This paper focuses on the development of a new tool, ViSual_IceD, a convolutional neural network (CNN) for improved sea ice detection. This tool builds upon the classic U-Net architecture and uses concurrent multispectral and Synthetic Aperture Radar (SAR) imagery, overcoming limitations inherent in using either type of imagery alone. The study found that ViSual_IceD outperformed other models, providing more accurate sea ice coverage detection, particularly in coastal regions. This could be a significant step forward in operational monitoring and analysis of large numbers of images to determine changes in polar ice due to changing climatic conditions.
Publication date: 12 Jan 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2401.06009
