Exploration of Resnet Variants in High Spatial Resolution Domain Adaptation

Abstract. When mapping land cover from airborne to spaceborne data, a problem arises, where the difference in sensors between the two shows a large spatial resolution inconsistency and spectral differences. Consequently, the same object may exhibit completely different features. This problem causes models trained from annotated airborne to be ineffective when applied to spaceborne. Cross-Sensor Land-COVER (LoveCS) shows good results in overcoming this problem. LoveCS leverages small-scale aerial image annotations to promote land cover mapping on large-scale spacecraft. LoveCS uses ResNet50 as its encoder. In recent years, many studies have tried to develop other variants of ResNet, such as ResNeXt, ResNeSt, Res2Net, and Res2NeXt. This variation of ResNet turned out to give better results in a variety of tasks compared to ResNet. Therefore, in this study we modified the LoveCS encoder by replacing ResNet50 with ResNet variants such as ResNeXt, ResNeSt, Res2Net, and Res2NeXt in an effort to improve LoveCS accuracy. We also offer LoveCS schemes with better accuracy based on the best encoders. Our evaluation shows that Res2Net50 as an encoder is able to improve LoveCS performance where the average F1 increases by 1.38%, OA by 1.96%, and Kappa by 2.75% from the baseline method.