Improving Satellite Imagery Segmentation Using Multiple Sentinel-2 Revisits

In recent years, analysis of remote sensing data has benefited immensely from borrowing techniques from the broader field of computer vision, including the use of shared models pre-trained on large datasets. However, satellite imagery has unique features that are not accounted for in traditional computer vision, such as the existence of multiple revisits of the same location. Here, we explore the best way to use revisits in the framework of fine-tuning pre-trained remote sensing models. As a test case, we focus on the applied research question of power substation segmentation, which is representative of applied uses of pre-trained models more generally. Through extensive tests of different multi-temporal input schemes across diverse model architectures, we find that fusing representations from multiple revisits in the model's latent space is superior to other methods of using revisits, including as a form of data augmentation. We also find that a SWIN transformer-based architecture performs better than U-Nets and ViT-based models. We verify the generality of our results on a separate building density estimation task. In total, this work equips remote sensing scientists with a general method for utilizing the unique temporal properties of their images.