Hyperspectral Data Analysis for Arid Vegetation Species : Smart & Sustainable Growth

Abstract

Hyperspectral Images are continuous narrow spectral bands provides a wealth of information which can be used in different applications. Advance developments in hyperspectral remote sensing technology since two decade have opened new opportunity to explore innovative ways to study vegetation species. In this research work, ground-based AISA Vis-NIR hyper spectral image system of 240 bands, wavelength range from 390 to 960 nm with 2.5 nm spectral resolution and 1cm spatial resolution at a distance of 10m was used for classification of prominent vegetation species (Cactus, Neem and Babool). Machine learning supervised classification algorithms are used to classifying the Hyperspectral data. In supervised classification, four methods have been used viz. Spectral Angle Mapper (SAM), Minimum Distance (MD), Support Vector Machine (SVM) and Spectral Information Divergence (SID) Classifier. Environment of Visualize Images (ENVI) software is used for processing and analysis of hyperspectral images for classification of vegetation species in Jodhpur study area. Accuracy assessments were also carried out for classified output images and estimate the performance of a classifier. The overall accuracy for SVM classification algorithm is best (81.2%) when 237 hyperspectral bands were used and SAM classification algorithm has provided a better overall accuracy (76.6%) when maximum noise function (MNF) 11 bands were used. This research demonstrated the efficient use of contiguous fine bands of Hyperspectral data in discrimination and classification of vegetation species.