Thematic Spatiotemporal Association Rules to Track the Evolving of Visual Features and Their Meaning in Satellite Image Time Series

Abstract

Satellite Image Time Series (SITS) is a set of images taken from the same satellite scene at different times. The mining of SITS is challenging task because it requires spatiotemporal data analysis. An example of the need for SITS mining is the analysis of solar flares and their evolving. Thematic Spatiotemporal Association Rules (TSARs) are associations that show spatiotemporal relationships among the values of the thematics attributes. By employing TSARs, we propose an approach to track the evolving of visual features of SITS images and their meaning. Our approach, called Miner of Thematic Spatiotemporal Associations for Images (MiTSAI), considers the data extracting and transformation, the thematic spatiotemporal association rule mining (TSARs), and the post-processing of the mined TSARs, that relate the visual features and their meaning. Our experiment shows that the proposed approach improves the domain expert team understanding of Solar SITS. Moreover, MiTSAI presented an acceptable time performance being able of extracting and processing TSARs using a long period of historical data faster than the period needed for the arrival of new data in the database.