Novel Spatial Approaches for Classification of Hyperspectral Remotely Sensed Landscapes

This paper proposes some novel approaches based on spatial homogeneous regions to improve hyperspectral remotely sensed landscape’s classification. Our proposed approaches are investigation of three segmentation techniques (watershed segmentation, hierarchical segmentation, partial clustering) in hyperspectral image and combination of spectral and spatial information in classification with majority vote rule. Proposed methods are compared with pixel-wise SVM and the ECHO, EMP and ML classification for University of Pavia and Indiana da-tasets. Empirical results showed all our proposed approaches yield higher accu-racies when compared to others and the hierarchical segmentation technique re-sulted best most accurate. The drawback of spectral-spatial classification ap-proaches consists in the fact that they smooth a classification map so our claim is just about remotely sensed landscape’s classification.