Distance insensitive concrete crack detection using convolutional neural network with controlled blurriness
Structural health monitoring system has been implemented to assess structural damage with minimal manpower. Most of research and development interests in structural damage detection have moved towards the use of artificial intelligence to aid in such process. Recent research has highlighted the use...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Language: | English |
Published: |
2019
|
Subjects: | |
Online Access: | http://eprints.utm.my/id/eprint/86088/1/NgSuFenMSKA2019.pdf |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
id |
my-utm-ep.86088 |
---|---|
record_format |
uketd_dc |
spelling |
my-utm-ep.860882020-08-30T08:56:03Z Distance insensitive concrete crack detection using convolutional neural network with controlled blurriness 2019 Ng, Su Fen TA Engineering (General). Civil engineering (General) Structural health monitoring system has been implemented to assess structural damage with minimal manpower. Most of research and development interests in structural damage detection have moved towards the use of artificial intelligence to aid in such process. Recent research has highlighted the use of convolutional neural network (CNN) as one of the powerful tools for accurate and effective image recognition. Nonetheless, the application of CNN on crack damage detection is limited by the inability of the method to detect crack autonomously without a given distance. In view of this, the present study developed a CNN based artificial intelligence for detecting concrete crack autonomously at various distance. The innovation of this study is the use of blurred and sharp images to train CNN. This idea is inspired from the fact that images taken from further distance Eire blurrier. Eight databases with different combination of datasets are then considered and trained on designed CNN. It is found that all networks recorded with at least 95 % accuracy. The robustness and adaptability of the network with the use of sharp images only are tested on twentythree images taken from Universiti Teknologi Malaysia under various conditions. Additionally, these eight networks are evaluated by classifying four different images taken in the distance of 0.5 m, 1.0 m, 1.5 m and 2.0 m, respectively. It is found that the most performing network across various image distances is the network solely made up of image with blurriness level 1. The results show that the presence of blurred images can potentially solve the image distance issue associated with CNN. 2019 Thesis http://eprints.utm.my/id/eprint/86088/ http://eprints.utm.my/id/eprint/86088/1/NgSuFenMSKA2019.pdf application/pdf en public http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:134342 masters Universiti Teknologi Malaysia, Faculty of Engineering - School of Civil Engineering Faculty of Engineering - School of Civil Engineering |
institution |
Universiti Teknologi Malaysia |
collection |
UTM Institutional Repository |
language |
English |
topic |
TA Engineering (General) Civil engineering (General) |
spellingShingle |
TA Engineering (General) Civil engineering (General) Ng, Su Fen Distance insensitive concrete crack detection using convolutional neural network with controlled blurriness |
description |
Structural health monitoring system has been implemented to assess structural damage with minimal manpower. Most of research and development interests in structural damage detection have moved towards the use of artificial intelligence to aid in such process. Recent research has highlighted the use of convolutional neural network (CNN) as one of the powerful tools for accurate and effective image recognition. Nonetheless, the application of CNN on crack damage detection is limited by the inability of the method to detect crack autonomously without a given distance. In view of this, the present study developed a CNN based artificial intelligence for detecting concrete crack autonomously at various distance. The innovation of this study is the use of blurred and sharp images to train CNN. This idea is inspired from the fact that images taken from further distance Eire blurrier. Eight databases with different combination of datasets are then considered and trained on designed CNN. It is found that all networks recorded with at least 95 % accuracy. The robustness and adaptability of the network with the use of sharp images only are tested on twentythree images taken from Universiti Teknologi Malaysia under various conditions. Additionally, these eight networks are evaluated by classifying four different images taken in the distance of 0.5 m, 1.0 m, 1.5 m and 2.0 m, respectively. It is found that the most performing network across various image distances is the network solely made up of image with blurriness level 1. The results show that the presence of blurred images can potentially solve the image distance issue associated with CNN. |
format |
Thesis |
qualification_level |
Master's degree |
author |
Ng, Su Fen |
author_facet |
Ng, Su Fen |
author_sort |
Ng, Su Fen |
title |
Distance insensitive concrete crack detection using convolutional neural network with controlled blurriness |
title_short |
Distance insensitive concrete crack detection using convolutional neural network with controlled blurriness |
title_full |
Distance insensitive concrete crack detection using convolutional neural network with controlled blurriness |
title_fullStr |
Distance insensitive concrete crack detection using convolutional neural network with controlled blurriness |
title_full_unstemmed |
Distance insensitive concrete crack detection using convolutional neural network with controlled blurriness |
title_sort |
distance insensitive concrete crack detection using convolutional neural network with controlled blurriness |
granting_institution |
Universiti Teknologi Malaysia, Faculty of Engineering - School of Civil Engineering |
granting_department |
Faculty of Engineering - School of Civil Engineering |
publishDate |
2019 |
url |
http://eprints.utm.my/id/eprint/86088/1/NgSuFenMSKA2019.pdf |
_version_ |
1747818493519069184 |