AUTOMATIZING THE LARGE-SCALE ANALYSIS OF UNDERWATER OPTICAL DATA

Underwater monitoring provides essential information to analyze the current condition and persisting trends of marine habitats. The optical data acquisition is a powerful solution to ensure both high-resolution and large-scale sampling of the seafloor. The use of autonomous data-driven robotics is making underwater imaging more and more popular. Nevertheless, video and image sequences are a trustworthy source of knowledge that remains partially unexploited: the human visual analysis of images is a very time-consuming task, which creates a bottleneck between data collection and extrapolation. This thesis presents a human-in-the-loop software solution, based on deep learning methodologies and computer vision, suitable for supporting and speeding up the analysis of visual data coming from underwater environmental monitoring activities.