1：Experimental Design and Method Selection:

The methodology involves using PolInSAR data to define a 29-dimensional feature vector based on H/A/α decomposition parameters, optimum coherence values, and span values. Two discrimination functions (H/A/α filter banks and probabilistic feature fusion) are proposed and trained with feature vector data to discriminate change types such as tree (TRE), low return (LRT), and ground (GRD).

2：Sample Selection and Data Sources:

PolInSAR image sets were collected in central New Mexico, USA, using the FARAD PhoeniX radar system on a DHC-6 airplane. Training and test data were selected from 15 image sets, with specific pixels chosen for each change type.

3：List of Experimental Equipment and Materials:

The primary equipment is the FARAD PhoeniX radar system (9.6-GHz center frequency) operated on a DHC-6 airplane. Data processing involved geometric matching, coregistration, and interferometric processing of SAR images.

4：6-GHz center frequency) operated on a DHC-6 airplane. Data processing involved geometric matching, coregistration, and interferometric processing of SAR images. Experimental Procedures and Operational Workflow:

4. Experimental Procedures and Operational Workflow: Steps include data acquisition, image formation, calibration, computation of coherence estimates, formation of feature vectors, training of discrimination functions using selected training data, and application to test data for performance evaluation.

5：Data Analysis Methods:

Performance was evaluated using receiver operating characteristic (ROC) curves, confusion matrices, and pass matrices. Qualitative analysis involved visualizing change discrimination images.