1

This paper presents a new method using a Convolutional Neural Network (CNN) to estimate sea ice concentration in the Baltic Sea by combining Sentinel-1 and AMSR2 data. The CNN architecture retains different resolution inputs without losing valuable information. By using a focal loss function and skip connection, the proposed model achieves a high R2 score of 90.6%.

This study introduces short-term and long-term vessel trajectory prediction methods to enhance maritime traffic management. Tested with Baltic Sea AIS data, the short-term method shows high accuracy, while the long-term method optimizes speed and memory, promising improvements in efficiency and safety.

Presenting a novel CNN model merging SAR and AMSR2 imaging for precise maritime ice-water detection. This fusion enables automatic sea ice chart generation, outperforming uni-modal SAR architecture with 94.60% pixel-wise accuracy and an F1-score of 94.99%.