Recent rapid technological advancements in pattern recognition and computer vision have led to great results on a wide range of applications. One of these applications is herbal plant species identification, as the proper automated system for the recognition of herbal plants is required for botanists to study therapeutic and nutritional uses of herbs. In literature, many studies have adopted classical machine learning approaches while some studies have adopted deep learning approaches, via the leaf images. For this work, we use the latest state-of-the-art framework, namely Mask R-CNN, to build such a classification system to identify a medicinal plant. In this paper, we demonstrate the development of the classification system using Mask RCNN and its backbone: region proposal network, RoI Pooling, RoI Align, and the network head: classification & detection, segmentation. The trained model achieved average accuracy of 95.7% for the identification of 30 medicinal plant species loaded from the Mendely Dataset. The model output is obtained as bounding box for object detection, mask, and class indicating a plant species.