A Decision Tree Approach for Identifying the Optimum Window Size for Extracting Texture Features from TerraSAR-X Data

Synthetic Aperture Radar (SAR) texture is an important derived variable for improving land cover classification accuracy from SAR data. However, a number of factors affect the amount and quality of texture information obtained from radar data and these include: the window size, data type, the size of grey level quantisation, displacement and the look direction. The main aim of this study was to determine the optimum window size for the extraction of texture features from TerraSAR-X (TSX) data and to study the effect of different window sizes on the classification accuracy of the selected land cover types. A new approach based on Decision Trees (DTs) was explored to determine the optimum window size for SAR texture analysis and the results compared with those obtained using Transformed Divergence (TD) and Jeffries Matusita (JM) statistical distance measures. In all the three approaches, a window size of 11 by 11 was found to be the most appropriate. Generally, the classification accuracy increased with the size of texture window. However, in all the three approaches, there was little increase in classification accuracy beyond a window size of 11 by 11.