of Timbre in Source Width Perception
No full textA study of rendering natural instrument sounds and vocals as an ensemble is presented. Using a single channel musical recording, a multi-channel signal is created by introducing delays and rendered in spatially distributed loudspeaker array. Three listeners participate in a listening experiment. We observed that the perceived width of the ensemble varies with the signal timbre. We find that spectro-temporal properties of the musical instruments play a significant role in the perception of source width. On an average, sustained sounds (like vocal) create wider percept than transients like percussion. Also, low pitched sounds (like male bass voice, saxophone, tuba) create wider percept than high pitched signals
Deep Reinforcement Learning Based Power Control for Wireless Multicast Systems
No full textWe consider a multicast scheme recently proposed for a wireless downlink 1. It was shown earlier that power control can significantly improve its performance. However for this system, obtaining optimal power control is intractable because of a very large state space. Therefore in this paper we use deep reinforcement learning where we use function approximation of the Q-function via a deep neural network. We show that optimal power control can be learnt for reasonably large systems via this approach. The average power constraint is ensured via a Lagrange multiplier, which is also learnt. In the longer version of the paper 2, we also demonstrate that our learning algorithm can be modified to allow the optimal control to track the time varying system statistics
Nanostructured Zn-Substituted Nickel Ferrite Thin Films: CMOS-Compatible Deposition and Excellent Soft Magnetic Properties
No full textNanostructured \textNi-x\textZn-1-xFe2O4 (x = 1, 0.5) films, about 1.5 μm thick on Si (100) substrates, were deposited using a low-Temperature (<150 C) microwave-Assisted solvothermal (MAS) technique that is compatible with back-end-of-The-line Si-CMOS processing. A nanocrystalline single-phase spinel structure with crystallite sizes of �4 nm for the nickel ferrite film (NF) and �6 nm for the zinc-substituted NF (ZNF) was obtained. The films demonstrate excellent surface smoothness and strong adherence to the substrate. Deconvolution of the A-1g vibration mode in Raman spectra of both films reveals a ''far-from-equilibrium'' crystallographic inversion induced by the MAS process. Its effect on the magnetic characteristics of the films is analyzed here. Both films exhibit in-plane (xy plane) isotropy with very low room-Temperature coercivities, 25 Oe for NF and 35 Oe for ZNF, which is essential for high-frequency, soft magnetic applications. The presence of interparticular dipolar interaction in both films is confirmed from temperature-dependent magnetization measurements made under different dc bias fields. The CMOS-compatible ferrite processing and superparamagnetic Ni-ferrite and NiZn-ferrite thin films presented here can meet upcoming technological needs in on-chip integrated passive devices
Phylogenetic diversity as a measure of biodiversity: Pros and Cons
No full textSpecies richness is predominantly used as one of the fundamental measures of biodiversity for prioritization of areas for conservation. However, species richness often underestimates true diversity, as it does not take into account the evolutionary histories of the species in an area. In this regard, phylogenetic diversity, which incorporates information regarding species relationships in calculating diversity, has been proposed as an alternative. Here we compare species richness and phylogenetic diversity of mammals in nine sanctuaries to explore the importance and use of evolutionary relationships in characterizing diversity. Our analyses suggest that even though species richness and phylogenetic diversity are correlated, they are often decoupled. Importantly, areas with low species richness might harbour high phylogenetic diversity and vice-versa. We recommend the use of both the diversity measures for a holistic understanding of biodiversity and for prioritization of areas for conservation
The orange red luminescence and conductivity response of Eu3+ doped GdOF phosphor: Synthesis, characterization and their Judd-Ofelt analysis
No full textPresent paper reports the synthesis of a series of Gd1-xEuxOF (0.01� x � 0.09) phosphor by solid state route at low temperature. Characterizations such as powder x-ray diffraction (PXRD), Rietveld refinement, scanning electron microscopy, UV-Visible absorption, AC conductivity and photoluminescence (PL) spectroscopy were carried out. Rietveld refinement of powder XRD data confirms the crystallization of the compounds in rhombohedral structure. Investigation of PL under 395 and 465 nm excitation wavelengths is done. Upon substitution Gd3+ by Eu3+ ions in GdOF, the PL spectra exhibited emission at 5D0 � 7F0 (579 nm), 5D0 � 7F1 (593 nm), 5D0 � 7F2 (612, 620 nm), 5D0 � 7F3 (651 nm) and 5D0 � 7F4 (705 nm) transitions. JO theory parameterization variation from emission spectra is used to compute the intensity parameters (Ω2, Ω4) of phosphors. Using the intensity parameters the radiative lifetime (�rad), radiative transition probabilities (AT), branching ratio (β) are evaluated. Dielectric and AC conductivity of Gd1�xEuxOF (0.01 � x � 0.09) is studied in the frequency range from 100 Hz to 5 MHz at a temperature of 300 K. Studies reveal that the AC conductivity increases gradually beyond 30 kHz up to100 kHz and rapidly increases at 300 kHz. At lower frequencies, it is found that the compound exhibit high values of dielectric loss and dielectric constant, which decreased drastically with increasing frequency. The result so obtained reveals that present phosphors could be a possible source for orange red constituent in white LEDs and is best suited for microwave device applications because of its low values of dielectric loss
Micro level analyses of environmentally disastrous urbanization in Bangalore
No full textIndian metropolitan (tier I) cities have been undergoing rapid urbanization during the post-globalization era with the unprecedented market interventions, which have led to the rapid land cover changes affecting the ecology, climate, hydrology, and local environment. The unplanned urbanization has given way to the dispersed, haphazard growth at the city outskirts with the lack of basic amenities and infrastructure as the planners lack advance information of sprawl regions. This has necessitated understanding and visualization of urbanization patterns for planning towards sustainable cities. The analyses of urban dynamics during 1973�2017 using temporal remote sensing data reveal 1028 increase in urban area with the decline of 88 vegetation and 79 of water bodies. Consequences of the unplanned urbanization are the increase in greenhouse gas emissions, decline in vegetation cover, loss of groundwater table (from 28 to 300 m), contamination of water sources, increase in land surface temperature, increase in disease vectors, etc. An attempt is made to understand the implications of unplanned growth at the micro level by considering the prime growth poles such as Peenya Industrial Estate (PIE), Whitefield (WF), Bangalore South Region (BSR). The spatial analyses reveal the decline of vegetation and open spaces with intense urbanization of 86.35 (in BSR), 87.39 (PIE) and 81.61 (WF) in 2017. WF witnessed the drastic transformation from agrarian ecosystem to a concrete jungle during the past four decades. Spatial patterns of urbanization were assessed through the landscape metrics and rule-based modeling which confirms intense urbanization with single class dominance. Specifically, NP metrics depicts PIE region had sprawl growth till 2003 with numerous patches and is transformed by 2017 it has become to a single dense urban patch. This necessitates appropriate planning strategies to mitigate further erosion of environmental resources and ensure clean air, water, and environment to all residents. © 2019, Springer Nature Switzerland AG
Performance evaluation of convolutional neural network at hyperspectral and multispectral resolution for classification
No full textConvolutional Neural Network (CNN) has established as an effective deep learning model for hyperspectral image classification by considering both spectral and spatial information. In this study, the performance of two-dimensional (2D) CNN architecture is evaluated at hyperspectral and multispectral resolution. Two types of multispectral data are analyzed viz., original and transformed multispectral data. Hyperspectral bands are transformed to spectral resolution of multispectral bands by averaging the reflectances of specific hyperspectral narrow bands which are falling within the spectral ranges of multispectral bands. The well-known Pavia University dataset and a new dataset of Pear orchard are investigated in this study. In case of Pear orchard dataset, classification is performed with both types of multispectral data. All the experiments are carried out with the same 2D CNN architecture. In case of Pavia University dataset, hyperspectral and transformed multispectral data achieve OA() of 94.29±1.28 and 94.27±2.01 respectively considering 20 samples as training. In case of Pear orchard dataset, hyperspectral, multispectral and transformed multispectral data achieve OA() of 91.59±0.89, 88.65±1.35, and 93.24±0.16 respectively considering 20 samples as training. It is evident that transformed multispectral data, which comprises of inherent hyperspectral information, provides similar or better performance compared to hyperspectral data. Further, with the use of 3D CNN architecture, classification performance improves in case of Pavia University dataset, whereas it remains statistically similar in case of Pear orchard dataset. The present promising results illustrates the performance of CNN even in small dataset which is comparable to several published state-of-the art results on the same dataset