This paper discusses existing methods for geological analysis of drill cores and describes the research and development directions of a machine learning framework for such a task. Drill core analysis is one of the first steps of the mining value chain. Such analysis incorporates a high complexity of input features (visual and compositional) derived from multiple sources and commonly by multiple observers. Especially the huge amount of visual information available from the drill core can provide valuable insights, but due to the complexity of many geological materials, automated data acquisition is difficult. This paper (i) describes the difficulty of drill core analysis, (ii) discusses common approaches and recent machine learning-based approaches to address the issues towards automation, and finally, (iii) proposes a machine learning-based framework for drill core analysis which is currently in development. The first major component, the registration of the drill core image for further processing, is presented in detail and evaluated on a dataset of 180 drill core images. We furthermore investigate the amount of labelled data required to automate the drill core analysis. As an interesting outcome, already a few labelled images led to an average precision (AP) of around 80%, which indicates that the manual drill core analysis can be made more efficient with the support of a Machine Learning/labeling workflow.