Plant leaf identification based on machine learning algorithms

Abstract

Classical plant identification process is timeconsuming and complicated. On the other hand, knowledge of plants and the ability to identify the plant species are depleting through generations. This lack of knowledge and drawbacks of manual identification were the underlying causes to develop this study. Hence, the main objective is to compare the performance of different machine learning algorithms and select the best algorithm to be used for further development of a mobile application to identify herbal, fruits, and vegetable plants available in Sri Lanka using their leaves. In this regard, this article focuses on pre-processing and effective classification of manually collected leaves datasets. In the pre-processing stage, noise handling, image enhancement, and transformation were done. Then, features were extracted with respect to shape, texture, and color. Subsequently, five machine learning algorithms were employed on the dataset for classification after normalizing the data. Finally, classification accuracies of the algorithms were obtained with accuracy and loss curves of the Multilayer Perceptron algorithm. The classification accuracies of Support Vector Machine, Multilayer Perceptron, Random Forest, K-Nearest Neighbors, and Decision Tree algorithms are 85.82%, 82.88%, 80.85%, 75.45%, and 64.39% respectively. According to the results, Support Vector Machine and Multilayer Perceptron algorithms exhibited satisfactory performance.