Comparison of classification accuracy of co-located hyperspectral & multispectral images for agricultural purposes

Abstract

The aim of this study is to compare the classification accuracy of multispectral Landsat 8 and hyperspectral EO-1 Hyperion satellite image data of the same region for agricultural purposes. Classification of hyperspectral remote sensing data is more challenging than multispectral data due to high amount of spectral information recorded in several image bands; therefore, Principal Component Analysis (PCA) was applied to these images for dimension reduction. Support Vector Machines (SVM) approach was used for classification of two different data considering the successive results obtained in latest research by applying SVM. Six different land cover classes, namely maize, cotton, urban, water, barren rock and other crop types were determined in this study and training areas were selected for each class during the training selection stage. 200 ground control points were selected within 135 km2 study area to conduct classification accuracy assessment. The overall classification accuracy of Hyperion image was found around 80%, whereas overall classification accuracy of Landsat image was found approximately 70%.