<![CDATA[Manual gemstone identification still faces several limitations, such as subjective assessment and strong dependence on expert experience, which may lead to misclassification, particularly for gemstones with similar visual characteristics. This study aims to apply a Convolutional Neural Network (CNN) for automatic visual-based gemstone image classification using a limited dataset. The dataset consists of three gemstone classes, namely Alexandrite, Almandine, and Amazonite, with a balanced class distribution. Image preprocessing includes image resizing, pixel value normalization, and data augmentation to increase data variability. The proposed CNN model is a custom architecture composed of three convolutional layers with ReLU activation, followed by max pooling, a fully connected layer with dropout, and a Softmax output layer. Model performance is evaluated using a confusion matrix and classification metrics, including accuracy, precision, recall, and F1-score. Experimental results show that the CNN model achieves a testing accuracy of 93.33% on the limited test dataset with relatively balanced performance across classes. However, analysis of the training and validation curves indicates the presence of overfitting, suggesting that the model’s generalization capability to unseen data remains limited. These findings highlight that the achieved accuracy is conditional on the specific and constrained dataset used. Therefore, future work is recommended to expand dataset size and diversity, apply more comprehensive data augmentation strategies, and explore transfer learning approaches to improve model stability and generalization performance.]]>
Copyrights © 2026
