Bollettino di Geofisica Teorica e Applicata
OGS Website
About the Journal
Contacts
To Authors
On-line Submission
Subscriptions
Forthcoming
On-line First
The Historical First Issue
Issues

2022 Vol. 63
1 / 2 / 3 / 4

2021 Vol. 62
1 / 2 / 3 / 4 / Suppl. 1 / Suppl. 2 / Suppl. 3

2020 Vol. 61
1 / 2 / 3 / 4 / Suppl. 1

2019 Vol. 60
1 / 2 / 3 / 4 / Suppl. 1 / Suppl. 2 / Suppl. 3

2018 Vol. 59
1 / 2 / 3 / 4 / Suppl. 1

2017 Vol. 58
1 / 2 / 3 / 4

2016 Vol. 57
1 / 2 / 3 / 4 / Suppl. 1

2015 Vol. 56
1 / 2 / 3 / 4

2014 Vol. 55
1 / 2 / 3 / 4

2013 Vol. 54
1 / 2 / 3 / 4 / Suppl. 1 / Suppl. 2

2012 Vol. 53
1 / 2 / 3 / 4

2011 Vol. 52
1 / 2 / 3 / 4 / Suppl. 1

2010 Vol. 51
1 / 2-3 / 4 / Suppl. 1

2009 Vol. 50
1 / 2 / 3 / 4

2008 Vol. 49
1 / 2 / 3-4 / Suppl. 1

2007 Vol. 48
1 / 2 / 3 / 4

2006 Vol. 47
1-2 / 3 / 4

2005 Vol. 46
1 / 2-3 / 4

2004 Vol. 45
1-2 / 3 / 4 / Suppl. 1 / Suppl. 2

2003 Vol. 44
1 / 2 / 3-4

2002 Vol. 43
1-2 / 3-4

2001 Vol. 42
1-2 / 3-4

2000 Vol. 41
1 / 2 / 3-4

1999 Vol. 40
1 / 2 / 3-4

1998 Vol. 39
1 / 2 / 3 / 4

1997 Vol. 38
1-2 / 3-4

1995 Vol. 37
145 / 146 / 147 / 148 / Suppl. 1

1994 Vol. 36
141-144 / Suppl. 1

1993 Vol. 35
137-138 / 139 / 140

1992 Vol. 34
133 / 134-135 / 136

1991 Vol. 33
129 / 130-131 / 132

 
 

Vol. 62, n.3, September 2021
pp. 455-466

Deep Learning for automatic classification of mineralogical thin sections

P. Dell'Aversana

Received: 26 April 2021; accepted: 7 July 2021; published online: 23 September 2021

Abstract

Deep Learning refers to a set of algorithms in machine learning aimed at learning through multiple levels of abstraction. It typically makes use of artificial neural networks characterised by many hidden layers, and has been producing excellent results in image recognition and classification. In this paper, I discuss and compare two different types of Deep Learning architectures: Fully Connected and Convolutional Residual Networks. Using a set of test images, I show that these neural networks can be applied, with different effectiveness, for classifying complex images extracted from mineralogical thin sections. Both techniques produce reliable results, although the learners are trained on a limited number of examples. However, in the case of Fully Connected Networks, the vanishing gradient problem represents a crucial limitation when increasing the number of hidden layers. In fact, the classification results remain unchanged, or even degrade, when additional neural layers are progressively stacked.
Instead, the same problem is overcome in Convolutional Residual Networks, through a technique based on shortcut connections. Hundreds of hidden layers are used for improving the classification performances, without incurring into the degradation of learning capabilities.



Download PDF complete


back to table of contents