Soil classification using airborne hyperspectral data employing various approaches

Abstract

Hyperspectral remote sensing technology is one of the advance technology for detailed land cover feature extraction. Hyperspectral datasets contain large number of contiguous spectral bands with a narrow spectral bandwidth which enables identification of peculiar absorption features for distinguishing different type of soils. The potential of Airborne Visible Infrared Imaging Spectrometer-Next Generation (AVIRIS-NG) data was tested for distinguishing black and red soils in the ICRISAT area near Hyderabad, Telangana. The AVIRIS-NG data captured in 432 narrow contiguous bands (346�2505 nm) with spectral sampling of 5 nm bandwidth and a 4m ground pixel size was used in this study. The dataset was first spectrally subsetted by identification and removal of bad bands and was atmospherically corrected by converting it to surface reflectance using FLAASH. The data was finally georeferenced using the Internal Geometry Module (IGM) parameters. Optimal spectral bands from the reflectance data were selected on the basis of different characteristics of various soils. Data dimensionality reduction technique Minimum Noise Fraction (MNF) was also performed to extract noise free components. Total five classes including red and black soils were considered for land cover classification. Pixel based classification techniques such as Spectral Angle Mapper(SAM) and Support Vector Machine (SVM) were performed on the reflectance as well as MNF transformed data. SVM was also performed on data containing noise free MNF components and the selected optimal spectral bands. In the resultant classified output of reflectance data, SVM classifier provided higher accuracy and was able to classify black and red soil in a better way than SAM technique. The results also suggested that use of MNF components and specific spectral bands altogether improvised the classification of black and red soil. � 2017 ACRS. All rights reserved.