Classification of Mineral Images from the Katanga Mining Area / DRC : Comparative Performance of Machine Learning Architectures

Abstract

Abstract

The aim of this study is to compare the performance of three Machine Learning models for mineral image classification: Multilayer Perceptron (MLP), Convolutional Neural Networks (CNN) and a transfer learning model based on VGG-16. The aim is to identify the most suitable machine learning techniques for mineral identification. For this analysis, we took into account key parameters such as batch size, number of epochs, filter size, accuracy, dropout techniques (in the event of overfitting) and the loss metric. The three models were trained on a sample of 5 mineral classes (chalcopyrite, cobaltocalcite, native copper, katangite and malachite). Each model was trained for 30 cycles with batches of 64 images. The results show that the model based on transfer learning with VGG-16 achieved 97.8% accuracy, outperforming MLP (75%) and CNN (96%). This performance underlines the importance of deep learning for processing complex images and represents a breakthrough for mineral resource exploration and identification.