A Model for Sea Ice Segmentation based on Feature Pyramid Network and Multi-head Self-attention

Sea ice constitutes a crucial component within the marine ecological system, holding significant implications for both the marine ecosystem and the broader climate system. Abbreviated ice seasons facilitate in-depth exploration of polar regions. However, the harsh sea ice conditions significantly impact maritime navigation, offshore operations, polar scientific expeditions, and various activities in high-latitude zones. The application of sea ice image processing enables precise detection, monitoring, and prediction of sea ice, providing insights into its morphological dynamics. This deepens our understanding of the influence of sea ice on both climate and the environment, establishing a scientific foundation for climate change prediction and mitigation. Consequently, this paper introduces a model that utilizes the Feature Pyramid Network (FPN) and a Multi-Head Self-Attention Mechanism for sea ice segmentation. Results from the Arctic Sea Ice Image Masking dataset highlight the superior performance of our proposed model, achieving the most favorable outcomes compared to its counterparts.