A BERT-based Model for Cloud Filling from Remote Sensing Data

Remote Sensing data are commonly used in analysis of geographical characteristic. One of the most frequent problems in Remote Sensing is the loss of data due to cloud-covered pixels. While traditional approaches took inspiration from DINEOF, recent advancements in Deep Learning and Machine Learning prompted a new paradigm to this well-known problem. In this paper, we proposed a Bert-based model for the cloud filling task named RoBERTaCF. Our method is compared to a recent Funk-SVD and our experiments indicated that RoBERTaCF achieved better performance in filling cloud scheme from remote sensing data.