Abstract:
"Traditionally, gem identification relies on specialized tools and expertise, making it challenging and prone to human error. Mistakes in detecting slight differences between gemstones can lead to serious consequences, such as exaggerating a gem’s value. These limitations highlight the need for a reliable methodology to improve gem testing procedures.
To address this, a mobile solution is proposed, utilizing a fine-tuned Convolutional Neural Network (CNN) model with progressive learning capabilities and enhanced explainability through Explainable Artificial Intelligence (XAI). The CNN model excels at pattern recognition in images, with its architecture comprising convolutional, pooling, and densely connected layers designed for optimal performance. Using a database of 12 gemstone categories, the model achieved high accuracy through a systematic process involving resizing, normalization, and data augmentation, minimizing overfitting and ensuring generalization.
The incorporation of XAI tools, such as LIME, enhances the system's transparency, providing users with clear insights into the classification process. This fosters trust and understanding, making the tool accessible to both beginners and professionals.
The result is a highly accurate system, boasting over 98% accuracy, coupled with the reliability of XAI. This innovative tool transforms gemstone identification, serving a wide range of users, from novices to professional gemologists and dealers, and marks a significant advancement in the field."
