An Unsupervised Classification and Feature Selection Approach For Discrimination of Ailanthus altissima (Mill.) Swingle Tree Leaves

Abstract

An unsupervised feature selection and classification approach is presented for automated discrimination of Ailanthus altissima (Mill.) Swingle tree leaves in this study. Experimental setup is consist of a total of 20 different tree specimens and 735 leaf images taken from Leafsnap Dataset where a binary mask is extracted from each of the leaves. 10 salient features defining shape and morphology are extracted from each of these masks. In this study, it was aimed to evaluate all different combinations of these features as subsets to find an optimal feature set for clustering of Ailanthus altissima (Mill.) Swingle tree leaves. Accordingly, two widely known unsupervised data clustering methods, Fuzzy C-means and K-Means are implemented as classifier. Multiclass and two class discrimination experiments are achieved via these methods and F-Score is utilized for objective evaluation of the performance. Experiments revealed that, it would be possible to combine extracted features for creating a subset that is effectively defining shape and morphology of the leaves. Also implemented clustering mechanisms are promising and may discriminate Ailanthus altissima (Mill.) Swingle tree leaves with high accuracy and sensitivity.