Showing 1–10 of 10 results for author: Martinez-Lorenzo, J A

Consensus Function from an $L_p^q-$norm Regularization Term for its Use as Adaptive Activation Functions in Neural Networks

Authors: Juan Heredia-Juesas, José Á. Martínez-Lorenzo

Abstract: The design of a neural network is usually carried out by defining the number of layers, the number of neurons per layer, their connections or synapses, and the activation function that they will execute. The training process tries to optimize the weights assigned to those connections, together with the biases of the neurons, to better fit the training data. However, the definition of the activatio… ▽ More The design of a neural network is usually carried out by defining the number of layers, the number of neurons per layer, their connections or synapses, and the activation function that they will execute. The training process tries to optimize the weights assigned to those connections, together with the biases of the neurons, to better fit the training data. However, the definition of the activation functions is, in general, determined in the design process and not modified during the training, meaning that their behavior is unrelated to the training data set. In this paper we propose the definition and utilization of an implicit, parametric, non-linear activation function that adapts its shape during the training process. This fact increases the space of parameters to optimize within the network, but it allows a greater flexibility and generalizes the concept of neural networks. Furthermore, it simplifies the architectural design since the same activation function definition can be employed in each neuron, letting the training process to optimize their parameters and, thus, their behavior. Our proposed activation function comes from the definition of the consensus variable from the optimization of a linear underdetermined problem with an $L_p^q$ regularization term, via the Alternating Direction Method of Multipliers (ADMM). We define the neural networks using this type of activation functions as $pq-$networks. Preliminary results show that the use of these neural networks with this type of adaptive activation functions reduces the error in regression and classification examples, compared to equivalent regular feedforward neural networks with fixed activation functions. △ Less

Submitted 30 June, 2022; originally announced June 2022.

Comments: 10 pages, 8 figures

Single-Frequency Imaging and Material Characterization using Reconfigurable Reflectarrays

Authors: Weite Zhang, Hipolito Gomez-Sousa, Juan Heredia-Juesas, Jose A. Martinez-Lorenzo

Abstract: In this work, a physical and geometrical optics based single-frequency imaging scheme is proposed for personal screening systems using multiple reconfigurable reflectarrays. This scheme is able to not only reconstruct profiles of potential threat objects on human body, but also identify their materials in terms of their complex relative permittivities. Both simulation and experiment are carried ou… ▽ More In this work, a physical and geometrical optics based single-frequency imaging scheme is proposed for personal screening systems using multiple reconfigurable reflectarrays. This scheme is able to not only reconstruct profiles of potential threat objects on human body, but also identify their materials in terms of their complex relative permittivities. Both simulation and experiment are carried out to detect dielectric objects at a microwave frequency of 24.16 GHz. The object profiles and complex relative permittivities are obtained with both high accuracy and computational efficiency, which show great potentials for security imaging where inspection of human body for threat materials, such as narcotics, explosives, and other types of contraband, is very common. △ Less

Submitted 23 August, 2019; originally announced August 2019.

Consensus and Sectioning-based ADMM with Norm-1 Regularization for Imaging with a Compressive Reflector Antenna

Authors: Juan Heredia-Juesas, Ali Molaei, Luis Tirado, Jose A. Martinez-Lorenzo

Abstract: This paper presents three distributed techniques to find a sparse solution of the underdetermined linear problem $\textbf{g}=\textbf{Hu}$ with a norm-1 regularization, based on the Alternating Direction Method of Multipliers (ADMM). These techniques divide the matrix $\textbf{H}$ in submatrices by rows, columns, or both rows and columns, leading to the so-called consensus-based ADMM, sectioning-ba… ▽ More This paper presents three distributed techniques to find a sparse solution of the underdetermined linear problem $\textbf{g}=\textbf{Hu}$ with a norm-1 regularization, based on the Alternating Direction Method of Multipliers (ADMM). These techniques divide the matrix $\textbf{H}$ in submatrices by rows, columns, or both rows and columns, leading to the so-called consensus-based ADMM, sectioning-based ADMM, and consensus and sectioning-based ADMM, respectively. These techniques are applied particularly for millimeter-wave imaging through the use of a Compressive Reflector Antenna (CRA). The CRA is a hardware designed to increase the sensing capacity of an imaging system and reduce the mutual information among measurements, allowing an effective imaging of sparse targets with the use of Compressive Sensing (CS) techniques. Consensus-based ADMM has been proved to accelerate the imaging process and sectioning-based ADMM has shown to highly reduce the amount of information to be exchange among the computational nodes. In this paper, the mathematical formulation and graphical interpretation of these two techniques, together with the consensus and sectioning-based ADMM approach, are presented. The imaging quality, the imaging time, the convergence, and the communication efficiency among the nodes are analyzed and compared. The distributed capabitities of the ADMM-based approaches, together with the high sensing capacity of the CRA, allow the imaging of metallic targets in a 3D domain in quasi-real time with a reduced amount of information exchanged among the nodes. △ Less

Submitted 13 November, 2018; originally announced November 2018.

Experimental Results of a 3D Millimeter-Wave Compressive-Reflector-Antenna Imaging System

Authors: Weite Zhang, Ali Molaei, Juan Heredia-Juesas, Luis Tirado, Katherine Graham, A. Bisulco, Hipolito Gomez-Sousa, Jose A. Martinez-Lorenzo

Abstract: This letter presents the first experimental results of our three-dimensional (3D) millimeter-wave (mm-wave) Compressive-Reflector-Antenna (CRA) imaging system. In this prototype, the CRA is 3D-printed and coated with a metallic spray to easily introduce pseudo-random scatterers on the surface of a traditional reflector antenna (TRA). The CRA performs a pseudo random coding of the incident wavefron… ▽ More This letter presents the first experimental results of our three-dimensional (3D) millimeter-wave (mm-wave) Compressive-Reflector-Antenna (CRA) imaging system. In this prototype, the CRA is 3D-printed and coated with a metallic spray to easily introduce pseudo-random scatterers on the surface of a traditional reflector antenna (TRA). The CRA performs a pseudo random coding of the incident wavefront, thus adding spatial diversity in the imaging region and enabling the effective use of compressive sensing (CS) and imaging techniques. The CRA is fed with a multiple-input-multiple-output (MIMO) radar, which consists of four transmitting and four receiving ports. Consequently, the mechanical scanning parts and phase shifters, which are necessary in conventional physical or synthetic aperture arrays, are not needed in this system. Experimental results show the effectiveness of the prototype to perform a successful 3D reconstruction of a T-shaped metallic target. △ Less

Submitted 28 July, 2018; originally announced August 2018.

Comments: 5 pages, 6 figures

Sensing Matrix Design via Capacity Maximization for Block Compressive Sensing Applications

Authors: Richard Obermeier, Jose Angel Martinez-Lorenzo

Abstract: It is well established in the compressive sensing (CS) literature that sensing matrices whose elements are drawn from independent random distributions exhibit enhanced reconstruction capabilities. In many CS applications, such as electromagnetic imaging, practical limitations on the measurement system prevent one from generating sensing matrices in this fashion. Although one can usually randomized… ▽ More It is well established in the compressive sensing (CS) literature that sensing matrices whose elements are drawn from independent random distributions exhibit enhanced reconstruction capabilities. In many CS applications, such as electromagnetic imaging, practical limitations on the measurement system prevent one from generating sensing matrices in this fashion. Although one can usually randomized the measurements to some degree, these sensing matrices do not achieve the same reconstruction performance as the truly randomized sensing matrices. In this paper, we present a novel method, based upon capacity maximization, for designing sensing matrices with enhanced block-sparse signal reconstruction capabilities. Through several numerical examples, we demonstrate how our method significantly enhances reconstruction performance. △ Less

Submitted 21 March, 2018; originally announced March 2018.

Generalized Optimization of High Capacity Compressive Imaging Systems

Authors: Richard Obermeier, Jose Angel Martinez-Lorenzo

Abstract: One of the greatest challenges in applying compressive sensing (CS) signal processing techniques to electromagnetic imaging applications is designing a sensing matrix that has good reconstruction capabilities. Compressive reflector antennas (CRA) are a class of antennas that have been shown to provide enhanced image reconstruction performance over traditional reflector antennas (TRA) when CS techn… ▽ More One of the greatest challenges in applying compressive sensing (CS) signal processing techniques to electromagnetic imaging applications is designing a sensing matrix that has good reconstruction capabilities. Compressive reflector antennas (CRA) are a class of antennas that have been shown to provide enhanced image reconstruction performance over traditional reflector antennas (TRA) when CS techniques are employed. In this paper, we present a unified CRA design method, which considers both the sensing capacity and efficiency of the antenna, and can be used for both compressive imaging and multiple-input multiple-output (MIMO) communication applications. The unified design method is assessed for a CRA configuration in which dielectric scatterers are added to the surface of a TRA. The design results demonstrate the ability of the unified design method to enhance the CS reconstruction capabilities of the CRA. △ Less

Submitted 21 March, 2018; originally announced March 2018.

Detecting Breast Cancer using a Compressive Sensing Unmixing Algorithm

Authors: Richard Obermeier, Jose Angel Martinez-Lorenzo

Abstract: Traditional breast cancer imaging methods using microwave Nearfield Radar Imaging (NRI) seek to recover the complex permittivity of the tissues at each voxel in the imaging region. This approach is suboptimal, in that it does not directly consider the permittivity values that healthy and cancerous breast tissues typically have. In this paper, we describe a novel unmixing algorithm for detecting br… ▽ More Traditional breast cancer imaging methods using microwave Nearfield Radar Imaging (NRI) seek to recover the complex permittivity of the tissues at each voxel in the imaging region. This approach is suboptimal, in that it does not directly consider the permittivity values that healthy and cancerous breast tissues typically have. In this paper, we describe a novel unmixing algorithm for detecting breast cancer. In this approach, the breast tissue is separated into three components, low water content (LWC), high water content (HWC), and cancerous tissues, and the goal of the optimization procedure is to recover the mixture proportions for each component. By utilizing this approach in a hybrid DBT / NRI system, the unmixing reconstruction process can be posed as a sparse recovery problem, such that compressive sensing (CS) techniques can be employed. A numerical analysis is performed, which demonstrates that cancerous lesions can be detected from their mixture proportion under the appropriate conditions. △ Less

Submitted 28 October, 2016; originally announced October 2016.

A computational method for modeling arbitrary junctions employing different surface integral equation formulations for three-dimensional scattering and radiation problems

Authors: Hipolito Gomez-Sousa, Oscar Rubiños-Lopez, Jose Angel Martinez-Lorenzo, Marcos Arias-Acuña

Abstract: This paper presents a new method, based on the well-known method of moments (MoM), for the numerical electromagnetic analysis of scattering and radiation from metallic or dielectric structures, or both structure types in the same simulation, that are in contact with other metallic or dielectric structures. The proposed method for solving the MoM junction problem consists of two separate algorithms… ▽ More This paper presents a new method, based on the well-known method of moments (MoM), for the numerical electromagnetic analysis of scattering and radiation from metallic or dielectric structures, or both structure types in the same simulation, that are in contact with other metallic or dielectric structures. The proposed method for solving the MoM junction problem consists of two separate algorithms, one of which comprises a generalization for bodies in contact of the surface integral equation (SIE) formulations. Unlike some other published SIE generalizations in the field of computational electromagnetics, this generalization does not require duplicating unknowns on the dielectric separation surfaces. Additionally, this generalization is applicable to any ordinary single-scatterer SIE formulations employed as baseline. The other algorithm deals with enforcing boundary conditions and Kirchhoff's Law, relating the surface current flow across a junction edge. Two important features inherent to this latter algorithm consist of a mathematically compact description in matrix form, and, importantly from a software engineering point of view, an easy implementation in existing MoM codes which makes the debugging process more comprehensible. A practical example involving a real grounded monopole antenna for airplane-satellite communication is analyzed for validation purposes by comparing with precise measurements covering different electrical sizes. △ Less

Submitted 19 October, 2015; originally announced October 2015.

Comparison of iterative solvers for electromagnetic analysis of plasmonic nanostructures using multiple surface integral equation formulations

Authors: Hipolito Gomez-Sousa, Oscar Rubiños-Lopez, Jose Angel Martinez-Lorenzo

Abstract: The electromagnetic behavior of plasmonic structures can be predicted after discretizing and solving a linear system of equations, derived from a continuous surface integral equation (SIE) and the appropriate boundary conditions, using a method of moments (MoM) methodology. In realistic large-scale optical problems, a direct inversion of the SIE-MoM matrix cannot be performed due to its large size… ▽ More The electromagnetic behavior of plasmonic structures can be predicted after discretizing and solving a linear system of equations, derived from a continuous surface integral equation (SIE) and the appropriate boundary conditions, using a method of moments (MoM) methodology. In realistic large-scale optical problems, a direct inversion of the SIE-MoM matrix cannot be performed due to its large size, and an iterative solver must be used instead. This paper investigates the performance of four iterative solvers (GMRES, TFQMR, CGS, and BICGSTAB) for five different SIE-MoM formulations (PMCHWT, JMCFIE, CTF, CNF, and MNMF). Moreover, under this plasmonic context, a set of suggested guidelines are provided to choose a suitable SIE formulation and iterative solver depending on the desired simulation error and available runtime resources. △ Less

Submitted 18 October, 2015; originally announced October 2015.

Model-based Optimization of Compressive Antennas for High-Sensing-Capacity Applications

Authors: Richard Obermeier, Jose Angel Martinez-Lorenzo

Abstract: This paper presents a novel, model-based compressive antenna design method for high sensing capacity imaging applications. Given a set of design constraints, the method maximizes the sensing capacity of the compressive antenna by varying the constitutive properties of scatterers distributed along the antenna. Preliminary 2D design results demonstrate the new method's ability to produce antenna con… ▽ More This paper presents a novel, model-based compressive antenna design method for high sensing capacity imaging applications. Given a set of design constraints, the method maximizes the sensing capacity of the compressive antenna by varying the constitutive properties of scatterers distributed along the antenna. Preliminary 2D design results demonstrate the new method's ability to produce antenna configurations with enhanced imaging capabilities. △ Less

Submitted 2 October, 2015; v1 submitted 20 July, 2015; originally announced July 2015.