A step towards automated TEM and SEM characterisation using deep learning : nano-tetrahedrons

Convolutional neural networks (CNNs) have attracted huge amount of attentions since the emergence of deep learning, because of their ability to learn and adapt features directly from the input data, and obtain accurate classification. Although CNNs have resulted in a variety of advances in fields pa...

Full description

Saved in:
Bibliographic Details
Main Author: Tang, Si An
Other Authors: Li Shuzhou
Format: Final Year Project
Language:English
Published: 2019
Subjects:
Online Access:http://hdl.handle.net/10356/78727
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-78727
record_format dspace
spelling sg-ntu-dr.10356-787272023-03-04T15:37:01Z A step towards automated TEM and SEM characterisation using deep learning : nano-tetrahedrons Tang, Si An Li Shuzhou School of Materials Science and Engineering Engineering::Materials Convolutional neural networks (CNNs) have attracted huge amount of attentions since the emergence of deep learning, because of their ability to learn and adapt features directly from the input data, and obtain accurate classification. Although CNNs have resulted in a variety of advances in fields particular for image recognition, microscopic images such as TEM and SEM images in materials science are essentially different problem compared to other applications of deep learning. In addition, limited research studies exist to direct the design of CNN architectures for such microscopic images, and it is unclear that whether CNN can be used for this application. In this work, we have investigated the design of a CNN model for images of common nanostructures, in particular, tetrahedron. We have shown that CNN is capable of classifying tetrahedron images out of similar shapes, which is cube and octahedron. Through the project, the important and commonly tuned parameters, such as batch size, number of convolutional layers, pooling layers, activation layers were exploited and discussed with the aim of getting an optimised model. Additionally, we have concluded the optimal number of iterations or epochs to train a CNN model for the application. Apart from that, we have proven that a dropout layer is necessary at the end of the model. We have also concluded that RMSprop is a better choice of optimiser compared to Stochastic Gradient Descend (SGD). This work suggested a CNN model for the classification of tetrahedrons, which serves as an extremely crucial starting point of realisation of automated characterisation in TEM and SEM equipment in the near future. Bachelor of Engineering (Materials Engineering) 2019-06-26T04:40:11Z 2019-06-26T04:40:11Z 2019 Final Year Project (FYP) http://hdl.handle.net/10356/78727 en Nanyang Technological University 38 p. application/pdf
institution Nanyang Technological University
building NTU Library
continent Asia
country Singapore
Singapore
content_provider NTU Library
collection DR-NTU
language English
topic Engineering::Materials
spellingShingle Engineering::Materials
Tang, Si An
A step towards automated TEM and SEM characterisation using deep learning : nano-tetrahedrons
description Convolutional neural networks (CNNs) have attracted huge amount of attentions since the emergence of deep learning, because of their ability to learn and adapt features directly from the input data, and obtain accurate classification. Although CNNs have resulted in a variety of advances in fields particular for image recognition, microscopic images such as TEM and SEM images in materials science are essentially different problem compared to other applications of deep learning. In addition, limited research studies exist to direct the design of CNN architectures for such microscopic images, and it is unclear that whether CNN can be used for this application. In this work, we have investigated the design of a CNN model for images of common nanostructures, in particular, tetrahedron. We have shown that CNN is capable of classifying tetrahedron images out of similar shapes, which is cube and octahedron. Through the project, the important and commonly tuned parameters, such as batch size, number of convolutional layers, pooling layers, activation layers were exploited and discussed with the aim of getting an optimised model. Additionally, we have concluded the optimal number of iterations or epochs to train a CNN model for the application. Apart from that, we have proven that a dropout layer is necessary at the end of the model. We have also concluded that RMSprop is a better choice of optimiser compared to Stochastic Gradient Descend (SGD). This work suggested a CNN model for the classification of tetrahedrons, which serves as an extremely crucial starting point of realisation of automated characterisation in TEM and SEM equipment in the near future.
author2 Li Shuzhou
author_facet Li Shuzhou
Tang, Si An
format Final Year Project
author Tang, Si An
author_sort Tang, Si An
title A step towards automated TEM and SEM characterisation using deep learning : nano-tetrahedrons
title_short A step towards automated TEM and SEM characterisation using deep learning : nano-tetrahedrons
title_full A step towards automated TEM and SEM characterisation using deep learning : nano-tetrahedrons
title_fullStr A step towards automated TEM and SEM characterisation using deep learning : nano-tetrahedrons
title_full_unstemmed A step towards automated TEM and SEM characterisation using deep learning : nano-tetrahedrons
title_sort step towards automated tem and sem characterisation using deep learning : nano-tetrahedrons
publishDate 2019
url http://hdl.handle.net/10356/78727
_version_ 1759857208824168448