A CNN-ASSISTED DECISION SUPPORT SYSTEM FOR ANCIENT WORLD COIN CLASSIFICATION

Abstract

The work is devoted to solving the problem of automating the classification of coins from the ancient world. The purpose of the work is to create a neural network model and a system for the classification of ancient coins. The object of research is photographs of coins from ancient cities in the polis of the northern Black Sea region. The subject of research is the process of coin classification. A system architecture, a neural network architecture, and an application for the classification of ancient coins are proposed. The results of the system's operation are demonstrated, and the quality of network training is evaluated. The models were trained using self-generated data samples consisting of sets of color images of coins (200 by 200 resolution) of several types and images of coins that play the role of all other coins and should be classified as not belonging to a given class. The relationship between the ratio of class coins to other coins was also found. For obverse sample 783-7831, 100% accuracy in ten epochs was achieved when the ratio between samples was 1:1, with ratios of 1:4 and 1:7, 30 epochs were required to achieve 100% accuracy. The following technologies were used to implement the Web system: a framework for Python – Django for writing the server part; a JavaScript programming language for writing the user part of the system; and a jQuery library to improve the interaction of the user part with the user. For interaction between the user and the server part of the applications, the AJAX request technology was used, which allows data to be transferred to the server part imperceptibly for the user without reloading the page. Sqlite3 was used as the main application database.