Hyperspectral Image Classification Based on Quadratic Fisher's Discriminant Analysis and Multi-class Support Vector Machine

ABSTRACT

In this paper, we present a hyperspectral image classification method based on Quadratic Fisher's Discriminant Analysis (QFDA) and Multi-class Support Vector Machine (M-SVM). QFDA has been utilized for feature extraction and dimensionality reduction of a high dimensional hyperspectral image and M-SVM is used for classification purpose. The suggested scheme is compared with Principal Component Analysis (PCA) with SVM scheme and it is observed that the proposed method is superior with respect to classification accuracy. Further, it is observed that QFDA preserves the class discriminatory information in the reduced features as compared to PCA.

Keywords:

• Fisher's Discriminant Analysis
• Hyperspectral Image
• Principal Component Analysis
• Support Vector Machine

Additional information

Notes on contributors

Rig Das

Rig Das was born in Kolkata, India, on 16 May 1985. He is a PhD research scholar in the Department of Computer Science and Engineering at National Institute of Technology, Rourkela, India. He is currently working on hyperspectral image processing. He had completed his Master's (MTech) in Computer Science and Engineering from North Eastern Regional Institute of Science and Technology (NERIST), India in 2012 and had worked as an assistant professor in the Department of Computer Science and Engineering at Assam Don Bosco University from July 2012 to April 2013. Previously, he has completed his Bachelor's degree (BTech) in Computer Science and Engineering in 2007 from West Bengal University of Technology, India. He has already published five international conference papers on his name in the field of Steganography and Steganalysis.

E-mail: [email protected]

Ratnakar Dash

Ratnakar Dash is an assistant professor in the Department of Computer Science and Engineering at National Institute of Technology, Rourkela, India since 2011. He had received his PhD degree in 2013 from NIT Rourkela, India and his research areas are signal processing, image processing, intrusion detection system, steganography, etc. He is a professional member of IE and has eight conference and journal papers on his name and he had guided several masters and bachelor degree projects.

E-mail: [email protected]

Banshidhar Majhi

Banshidhar Majhi is the Dean Academic and Professor in the Department of Computer Science and Engineering at NIT Rourkela, India. He has completed his PhD in 2001 from Sambalpur University, Orissa, India. His research areas are image processing, data compression, cryptography and security, parallel computing and soft computing. He is a professional member of FIETE, LMCSI, AMIE (India). He had successfully guided five PhD scholars in the field of image processing. He is the author or co-author of more than sixty international conference and journal papers.

E-mail: [email protected]