Sea Ice Segmentation From SAR Data by Convolutional Transformer Networks
This work addresses the need for automated, efficient sea ice monitoring for climate research, though it appears incremental as it combines existing convolutional and transformer methods.
The paper tackles sea ice segmentation from SAR satellite imagery by proposing a hybrid convolutional transformer network, achieving a mean intersection over union of 63.68% on the AI4Arctic dataset with an inference time of 120ms for a 400x400 km product.
Sea ice is a crucial component of the Earth's climate system and is highly sensitive to changes in temperature and atmospheric conditions. Accurate and timely measurement of sea ice parameters is important for understanding and predicting the impacts of climate change. Nevertheless, the amount of satellite data acquired over ice areas is huge, making the subjective measurements ineffective. Therefore, automated algorithms must be used in order to fully exploit the continuous data feeds coming from satellites. In this paper, we present a novel approach for sea ice segmentation based on SAR satellite imagery using hybrid convolutional transformer (ConvTr) networks. We show that our approach outperforms classical convolutional networks, while being considerably more efficient than pure transformer models. ConvTr obtained a mean intersection over union (mIoU) of 63.68% on the AI4Arctic data set, assuming an inference time of 120ms for a 400 x 400 squared km product.