Image fusion based approach of water extraction from spectrally mixed water regions belonging to the sources of varying nature

Remote sensing devices are keeping eyes on the earth surface to acquire information directly and quickly. A remote sensing image carries information about the biophysical materials of the million acres of land and hence identification and extraction of earth features from the remote sensing big data is the major application of remote sensing. In the present work Landsat-8 and Sentinel-2 data are used to extract water bodies in Prayagraj district, Uttar Pradesh, India. There are two problems observed in dealing with panchromatic band of Landsat-8 in the study area i.e., meet but don’t mix and feature mixing. The meet but don’t mix problem refers to the classification of water regions of two confluencing rivers into two different classes. On the other hand, feature mixing refers to the mixing of water regions with neighboring vegetation regions during classification. It is claimed in past research that the Near InfraRed (NIR) band is suitable for water enhancement as the other Land Use Land Cover (LULC) features are suppressed. So, the Panchromatic (PAN) band is fused with Infra-Red band using Principal Component Analysis (PCA). The fused images are classified and it is observed that the aforesaid two problems are resolved and the overall classification accuracy is improved from 74.23% to 89.1% after fusion.