Electrical and Computer Engineering ETDs

Publication Date

Summer 8-1-2022

Abstract

This research focuses on implementing four different applications of machine learning on images. The various categories of digital images considered for these applications are grayscale, RGB, and infra-red images. The first framework uses an unsupervised learning strategy for detecting fire and smoke from an infra-red image dataset. This problem was solved using a classical machine learning algorithm since the dataset was small and unlabeled. Next, a semi-supervised deep learning model was used for facial expression recognition. Here we detect emotions from a moderately large dataset containing labeled and unlabeled grayscale images. The third application focused on single image superresolution, which was developed to increase the resolution of the input RGB image. The approach used deep learning due to the availability of sufficiently large, labeled data. The final structure consisted of a multimodal deep learning-based system for solar irradiance forecasting for cloudy days. The data used for these former applications consists of only images, whereas this work uses multimodal data, i.e., time-series data and infra-red images. Therefore, this research aims to consolidate heterogeneous, disconnected data from various sources to produce robust predictions. The performance evaluations of all the algorithms were done by comparing them with the state-of-the-art methods, and it was found that the proposed methods outperform the standard models.

Keywords

deep learning, machine learning, image processing, computer vision

Document Type

Dissertation

Language

English

Degree Name

Electrical Engineering

Level of Degree

Doctoral

Department Name

Electrical and Computer Engineering

First Committee Member (Chair)

Manel Martinez-Ramon

Second Committee Member

Sandra Biedron

Third Committee Member

Ali Bidram

Fourth Committee Member

Huiping Cao

Download

Share

COinS