Object oriented post-classification of change images

Commercial satellite images have long been used for environmental monitoring. The improvements in spatial and spectral resolution bring with them new applications in different fields. We have already investigated the use of medium-resolution LANDSAT TM5 images for the routine nuclear verification, based on recently published visualization and change detection algorithms: canonical correlation analysis to enhance the change information in the difference images and Bayesian techniques for the automatic determination of significant thresholds. Now, the high spatial ground resolution of IKONOS and other future satellites provides a good basis for recognizing and monitoring of small-scale structural changes and for planning of routine and/or challenge inspections of nuclear sites. Aside from the advantages of the improved spatial resolution some problems due to sensor and solar conditions exist: Shadow formation and off-nadir images make it more difficult to interpret the complex changes. In order to solve these problems, we supplement the pixel-based change detection analysis with a supervised, object-oriented post-classification of change images carried out with the image analysis system eCognition. Defining of different object classes of the change pixels helps to distinguish between the different man-made, vegetation and other changes. By means of semantic relations between the object classes of changes and other classes it is possible to exclude shadow affected regions and to concentrate on specific areas of interest.