Tree-Based Classifier for Hyperspectral Image Classification via Hybrid Technique of Feature Reduction

Remote sensing by hyperspectral imaging (HSI) has proved to be an emerging technology for researchers with broad appeal in legionary branches. Imaging spectroscopy, another name of hyperspectral remote sensing, is mainly employed to identify ground cover. HSI contains eminent spectral details, leading to a high dimension which results in a menace for researchers. A viable resolution to this crux is feature reduction (FR). Hither, a hybrid technique of FR, mainly coupling of feature extraction (FE) with feature selection (FS) has been preferred. First, features are extracted via Principal Component Analysis (PCA), then selected via Mutual Information (MI). At the last step, classification has been carried out using tree-based classifier. Here decision tree (DT), random forest (RF) and extra tree (ET) have been chosen as the classifier. For this study, AVIRIS which is also known as the Indian Pine dataset has been picked. This proposed approach offers classification accuracy(CAr) of 93.48% for DT, 96.83% for RF and 97.93% for ET.