OBJECT-BASED FOREST CHANGE DETECTION USING HIGH RESOLUTION SATELLITE IMAGES
- 1G&E Laboratory, ENSEGID, Bordeaux University, 1 Allée F. Daguin, 33607 Pessac Cedex, France
- 2EPHYSE Laboratory, INRA, 33140 Villenave d'Ornon, France
Keywords: Multitemporal classification, segmentation, feature selection, change detection, forest damage
Abstract. An object-based approach for forest disaster change detection using High Resolution (HR) satellite images is proposed. An automatic feature selection process is used to optimize image segmentation via an original calibration-like procedure. A multitemporal classification then enables the separation of wind-fall from intact areas based on a new descriptor that depends on the level of fragmentation of the detected regions. The mean shift algorithm was used in both the segmentation and the classification processes. The method was tested on a high resolution Formosat-2 multispectral satellite image pair acquired before and after the Klaus storm. The obtained results are encouraging and the contribution of high resolution images for forest disaster mapping is discussed.