Hyperspectral Image Classification using Support Vector Machine with Guided Image Filter

Shambulinga M,G Sadashivappa

doi:10.14569/ijacsa.2019.0101038

Shambulinga M, G Sadashivappa

Open Access

https://doi.org/10.14569/ijacsa.2019.0101038

Copy DOI

Abstract

Hyperspectral images are used to identify and detect the objects on the earth’s surface. Classifying of these hyperspectral images is becoming a difficult task, due to more number of spectral bands. These high dimensionality problems are addressed using feature reduction and extraction techniques. However, there are many challenges involved in the classification of data with accuracy and computational time. Hence in this paper, a method has been proposed for hyperspectral image classification based on support vector machine (SVM) along with guided image filter and principal component analysis (PCA). In this work, PCA is used for the extraction and reduction of spectral features in hyperspectral data. These extracted spectral features are classified using SVM like vegetation fields, building, etc., with different kernels. The experimental results show that SVM with Radial Basis Functions (RBF) kernel will give better classification accuracy compared to other kernels. Moreover, classification accuracy is further improved with a guided image filter by incorporating spatial features.

Highlights

Remote Sensing is the study of collecting the earth’s surface data and information by measuring the reflected signal form objects at long distances
The results of the support vector machine (SVM) classification accuracy of different kernels and guided image filter are shown in Table I and Table II, respectively
A spectral and spatial classification method was developed for the hyperspectral image classification

Summary

INTRODUCTION

Remote Sensing is the study of collecting the earth’s surface data and information by measuring the reflected signal form objects at long distances. PCA and ICA are unsupervised methods do not require labels and LDA is the supervised method requires labels for extraction of a feature vector These feature extraction techniques preserve the spectral signature information of different class by projecting a feature vector with more dimension vector to less dimension vector. Different types of methods developed based on distance and information theory [5] for the spectral band selection in hyperspectral images. Experimental results of these feature selection methods have shown better performances, but these methods have some theoretical difficult in finding the best subset combination of features for classification.

PRINCIPAL COMPONENT ANALYSIS

METHODOLOGY

Experimental Setup

Classification Results

CONCLUSIONS