Unsupervised Functional Information Clustering in Extreme Environments from Filter Banks and Relative Entropy
Regroupement non supervisé d'informations fonctionnelles dans des environnements extrêmes à partir de banques de filtres et d'entropie relative
Atto, Abdourrahmane M. ; Karbou, Fatima ; Giffard-Roisin, Sophie ; Bombrun, Lionel
This chapter addresses feature extraction from wavelets and convnet convolutional neural network filters for unsupervised image time series analysis. It proposes the ability of wavelets and neuro-convolutional filters to capture non-trivial invariance properties, as well as the new centroid solutions, for high-level relative entropy-based feature analysis. The chapter considers unsupervised approaches without any requirement of the availability of training labels. Indeed, remote sensing imagery is a large source of images acquired in extreme geographical areas where collecting ground truths are usually very difficult. The chapter deals with a model selection framework summarizing the feature extraction. The texture analysis is performed over small image patches that are normalized to have zero-mean in the multivariate Gaussian case and unitary scale parameter in the Weibull-based modeling. The method proposed for functional image time series clustering relates to finding similarities between pixel evolution paths.</p>
Accès à la notice sur le site du portail documentaire de Météo-France