A Novel Approach to Multi-Class Crack Classification: ResNet50 Enhanced with Squeeze-and-Excitation Blocks

This study focuses on introducing a novel multi-class crack classification framework for concrete structures by integrating Squeeze-and-Excitation (SE) blocks into the ResNet50 architecture. Unlike conventional models, which lack feature recalibration capabilities, the embedded SE block adaptively enhances channel-wise feature importance, significantly improving crack detection performance under complex conditions such as lighting variations and background noise. The proposed model is trained on a comprehensive dataset of 7,147 images, derived from the Özgenel dataset, the Concrete Crack Conglomerate Dataset, and additional images collected from buildings at Universiti Teknologi Malaysia (UTM). Experimental results demonstrate that ResNet50-SE achieves a remarkable accuracy of 99.88%, outperforming the standard ResNet50 (93.25%) and surpassing existing multi-class crack classification models. Furthermore, the optimized architecture ensures computational efficiency, making it well-suited for real-time structural health monitoring applications. These findings validate the effectiveness of SE-enhanced architectures in improving feature extraction and classification accuracy, marking a significant advancement in automated crack detection.