The Eight Great Stupa Image Classification Using Machine Learning

Abstract

In this project, a machine learning based system was developed for classifying the stupa images into eight distinct categories namely, Descent from God Realm, Enlightenment Stupa, Heart Lotus, Miraculous Display, Nirvana Stupa, Reconciliation Stupa, Turning of the Wheel and Victory Stupa. For feature extraction combination of Local Binary Pattern (LBP), and Color Histogram (HSV) was applied. Principal Component Analysis (PCA) was employed for dimensionality reduction to improve computational efficiency and model performance. Three machine learning classifiersSupport Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF)were trained and evaluated. The SVM classifier, using the RBF kernel, achieved the highest performance with a cross-validation score of 90.38% and a test accuracy of 94%, both KNN and Random Forest models also demonstrated strong performance by achieving 92% accuracy. Per-class F1-score analysis showed consistently high classification performance across most classes, with SVM slightly outperforming the others. The experiment results demonstrate that traditional machine learning methods combined with effective
feature extraction are highly suitable for stupa image classification.