A Maximum Entropy Approach for Mapping Falcata Plantations in Sentinel-2 Imagery

Abstract

Mapping tree species is essential for monitoring, planning, and better managing industrial tree plantations (ITP). Due to the intensive procedure of field sampling and multi-class manual training data collection for image classification, an approach that allows fewer data would be efficient. This study evaluated the performance of a one-class classifier called Maximum Entropy (MaxEnt) for mapping Falcata (Paraserianthes falcataria) in Sentinel-2 imagery. Two MaxEnt parameters were tested, namely sample size and binary threshold. Using a default threshold of 0.5, MaxEnt can provide classification accuracies ranging from 89.41-92.84% using sample sizes as small as 30 and as high as 500. A 0.3 binary threshold applied to MaxEnt logistic output with 500 samples were the best parameter values for classifying Falcata using Sentinel-2 imagery.