Pavement crack classification using deep convolutional neural network
Loading...
Supplementary material
Other Title
Authors
Osman, M.K.
Mohammed Zamree, M.E.A.
Idris, M.
Ahmad, K.A.
Mohamed Yusof, N.A.
Ibrahim, A.
Hasnur Rabiain, A.
Bahri, Intan
Mohammed Zamree, M.E.A.
Idris, M.
Ahmad, K.A.
Mohamed Yusof, N.A.
Ibrahim, A.
Hasnur Rabiain, A.
Bahri, Intan
Author ORCID Profiles (clickable)
Degree
Grantor
Date
2021-11-15
Supervisors
Type
Journal Article
Ngā Upoko Tukutuku (Māori subject headings)
Keyword
crack severity
road maintenance
flexible pavement
deep convolution neural network (DCNN)
road maintenance
flexible pavement
deep convolution neural network (DCNN)
ANZSRC Field of Research Code (2020)
Citation
Osman, M.K., Mohammed Zamree, M.E.A., Idris, M., Ahmad, K.A., Mohamed Yusof, N.A., Ibrahim, A., Hasnur Rabiain, A., & Bahri, I. (2021). Pavement crack classification using deep convolutional neural network. Journal of Mechanical Engineering, 10(1), 227-244. https://hdl.handle.net/10652/6113
Abstract
Road safety is one of the more difficult aspects concerning the field of civil engineering. Manual road inspection and distress detection by a road surveyor is a time-consuming, dangerous, and laborious process. This paper proposes an automated method to classify three common types of road distress; namely crocodile, longitudinal and transverse cracks using a deep convolution neural network. Four processes are involved to include data collection, cracked photo enhancement, cracks classification and performance evaluation. The first process of data collection involves capturing pavement crack images using a digital camera. Secondly, the crack images are labelled according to their group and their contrast further improved using the contrast limited adaptive histogram equalization (CLAHE) method. The third process involves training the deep convolutional neural network (DCNN). In this process, two (2) DCNN models are devised which are VGG16 and 9-Layer CNN models. Simulation results show that VGGG-16 with CLAHE enhancement were able to classify pavement cracks with high accuracy, precision, recall and F1-scores of 99.5%, 98.5%, 99.5% and 98.99% respectively. Through deep learning techniques, the VGGG-16 with CLAHE has demonstrated promising potential in classifying pavement cracks.
Publisher
College of Engineering, Universiti Teknologi MARA (UiTM)
Permanent link
Link to ePress publication
DOI
Copyright holder
Copyright notice
© 2021 College of Engineering, Universiti Teknologi MARA (UiTM), Malaysia