Many scholars have conducted a number of studies on classifying PolSAR images by the semi-supervised classification algorithm.
Therefore, this study focuses mainly on the issue of hardness when selecting training samples and low classification accuracy in SAR images; research for the classification scheme, which is similar to the semi-supervised algorithm carried out on PolSAR images; and a new classification scheme proposed by combining the improved adaptive region growing algorithm with the Wishart maximum likelihood classification algorithm.
Semi-supervised learning follows the state of PolSAR image classification.
The Wishart distance is used to measure the distance between each unknown pixel in the whole PolSAR image and each cluster center, the unknown pixel is assigned to the class with the smallest distance.
Using the trial-and-error approach to compare different information extracted in the entire PolSAR image (including span, entropy, scattering angle, anisotropy, and so on), the main diagonal elements of the polarimetric coherency matrix are found to have apparent differences between various land cover types and the background.