Classical radar imaging generally considers targets as set of isotropic independent sources with a constant response in the measured frequency band. Nevertheless, new radar capabilities, in terms of signal bandwidth and angular excursion, may challenge this bright point model. Studies based on multidimensional time-frequency (TF) analysis, describing the angular and frequency behaviour of a scene-s reflectivity, showed that some scatterers may have anisotropic and dispersive responses. Polarisation diversity is an interesting additional source of information in radar imaging, and provides indicators closely linked to some geometric and electromagnetic properties of the observed objects. In this study, a fully polarimetric TF analysis is proposed for radar imaging (SAR, ISAR) that characterises the anisotropic and dispersive behaviour of the polarimetric response of deterministic targets. This method is based on the hyper-image concept, which describes the response of scatterers as a function of the observation angle, the emitted frequency and polarimetric canonical behaviours. Polarimetric hyper-images point out that non-stationary behaviours can be related to physical properties of the target (geometrical shape, relative orientation) and allow a better understanding of the scattering mechanisms. This polarimetric hyper-image representation is then used to detect non-stationary scatterers and to classify their behaviour.