Evaluation of the Spectral Angle Mapper "SAM" Classification Technique using Hyperion Imagery
Lamine, Salim¹; Petropoulos, G.P^2
1Mediterranean Agronomic Institute of Chania, GREECE; ²University of Aberystwyth, UNITED KINGDOM

Hyperspectral remote sensing technology demonstrates the capacity for accurate vegetation identification. The aim of our study was to evaluate Spectral angle mapper (SAM) classifier using Hyperion hyperspectral imagery in order to show up the potential of Hyperion data for land use/cover (LULC) characterization in the North of Greece. Hyperion data pre-processing was effectuated to reduce the Hyperion dataset and to eliminate bands giving redundant spectral information. The pixel-based supervised classification implemented the spectral angle mapper SAM classifier performed through ENVI 4.7 software. The results were evaluated using the validation training sites obtained from the Advanced Land Imager (ALI) sensor and the CORINE2000 map in conjunction with the Hyperion data. The classification results mainly appeared to suffer from the relatively low spatial resolution of the Hyperion sensor and the spectral similarity among some land cover classes. The results showed that the procedure of data reduction permitted to obtain more accurate and better classification results after selecting the optimal Hyperion band combination. The optimal overall accuracy of the SAM pixel-based classification was 77.28% for the scenario 02 that adopted the 30 selected bands from both VNIR and SWIR Hyperion bands and maximum angle of 0.3, and the overall KHAT value was 0.7278 which indicates a good agreement between the classification and the reference data. Our results highlight the potential of hyperspectral remote sensing data as well as the SAM classifier approach for mapping heterogeneous land use/cover in the study site.