Spectral feature-based hyperspectral RS image retrieval

Oriented to the demands of vast RS information management for RS image retrieval, the applications of spectral features are discussed by taking hyperspectral RS image as an example. It is proposed that spectral features-based retrieval includes two modes: retrieval based on point mask and polygon mask. The most key issues in retrieval are spectral features extraction and similarity measure. The spectral vector can be used to retrieval directly, and the spectral angle and spectral information divergence ( SID) are effective in similarity measure. The local maximum and minimum in reflectance spectral curve, corresponding to reflectance apex and absorption apex, can be used to retrieval also, but effective matching strategy should be adopted. The quantitative indexes for spectral curves such as moment, fractal and entropy are not suitable to retrieval because of poor similarity measure performance.