Deep Learning untuk Identifikasi Daun Tanaman Obat Menggunakan Transfer Learning MobileNetV2

Abstract

Medicinal plants are plants used as alternative medicines for healing or preventing various diseases due to their active substances. The utilization of medicinal plants in Indonesia has been widespread among the community since ancient times and is a heritage passed down from ancestors. Medicinal plants have leaf structures that are almost similar between one plant and another, which can lead to confusion for some people and require precision in identifying the leaves of medicinal plants. Incorrect identification can have negative consequences for the users. In recent years, deep learning has been used to identify objects because of its ability to interpret images. This study used a transfer learning method to identify medicinal plants. Transfer learning utilizes a pre-trained model to learn and perform new tasks, making it suitable for smaller datasets. The pre-trained model used in this study is MobileNetV2. MobileNetV2 has a lightweight architecture and high accuracy. Fine-tuning techniques were applied in this study to improve the model's performance. Several experiments were conducted with parameters such as epochs and fine-tuning layers to obtain the best results. The research yielded a training accuracy of 97%, validation accuracy of 96%, and testing accuracy of 93%.