Systems that use Artificial Intelligence (AI) are becoming increasingly popular, and the development of these systems can, in theory, be used to automate processes that are performed manually, for example, using computer vision techniques for pattern recognition. Particularly, ore beneficiation processes go through classification steps that are done by manual process. The challenge lies in choosing the AI algorithms that are best suited for automating these processes. In this work, we carried out a study of a computer vision technique (Mask R-CNN) that can be used for classification and segmentation of images of iron ore particles obtained by a low resolution microscope. Aspects of average precision of the model segmentation (mAP) and accuracy of the classification of iron and mixed particles were evaluated. The objective was to verify the feasibility of using image segmentation and classification techniques in the iron ore concentration process from ore samples obtained with a low resolution microscope using deep learning models. It was verified that it is feasible to use segmentation and classification techniques based on deep learning algorithms for the particles in question.