A New Paradigm for the Exploitation of the Semantic Content of Large Archives of Satellite Remote Sensing Images

This paper presents a new paradigm for extracting information from large databases of remote sensing images. It aims at improving any task applied to image time series by exploiting properties related to their temporal cross-dependence. Images part of the same time series are casually related to each other. As a consequence, the results of the tasks are mutually entangled. The proposed paradigm exploits this property and validates the results of the tasks one to each other to improve the overall performance. The paradigm is general and has relevant implications in Big Data analysis because it is suitable to archives containing not only Earth Observed images but any time-varying quantity or feature. Preliminary results show that change detection accuracy improves after the evaluation of the conservative property within the image time series.