Measures to improve crop classification using remotely sensed hyperion hyperspectral imagery

摘要

Hyperion- a hyperspectral sensor is carried on NASA's EO1 satellite. This study was carried out for Lonar area of Jalna district, Maharashtra using data of January 2008. Hyperion data contains 242 spectral bands ranging from 356 to 2577 nm out of which 196 calibrated bands (bands: 8–57 and 79–224) are used for further processing. Level 1 product (.L1R) for which only radiometric correction was applied is used for this study. To get the complete advantage of hyperspectral data atmospheric correction is essential. FLAASH, a very effective code for hyperspectral data available in ENVI is applied for atmospheric correction. The atmospherically corrected image contains 168 bands after removing absorption bands. As a first measure principal component and band correlation analysis based spectral subset is applied for optimum band selection for vegetation application. Field study was conducted in January 2009 to collect field spectra. Spectral library was built for major three crops of the study area i.e. chana, jawar and wheat by spectra collected from the field. As a second measure before classification NDVI value based mask is applied to differentiate agricultural areas from other vegetated areas and non vegetated area. After discarding other areas, crop classification is carried out only in the agricultural area. Spectral Angle Mapper (SAM) a very popular algorithm for hyperspectral image classification is applied for image classification and accuracy assessment is carried out.