Predicting characteristics of cracks in concrete structure using convolutional neural network and image processing

Qayyum, Waqas; Ehtisham, Rana; Bahrami, Alireza; Mir, Junaid; Khan, Qaiser uz Zaman; Ahmad, Afaq; Özkılıç, Yasin Onuralp

doi:10.3389/fmats.2023.1210543

Cited by 4 publications

(1 citation statement)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They employed the SVM as a substitute for the SoftMax layer to improve sorting abilities and achieved an approximate 86% accuracy rate. The performance of pre-trained CNN models for classification depends on the number of images used to train the models [29].…”

Section: Literature Reviewmentioning

confidence: 99%

Classification of Wall Defects for Maintenance Purposes Using Image Processing

Qayyum,

Ehtisham,

Plevris

et al. 2023

Proceedings of the 8th International Conference on Computational Methods in Structural Dynamics and Earthquake Engineering (COM

View full text Add to dashboard Cite

Stakeholders are increasingly interested in finding efficient methods for regularly surveying and reporting on the state of their building assets, focusing on accuracy, consistency, and ease of use. High-rise buildings can exhibit wall surface defects and flaws such as cracks and spalls, which can significantly affect the structures' safety and appearance. Such problems need to be taken care of in a timely manner, before they become too hazardous or costly to fix. In this research work, images of several types of wall damage are classified into three main categories: (i) Undamaged; (ii) Cracked; and (iii) Miscellaneous. In total, 6000 images were used in the dataset, equally subdivided into the three categories. In machine learning, convolutional neural networks (CNNs) stand out as a form of neural network that excels at image classification. A transfer learning approach was implemented to classify wall surface defect images using three pre-trained CNN models, namely ResNet-50, ResNet-101, and Inception V3. 70% of the data set was used for training purposes, and the remaining 30% was used for validation. Several metrics including accuracy, precision, recall, and F1-score were computed for each model, in an attempt to find the best model for the damage classification task at hand. According to the results, Inception V3 demonstrated superior performance compared to the ResNet-50 and ResNet-101 models, achieving an overall accuracy of 87.1%. In contrast, ResNet-101 and Res-Net-50 obtained overall accuracies of 85.3% and 78.3%, respectively. The suggested methodology offers several benefits and a clear potential for broader adoption in the future as it can significantly reduce the time and effort required for manual inspection and classification of defects, allowing for more efficient maintenance and repair processes.

show abstract

Section: Literature Reviewmentioning

confidence: 99%