Date of Award

Fall 12-2025

Document Type

Doctoral Project

Degree Name

Doctor of Engineering (D Eng)

Department

Electrical & Computer Engineering

Program/Concentration

Electrical and Computer Engineering

Committee Director

Khan Iftekharuddin

Committee Member

Jiang Li

Committee Member

Mecit Cetin

Committee Member

Norou Diawara

Abstract

Reliable flood-level estimation using aerial UAV (Unmanned aerial vehicle) imagery is essential for effective post-disaster assessment and rapid emergency response. This study utilizes a UAV-based dataset, referred to as the UVA dataset, which integrates multiple public datasets and manually labeled UAV images, together with a multi-stage vehicle-centric framework for flood-depth estimation. The UVA dataset integrates images from multiple public sources, including Unmanned Drone Water Assessment (UDWA), Unmanned Aerial Vehicle Detection and Tracking (UAVDT), Car Parking Lot (CARPK), and additional UAV-view images collected from the internet, followed by manual annotation for water depth, viewing angle, and altitude labels. In the proposed framework, vehicles are first detected and cropped using a Mask R-CNN model. Each cropped vehicle image is then standardized through zero padding and size normalization to preserve geometric consistency. Three deep learning models are subsequently trained: (1) a view classification model to identify suitable perspectives, (2) a height classification model to filter out high-altitude samples, and (3) a water-depth classification model that incorporates Sobel edge detection and attention mechanisms, where Sobel reveals boundary features and attention strengthens the network’s focus on these discriminative regions. The first two models act as pre-filters, removing images unsuitable for depth estimation, thereby enhancing the performance of the final classification. Experimental results show that the Sobel–Attention model achieves consistent improvements over the baseline classifier, and that the proposed view and height filtering stages further enhance flood-depth estimation by reducing the influence of unsuitable perspectives and high-altitude imagery. The UVA dataset and the multi-stage pipeline therefore offer a practical and effective framework for UAV-based flood-scene analysis under diverse real-world conditions.

Rights

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

DOI

10.25777/5k5r-je24

Download

Share

COinS