About OpenKnowledge@NAU | For NAU Authors

An evaluation of machine learning to detect recent moisture in street level images

Couey, Benjamin Hoak (2022) An evaluation of machine learning to detect recent moisture in street level images. Masters thesis, Northern Arizona University.

[thumbnail of Couey_2022_evaluation_machine_learning_detect_recent_moisture_street_l.pdf] Text
Couey_2022_evaluation_machine_learning_detect_recent_moisture_street_l.pdf - Published Version
Restricted to Repository staff only

Download (7MB) | Request a copy


Floods are major disasters which threaten communities across the globe, and predicting flooding is an importantpart of mitigation strategies. Flood prediction models benefit from having access to ground-level observations of where flooding is occurring and extending the lead time on these observations as much as possible is also valuable. This project complements the FloodAware research project, which is based on using image processing and existing networks of static cameras common in modern urban environments to determine water levels around a city. Where FloodAware focuses on detecting standing water, this project seeks to explore whether a machine learning model can determine the presence of significant moisture, indicative of recent rain, in the images gathered by these cameras to increase the lead time offered by the system. Because these cameras are static and observing a scene with a consistent perspective, this project presents a study to analyze if training a model for a single camera site could improve its performance, as well as address other practical questions regarding the implementation of this technique. To accomplish this, this project ran experiments on three convolutional neural network models of varying depth, a K-nearest neighbors model, and a linear classifier both across all camera sites and on a site by site basis. It found that these models achieved better performance when trained on a site by site basis, but that there are suggestions a general model could more easily be extended to entirely new sites.

Item Type: Thesis (Masters)
Publisher’s Statement: © Copyright is held by the author. Digital access to this material is made possible by the Cline Library, Northern Arizona University. Further transmission, reproduction or presentation of protected items is prohibited except with permission of the author.
Keywords: Flood forecasting; Flood management; Image classification; Machine learning
Subjects: G Geography. Anthropology. Recreation > GB Physical geography
NAU Depositing Author Academic Status: Student
Department/Unit: Graduate College > Theses and Dissertations
College of Engineering, Informatics, and Applied Sciences > School of Informatics, Computing, and Cyber Systems
Date Deposited: 08 Jun 2023 17:22
Last Modified: 08 Jun 2023 17:22
URI: https://openknowledge.nau.edu/id/eprint/5994

Actions (login required)

IR Staff Record View IR Staff Record View


Downloads per month over past year