Abstract : This PhD Habilitation is devoted to the analysis of time series of Low Spatial Resolution (LSR) and Very High Spatial Resolution (VHSR) remote sensing images. Events of interest are related to meteorology and oceanography (for LSR data) and to agriculture and urban applications (for VHSR data). The rate of acquisition of satellite data is inversely proportional to their spatial resolution. Therefore with LSR images, the cadence is very high (for instance one image every $15min$ with MSG --Meteosat Second Generation) and enables an analysis of the turbulent atmospheric flows observed through the motion of the clouds, the oceanic circulation, ... Related computer vision problems concern motion estimation, curve tracking and missing data interpolation. On the other hand, with VHSR images, the time between two data can be from several weeks to several months. The related studies are rather concerned with the definition of advanced change detection techniques to highlight the main structural changes between images. From a methodological point of view, we are based on the introduction of physical knowledge in computer vision tools in order to define specific techniques for the analysis of structures in LSR data. A large part is devoted to variational assimilation approaches. As for VHSR images, we propose original descriptors to characterize the textured areas and applied them for numerous problems (segmentation, classification, orientation estimation, textured front detection). Finally, a chapter is especially devoted to the change detection issue where we introduce techniques for binary and multi-label change detection.