Date of Award
12-2025
Degree Type
Thesis
Degree Name
M.S.
Degree Program
Computer Science
Department
Computer Science
Major Professor
Dr. Tamjidul Hoque
Second Advisor
Dr. Abdullah Al Redwan Newaz
Third Advisor
Dr. Ben Samuel
Abstract
Levee encroachments are human made structures or vegetation growth that intrude upon levee systems and pose a significant threat to their structural integrity and effectiveness during the time of flood. Levee monitoring involves inspecting and evaluating encroachments that can include a variety of objects like fences, walls, concrete pipes, trees, large bushes, etc. along with other deficiencies. While there has been recent development in detecting levee deficiencies like sand boils and seepages with deep learning approaches, detecting encroachments is challenging due to the lack of training data and the fact that encroachments are defined by their position relative to the levee.
This thesis presents a synthetic data generation pipeline using Unity, a game development platform capable of rendering photorealistic images and procedural generation aimed to solve the issue of data scarcity of levee encroachments. 3D virtual environments were developed in Unity to simulate diverse landscapes under varying environmental conditions, lighting and encroachment types. With this system a large-scale dataset of labeled RGB images was produced, capturing variety of encroachment scenarios. To assess the validity of the generated synthetic dataset, selected deep learning models for semantic segmentation were trained on the dataset and evaluated. The results were compared with the same models trained on real dataset of levee encroachments. Comparison showed that models trained on the synthetic dataset had better Intersection over Union (IOU) scores and accuracy than those trained on the real dataset. This underscores the viability of virtual environments as scalable and controllable sources of training data, contributing to the advancement of automated levee inspection systems.
Recommended Citation
Katwal, Anav, "Synthetic Data Generation with Unity for Semantic Segmentation of Levee Encroachments" (2025). University of New Orleans Theses and Dissertations. 3314.
https://scholarworks.uno.edu/td/3314
Rights
The University of New Orleans and its agents retain the non-exclusive license to archive and make accessible this dissertation or thesis in whole or in part in all forms of media, now or hereafter known. The author retains all other ownership rights to the copyright of the thesis or dissertation.