Dealing with Unbalanced Data in Leaf Disease Detection: A Comparative Study of Hierarchical Classification, Clustering-based Undersampling and Reweighting-based Approaches.