Comparative study of hyperspectral imagery classification with SVM and ensemble machine learning methods

Abstract

Hyperspectral images are usually high-dimensional data consisting of hundreds of spectral
bands. Thanks to the spectral details they provide, they are preferred in many ground
observation tasks such as forest areas, vegetation, and harvest forecasting. With the
widespread use of artificial intelligence in many areas, the use of machine learning algorithms
in highly complex data such as hyperspectral data continues to increase. In this study, the
Indian Pines dataset was classified using three different machine learning algorithms. In the
experiments, the performance of the Support Vector Machines algorithm was compared with
the performance of the Random Forest and XG Boost ensemble methods. According to the
results obtained, the highest performance was obtained with the XG Boost algorithm as
90.88%. The worst result was obtained with Random Forest as 79.61%. The SVM algorithm,
on the other hand, took second place in the performance obtained with an accuracy of 85.12%.
The results obtained are presented together with the visuals and the performance metrics are
also evaluated as precision, recall, and F1 score.