Author(s): Jiwan Hong; Young Ho Kwon; Sehoon Oh; Joon Heo
Linked Author(s):
Keywords: Dam Crack Detection UAV-based RGB Imagery Deep Learning Comparative Analysis
Abstract: This research proposes a robust framework for dam crack detection by integrating UAV-acquired RGB images into existing open-source datasets. First, the performance of existing studies on dam crack detection was thoroughly analyzed to establish a comprehensive baseline. A comparative review of state-of-the-art methodologies was conducted to identify key challenges and limitations in crack detection, providing insights for designing an improved framework. Subsequently, the dataset was augmented with high-resolution UAV-based RGB data, enhancing the model's ability to accurately identify fine crack features. A deep learning model tailored specifically for dam crack detection was designed by incorporating hierarchical attention mechanisms and multi-scale feature extraction techniques. These innovations strengthened the model’s robustness, enabling consistent crack detection under varying environmental conditions. Experimental results demonstrated that the integration of UAV-based data and the specialized model design significantly improved crack detection accuracy and robustness. This study is expected to advance the development of reliable deep learning-based monitoring systems for dam structural integrity.
Year: 2025