Time-Series Analysis of Massive Satellite Images: Application to Earth Observation

GAIA team in GIPSA-lab, Grenoble

Date(s) : 04/04/2022   iCal
14 h 30 min - 15 h 30 min

This work takes place in the context of the processing of the Satellite Image Time-Series (SITS) data from Sentinel-2 mission, launched in 2015 by the European Space Agency. It focuses on the classification task, i.e. land use or land cover maps that can be produced using spectro-temporal aspect of the Sentinel-2 SITS.

The main difficulty is the acquisition noise (clouds, shadows) combined with the orbital path of the satellites resulting in irregular and unevenly sampled time-series. Conventional approaches re-sample time-series to a set of time stamps, then they use machine learning techniques to classify vectors at a large-scale (national scale). The main disadvantage of this two-step processing approach is that it increases the number of operations applied to the SITS, implying a more difficult transition to massive amount of data. To a lower extent, the re-sampling step may slightly alter the temporal characteristics of the data.

To tackle this concern, we introduce a novel model-based approach with the ability to classify irregularly sampled time-series based on a mixture of multivariate Gaussian processes using spectral and temporal information from SITS thanks to a Kronecker structure of the covariance operator of the Gaussian process. The model allows jointly the imputation of missing values and is scalable to large-scale data-sets. It is evaluated numerically on simulated data-sets to illustrate the importance of taking into account spectral correlation and on Sentinel-2 SITS in terms of classification and imputation accuracy and is compared with conventional approaches. Analyses of the results illustrate the relevance of the model and the benefit of using interpretable parametric models.


Retour en haut 

Secured By miniOrange