Image classification using higher-order statistics-based ICA for NOAA multispectral satellite image

摘要

The main objective of this research is an object classification with reduced bands of a multispectral National Oceanic and Atmospheric Administration image by using a higher-order statistics-based independent component analysis and clustering method. Enhancement is not only improving the spatial resolution, but should be preserving the spectral information. The ICA is used for dimensional reduction of multispectral images and enhancement of techniques for improving the spectral and spatial values. This integrated composite image is classified by using a K-means clustering algorithm, and the objects are separated based on homogeneity feature levels with pixel values. This unsupervised classification can be used for extracting land, water and clouds with high accuracy and kappa coefficient values compared with band calibration values of NDVI and surface temperature. The image has low colour distortion, high resolution, improved visual quality, and accurate information with good statistical parameter values.