CSgI: A Deep Learning based approach for Marijuana Leaves Strain Classification

This paper proposes a novel approach that classifies the images of various marijuana/cannabis leaves into their respective classes of strains and types. The proposed architecture works on a two-fold technique which when implemented in the requisite sequence delivers phenomenal results to the classification problem statement. The first fold, being the segmentation or foreground extraction in the images, focuses on extracting the RDI (Region of Interest) using a robust segmentation algorithm which can suitable separate the foreground from the image; and the second fold, being the Deep Learning aspect focuses on the result classification task. This literature gives a quantitative analysis of implementing this classification problem vide a transfer learning paradigm (for application instances with less training data in hand) & training the entire CNN archetype from scratch (for application instances with sufficient training data in hand). Thus, altogether the proposed methodology distinctively deploys ConvNets for the posed classification problem having dual aspects of approaches & implementation wiz: a) Transfer Learning & b) Training the entire CNN from scratch. The novelty of the proposed work can be counted upon as the construction of a robust algorithm very first of its kind in this respective application domain which is potent enough to render the correct class label of the strain/type of marijuana or cannabis leaf image when fed to the system for classification.