Computational Technology Resources - CCC

Y.H. Wei^1,2 and Y.Q. Ni^1,2

¹Department of Civil and Environmental Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong
²Hong Kong Branch of Chinese National Engineering Research Center on Rail Transit Electrification and Automation, Hong Kong

doi:10.4203/ccc.6.5.2

Y.H. Wei, Y.Q. Ni, "An Unsupervised Crack Detection Approach Based on a Sliding Window Variational Autoencoder", in P. Ivanyi, J. Kruis, B.H.V. Topping, (Editors), "Proceedings of the Seventeenth International Conference on Civil, Structural and Environmental Engineering Computing", Civil-Comp Press, Edinburgh, UK, Online volume: CCC 6, Paper 5.2, 2023, doi:10.4203/ccc.6.5.2

Keywords: crack detection, variational autoencoder, sliding windows, serialized input, anomaly detection, robustness, unsupervised learning.

Abstract

The current investigation presents a novel approach to detect cracks using the variational autoencoder (VAE). In this method, the input image is first divided into multiple segments using sliding windows and then fed into the VAE sequentially. The use of sliding windows effectively limits the number of neural nodes in the input layer of the VAE, which enhances the method's robustness. Additionally, the sliding window technique allows for the image information to be viewed as a time series, with cracks being treated as anomalies in the time series. By using the sliding window VAE (SW-VAE) with robust properties, such anomalies can be discarded during the reconstruction process. As a result, the detection of cracks can be achieved by comparing the difference between the input and output of the SW-VAE. Notably, this technique does not require positive sample training or learning image features specific to cracks, thus avoiding the challenge posed by the lack of training data or imbalanced datasets.

download the full-text of this paper (PDF, 6 pages, 238 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description

	Computational & Technology Resources an online resource for computational, engineering & technology publications
	not logged in - login
Front Page Browse CCC CCP CSETS CTR IJRT Other Authors Search Purchase Guide FAQ Contact us	Civil-Comp Conferences ISSN 2753-3239 CCC: 6 PROCEEDINGS OF THE SEVENTEENTH INTERNATIONAL CONFERENCE ON CIVIL, STRUCTURAL AND ENVIRONMENTAL ENGINEERING COMPUTING Edited by: P. Ivanyi, J. Kruis and B.H.V. Topping Paper 5.2 An Unsupervised Crack Detection Approach Based on a Sliding Window Variational Autoencoder Y.H. Wei^1,2 and Y.Q. Ni^1,2 ¹Department of Civil and Environmental Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong ²Hong Kong Branch of Chinese National Engineering Research Center on Rail Transit Electrification and Automation, Hong Kong doi:10.4203/ccc.6.5.2 Full Bibliographic Reference for this paper Y.H. Wei, Y.Q. Ni, "An Unsupervised Crack Detection Approach Based on a Sliding Window Variational Autoencoder", in P. Ivanyi, J. Kruis, B.H.V. Topping, (Editors), "Proceedings of the Seventeenth International Conference on Civil, Structural and Environmental Engineering Computing", Civil-Comp Press, Edinburgh, UK, Online volume: CCC 6, Paper 5.2, 2023, doi:10.4203/ccc.6.5.2 Keywords: crack detection, variational autoencoder, sliding windows, serialized input, anomaly detection, robustness, unsupervised learning. Abstract The current investigation presents a novel approach to detect cracks using the variational autoencoder (VAE). In this method, the input image is first divided into multiple segments using sliding windows and then fed into the VAE sequentially. The use of sliding windows effectively limits the number of neural nodes in the input layer of the VAE, which enhances the method's robustness. Additionally, the sliding window technique allows for the image information to be viewed as a time series, with cracks being treated as anomalies in the time series. By using the sliding window VAE (SW-VAE) with robust properties, such anomalies can be discarded during the reconstruction process. As a result, the detection of cracks can be achieved by comparing the difference between the input and output of the SW-VAE. Notably, this technique does not require positive sample training or learning image features specific to cracks, thus avoiding the challenge posed by the lack of training data or imbalanced datasets. download the full-text of this paper (PDF, 6 pages, 238 Kb) go to the previous paper go to the next paper return to the table of contents return to the volume description
Back to top	©Civil-Comp Limited 2023 - terms & conditions