Automatic Concrete Surface Crack Recognition Using EfficientNetV2 Variants

Building health and structural integrity are important to ensure occupant safety and long-term durability. Monitoring and assessing building condition regularly ensures that potential problems are identified early. This research aims to develop and implement a deep-learning (DL) supported tool to examine the concrete surface images collected using digital camera. The data gathering forms the first stage and the chosen DL algorithm is used for feature extraction followed by the feature selection and then fusion is done and finally a fivefold cross verification schema is applied for a binary classification. This work considered the EfficientNetV2 (ENV2) model with variants like small, medium and large for the examination. The investigation is implemented using individual and fused deep-features and the performance of the developed tool is confirmed using the achieved accuracy. The outcome of this research confirms that the proposed tool produces a betterment of >96% accuracy with individual-features and 100% accuracy with fused-features. These results confirm that the proposed scheme provides significant result during the crack detection.