Quantitative Analysis of Statistical Models for Typical Terrains in SAR Data

Many approaches to problem of detection and classification from the SAR image data are based on statistical models for the observations. The objective of this paper is to provide a method of modeling the SAR image. Firstly, the statistical properties of SAR image data are analyzed. The multi-model notion, which includes some different model fitting the different terrains, is introduced to describe accurately the local region in the large scene. The statistics for testing good-of-fit between the histogram of the sample data and the known distribution under the model assumptions are proposed. The validity of those assumed models for typical terrains are measured by the Radarsat data with 9-m resolution and X band SAR data with 0.5-m resolution. Finally, the good models matching with the typical terrains are collected as a reference