Application of Augmentation Method on Pharmacognosy Dataset Using Horizontal and Vertical Flip Technique

Abstract

This study aimed to apply image augmentation techniques, namely horizontal flip and vertical flip, to a pharmacognosy dataset to increase the diversity of training data in a pharmacognosy image recognition system. By applying these two techniques, this study focused on finalizing a pharmacognosy image dataset that could be used to train machine learning models. The application of these augmentation techniques improved the accuracy and generalization ability of the model in recognizing pharmacognosy images taken from various viewpoints and orientations. This study used two image augmentation techniques, vertical flip augmentation (VFA) and horizontal flip augmentation (HFA), to expand the pharmacognosy image dataset. Each augmentation technique produced four times the number of modified images from the original images with more and more diverse data variations. With the application of the vertical flip augmentation technique, the training dataset consisted of 2,400 images, a validation dataset of 300 images, and a testing dataset of 300 images, for a total of 3,000 data sets. Similarly, the horizontal flip augmentation technique yielded the same amount of data: 2,400 data points for training, 300 data points for validation, and 300 data points for testing. These two techniques increased the total number of training and testing data points to 3,000.