Deep Learning-Based Classification of Archaeological Artifacts Using Image Processing

Abstract

The classification of archaeological artifacts according to their material types is important for archaeological documentation and analysis. This study evaluates five convolutional neural network (CNN) architectures for the classification of archaeological artifact images. A balanced dataset containing 1,000 images from four material classes, namely ceramic, glass, metal, and stone, was used. Five CNN architectures, SqueezeNet1.1, MNASNet0.5, RegNetX-200MF, FasterNet-T0, and InceptionNeXt-Tiny, were evaluated using stratified five-fold cross-validation. Model performance was assessed using accuracy, precision, recall, F1-score, and confusion matrices. The experimental results showed that InceptionNeXt-Tiny achieved the highest average accuracy of 90.70 ± 0.76% and F1-score of 90.66 ± 0.76%. RegNetX-200MF and FasterNet-T0 achieved average accuracies of 87.50% and 86.20%, respectively. Confusion matrix analysis showed that classification errors varied across material classes, particularly between glass and metal and between ceramic and stone. The results demonstrate the potential of CNN architectures for automated material classification of archaeological artifacts.