Enhanced Mineral Image Classification Using YOLOv8-CLS With Optimized Feature Extraction and Dataset Augmentation

Abstract

With the increasing global emphasis on ecological conservation and the sustainable use of resources, the effective utilization of mineral resources has become a pressing priority. Accurate mineral classification helps reduce resource waste, mitigate ecological impact, and improve processing efficiency. This paper proposes a deep learning model for mineral image classification based on the YOLOv8-CLS architecture, specifically targeting seven minerals: bornite, quartz, malachite, pyrite, muscovite, biotite, and chrysocolla, which are the focus of this study. The model was trained and tested on an open-source mineral image dataset, achieving Top-1 and Top-5 accuracy rates of 0.92053 and 0.99399, respectively, after 120 training epochs. During the testing phase, the model was evaluated on the test set, achieving a Top-1 accuracy of 0.90681 and a Top-5 accuracy of 0.99283, demonstrating high accuracy and stability. Although a decline in Top-1 accuracy to 0.80219 was observed when testing the model against a new batch of data and comparing it to the classic ResNet50 and ResNet101 models, the YOLOv8-CLS model still outperforms these models by 0.00684 and 0.04278, respectively, while also having lower performance overhead. Despite some remaining flaws, this study demonstrates that the YOLOv8-CLS model is more efficient than traditional models in intelligent mineral classification, contributing to resource efficiency and promoting the development of sustainable mining practices.