Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Structural Defects Classification and Detection Using Convolutional Neural Network (CNN): A Review
Assessing the structures’ condition is mostly dependent on physical site reconnaissance and expert opinion, which involves systematical collection and storage of data and interpretation of data. With the increasing demand of clients, especially during the time of the pandemic, stakeholders are more focused on time-saving, minimized physical presence, and cost-effective means in decision making, before these structures can endanger the community and the occupants. In traditional ways, structural investigations are laborious, time-consuming, expensive, and in some cases health hazardous. This study aims at reviewing supervised learning of Convolutional Neural Networks (CNN), which can work in an automated manner to detect, identify and localize the defects, e.g., crack, surface deterioration, spalling, moisture damage, corrosion, etc., from images in a safe and budget-friendly manner, employing built-on and pre-trained CNN classifier models. While using the different type of CNN classifier models, prior studies have evaluated the models performance mostly based on accuracy, precision, recall, Intersection over Union (IoU), root mean square value (RMSE). This paper summarizes the important aspects of a CNN architecture for damage detection such as dataset preparation, random weight initialization, different learning rate, fine-tuning the hyper-parameters, and the working method with fast-forward pass and back-propagation. Finally, this paper will outline some of the challenges of using CNN in structural condition assessment and identify some future scopes.
Structural Defects Classification and Detection Using Convolutional Neural Network (CNN): A Review
Assessing the structures’ condition is mostly dependent on physical site reconnaissance and expert opinion, which involves systematical collection and storage of data and interpretation of data. With the increasing demand of clients, especially during the time of the pandemic, stakeholders are more focused on time-saving, minimized physical presence, and cost-effective means in decision making, before these structures can endanger the community and the occupants. In traditional ways, structural investigations are laborious, time-consuming, expensive, and in some cases health hazardous. This study aims at reviewing supervised learning of Convolutional Neural Networks (CNN), which can work in an automated manner to detect, identify and localize the defects, e.g., crack, surface deterioration, spalling, moisture damage, corrosion, etc., from images in a safe and budget-friendly manner, employing built-on and pre-trained CNN classifier models. While using the different type of CNN classifier models, prior studies have evaluated the models performance mostly based on accuracy, precision, recall, Intersection over Union (IoU), root mean square value (RMSE). This paper summarizes the important aspects of a CNN architecture for damage detection such as dataset preparation, random weight initialization, different learning rate, fine-tuning the hyper-parameters, and the working method with fast-forward pass and back-propagation. Finally, this paper will outline some of the challenges of using CNN in structural condition assessment and identify some future scopes.
Structural Defects Classification and Detection Using Convolutional Neural Network (CNN): A Review
Lecture Notes in Civil Engineering
Walbridge, Scott (Herausgeber:in) / Nik-Bakht, Mazdak (Herausgeber:in) / Ng, Kelvin Tsun Wai (Herausgeber:in) / Shome, Manas (Herausgeber:in) / Alam, M. Shahria (Herausgeber:in) / el Damatty, Ashraf (Herausgeber:in) / Lovegrove, Gordon (Herausgeber:in) / Arafin, P. (Autor:in) / Billah, A. H. M. M. (Autor:in)
Canadian Society of Civil Engineering Annual Conference ; 2021
Proceedings of the Canadian Society of Civil Engineering Annual Conference 2021 ; Kapitel: 27 ; 281-293
18.05.2022
13 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Automated classification of civil structure defects based on convolutional neural network
| Springer Verlag | 2021
Gemstone Classification Using Deep Convolutional Neural Network
| Springer Verlag | 2024
Gemstone Classification Using Deep Convolutional Neural Network
| Springer Verlag | 2024