Object model and two-stage classification for automated object-based analysis of remote sensing imagery

With IMALYS, a software prototype that integrates various methods of object-based image analysis is introduced. Two key concepts — image segmentation and classification — are focused. With regard to image segmentation, IMALYS implements a method that was developed as combination of region-growing and watershed transformation approaches and is able to conduct image segmentation solely based on image parameters. Concerning classification, IMALYS applies a two-stage process combining an unsupervised method based on the concept of self-organizing maps (SOM) and a supervised method applying principles of support vector machines (SVM) in order to extract real-world objects and information from previously segmented remote sensing imagery. During this process an object model is utilized that examines (1) adjacent pairs of image segments, (2) the spatial proximity of image segments and (3) the context of image segments in regard to the desired classification scheme. By this means, IMALYS is currently developed to provide automated analysis procedures for up-to-date retrieval of thematic information from remotely sensed data.