Implementation of Dataset Augmentation on Ethnofimedicinal Images Using Rotation and Channel Shift Techniques

Abstract

This study aimed to increase the quantity and variety of ethnopharmacological image datasets using image augmentation techniques, specifically rotation range augmentation (RRA) and channel shift range augmentation (CSA). The dataset augmentation was conducted to enrich the training data for the development of machine learning models used to recognize medicinal plant images. The RRA technique rotated images by random angles, providing variations in object orientation, while CSA altered the color channel values to simulate changes in lighting and the natural colors of plants. The research process included dataset collection, data preprocessing, application of both augmentation techniques, and division of the dataset into training, validation, and testing data. The results showed that the CSA technique produced 2,400 training data, 300 validation data, and 300 testing data, while the RRA technique produced the same amount of data. Therefore, the total data generated from both augmentation techniques amounted to 6,000 images, which could improve the accuracy and performance of deep learning models in recognizing ethnopharmacological images.