Paper Title
Application of Convolutional Neural Network on Chinese Medical Flower Image Classification

Abstract
Background and objective: Due to the large variety of traditional Chinese medicines and their similar appearance, the shapes and colors of different Chinese medical flowers may be similar and confusing. Normally, the identification and classification of Chinese medical flowers are highly depended on experienced experts through visual assessment, which is error-prone and time-consuming. There are not many kinds of edible Chinese medicinal flowers, thus proper identification is crucial. Method: The original Chinese medical flower images are augmented including the methods of Gaussian filtering, brightness enhancement, brightness reduction, mirror flip, noise addition, 90° rotation, 180° rotation and 270° rotation methods, etc. We compared the classification results on Chinese medical flowers from two convolutional neural network (CNN) models Net and Inceptionv3 before data augmentation and after data augmentation. Results: Before data augmentation, the accuracy, specificity, recall, and F1-score from the Alex Net model were 93.57%, 92.98%, 94.52%, and93.62%, respectively. The accuracy, specificity, recall, and F1-score from the Inceptionv3 model were 89.18%, 88.21%, 90.06%, and 88.79%, respectively. After data augmentation, the accuracy, specificity, recall, and F1-score from the Alex Net model were 98.53%, 98.41%, 98.50%, and 98.45%, respectively. The accuracy, specificity, recall, and F1-score from the Inceptionv3 model were 98.61%, 98.61%, 98.55%, and 98.58%, respectively. Conclusions: The classification performance from models of AlexNetInceptionv3 are close with less execution time found from Alex Net model. Keyword – Convolutional Neural Network, Chinese Medical Flower, Data Augmentation.