Performance of various Wavelet based Features on AVIRIS Data Classification

Abstract

Remote Sensing is the technique which is used for obtaining the information about an earth
surface and identification of earth surface features to estimate the geographical properties
using electromagnetic radiation. Hyperspectral Image consists of hundreds of spectral bands
which provide detailed information and this can be used for land cover classification. In this
paper Feature Extraction is done by using various Discrete Wavelet Transform (DWT) and
Co-occurrence features are extracted by using this transformed co-efficient. DWT consists of
many wavelet families such as Daubechies, Symlet etc. Such wavelets are used for extracting
co-occurrence features. Image classification is done by using SVM classifier. Results
obtained from the different wavelet families are compared. In this paper Hyperspectral
dataset obtained by an AVIRIS sensor is used. Accuracy for Haar is 75.32%, DB 4 is 82.39%,
DB8 74.45%and for Sym4 and Sym8 is 64.59% and 70.65% respectively.