Hyperspectral imaging spectrometers have been extensively researched in the past a few decades. They can measure electromagnetic energy in their instantaneous field of view in hundreds of wavelengths. With such high spectral resolution less than 10 nanometers, it is possible to distinguish materials of subtle difference in spectrum. Recently, a two-dimensional concave grating hyperspectral spectrometer has been developed under the support of National Space Organization in Taiwan. This airborne system is integrated with two subsystems of visible-near infrared (VNIR) and short-wave infrared (SWIR) bands with 3.5 and 10 nanometers spectral resolution respectively. With the design fly altitude of 2000 meters, the spatial resolution is about 70 cm. In this study, linear and nonlinear classification methods of Fully Constrained Least Squares (FCLS) and Back Propagation Neural Network (BPNN) for agriculture crops are discussed. Based on the ground truth, four crops are selected in the study site, including chives, broccoli, rape and pea. The experimental results indicate the classification accuracy of BPNN can exceed 90% and outperforms classification results of FCLS.