A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Automatic image-based brick segmentation and crack detection of masonry walls using machine learning
Abstract This paper aims to improve automation in brick segmentation and crack detection of masonry walls through image-based techniques and machine learning. Initially, a large dataset of hand-labelled images of different in colour, texture, and size of brickwork masonry walls has been developed. Then, different deep learning networks (U-Net, DeepLabV3+, U-Net (SM), LinkNet (SM), and FPN (SM)) were utilised and their quality was assessed. Furthermore, the ability to generate geometric models of masonry structures and the evaluation of the geometric properties of detected cracks was also investigated. Additional metrics were also developed to compare the CNN output with other image-processing algorithms. From the analysis of results it was shown that the use of machine learning, for brick segmentation, provides better outcome than typical image-processing applications. This implementation of deep-learning for crack detection and localisation of bricks in masonry walls highlights the great potential of new technologies for documentation of masonry fabric.
Highlights Automatic brick segmentation from images using machine learning. Automatic crack quantification from images using machine learning. The brick detection model obtained a validation-accuracy of 96.86%. The proposed procedure paves the way for automated documentation and inspection of full-scale masonry structures.
Automatic image-based brick segmentation and crack detection of masonry walls using machine learning
Abstract This paper aims to improve automation in brick segmentation and crack detection of masonry walls through image-based techniques and machine learning. Initially, a large dataset of hand-labelled images of different in colour, texture, and size of brickwork masonry walls has been developed. Then, different deep learning networks (U-Net, DeepLabV3+, U-Net (SM), LinkNet (SM), and FPN (SM)) were utilised and their quality was assessed. Furthermore, the ability to generate geometric models of masonry structures and the evaluation of the geometric properties of detected cracks was also investigated. Additional metrics were also developed to compare the CNN output with other image-processing algorithms. From the analysis of results it was shown that the use of machine learning, for brick segmentation, provides better outcome than typical image-processing applications. This implementation of deep-learning for crack detection and localisation of bricks in masonry walls highlights the great potential of new technologies for documentation of masonry fabric.
Highlights Automatic brick segmentation from images using machine learning. Automatic crack quantification from images using machine learning. The brick detection model obtained a validation-accuracy of 96.86%. The proposed procedure paves the way for automated documentation and inspection of full-scale masonry structures.
Automatic image-based brick segmentation and crack detection of masonry walls using machine learning
Loverdos, Dimitrios (author) / Sarhosis, Vasilis (author)
2022-05-26
Article (Journal)
Electronic Resource
English
CNN-Based Watershed Marker Extraction for Brick Segmentation in Masonry Walls
| British Library Conference Proceedings | 2019
Modeling Unreinforced Brick Masonry Walls
| British Library Conference Proceedings | 1995
Permeability of brick masonry walls -- Hypothesis
| Engineering Index Backfile | 1933