Performance evaluation of FFT_PCA Method based on dimensionality reduction algorithms in improving classification accuracy of OLI data


1 Associated professor of remote sensing and GIS, Faculty of Geography, Kharazmi University

2 MA of remote sensing and GIS, Kharazmi University


Fusions of panchromatic and multispectral images create new permission to gain
spatial and spectral information together. This paper focused on hybrid image fusion
method FFT-PCA, to fuse OLI bands to apply Dimensionality Reduction (DR)
methods (PCA, ICA and MNF) on this fused image to evaluate the effect of these
methods on final classification accuracy. A window of OLI images from Ardabil
County was selected to this purpose and preprocessing method like atmospheric and
radiometric correction was applied on this image. Then panchromatic (band8) and
multispectral bands of OLI were fused with FFT-PCA method. Three dimensionality
reduction algorithms were applied on this fused image and the training data for
classification were selected from DRs Output. A total of eight classes include bare
land, rich range land, water bodies, settlement, snow, agricultural land, fallow and
poor range land were selected and classified with support vector machine algorithm.
The results showed that classification based on dimensionality reduction algorithms
was quite good on OLI data classification. Overall accuracy and kappa coefficient of
classification images showed that ICA, PCA and MNF methods 86.9%, 89%, 96.8%
and 0.84, 0.91, 0.96 respectively. The MNF based image classification has higher
classification accuracy between two others. PCA and ICA have lower accuracy than
MNF respectively.