Date of Award
Engineering and Applied Science - Computer Science
Hoque, Md. Tamjidul
Newaz, Abdullah Al Redwan
Levees are earthen structures constructed to mitigate flooding in low-lying areas. Although levee systems can reduce flood risks, they cannot completely eliminate them. Failures within flood control systems due to inadequate maintenance or strong water currents can lead to significant property damage and catastrophic loss of life, as was seen during Hurricane Katrina. Consequently, regular inspections are essential to identify and address any issues with the levees promptly. However, current inspection methods rely on manual techniques that are time-consuming, labor-intensive, and prone to human error. Therefore, this study proposes using deep learning models for more efficient and frequent assessment of levee systems.
The research suggests employing deep-learning architectures to detect three primary deficiencies: cracks, sand boils, and seepages using a semantic segmentation approach. The proposed models focus on lightweight designs, incorporating transfer learning techniques and enhancing feature representation to improve the accuracy and efficiency of fault segmentation using images within flood control systems. Moreover, these models are evaluated with real-world data specific to levees, allowing for their effectiveness in handling various faults exhibiting variations and complexity. As a result, the research enables pixel-level fault detection for AI-based inspection of levee systems, providing valuable support to existing assessment procedures. Furthermore, the study's findings have the potential to extend beyond levee inspection, with implications for broader applications of deep learning in semantic segmentation for real-world scenarios.
Keywords: Levee; Faults; Deep learning; Images; Detection; Segmentation
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Panta, Manisha, "Faults Segmentation in Levee Systems Using Deep Learning Approaches" (2023). University of New Orleans Theses and Dissertations. 3107.
Available for download on Saturday, August 01, 2026