DIMENSION REDUCTION WITH PRINCIPAL COMPONENT ANALYSIS IN HYPERSPECTRAL IMAGE CLASSIFICATION USING MACHINE LEARNING

Hyperspectral imagery technologies are one of the future trend investigations. In recent years, numerous studies have been applied to hyperspectral imagery for many purposes. Feature extraction is more complex and time-consuming when compared to other data sources. However, dimension reduction algorithms can help in processing and extracting this manner. Principal Component Analysis (PCA) algorithm has many advantages for hyperspectral imagery. In the study, PCA algorithm was used in the Support Vector Machine (SVM), Random Forest (RF), and Multilayer Perceptron (MLP) methods for hyperspectral image classification on Pavia University dataset. The dataset includes nine classes asphalt, meadows, gravel, trees, painted metal sheets, bare soil, bitumen, self-blocking bricks, and shadows. The geometric resolution of the images is 1.3 meters. The study conducted different PCA band combinations using proper machine learning algorithm parameters. Different band combinations are used in the experiments as 25 and 50 bands. The results are compared quantitively in the meaning of accuracy and time. General accuracies have been seen at over % 85 for two band combinations, too.