Iron ore image classification using deep learning

This study presents a method for classifying microscopic images of iron ore using convolutional neural networks (CNNs). The dataset used in the experiment consisted of microscopic images of iron ore samples gathered from Kiriburu Iron Ore Mines, Jharkhand, India. Iron ore samples with varying hematite content labelled as hard laminated, soft laminated and lateritic were selected for the study. The CNN model with 3 hidden layers gives test accuracy of 92.16% on the test set. MobileNet-v2 which is pre-trained model yields test accuracy of 98.03%. The results show that CNNs can be effectively used for classifying microscopic images of iron ore and can potentially be applied in the mining industry for quality control and mineral identification.