Gemstone Identification and Authentication Using CNN and KNN: A Hybrid Classification Approach

In this study, a classification hybrid model for the recognition of seven different types of gemstones is developed by using CNNs and KNN. The model classifies Kyanite, Emerald, Coral, Benitoite, Amber, Topaz, and Zoisite. The proposed model achieved excellent performances in precision, recall, and F1-score across all classes. 10,362 images make up the data set applied in this study, and the model’s overall accuracy was 98.34%. Precisely, the precision for each of the gemstones ranged between 98.52 % for Kyanite to 98.62 % for Topaz and recall values were all high with the highest value from Emerald of 98.41 %. F1-scores also achieved high values between 98.38 % for Kyanite to 98.52 % for Topaz as shown by the model’s ability to balance recall with precision. The support for each class was well-distributed, where the number of instances per class varied from 1425 for Kyanite to 1524 for Topaz; thus, balanced performances were ensured across the classes. The proportion of support for each class stayed the same, varying from 0.14 to 0.15; thus, a stable dataset representation was ensured. The model’s overall accuracy stood perfect at 1.00 for each class, meaning all the instances of the gemstone were correctly classified without any misclassification. The hybrid CNN-KNN model, holding high performance on all evaluation metrics, suggests robustness and reliability in classification tasks concerning gems and stones. It will be able to identify and classify gemstones with accuracy and is a tool that has a good potential application in not-so-negligible fields such as jewelry, mineralogy, and authentication of stones requiring accuracy and efficiency in a high range. Results demonstrate the capability of the model to automatically classify the gemstones, thereby minimizing human intervention and having a reliable quality assessment and authentication method.