報(bào)告人：Thomas Corpetti 教授

　　時(shí)間：2017年12月20日 星期三 下午14:00

　　地點(diǎn)：遙感地球所奧運(yùn)園區(qū) A503

　　主要內(nèi)容：Hyperspectral images are commonly used in the context of earth observation.

　　Among classical problems in the analysis of hyperspectral images, the unmixing one, which aims at solving the inverse problems of determining the composition of pure elements inside a pixel, is crucial. Today, large image sizes, strong non-linearities in the mixing and presence of outliers are still open problems in unmixing algorithms. In this talk we present a new method (SAGA+) that scales favorably with mentioned problems. It performs an unsupervised unmixing jointly with outliers detection capacities, and has a global linear complexity. Non linearities are handled by decomposing the hyperspectral data on an overcomplete set of spectra, combined with a specific sparse projection, which guarantees the interpretability of the analysis.

　　報(bào)告人簡(jiǎn)介：Thomas Corpetti is a Senior Researcher with CNRS (French national center for scientific research). His main research interests concern remote sensing image processing for environmental applications. He has mainly been involved in mixing physical models with computer vision during the past 10 years (data assimilation for wind estimation, turbulence analysis, super-resolution, crop monitoring). A the moment, his research interests mainly concern unmixing and urban vegetation monitoring with various high resolution sensors (LiDAR, PLEIADES, Sentinel-2). He works with LETG (littoral, environment, remote sensing and geomatics) Lab and from 2009 to 2012, he was in Beijing, China, working with LIAMA (sino- french lab on computer sciences and applied mathematics).