Abstract
Earthen embankments are built to prevent flooding and protect communities from the dangers of floods and high water levels. However, these geotechnical structures may not always remain serviceable and can fail due to long-term seepage and ponding. For instance, erosion causes the earthen structure to weaken and eventually fail, which may be due to several factors, including the velocity of the water, soil water characteristics, fine content, and gradation of the soil. The presented research explores an advanced approach to address the critical issue of identifying the seepage and ponding through the embankment by assimilating the passive infrared thermographic imageries with Deep Learning (DL) algorithms. To facilitate the development and validation of developed DL frameworks, a physical experimentation setup at the model scale is developed. This platform enabled the generation of a comprehensive dataset of thermal images across various environmental scenarios, including vegetation coverage and rainfall. Multiple DL frameworks were initially explored within the framework and the models were designed to process sequences of thermal images and predict the extent of seepage and ponding. This research builds upon effectively transforming the complex task of embankment leakage identification into an image classification problem. Moreover, the developed framework demonstrates that mapping of seepage and ponding can be achieved with great accuracy and is vital in enhancing embankment safety and disaster prevention strategies in flood-prone areas.