Automated Forest Types Classification using Ensemble Machine Learning in Hyperspectral Satellite Images

The forest is a major factor in the changes in climatic conditions. As the population is increasing the needs of humans are also growing, where numerous of forest areas are destroyed for civilization. The Ensemble Machine Learning (ML) classifier is proposed for forest type classification in hyperspectral satellite images. The ensemble ML classifiers such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Random Forest (RF) leads to enhance the robustness and accuracy. The SVM provides better generalization, KNN captures local patterns and RF includes robustness to overfitting thereby enhancing classification performance. The Local Ternary Pattern (LTP) is used for feature extraction which enhances the robustness to noise and it captures texture data through encoding three distinct states that assist to classify forest types. The median filter is used as preprocessing which effectively removes noise without reducing image sharpness. The ensemble ML got better performance of 98.4% recall, 98.7% precision, 99.2% accuracy and 98.6% f1-score for Sentinel-2 dataset compared to VGG19 and ResNet50.