摘要：Summary Hyperspectral sensors enable the acquisition of data with increased number of spectral bands at a higher spectral resolution. Data acquired through such sensors has been widely utilized in many remote sensing applications including mineral mapping. Development of classification algorithms is also continued along with hyperspectral sensors. Choice of an algorithm is extremely important for proper classification. To the best of our knowledge, there exists limited knowledge on the relative performance of similarity measures on a mineral classification. This study uses three statistics namely spectral discriminatory probability (SDP), spectral discriminatory entropy (SDE) and spectral discriminatory power (SDPW) to assess the performance of various similarity measures. Similarity measures chosen are Spectral Angle Mapper (SAM), Spectral Information Divergence (SID), Jeffries–Matusita distance (JM) and their hybrid combinations of SID–SAM, SID–SCA, and JM–SAM. All the similarity measures and statistics were developed on MATLAB® platform and evaluated the same using freely available AVIRIS mineral data from U.S. Geological Survey spectral library. Analysis of statistical results collectively revealed that among the chosen algorithms SID–SAM and SID–SCA outperform the other similarity measures when tested on mineral data. This result has an important implication on choosing of appropriate similarity measure for mineral classification.
