Aiming at weak sea-surface target detection, this paper proposes a detector based on feature selection and construction. Firstly, multi-dimensional features were extracted from multiple dimensions such as polarization domain, phase domain, time domain, frequency domain and time-frequency domain, and the SHAP value analysis method was used to quantitatively evaluate the importance of multi-dimensional features, and the three features with the highest importance were selected to reduce the dimension of the feature space and the computational complexity. Secondly, to fully utilize information from the polarization domain, the normalized importance of each feature under different polarization modes is used as a linear weighting coefficient for constructing polarization-based features. This effectively compresses high-dimensional polarization features into a 3D feature vector. Finally, anomaly detection using the XGBoost algorithm is applied on these obtained 3D features to obtain corresponding results. Experimental results on a measured database demonstrate that our proposed phase feature detector outperforms existing three-feature detectors.
