Abstract: The detection of floating small target in sea clutter is a hotspot and difficult problem in radar community. The traditional adaptive coherent integration detection methods cannot detect these targets effectively because of the weak energy and overlapping Doppler domain with clutter. The target detection based on characteristic is an effective method to solve sea floating small target detection problem. In this paper, firstly the time domain and frequency domain characteristics (the relative average amplitude, the relative Doppler peak height and the relative vector-entropy) are extracted from four different polarization channels. Then they are fused to four-polarization-channels characteristics in polarization dimensionality. Finally the fast convexhull learning algorithm is used to obtain the decision region of sea clutter and detect targets' returns. The experiments based on real datasets show that the new method is effective and has a good detection performance.