ORCID

0000-0002-8073-0817

Keywords

Image Dehazing, Near-Infrared, Haar Wavelets, Image Fusion, Deep Learning, Photovoltaic Power

Subject Categories

Artificial Intelligence and Robotics | Computer Sciences

Abstract

Weather-induced variability poses significant challenges to the reliability and performance of modern computational systems, particularly those relying on visual perception and environmental prediction. This dissertation focuses on enhancing computer vision and machine learning based predictive models that operate under varying atmospheric conditions. Two representative weather-impacted applications are investigated: image dehazing and solar photovoltaic (PV) power output forecasting. Image dehazing focuses on the restoration of clear, unobstructed visuals from hazy or foggy images, a task that is vital for various applications. On the other hand, photovoltaic (PV) power forecasting aims to predict future solar energy generation based on historical sky images and PV output data, which depends on variable atmospheric conditions such as cloud coverage, haze, fog, humidity, etc. In the first half of this dissertation, we investigate the dependency of haze on the spectral wavelength of light. First, we develop an enhanced dehazing method based on RGB and Near Infrared (NIR) images that improves visibility and contrast under heavy haze. To account for the limited availability of NIR images in practice, we then propose an end-to-end deep neural network based on the U-Net architecture, specifically designed for RGB images to address the challenges of non-homogeneous haze and color accuracy. The second half of this dissertation is focused on PV power forecasting under variable weather conditions, addressing fluctuations in solar irradiance due to transient effects such as moving clouds, but also persistent conditions such as humidity and haze. We propose a multi-horizon approach that improves existing deep learning-based forecasting models through joint optimization of minute-by-minute forecasting of PV output within the forecast horizon. While this dissertation takes initial steps in casting light on major challenges involved in weather-dependent prediction problems, it opens new opportunities by proposing future directions of connecting visual information to latent energy in light and its spectral dependency.

Completion Date

2026

Semester

Summer

Committee Chair

Foroosh, Hassan

Degree

Doctor of Philosophy (Ph.D.)

College

College of Engineering and Computer Science

Department

Computer Science

Format

PDF

Document Type

Dissertation

Language

English

Download

Share

COinS
 

Accessibility Statement

This item was created or digitized prior to April 24, 2027, or is a reproduction of legacy media created before that date. It is preserved in its original, unmodified state specifically for research, reference, or historical recordkeeping. In accordance with the ADA Title II Final Rule, the University Libraries provides accessible versions of archival materials upon request. To request an accommodation for this item, please submit an accessibility request form.