Abstract:
"This research addresses the challenge of accurately identifying and authenticating gemstones using a vision transformer model. Traditional methods often depend on expert knowledge and specialized tools, which may be inaccessible for general use. To overcome these limitations, this study leverages advanced AI techniques to develop a more practical and universally accessible solution for gemstone verification.
The methodology involves preprocessing gemstone images to improve quality and then utilizing the vision transformer model to classify gemstones based on type and authenticity. This approach enhances classification precision and reliability compared to existing AI models, demonstrating the practical application of deep learning in real-world scenarios.
Initial testing results show promising performance, with a training accuracy of 68.25% and testing accuracy of 75.87%. The model also achieved an F1 score of 0.7231, precision of 0.7635, and recall of 0.7287. These metrics highlight the model’s ability to effectively identify and classify gemstones while indicating areas for further optimization. Future improvements aim to enhance accuracy and reliability, making the system even more robust for gemstone authentication."
