A novel two-tier paradigm for labeling water bodies in supervised satellite image classification

Abstract

The satellite images are the high dimensional data with huge spatial details which needs intelligent interpretation methods. The use of satellite images is more prevalent nowadays in many real time surveillance applications and it leads to the need of simple and accurate processing. The advent of machine learning classifiers improved the efficiency of satellite image applications. However, the ground truth data availability and spatial data labeling are certain serious limitations for supervised satellite image classification. Thus change monitoring models are highly dependent on unsupervised clustering methods with the compromised level of accuracy. This paper is focused on addressing the steps to overcome the above said ground truth validation, labeling and other feature discrimination issues in a supervised soft classifier model for simple change recognition of seasonal water bodies in satellite images.