Klasifikasi Kerusakan Uang Rupiah Menggunakan CNN Dengan Arsitektur VGG16

Abstract

This study developed a deep learning model using a Convolutional Neural Network (CNN) architecture with VGG16 to classify the level of damage to rupiah banknotes. Previous studies have focused more on recognizing denominations and detecting counterfeit money using CNN and transfer learning, while the classification of physical damage to rupiah banknotes is still limited, both locally and internationally, and often relies on special acquisition devices or template registration. The dataset used consists of images of rupiah banknotes grouped into three damage categories: >20%, >40%, and >50%. This dataset is divided into 80% for training data (537 images) and 20% for test data (135 images). To enrich the data variety, this study applied on-the-fly data augmentation techniques with rotation, zoom, and flipping during the training process. The experimental results show that this model achieves an accuracy of 93.33%, with excellent precision, recall, and F1-score values, especially in the >50% damage category. The use of the ADAM optimizer with a learning rate of 1e-3 proved to provide more stable and efficient training. Overall, this study shows that the application of CNN with the VGG16 architecture is effective in classifying rupiah currency damage and can contribute to the development of image processing technology, particularly for evaluating currency feasibility in real-world scenarios.