A fusion method of hyperspectral image based on spectral high fidelity applied in spectrum retrieval of vegetation species

Urban grass is the interference object of vegetable species recognition. Therefore choose an instance of urban grass to retrieve the spectrum curve of interference vegetation. The spectrum retrieval of vegetation species includes three steps, 1) the Hyperspectral image preprocessing, 2) the high fidelity image fusion, and 3) the purity endmember extraction. Firstly, the Hyperspectral image is preprocessed including the removal of bad bands, the radiance calibration, and the FLAASH atmospheric correction. Secondly, the Gram-Schmidt fusion method which has an advantage of spectral high fidelity was employed to fuse the Hyperspectral image and the high spatial panchromatic image. Thirdly, the grass reference vectors was applied in masking the fusion image and then the minimum noise fraction was used to forward and inverse transform the masking image. The pixel purity index of image was calculated after de-noising and then the threshold range was determined to obtain the region of interest that has high purity. The principal component analysis was adopted to forward transform the visible, near infrared, shortwave infrared channels respectively and then the first and second bands of each channel were selected. The optimum index factor was used to acquire the eigenvalues of optimum bands combination and then the N-dimensional visualization was applied in extracting study area endmember of grass species. Finally the spectrum curve of urban grass was retrieved from the average endmember spectral of original fusion image.