Marine mucilage detection from satellite images with instance segmentation

In recent years, the escalating sea surface temperature, exacerbated by global warming, has given rise to numerous issues. One of these challenges is the proliferation of marine mucilage. The first occurrence of mucilage in the Sea of Marmara was observed in the 1990s. However, in May 2021, it resurfaced in the Sea of Marmara, intensifying in the region and evolving into a significant problem threatening both marine natural life and the economy. Mucilage, spreading across extensive areas on both the surface and depths of the sea, poses a serious threat to the ecosystems in this area. Early detection of mucilage formation is crucial to accelerate intervention and prevent its further spread. This study focuses on a portion of the Sea of Marmara due to the heightened mucilage activity observed in the year 2021. Both with and without Mucilage-infested images were obtained from the Planet's satellite archive between March 14, 2021, and July 4, 2021, to create the dataset. The acquired images were manually labeled to suit the model training requirements. Using this dataset, a model was trained and analyzed, based on the YOLOv5 network, a version designed for segmentation to automatically detect mucilage. The analysis of the designed model revealed promising performance in mucilage detection.