Abstract:
The demand for automated and precise currency identification systems is on the rise,
fueled by advances in banking operations, retail market activities, and accessibility
opportunities for people with visual impairments. This research focuses on the detection and
classification of the 2016 Indonesian Rupiah coins, which creates a particular challenge, as
these coins have a high degree of similarity within class and have significant variation in
environmental conditions (lighting and occlusion).
The writer proposes a robust solution utilizing transfer learning on the state-of-the-art
YOLOv8 (You Only Look Once version 8) object detection model. This will incorporate the
fine-tuning of a pre-trained YOLOv8 model on our custom dataset of 2016 Rupiah coins with
high rates of precision and recall. This study will provide a complete image dataset of the
Rupiah coin from numerous real-world scenarios, and then undergo a rigorous model training
and validation process. The intention is not to create yet another model as output, but to obtain
an accurate and efficient model working in real-time immediately on coins, and to demonstrate
the effectiveness of transfer learning as a potential pathway for hyper-local adaptation of
advanced deep learning models for classification and detection of particular regions and
objects. The results will help to contribute to the realm of applied computer vision and support
a significant real-world problem of developing automated systems for handling coins in
Indonesia.
