Abstract: With the rapid development of fruit industry in southern Xinjiang， the classification and extraction of fruit tree has become a hot topic. As the experimental object， this paper selected five kinds of fruit in the southern basin of Xinjiang. In order to find a fast， accurate， large scale， timely， dynamic monitoring of fruit trees and classification method， hyperspectral reflectance data of different leaf species were measured by spectrometer. It would provide necessary technical support to species recognition at the level of spaceborne hyperspectral remote sensing. Based on analysis on the spectral characteristics of five fruit tree species， it employed a way of smoothing filter value treatment under different step sizes and five data transformation methods. The results showed that the species recognition accuracy was the highest after the first derivative transformation， which reached 99.3% under smoothing processing with the step interval of 5 nm. It indicated that this method of remote sensing identification provided a new way for the main tree species in southern Xinjiang.

Key words: hyperspectral, leaf, Southern Xinjiang, feature bands, species identification