<![CDATA[Cash still dominates transactions in Indonesia, but the process of classifying banknote denominations in automated systems such as ATMs and digital cashiers still faces challenges in terms of accuracy and efficiency. Misidentification of banknote denominations can reduce the reliability and operational efficiency of the system. Therefore, selecting the right algorithm method is crucial. This study compares the performance of VGG16 and MobileNetV2 using transfer learning. The dataset contains 2,100 Indonesian banknote images from seven classes. Both models were trained using pre-trained weights from ImageNet and evaluated using accuracy, precision, recall, F1-score, and computation time. The results showed that MobileNetV2 outperformed VGG16, achieving 94.76% accuracy, 94.99% precision, 94.76% recall, and 94.74% F1-score, with a training time of 569 seconds. In comparison, VGG16 achieved 87.62% accuracy, 89.04% precision, 87.62% recall, and 87.67% F1-score, requiring a training time of 1456 seconds. These results indicate that MobileNetV2 extracts features more effectively and generalizes better. This study demonstrates that MobileNetV2 can be used as an optimal solution for developing an accurate and efficient image-based banknote classification system.]]>
Copyrights © 2026
