2025 Volume 4 Issue 1 Pages 285-293
In this study, the impact acoustics method was applied during pavement replacement work on RC slabs of highway bridges. The scalograms obtained were automatically classified using a defect classification model based on a pre-trained convolutional neural network constructed with RC slab specimens. The VGG19_bn model served as the pre-trained model, with fine-tuning employed during training. Classification of the scalograms obtained from the RC slabs achieved an accuracy of 98.4% for sound parts, while the accuracy for defective parts was lower, at 46.8%. This highlighted challenges regarding the accuracy of the defect classification model, particularly in identifying defective areas. In the future, improvements are necessary to ensure sufficient accuracy by incorporating data from RC slabs of highway bridges into the model construction process.
