Tri-dimensional vessel classification patterns inferred via single–pass Polarimetric SAR Interferometry
Gerard Margarit-Martin(1) , Jordi J. Mallorqui(1) , and Xavier Fabregas(1)
Universitat Politecnica de Catalunya (UPC),
CJordi Girona 1-3,
Single-pass polarimetric SAR interferometry (POLInSAR) has shown notable performances when applied to distributed targets, especially forested areas. The capability to distinguish geophysical parameters from volumes with a vertical distribution of scattering mechanisms has allowed feasibly estimate, among others, tree height information. Such information is essential to evaluate biomass in a global scale, an issue that it was not possible since the first POLInSAR systems were boarded in airborne platforms. The potentialities of POLInSAR techniques are not only restricted to distributed targets, but also they could be extended to other kind of targets, as for example ships. In this case, the possibility to have at one’s disposal the third coordinate of scatters’ location would represent a significant progress in the diverse classification approaches currently available. Among them, the most interesting one is the analysis of polarimetric SAR (POLSAR) data via Coherent Target Decompositions (CTD), which decomposes the complex polarimetric behaviour in a mixture of elemental scattering mechanisms that can be connected with specific parts of vessel’s structure. The distribution and significance of those mechanisms in SAR images could allow build a classification pattern.
However, despite the potentialities of this method, there are some limitations that could mismatch data interpretation. Certainly, CTD consider the same centre of phase for all possible scatters within a resolution cell. For those situations where this assumption is not accomplished, CTD provide unpredictable results that are extremely dependent on the observation conditions and can not fit the observed geometry. In vessel SAR images, this situation is unfortunately very often due to the ships’ geometry and the resolution provided by current sensors. But even more worsening are the azimuth distortions induced by the vessel motions due to ocean waves. Indeed, the sea state can change the target’s geometry observed by the sensor almost in a random way causing significant mismatches in the polarimetric information. This situation avoids the practical usage of POLSAR images for ship classification.
An efficient way to overcome those drawbacks lies on basing classification in a quantitative measurement of target’s geometry far from the qualitative analysis done in the previous case. This could be achieved by inferring scatters’ height via single-pass interferometry. In this sense, the current paper presents a simple but robust approach for vessel classification based on combining polarimetry with across-track interferometry. It consists on derive height maps for each channel of the Pauli decomposition, which could be also able to retrieve sub-pixel heights for the same cell under specific conditions. The selection of this theorem is motivated by the orthogonality of its basis, its simplicity and the polarimetric scattering studies done with simulated images that show the most important scattering centres of most vessels could be summarized by trihedral and dihedral behaviours. The performances of this method will be tested with an orbital high-frequency POLInSAR simulator for different vessel models and scenario configurations. Special attention will be paid for the sea state and its influence in data interpretation. The results will show the global height map derived by combining the information of each Pauli channel is more robust against the observation conditions, specially the distortions of ocean waves and, in turn, it can feasibly be used as a classification pattern.