SAR image wetland classification method based on feature optimization

The invention discloses an SAR image wetland classification method based on feature optimization. The method comprises the steps of obtaining a polarization scattering matrix, a Pauli conversion vector and a coherence matrix or a covariance matrix of an SAR image; extracting co-polarization phase difference characteristics of the wetland ground objects under a single time phase by using the polarization scattering matrix; fully-polarized coherence optimization improvement features of the wetland ground features under different time phases are extracted by using the Pauli conversion vector; performing polarization decomposition on the coherence matrix or the covariance matrix to obtain polarization decomposition characteristics of the wetland ground features; constructing an initial feature set of the wetland ground features by using the features; optimizing the initial feature set by using a CNN model to obtain an optimal feature set; and classifying the optimal feature set by using a deep random forest metho