A spectral-lib based algorithm to pick up pure crop pixels from hyperspectral image

Crop growth monitoring is one of major directions of remote sensing applications. A widely used way is to draw out the NDVI curve of the interesting region in crop growth seasons, then determine whether crop is good or not according to the characters of the NDVI curve and some empirical knowledge. Whether the pixel used to draw NDVI curve exactly describes the target crop species strictly affects the accuracy of the result. The objective of this research is to design an algorithm to find out pure pixel of target crop species from hyperspectral image. The algorithm is based on the spectral library, and it will obtain the sample spectra from the library. If the library returns zero sample canopy spectra, the algorithm will automatically simulate the sample spectra. Then it will aggregate narrow bands into broad bands to match the sensor bands, and at last compare the pixel spectra with the sample spectra. In this research, we use Hyperion, OMIS and MODIS data of different spatial and spectral resolution and use different methods to calculate distance