GC3558: An open-source annotated dataset of Ghana currency images for classification modeling.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning has advanced various fields, including banking, but currency identification remains challenging due to intraclass variation and illumination.
• Existing currency datasets primarily focus on currencies like Indian, Thai, Chinese, and U.K. currencies, highlighting a need for more diverse datasets, particularly for African currencies.

Purpose of the Study:

• To introduce the Ghana Currency image dataset (GC3558), a novel resource for AI researchers.
• To facilitate the development and evaluation of machine learning models for identifying Ghana's currency (cedi).

Main Methods:

• A high-resolution camera was used to capture 3558 color images of genuine Ghana currency.
• The dataset comprises 13 distinct classes, including various denominations of coins and paper notes (e.g., 10 pesewas coin to 200 cedis note).
• All images were de-identified and validated before being made publicly available.

Main Results:

• The GC3558 dataset provides a comprehensive collection of Ghana currency images.
• The dataset is suitable for training and testing AI models designed for currency recognition tasks.
• It addresses the gap in available datasets for African currencies.

Conclusions:

• The Ghana Currency image dataset (GC3558) is a valuable resource for the AI research community.
• It will enable researchers to develop more robust and accurate currency identification systems.
• The availability of this dataset promotes further research in cross-currency recognition and deep learning applications in finance.