Raman spectroscopy is a well-established technique for rapid analysis for Cultural Heritage characterization. It represents one of the most used and reliable technique because non invasive and usable, even on-site, without any sample preparation. Overall in the last decade, several methods have been tested with the purpose of obtaining an automatic classification process of spectra of unknown materials. Machine learning methods and in particular Neural Networks are the most recently exploited on this topic. Unfortunately, the availability of real Raman spectra is often limited considering the amount of data necessary for the use of these approaches. In this work we propose a pipeline for augmenting data with a GAN reinforcement, with the aim of creating a dataset large enough to take advantage of such. Furthermore, three types of neural networks (binary-NN, categorical-NN and CNN) are trained with the aforementioned dataset and compared with each other, taking into consideration their accuracy. The results show that the enhanced data augmentation with GANs significantly increases the accuracy of the networks. © 2020 IEEE.

Enhanced Data Augmentation using GANs for Raman Spectra Classification

De Santis E;Pomante L.
2020

Abstract

Raman spectroscopy is a well-established technique for rapid analysis for Cultural Heritage characterization. It represents one of the most used and reliable technique because non invasive and usable, even on-site, without any sample preparation. Overall in the last decade, several methods have been tested with the purpose of obtaining an automatic classification process of spectra of unknown materials. Machine learning methods and in particular Neural Networks are the most recently exploited on this topic. Unfortunately, the availability of real Raman spectra is often limited considering the amount of data necessary for the use of these approaches. In this work we propose a pipeline for augmenting data with a GAN reinforcement, with the aim of creating a dataset large enough to take advantage of such. Furthermore, three types of neural networks (binary-NN, categorical-NN and CNN) are trained with the aforementioned dataset and compared with each other, taking into consideration their accuracy. The results show that the enhanced data augmentation with GANs significantly increases the accuracy of the networks. © 2020 IEEE.
9781728162515
File in questo prodotto:
File Dimensione Formato  
DiFrischia_S18202.pdf

non disponibili

Tipologia: Documento in Pre-print
Licenza: Dominio pubblico
Dimensione 1.55 MB
Formato Adobe PDF
1.55 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11697/167351
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? 1
social impact