Comparative study of classifiers for leaf classification and extraction of medicinal properties

In this paper, we investigate the relative efficiencies of image classifiers - namely, a K-Nearest Neighbours (KNN) classifier, a Support Vector Machine (SVM) and a Convolutional Neural Network (CNN) in leaf classification. The primary purpose of classifying leaves is to identify local plant species commonplace in and around households so as to extract and derive utility from the medicinal properties associated with them (if any). The input image acquired from the user undergoes appropriate image processing, after which defining features of the leaf are extracted in the form of numerical feature vectors. These are used to train the SVM and the KNN, enabling them to perform multi-class classification of the genus and species of the plant. CNNs have been attracting considerable amounts of attention in the field of image processing. The principal characteristic is that classification by CNNs occurs sans feature extraction. Medicinal properties corresponding to the plant and their directions for use are web scraped from the internet. Traditionally, the medicinal properties of plants are held in high regard for the treatment of common ailments and enhancing general health. A striking feature of this study is to equip the general populace with awareness of these properties to derive the benefits of their utility. Comparing the performance of the classifiers mentioned above leads us to conclude that CNN yields the best results.