Research Information System
Advances in Engineering Software, vol.207, 2025 (SCI-Expanded)
Durability cracks in concrete and mortar surfaces are caused by environmental, mechanical and chemical factors. These cracks can adversely affect the long-term durability and strength of concrete. Therefore, it is important to detect cracks in concrete and mortar surfaces at early ages and take precautions. In this scope, in this study, a Graphical User Interface (GUI) was developed for segmentation of cracks in concrete and mortar surfaces using deep learning models. The Alkali-Silica Reaction (ASR) test, which is a durability problem, was performed and a total of 600 images were taken on the test specimens at the end of the 28th day. The images were then augmented eight times using image augmentation techniques to generate a comprehensive dataset. Models with different U-Net network structures such as Classical U-Net, U-Net++, Attention U-Net, Residual U-Net (ResU-Net), Recurrent Residual U-Net (R2U-Net), LinkNet, Recurrent Residual Attention U-Net (R2AU-Net) and DenseU-Net were trained using this database and performance metrics were comparatively investigated. Among the eight different models, R2U-Net is the most powerful model with an F1-Score of 0.901, Intersection over Union (IoU) of 0.867 and AUC of 0.958. Then, to determine the test performance of the model, crack images were taken from different concrete and mortar surfaces and tested in the R2U-Net model. The results of the analysis showed a high accuracy rate between the predicted masks and the true masks. Finally, a GUI was designed to provide the crack segmentation mask image and crack area for easier observation and analysis of crack segmentation. It is expected that this interface will have important contributions to future applications of crack segmentation.