The ASSIST project focuses on developing advanced self-supervised deep learning techniques for the analysis of side-scan sonar images, with the goal of real-time seafloor mapping.
By using Vision Transformers and multimodal data that combine acoustic and optical sensors, the project aims to overcome current limitations that require tedious manual annotation for seafloor classification.
ASSIST explores self-supervised learning strategies to reduce the dependence on human annotations, while also seeking to improve the semantic classification of different types of seabed. Furthermore, the system is expected to be integrated into an autonomous underwater vehicle (AUV), the Girona 1000, enabling autonomous planning and navigationin adaptive missions.