<![CDATA[Rice is the main staple food for the majority of the Indonesian population; therefore, rice quality plays an important role in maintaining national food security. However, rice quality assessment in Indonesia is still largely conducted manually, making it subjective and inconsistent. Consequently, the application of image processing techniques and deep learning algorithms is required to achieve a more objective and accurate classification process. This study aims to compare the performance of four Convolutional Neural Network (CNN) architectures, namely VGG16, VGG19, MobileNet, and MobileNetV2, in classifying five types of rice: Arborio, Basmati, Ipsala, Jasmine, and Karacadag. The dataset used consists of 4,000 images that were processed directly without additional preprocessing stages. The experimental results show that MobileNet achieved the highest accuracy of 99% with a training time of 756 seconds. Meanwhile, MobileNetV2 and VGG16 achieved accuracies of 98% with training times of 728 seconds and 1,502 seconds, respectively, while VGG19 produced the lowest accuracy of 97% with a training time of 1,038 seconds. Based on these results, it can be concluded that the MobileNet architecture demonstrates the best performance in classifying the five rice varieties in the dataset used.]]>
Copyrights © 2026
