Deep learning for Satellite Image Time Series Analysis
Keywords: Satellite Image Time Series, Deep learning
Abstract: An increasing number of satellites measure Earth’s surface properties at high spatial resolution in regular and frequent time intervals. This creates an unprecedented quantity of satellite image time series data that describe the Earth’s surface dynamics. Extracting knowledge from this massive volume of complex data opens opportunities and poses challenges to the methods. The spatial, spectral and temporal characteristics of satellite data require models to learn relevant and discriminative features for a given use-case. Additionally, the lack of labelled data makes classical hard-supervised learning difficult. Scaling-up methods to global scale imposes further constraints to the model design.
This ISPRS 2021 Thematic Session focuses on the use of deep learning techniques originated from related fields, like computer vision (e.g., convolutional neural networks) or natural language processing (e.g., self-attention models, recurrent neural networks) that have proven to be useful for satellite image time series analysis. We welcome method contributions for various applications including vegetation modeling, land cover land use mapping and change detection, climate forecasting, and precipitation nowcasting,