Land cover classification using satellite data is one of the most important applications of remote sensing. A great deal of ground information is usually required to generate high quality land cover classification. However, in complex natural areas, collecting information on the ground can be timeconsuming and extremely expensive. Nowadays, multiple sensor technologies have gained great attention in land cover classification. They bring different and
complementary in formation spectral characteristics that may help to overcome the limitations caused by inadequate ground information. In our research, we focus on the fusion of heterogeneous information from different sources. The combination system aims to solve the prob lems caused by limited labeled samples and can thus be used in land cover classification for hard to access areas. We propose a new fusion framework based on belief
functions that can combine heterogeneous information and manage uncertainty. This fusion fr amework includes a transformation
method that can transfer heterogeneous information in the same framework and an iterative fusion strategy that takes the most reliable information from each classification result and to combine in a cautious way. We evalua te this fusion framework on real remote sensing data and it shows satisfying performance to improve the accuracy of land cover classification when training samples are limited or mislabeled. 

The project initiative MSTD aims to develop vehicle perception systems based on deep learning techniques/multisensor data fusion making it possible to reliably detect nearby vulnerable people I am highly interested in the applications related to autonomous driving especially object detection/segmentation in urban environments and multimodal information fusion. I also enjoy broadly collaborating with multi partners in different national/ international environments. 

Na Li, PhD en informatique (Université de Rennes 1), est Ingénieure en Systèmes interactifs et robotique (Centralesupélec)

Encadrant : Franck Davoine (CR, Heudiasyc)

Laboratoires impliqués dans le post doc :

Heudiasyc - Heuristique et Diagnostic des Systèmes Complexes
ISIR - L'institut des systèmes intelligents et de robotique