HYPERSPECTRAL ANALYSIS FOR RECOGNITION AND CLASSIFICATION OF EARTH'S SURFACE MATERIALS

DOI 10.31673/2412-4338.2025.022314

Authors

  • Андрій Анатолійович Стражніков, (Strazhnikov Andrii) State University of Information and Communication Technologies, Kyiv https://orcid.org/0009-0000-3492-2968
  • Тетяна Миколаївна Кисіль, (Kysil Tetiana) State University of Information and Communication Technologies, Kyiv https://orcid.org/0000-0002-5123-0768
  • Софія Веніамінівна Халапова, (Halapova Sofiia) State University of Information and Communication Technologies, Kyiv
  • Вікторія Вікторівна Жебка, (Zhebka Viktoriia) State University of Information and Communication Technologies, Kyiv https://orcid.org/0000-0003-4051-1190

Abstract

This paper analyzes modern approaches to hyperspectral image classification, including machine learning methods that demonstrate high efficiency. Hyperspectral analysis enables precise identification of Earth's surface materials by utilizing the spectral characteristics of objects. However, hyperspectral image classification remains challenging due to the high dimensionality of data, the lack of labeled samples, and significant computational costs.
The authors examine the XGBoost and Random Forest algorithms, which are used for accurate material recognition based on their spectral properties. These methods allow the development of an effective classification system of Earth’s surface objects, while reducing computational costs and maintaining recognition accuracy. A comparative performance analysis of the two selected models is conducted based on metrics such as Confusion Matrix, Accuracy, Sensitivity, Specificity, Precision, and Recall.
Particular attention is paid to the problem of dimensionality reduction. Such data compression is a crucial step in hyperspectral image classification. The application of appropriate methods helps reduce the computational complexity and improves algorithm performance. Additionally, the study explores the adaptability of classification models to different datasets, which is essential for their practical application in the real-world scenarios.
The research results show that the proposed classification system ensures accurate identification of Earth’s surface materials. The findings can be applied in the military intelligence for object detection, environmental change monitoring, and the recognition of potentially hazardous materials. The proposed methods and models enhance the efficiency of hyperspectral data analysis, providing new opportunities for further research in this field.

Keywords: Hyperspectral analysis, ROSIS, XGBoost, RandomForestClassifier, classification task, image recognition.

Downloads

Published

2025-06-25

Issue

No. 2 (2025)

Section

Articles