1.Azim Khan
Abstract:
Accurately perceiving and interpreting environmental data is crucial in many essential sectors, such as military operations, emergency response, and autonomous systems. Traditional approaches to environmental perception mostly rely on visual data, but these methods can falter in challenging conditions such as low visibility or adverse weather. The absence of recognizable visual characteristics, the presence of noise, low-temperature contrast scenes, sensor noise under low-light conditions leading to grainy visuals, and the loss of information inherent in images frequently impede their integration into deep learning models.
The Discrete Fourier Transform (DFT) offers an alternative by capturing important patterns and structures that may not be easily discernible in the spatial domain. To address these challenges, this thesis introduces a novel Low-Rank Discrete Fourier Transform (DFT) based method specifically designed for noisy images, encompassing both thermal and RGB images.
This research contributes to the field by developing innovative techniques that improve the contextual understanding of images. This enables better deployment of deep learning models in applications that need precise environmental perception. We validate our method within the context of deep learning-based situational awareness systems, aiming to demonstrate that our Low-Rank DFT approach maintains the critical features necessary for accurate perception and decision-making.
