Herbal leaf authentication using convolutional neural network on Raspberry Pi 3

Abstract

Currently, there are many methods for classifying herbal plants based on leaf authentication. Basically, the method of leaf authentication is a visual comparison of images taken by a camera with a reference visual image. This paper aims to identify herbal leaves using the artificial intelligence method i.e. the convolutional neural network (CNN) applied to the Raspberry Pi. The advantage of CNN is that it does not need feature extraction because it contains an automatic feature extraction process. In this paper, there are 10 types of leaves from different herbal plants which are divided into two-third training data and one-third testing data. The results of the identification system using CNN will be validated with other data that are not included in training and testing. Furthermore, it will be tested with different types of leaves that are outside ten types of leaves in the experiment. The accuracy using this CNN method is above 90%. These results indicate that the CNN method is more accurate in herbal leaf authentication.