Experience Gained with Texture Modeling and Classificationof 1 Meter Resolution SAR Images

摘要

For many years, high resolution SAR (Synthetic Aperture Radar) imaging was limited to airborne instruments. Nowadays, the analysis of spaceborne high resolution SAR images with up to 1 meter spatial resolution has be-come possible with the advent of German, Italian, and Canadian missions and their subsequent data distribution. For instance, compared to previous missions with much lower resolution, the German TerraSAR-X data allow us to analyze SAR images containing an increased amount of details and information content. As a consequence, a robust detection and recognition of small scale man-made structures representing buildings, roads, harbors, bridges, etc has become a new challenging task. An important property of SAR data is the presence of speckle phenomena which, in most cases, precludes an automated interpretation of SAR images. Therefore, we use a Bayesian approach relying on models and their parameters to fit the data. We suggest an automated method being able to extract and interpret the genuine information contained in high resolution SAR images. Our solutions are provided for optimal processing both for visual and automated data interpretation. The image information content is extracted using model-based methods based on Gibbs Random Fields combined with a Bayesian inference approach. The approach enhances the local adaptation by using a prior model, which learns the image structure; it enables despeckling with minimum loss of resolution and simultaneously estimates the local description of the structures. Form these we may obtain detection, classification, and recognition of the image content. In the following, we present typical texture description and classification examples of 1 meter resolution TerraSAR-X images taken in spotlight mode. In particular, we describe how well speckle can be removed, how well local texture parameters of the data can be estimated using dedicated model-based methods, and what can be expected from automated classification. For our work, we use the Knowledge-based Information Mining system called KIM, which includes a graphical user interface for data handling, image inspection, and semantic image annotation.