Image Classification Based on Pixel Scattering Characteristics Using Freeman Decomposition

Polarimetric SAR data, due to its unique imaging mechanism, provides rich scattering information about ground targets. During the processing of polarimetric SAR data for the San Francisco coastline, pixel-based scattering characteristic analysis serves as the key to achieving accurate classification.

Freeman decomposition is a classical polarimetric target decomposition method that breaks down each pixel's scattering characteristics into three fundamental mechanisms: surface scattering, double-bounce scattering, and volume scattering. This decomposition approach enables deeper understanding of different ground objects' scattering behaviors, providing reliable feature basis for subsequent classification. In implementation, this typically involves processing the covariance or coherency matrix for each pixel and calculating the contribution percentages of each scattering mechanism.

After completing Freeman decomposition, the Wishart clustering algorithm is applied to cluster the decomposition results. Based on the complex Wishart distribution, Wishart clustering effectively utilizes the statistical properties of polarimetric SAR data to enhance classification accuracy. The method considers not only pixel scattering characteristics but also fully incorporates the coherency matrix information of polarimetric SAR data. Algorithm implementation often involves iterative clustering where pixels are assigned to classes based on minimum distance to cluster centers in the feature space.

The entire processing pipeline ensures that classification results preserve the polarimetric properties of SAR data while accurately reflecting scattering mechanism differences among various ground objects. The final classification map clearly distinguishes different ground types along the San Francisco coastline, such as water bodies, vegetation, and man-made structures, providing a reliable foundation for subsequent target identification and analysis. The classification workflow typically involves preprocessing, feature extraction using Freeman decomposition, and unsupervised classification using Wishart clustering with optimal parameter selection.