Design and Implementation of Ore Classifier Using Transfer Learning Approach

Nickel and iron ore are the backbone of metal manufacturing and industrial operations. Nickel and iron ore production is highly dependent on the speed of ore classification. The slow and repetitive process of ore classification, as well as the manual listing of trip inventory, significantly impact the speed of ore production in mining companies. Thus, the researchers have developed an alternative ore classifier device that utilized a transfer learning approach in the ResNet- 50 model and is equipped with RFID technology for automated trip inventory listing. A total of 1549 images were captured for training the ore classification model. The evaluation of the model's performance was based on training and validation accuracy and loss. Results indicated that the model learned well from the training data, demonstrated good generalization on the validation set, and avoided overfitting. The device's performance was evaluated using accuracy, precision, recall, and F-1 score. The prototype achieved an overall accuracy of 96.7% in accurately identifying three different classes of nickel and iron ore, including waste samples. These findings indicate that the device can enhance ore production efficiency by reducing the time and labor required for ore classification.