Knowledge-based approach for VHR satellite image time series analysis

As satellite data volumes are growing thanks to technological evolution, there are more needs to automatic approaches for Satellite Image Time Series (SITS) analysis. In this article, we propose a new approach based on experts knowledge for land cover monitoring. This approach helps geoscientists to overcome direct interpretation difficulties by modeling expert knowledge. As a first step, using predefined nomenclature and a prior information, concepts are set. Then, each SITS region temporal evolution, represented by a graph, is assigned to the most similar reference one in the knowledge database using the marginalized graph kernel similarity measure.