Multispectral and digital elevation model information fusion for classification and change detection in earth observation

Kurzfassung

A review of the state-of-the-art in land cover classification with satellite imagery of the Earth is provided. Information fusion of different typologies (in terms of processing, operation levels and information sources) for classification and change detection purposes is also surveyed. Three different approaches for automatic unsupervised classification of land cover have been implemented to exploit the fusion of multispectral and digital elevation model information for subsequent change detection processes. In broad terms, the first method is based on feature extraction and clustering, the second makes use of a hierarchical decision tree, and the third one constitutes a hybrid, incorporating further techniques. Different types of features have been extracted, at several levels of the process, and merged also at different fusion levels. The algorithms have been tested on two case scenarios, defined by three datasets. A comprehensive set of evaluation measures has been used for the assessment of the classifications, in the three cases scoring high, and especially with the third method, with completeness and correctness values of 97.1877% and 97.3453%. The potential of subsequently applying post-classification change detection techniques with the obtained results is demonstrated, allowing not only for the detection of changes, but for their identi�cation. Implementation of the algorithms and the software system and graphical interface generated has been undertaken in Matlab v.R2010a and IDL v.6.2.