Loading...

Machine-Vision Based Visual Inspection of Masonry Structures Using Optical Measurements on Still Images

Ghorbanian, Mohammad Javad | 2021

631 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 54170 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Rahimzadeh Rofooei, Fayyaz; Mahdavi, Hossein
  7. Abstract:
  8. Cracks are the earliest signs of damage in structures which are needed to get detected through structural inspection before causing failure. Regular inspection and post-disaster inspection of structures are usually conducted manually by engineers. These kinds of visual inspections could be expensive, subjective, not reliable, and most importantly inefficient. Computer vision developments help automate this process by analyzing digital images captured from the surface of structures. Despite the fact that masonry structures have more complex surfaces than other kinds of structures, limited studies have focused on presenting machine-vision-based techniques for crack detection in those structures. This study specifically aims to investigate the application of convolutional neural network models in patchwise classification and segmentation of cracks on a still image dataset of a masonry brick heritage building. The collected dataset includes 256 still images containing cracks and 474 still images representing the intact surface of the structure, all in 128*128 resolution. A pretrained EfficientNet-bo CNN architecture, which is presented and trained on 75% of the dataset using a cross validation technique, achieves 96% accuracy in classifying the validation part of the dataset into the crack or intact images. Also, a pretrained Unet architecture with a EfficientNet-bo encoder achieves an 85.1% F1-Score in detecting cracks at a pixel level. While the models are trained on a limited dataset which includes some pretty noisy and complex still images containing some different patterns of cracks in a background of bricks and mortar, the accuracy and efficiency of the models in detecting cracks are improved in comparison with the existing models in the literature
  9. Keywords:
  10. Convolutional Neural Network ; Crack Detection ; Automatic Identification ; Masonry Buildings ; Visual Inspection ; Machine Vision ; Deep Learning

 Digital Object List

 Bookmark

...see more