seminar Inria F107 - 11am, December 17th (tomorrow) - Hernan Dario Benitez Restrepo

[Julien Mairal]

Dear all,

tomorrow, we will have the pleasure to welcome Hernan Dario Benitez Restrepo from Pontificia Universidad Javeriana Sede Cali, Columbia.

The seminar will take place at 11am, F107, Inria, and will be about

Change detection and temporal interpolation in multispectral images

Change detection (CD) refers to the task of analyzing remote sensing images acquired over an area of interest at different times, which allows to quantify the magnitude of a natural disaster (i.e flooding) or changes generated by human activities. Issues such as the sensitivity to multispectral image noise and the need for high accuracy estimation of the difference image probabilistic distribution, limit current CD techniques based on machine learning and probabilistic thresholding. Furthermore, sensor malfunction and difficult

atmospheric conditions provoke information missing in optical remote sensing data that hinders interpretation. Also, Gaussian processes (GP) are well-known state-of-the-art methods to recognize patterns and predict data.  In this talk, we present a graph based data fusion approach applied to CD whose main contribution is the extension of the graph-based model originally developed for image segmentation. Furthermore, we show preliminary results on the application of GP in the prediction of missing data in multispectral satellite time series. The results of these novel approaches present performance comparable to state-of-the-art methods. 

Best,

Julien