Archaeological and Experimental Lithic Microwear Classification Through 2D Textural Analysis and Machine Learning

Abstract

The paper focuses on introducing 2D texture analysis as a quantitative method for functional analysis in archaeology. The paper aims to demonstrate the validity of this method for quantifying use-wear analysis and to evaluate different processing, extraction, and classification techniques. The method presented relies on five techniques of quantitative feature extraction from photographic images and nine classification techniques through machine learning algorithms. After creating a training dataset with experimental traces, machine learning models were validated through experimental and archaeological image classification. The best result achieved a classification accuracy of 80%, suggesting convolutional neural network and grey level co-occurence matrix as the best quantification options and neural networks as the best classification algorithm. The paper proposes to use the method as a fundamental tool in functional analysis to remove subjectivity criteria from traditional analysis and to address issues related to the credibility of the discipline, calibration, standardisation, and reproducibility of methods and results.