Heritage Coin Identification using Convolutional Neural Networks: A Multi-Classification Approach for Numismatic Research
Shiva Mehta, Vinay Kukreja, Dibyahash Bordoloi
Abstract
Heritage coin identification and categorization is a complex undertaking that requires much knowledge. For the automated recognition and categorization of six heritage coins, including ancient Greek, Roman, Byzantine, Islamic, Indian, and Chinese coins, we suggest a deep learning-based method in this research study employing Convolutional Neural Networks (CNN). To boost the size and variety of the dataset, we pre-processed a dataset of 8230 photos of historical coins by normalizing the pixel values and using several data augmentation methods. For the CNN model, a modified VGG-16 architecture is implemented and it has achieved a 98.7% accuracy on the validation set. Calculations of the model's accuracy, recall, and F1 score for every class showed that it performed well across all of them. The proposed method offers a responsible and effective way for identifying and classifying legacy coins, which has essential applications in numismatics and the protection of cultural heritage. The automated detection of intricate patterns and characteristics of historical coins is made possible by using CNN in the proposed method, making it a valuable tool for archaeologists, historians, and coin collectors.