Neuromorphic GAN-assisted Computing Architecture for Image Dehazing in Underwater Navigation

Abstract

This research aims to develop a real-time imaging system that enhances the visual quality and restores the true color of underwater image scenes. Images captured underwater suffer from low visibility due to the effects of absorption and scattering, resulting in haze and reducing the accuracy of underwater visual-based navigation and localization. To address these challenges, we propose a novel SpikingGAN-assisted computing architecture for image dehazing in the underwater environment. By developing and comparing two distinct methods for spiking-based image dehazing—weight transfer from conventional GANs and hybrid SpikingGAN direct training—we evaluate their performance using algorithmic and hardware metrics defined in this work. These methods can be then deployed on real-time hardware platforms, enabling more effective, energy-efficient underwater navigation through real-time image dehazing, supporting improved operational performance in challenging underwater conditions.

Date of Award	15 Dec 2024
Original language	American English
Supervisor	JORGE Dias (Supervisor)