A review of deep leaning in image classification for mineral exploration

Efficient sorting and optimal utilization have become the common core issues for today's mining industry. Vision-based sorting technology provides a powerful answer to this problem. Deep learning, as one of the outstanding branches of machine learning, makes it possible to design, deploy and apply the high-performance vision models, and the architecture represented by convolutional neural networks again improves the accuracy and speed of mineral sorting. Based on this, we carefully review and survey the research and applications of deep learning in digital mineral image classification between 2017 and 2022, and state the evolution and innovation of it in terms of mineral types, data qualities, and model architectures. The scope considered in this review is to predict one class label for the entire image rather than the segmentation task of pixel-by-pixel classification. This work further outlines several problems that still stand out in current industrial implementation, summarizes the relevant solutions, and discusses the application potentials of deep learning in mining domain.