Hyperspectral datasets provides explicit ground covers with a number of contiguous bands. Filtering of high - dimensional hyperspectral datasets will pave way for further processing with the dataset in a better way. In order to discriminate surface features potentially, the number of spectral bands need to be minimized without losing the original information from the hyperspectral dataset. This technique is termed as ‘dimensionality reduction’. Several approaches are available for reducing higher order dimension to low order dimension of hyperspectral sensor datasets. In this paper, two major dimensionality reduction techniques such as Principal Component Analysis (PCA) and Minimum Noise Fraction (MNF) have been applied to reduce the dimension of spectral features. The dataset used for the study is AVIRIS NG hyperspectral imagery having a total number of 425 spectral bands covering the wavelengths from 356 nm to 2500 nm. Dimensionality reduction techniques are applied to acquire highly informative bands favoring urban landscape of Kalaburagi, Karnataka. The redundant bands are based on a fact that neighboring bands are highly correlated with each other thus sharing similar information. The benefits of utilizing dimensionality reduction methods are to slacken the complexity of the data during processing and transform original data to remove correlation among the bands. Performance evaluations for dimensionality reduction techniques are assessed and bands that are highly uncorrelated are considered for further processing.
Previous Article in event
Next Article in event
Integrated hydrological modelling over upstream catchments of Himalayan rivers and assessment of extreme hydrological eventsNext Article in session
Comparison and evaluation of dimensionality reduction techniques for hyperspectral data analysis
Published: 05 June 2019 by MDPI in 2nd International Electronic Conference on Geosciences session Earth Sciences through Earth Observation
Keywords: Dimensionality reduction; PCA; MNF; AVIRIS NG; Kalaburagi