A0521
Title: Statistical copulas approach for dependence in remote sensing problems
Authors: Cristiano Tamborrino - University of Bari (Italy) [presenting]
Abstract: Remote sensing data acquired from Earth by satellites and airplanes have become increasingly important in various environmental and ecological contexts (e.g. agriculture, oceanographic or urban areas) over the last decade. The constellations of satellites and sensors on board are numerous and with ever-increasing technologies. This allows the acquisition of Hyperspectral (HS), Multispectral (MS) and Synthetic Aperture Radar (SAR) images with a very high spatial and spectral resolution, moreover, it is possible to collect a large amount of historical data, with a very short time interval, in different areas of the planet. The analysis of this large amount of data requires ever more precise and fast methods that must take into account not only the dependence on the spectral characteristics of every single image, but also on the temporal ones. Copulas are an excellent statistical tool, capable of modeling joint distributions between any random variable. Recently, it has been seen that the use of copulas alongside the classic machine learning algorithms for classification, clustering or anomaly detection leads to a more precise and robust quantitative analysis even with respect to the latest developments with Neural Network architectures. We will apply this tool in different areas of remote sensing data analysis, the proposed approaches will be tested on hyperspectral and multispectral images and the results are compared with the most advanced methods.
