High Impact Factor : 4.396 icon | Submit Manuscript Online icon |

A Novel Method for Hyperspectral Image Classification Using CNN Based Laplacian Eigenmap Pixels Distribution Flow

Author(s):

R.KIRTHIKA , THANTHAI PERIYAR GOVT. INSTITUTE OF TECHNOLOGY; DR.T.K.SHANTHI, THANTHAI PERIYAR GOVT. INSTITUTE OF TECHNOLOGY; G.MANIKANDAN, THIRUVALLUVAR COLLEGE OF ENGINEERING AND TECHNOLOGY

Keywords:

Hyper Spectral Image, Laplacian eigen map, Image classification

Abstract

The problems of under classification, in hyperspectral imagery (HSI) and the high complexity of computing Eigen value problem for searching the nearest neighbouring pixel still exist in the nonlinear dimensionality reduction with LE-PD for classification. Therefore, this paper proposes an innovative graphical method namely CNN (Condensed Nearest Neighbour) based LEPD flow to solve the above two problems. First, data reduction and KNN (K Nearest Neighbour) is introduced in the LEPD construction to preliminarily reduce the dimension of HSI data. It aims to improve the speed of nearest neighbour searching. Then, based on the KNN graph, we get a connection matrix consisting of useful points for classification. After graph construction, the adjustment of mapping should be done for better visualization of results and finally the image is classified using threshold methods based on the KNN classification map with CNN extracted prototypes. The experimental result shows that 88% classification accuracy and provides better classification results and high computational speed.

Other Details

Paper ID: IJSRDV2I2197
Published in: Volume : 2, Issue : 2
Publication Date: 01/05/2014
Page(s): 833-836

Article Preview

Download Article