Hyperspectral classification of ancient cultural remains using machine learning

Abstract

The application of remote sensing in archaeology has recently gained widespread recognition, leading to the discovery of numerous significant cultural remains. However, the lack of theoretical data on spectral classification severely constrains the practicability of remote sensing archaeological investigations. In this study, we have collected a comprehensive dataset comprising over 15,000 spectral curves acquired from eight distinct categories of typical archaeological remains in Central China. Machine learning is utilized to conduct an in-depth analysis and classification of the hyperspectral attributes of cultural remains. The feature spectra are preprocessed using the Standard Normal Variable Transform (SNV) and Principal Component Analysis (PCA). A spectral classification model is proposed to improve the accuracy of typical archaeological remains using Support Vector Machines (SVM). The evaluation demonstrated that the SVM exhibited the highest classification accuracy of 99.82%. It was ultimately determined that the most distinguishable bands from ancient cultural remains were in the ranges of 524–553 nm, 663–686 nm, 974–1000 nm, 1092–1114 nm, and 2161–2185 nm. The research provides an important theoretical basis and a scientific method for remote sensing archaeology investigations, which is of great significance in understanding the past and facilitating present sustainable development.