Automatic recognition of hyperspectral image based on spectral knowledge

In order to solve the problems of current methods for mineral recognition from hyperspectral data, such as the requirement for prior information, the failure to make full use of absorption features and the lack of the automation of recognition process, an automatic recognition approach based on the spectral knowledge was proposed. The spectral knowledge library including the spectral information and absorption features was generated as the recognition standard, in which the absorption features were enhanced by removing the continuum of image spectra and library spectra as well. The decision method was proposed based on the major and minor absorption features, and a multi-constraint criterion was established to improve the recognition accuracy and avoid the false recognition. The accuracy evaluation of the proposed approach was performed on the simulated data and the airborne visible/infrared imaging spectrometer (AVIRIS) data as well. Experimental results show that the recognition accuracy reaches 80.3% when the signal-to-noise of image is higher than 200. Fine results with the high accuracy are obtained by the proposed approach, and the mineral automatic recognition from hyperspectral data is achieved simultaneously.