Abstract
This study presents the band arithmetic approach to identify land cover types such as vegetation, bare land, water, and built-up area in the area of Nanded district and the Purna region of Parbhani district, India. Rather than boosting intelligence to the classifier end, the principal purpose of the proposed research work is to produce various formulations for the feature enhancement. The classifier could classify the formulae transformed images and enhance the accuracy of classification. These formulations are the estimated arithmetic between the various bands of the multispectral satellite imagery of the Sentinel-2. We carried out estimation using knowledge of the spectral reflectance curve. Different arithmetic formulations among the bands of the same date scene covered by satellite are proposed and applied at the pixel level. We have tested the efficacy of the proposed methods using a random forest (RF) classifier. Proposed methods provide enhanced results in terms of accuracy. The overall accuracy reaches up to 95 percent with a kappa coefficient 0.93 for the Nanded site and 91 percent with a kappa coefficient 0.88 for the Purna site.
Acknowledgements
We are thankful to the ESA and USGS for providing satellite imagery for our selected regions and durations. We thank anonymous reviewers provided helpful comments on earlier drafts of the manuscript.
Funding
We acknowledge the funding from All India Council for Technical Education (AICTE), New Delhi, India, for this research project.