Supervised landmask estimation using contextual information in SAR data

Synthetic Aperture Radars are powerful observation tools in cases where the utilization of optical data is restricted. As one of the main applications of these systems is the control of maritime traffic, a land mask needs to be estimated. In this paper two different processing schemes are proposed in order to perform the land mask estimation on a TerraSAR-X acquired SAR image. The first one consists on an unsupervised edge detector based on the wavelet transform modulus maxima, while the second one performs a supervised detection based on SVMs. Both processing schemes apply a blocktracing algorithm after the edge detection stage. The edge detector based on the wavelet transform finds quite a lot of edges over the sea area, missclassifying a big region of water as land. Thanks to contextual information and the supervised training, the edge detector based on SVMs can outperform the edge detector based on the wavelet transform in the classification of sea areas obtaining a better landmask.