Vietnamese herbal plant recognition using deep convolutional features

Abstract

Herbal plant image identification is able to help users without specialized knowledge about botany and plan systematics to find out the information of herbal plans, thus it has become an interdisciplinary focus in both botanical taxonomy and computer vision. A computer vision aided herbal plan identification system has been developed to meet the demand of recognizing and identifying herbal plants rapidly. In this paper, the first herbal plant image dataset collected by mobile phone in natural scenes is presented, which contains 10,000 images of 10 herbal plant species in Vietnam. A VGG16-based deep learning model consisting of 5 residual building blocks is used to extract features from the images. A comparative evaluation of seven classification methods using the same deep convolutional feature extraction method is presented. Experiments on our collected dataset demonstrate that deep learning features worked well with LightGBM classification method for herbal plant recognition in the natural environment with a recognition rate of 93.6%.