Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userWide Field Imaging. I. Applications of Neural Networks to object detection and star/galaxy classification
152
0
0.0
(
0
)
Authors
S. Andreon
Ask ChatGPT about the research
No Arabic abstract
[Abriged] Astronomical Wide Field Imaging performed with new large format CCD detectors poses data reduction problems of unprecedented scale which are difficult to deal with traditional interactive tools. We present here NExt (Neural Extractor): a new Neural Network (NN) based package capable to detect objects and to perform both deblending and star/galaxy classification in an automatic way. Traditionally, in astronomical images, objects are first discriminated from the noisy background by searching for sets of connected pixels having brightnesses above a given threshold and then they are classified as stars or as galaxies through diagnostic diagrams having variables choosen accordingly to the astronomers taste and experience. In the extraction step, assuming that images are well sampled, NExt requires only the simplest a priori definition of what an object is (id est, it keeps all structures composed by more than one pixels) and performs the detection via an unsupervised NN approaching detection as a clustering problem which has been thoroughly studied in the artificial intelligence literature. In order to obtain an objective and reliable classification, instead of using an arbitrarily defined set of features, we use a NN to select the most significant features among the large number of measured ones, and then we use their selected features to perform the classification task. In order to optimise the performances of the system we implemented and tested several different models of NN. The comparison of the NExt performances with those of the best detection and classification package known to the authors (SExtractor) shows that NExt is at least as effective as the best traditional packages.
rate research
Read More
This is the first paper of a series that will present data and scientific results from the WINGS project, a wide-field, multiwavelength imaging and spectroscopic survey of galaxies in 77 nearby clusters. The sample was extracted from the ROSAT catalogs with constraints on the redshift (0.04<z<0.07) and distance from the galactic plane (|b|>20). The global goal of the WINGS project is the systematic study of the local cosmic variance of the cluster population and of the properties of cluster galaxies as a function of cluster properties and local environment. This data collection will allow to define a local Zero-Point reference against which to gauge the cosmic evolution when compared to more distant clusters. The core of the project consists of wide-field optical imaging of the selected clusters in the B and V bands. We have also completed a multi-fiber, medium resolution spectroscopic survey for 51 of the clusters in the master sample. In addition, a NIR (JK) survey of ~50 clusters and an H_alpha + UV survey of some 10 clusters are presently ongoing, while a very-wide-field optical survey has also been programmed. In this paper we briefly outline the global objectives and the main characteristics of the WINGS project. Moreover, the observing strategy and the data reduction of the optical imaging survey (WINGS-OPT) are presented. We have achieved a photometric accuracy of ~0.025mag, reaching completeness to V~23.5. Field size and resolution (FWHM) span the absolute intervals (1.6-2.7)Mpc and (0.7-1.7)kpc, respectively, depending on the redshift and on the seeing. This allows the planned studies to get a valuable description of the local properties of clusters and galaxies in clusters.
The ongoing effort to implement compact and cheap optical systems is the main driving force for the recent flourishing research in the field of optical metalenses. Metalenses are a type of metasurface, used for focusing and imaging applications, and are implemented based on the nanopatterning of an optical surface. The challenge faced by metalens research is to reach high levels of performance, using simple fabrication methods suitable for mass-production. In this paper we present a Huygens nanoantenna based metalens, designed for outdoor photographic/surveillance applications in the near-infra-red. We show that good imaging quality can be obtained over a field-of-view (FOV) as large as +/-15 degrees. This first successful implementation of metalenses for outdoor imaging applications is expected to provide insight and inspiration for future metalens imaging applications.
As part of a global analysis of deep star counts to constrain scenarii of galaxy formation and evolution, we investigate possible links between the galactic spheroid and the dark matter halo. A wide set of deep star counts at high and intermediate galactic latitudes is used to determine the large scale density law of the spheroid. Assuming a power density law, the exponent, flattening, local density and IMF slope of this population are estimated. The estimation is checked for robustness against contamination of star counts by the thick disc population. Contamination effects are derived from a model of population synthesis under a broad variety of thick disc parameters. The parameter fit is based on a maximum likelihood criterion. The best fit spheroid density law has a flattening of 0.76, a power index of 2.44. There is a significant degeneracy between these two parameters. The data are also compatible with a slightly less flattened spheroid (c/a = 0.85), in combination with a larger power index (2.75). A flatter spheroid (c/a = 0.6) with a power index of 2 is not excluded either. We also constrain the spheroid IMF slope alpha to be 1.9 +/- 0.2, leading to a local density of 1.64 10$^{-4}$ stars pc$^{-3}$ and a mass density of 4.15 10$^{-5}$ Msun pc$^{-3}$. With this slope the expected mass density of brown dwarfs in the halo makes a negligible part of the dark matter halo, as already estimated from microlensing surveys. So, as star count data progresses in depth and extent, the picture of the spheroid star population that comes out points to a shape quite compatible with what we know about the distribution of baryonic dark matter if it is made of stellar remnants, suggesting a common dynamical origin.
Resonant frequencies of the two-dimensional plasma in FETs increase with the reduction of the channel dimensions and can reach the THz range for sub-micron gate lengths. Nonlinear properties of the electron plasma in the transistor channel can be used for the detection and mixing of THz frequencies. At cryogenic temperatures resonant and gate voltage tunable detection related to plasma waves resonances, is observed. At room temperature, when plasma oscillations are overdamped, the FET can operate as an efficient broadband THz detector. We present the main theoretical and experimental results on THz detection by FETs in the context of their possible application for THz imaging.
We consider the training process of a neural network as a dynamical system acting on the high-dimensional weight space. Each epoch is an application of the map induced by the optimization algorithm and the loss function. Using this induced map, we can apply observables on the weight space and measure their evolution. The evolution of the observables are given by the Koopman operator associated with the induced dynamical system. We use the spectrum and modes of the Koopman operator to realize the above objectives. Our methods can help to, a priori, determine the network depth; determine if we have a bad initialization of the network weights, allowing a restart before training too long; speeding up the training time. Additionally, our methods help enable noise rejection and improve robustness. We show how the Koopman spectrum can be used to determine the number of layers required for the architecture. Additionally, we show how we can elucidate the convergence versus non-convergence of the training process by monitoring the spectrum, in particular, how the existence of eigenvalues clustering around 1 determines when to terminate the learning process. We also show how using Koopman modes we can selectively prune the network to speed up the training procedure. Finally, we show that incorporating loss functions based on negative Sobolev norms can allow for the reconstruction of a multi-scale signal polluted by very large amounts of noise.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
