Development of Statistical Based Decision Tree Algorithm for Mixed Class Classification with Sentinel-2 Data

Abstract

Nowadays various types of satellite data are available which have different spectral bands. These spectral bands have special characteristics to interact with particular targets. Therefore, it is also important to use these bands efficiently. The prime objective of this paper is to identify the various land cover classes namely bare soil, moist soil, urban, tall vegetation, short vegetation, pasture, water with sentinel-2 data. The task takes up a challenging edge because various classes among the aforementioned list have characteristic similarities making it cumbersome to classify them effectively with different bands of sentinel-2 data. The classes like urban and bare soil, short vegetation and pasture, tall vegetation and short vegetation, moist soil and bare soil have similarities in their respective characteristics in different bands of the sentinel-2 and are often referred as mixed classes. For this purpose, 10 bands of sentinel-2 were used and results were compared with commonly used techniques like support vector machine. It is observed that the proposed statistical based decision tree algorithm with 10 bands of sentinel-2 has good capability to segregate pastures and moist soil as well as other classes with good accuracy. © 2020 IEEE.