Automatic-Zooming-Type Window Size Optimization for Polsar Data Interpretation

In our previous works, we have found the optimal window size for ALOS2-PALSAR2 data is 7 ×7 (or other window size settings including around 50 pixels). Such an optimal window size is fixed for the whole observation area. However, the optimal window size for different types of targets are actually different. In order to improve general interpretation performance, we propose a “target-wise” window size optimization algorithm. The process is like automatically zooming and finding the particular optimal window sizes for different target areas. The results show that the proposed algorithm has higher interpretation performance than fixed-type window size optimization.