Enhancing Posidonia Oceanica Meadows Monitoring Through Semantic Occupancy Mapping Strategy

Posidonia Oceanica (PO) meadows play a crucial role in marine ecosystems, providing essential services such as carbon sequestration, coastal protection, and biodiversity support. Monitoring the distribution of these meadows is vi-tal for conservation and management efforts, yet traditional methods often face limitations in scale, accuracy, and temporal resolution. This paper introduces a novel semantic occupancy mapping strategy to enhance the monitoring of PO meadows. By integrating data coming from the sensors mounted on board of an Autonomous Underwater Vehicle (AUV) with machine learning algorithms, the proposed strategy allows for the generation of semantically enriched maps that capture the spatial extent of PO meadows. The method employs semantic segmentation to distinguish between areas containing PO and areas with other subjects (for example, rocks, sand, and other plant species). These data are fused with an approximate bathymetry provided by the elaboration of the measurements coming from a Doppler Velocity Log (DVL), to obtain a 3D semantic reconstruction of the inspected environment. The elaboration of data acquired from a coastal area near Cecina, Tuscany (Italy) demonstrate the efficacy of the approach, highlighting its potential to improve monitoring accuracy, to evaluate the growth/decrease of the meadows over time and contribute to the long-term preservation of these vital ecosystems.