Evaluating the Effect of Incident Angle on Sea Ice Classification in SAR Images Based on a Deep Learning Model

Abstract

Accurate classification of different types of Arctic sea ice is crucial for safe marine navigation. Synthetic aperture radar (SAR) images are widely used for this purpose, but the backscattering coefficient intensity, which is critical for sea ice classification accuracy, is influenced by the incident angle (IA) of the SAR image. In this study, we investigated the impact of SAR IA on sea ice classification using a U-Net deep learning model. We collected 14 Sentinel-1 A/B Extended Wide (EW) mode images as testing datasets and conducted sensitivity experiments to compare the accuracy of sea ice classification with and without IA input, as well as with SAR images after IA correction. Our results indicate that the highest classification accuracy was achieved with SAR images that underwent IA correction as the model's input. Therefore, it is essential to correct the SAR images with IA in the sea ice classification model based on deep learning to improve the accuracy of sea ice type identification.