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.
