Skip to Main content Skip to Navigation
Preprints, Working Papers, ...

Detection of defects in a micro-structure: recurrent neural networks and sparsity-constrained inversion

Abstract : Detection of an unknown number of missing rods within a micro-structure composed of a grid-like, finite set of infinitely long circular cylindrical dielectric rods under sub-wavelength condition is investigated. Sub-wavelength distances between adjacent rods and sub-wavelength rod diameters lead to a severe challenge due to need of super-resolution within the present micro-structure, beyond Rayleigh criterion. Two different methods are proposed to realize the detection: One is a data-driven method, which builds upon a framework of convolutional neural networks (CNN) and recurrent neural networks (RNN). Another one is an analytical method, which builds upon the multiple scattering expansion (MSE) with sparsity-prior information, which needs a good understanding of physics behind the system. Comprehensive numerical simulations of interest for both methods are proposed, in terms of missing rod number, shape, frequency of observation and configuration of structure. Comparison between the two methods are also presented, both of them achieving detection under certain conditions as discussed.
Complete list of metadatas

https://hal-centralesupelec.archives-ouvertes.fr/hal-02905435
Contributor : Dominique Lesselier <>
Submitted on : Thursday, July 23, 2020 - 2:06:33 PM
Last modification on : Wednesday, September 16, 2020 - 5:52:13 PM

Identifiers

  • HAL Id : hal-02905435, version 1

Citation

Peipei Ran, Dominique Lesselier, Mohammed Serhir. Detection of defects in a micro-structure: recurrent neural networks and sparsity-constrained inversion. 2020. ⟨hal-02905435⟩

Share

Metrics

Record views

38