Analysis of Classification Models for Identifying Cracks in Concrete Structures

Purpose: The current study was used to propose a model for identifying cracks in concrete structures by comparing and analyzing various models using Convolutional Neural Networks (CNNs).BR/Methods: Two CNN-based classification models, VGG-16 and ResNet-50, were developed, compared, and analyzed. A confusion matrix was employed as a performance indicator to evaluate their performance in individual instances.BR/Results: The comparative analysis indicated that ResNet-50 outperformed VGG-16 in performance metrics. Additionally, the inference speed based on test data revealed a significant difference, with ResNet-50 requiring 35 seconds compared to VGG-16's 77 seconds.BR/Conclusion: The ResNet-50 showed excellent performance in confusion matrix-based performance indicators and inference speed. It shows strong potential for practical applications in identifying concrete crack structures in real-world scenarios.