A BAND SELECTION METHOD FOR CROP CLASSIFICATION BASED ON SPATIAL AND SPECTRAL CORRELATION USING HYPERSPECTRAL IMAGE

It has been proved that hyperspectral images can classify objects more effectively than any other optical remote sensing data. However, the selection of the most informative and distinct bands is an inevitable challenge. We propose a band selection method that selects bands based on their spatial patterns and spectral values. We have compared with and without including spatial relationships among the different bands. Experiments on the Salinas and Indian Pines datasets demonstrate that the proposed model has significantly better performance while considering spatial relationships along with spectral data.