Receive Date:

Revise Date:

Accept Date: Timely crack detection in concrete structures plays a crucial role in ensuring safety, enhancing durability, and reducing infrastructure maintenance costs. Despite the widespread use of traditional methods such as visual inspection and non-destructive testing, these approaches face limitations in accuracy, speed, and scalability. In this study, an artificial intelligence-based framework is proposed for crack detection by integrating visible (RGB) and thermal images. The fused images are processed using a Convolutional Neural Network (CNN), while a Generative Adversarial Network (GAN) enhances feature quality and training data diversity. Evaluation of six fusion ratios demonstrated that a combination of 65% RGB and 35% thermal imagery achieved the best performance. The proposed model attained an accuracy of 98.60% and a precision of 99.54%, outperforming approaches based solely on RGB or thermal images. Ablation studies confirmed that the GAN component contributed a 2.30% improvement in accuracy and a 2.20% improvement in precision compared to the baseline CNN. Furthermore, the results indicate that multimodal data fusion improves system robustness under varying environmental conditions. The proposed framework also demonstrates strong generalization capabilities. The findings highlight the potential application of the proposed framework in structural health monitoring systems and the inspection of bridges and concrete buildings, paving the way for intelligent inspection systems