Innovations in Marine Unmanned Vehicles: Image Quality Enhancement with Segmented Autoencoders

This paper introduces a novel image enhancement technique for marine unmanned platforms utilizing segregated autoencoders. Given the limited computing resources on board unmanned aerial vehicles and communication interference, an optimized solution is proposed that finds a compromise between data compression and reconstruction. The system employs an autoencoder initially trained as a unified model, which is later split, with the encoder part deployed on the unmanned vehicle and the decoder at the receiving station. This approach allows for efficient data transmission and subsequent image enhancement, enabling real-time, high-quality imaging under challenging marine conditions.