Traditional Vietnamese Herbal Medicine Image Recognition by CNN

Abstract

The use of computer vision in traditional medicine is crucial, and it might be beneficial to automatically recognize two-dimensional images of Vietnamese herbs. With the help of potent approaches applied to the field of automatic identification, we give a dataset of dried herbal images and identification results. Deep feature and transfer learning were the two methods employed in the study; the findings indicate that SOTAs is a quick and easy method with lots of application potential for VTM picture identification. As a consequence, all 100 therapeutic herbs can be identified with an average accuracy of 99.275% by current convolutional neural networks state of the art model begin with VGG16 and end by Xception. Future applications can also benefit from the accuracy of classification algorithms like SVM and RF on manually extracted deep features.