Classification of Medicinal Leaves Using Support Vector Machine, Convolutional Neural Network and You Only Look Once

Plants are a vital part of the ecological system. The identification of medicinal plants is the need of the hour. The main objective of this work is to classify seven classes of medicinal plants based on their leaves using three different techniques-Support Vector Machines, Transfer Learning - VGG16 Model, and a real-time object detection algorithm You Only Look Once. The dataset containing more than 7000 leaves are collected from a vast repository of leaves. Each class includes 1000 leaves naturally collected from the repository. The dataset is also extended by adding around 150 more leaves to the individual class using the data augmentation techniques-random rotation, horizontal flip and random noise to the data. The performance of each method is shown using the classification metrics. After the implementation of algorithms, for the test data, transfer learning achieved the overall accuracy of 98%, SVM obtained 97% accuracy after tuning hyperparameters with GridSearchCV, and You Only Look Once obtained around 84% accuracy. The statistical analysis is also performed using a non-parametric test Chi-Square goodness of fit on test data to identify the model that best fits the data.