The discrepancies caused by different cluster merging algorithms in fully polarimetric SAR classification

The discrepancies caused by different cluster merging algorithms in fully polarimetric SAR classification are analyzed here. There are two often-used merging schemes, i.e., merging first to desirable cluster numbers and then iterative clustering and, the agglomerative hierarchical clustering, both using three different between-cluster distance measures herein. One sub-image of RadarSat-2 SAR SLC image is used here. The results illustrate that (1) The choice of between-cluster distance measures in merging scheme one affects the merging results obviously. (2) the agglomerative hierarchical clustering, the merging scheme two can significantly alleviate these discrepancies caused by different between-cluster distance measures and get almost the same merging results. (3) the agglomerative hierarchical clustering also will gain the stability of merging sequence when Pct is small enough.