CNN Based on Transfer Learning Models Using Data Augmentation and Transformation for Detection of Concrete Crack

Cracks in concrete cause initial structural damage to civil infrastructures such as buildings, bridges, and highways, which turn causes further is thus regarded a serious safety concern. Early detection of it can assist preventing enable advance by avoiding any possible accident caused while using those infrastructures. Machine learning-based gaining favor over time-consuming classical approaches that only fulfill the objective early detection. To identify surface cracks from images, this research developed transfer learning approach (TL) based on Convolutional Neural Networks (CNN). This work employs strategy leveraging four existing deep (DL) models named VGG16, ResNet18, DenseNet161, AlexNet with pre-trained (trained ImageNet) weights. validate performance each model, indicators are used: accuracy, recall, precision, F1-score. Using publicly available CCIC dataset, suggested technique outperforms testing accuracy 99.90%, precision 99.92%, recall 99.80%, F1-score 99.86% for crack class. Our validated an external BWCI, Kaggle. achieved 99.60%, 99.90% respectively. proposed method, CNN demonstrated be more effective at detecting structures also applicable other tasks.