Implementasi dan Evaluasi Pendekatan Tuning Hiperparameter YOLOv11 untuk Deteksi Nama Obat Berbasis Citra

The advancement of computer vision technology has created opportunities for object detection systems to assist in identifying important information on medicine packaging, particularly the medicine name, which often appears as small text and is placed in varying positions on the package. This study aims to implement and evaluate a hyperparameter tuning approach on the YOLOv11 model to detect medicine name regions from image-based data. The dataset used consists of medicine packaging images that have been annotated with bounding boxes around the medicine name area. The research process includes stages of data preprocessing, model training, and performance evaluation using the mean Average Precision (mAP) metric. The experimental results indicate that the tuned hyperparameter configuration improves model performance compared to the default configuration, with the best model achieving an [email protected]:0.95 score of 0.7715 on the test set, demonstrating the model’s ability to localize medicine name regions more precisely across various Intersection over Union (IoU) thresholds. The focus on the [email protected]:0.95 metric is important because it better reflects the accuracy of the model in determining bounding box positions under stricter IoU conditions. Therefore, this study shows that applying hyperparameter tuning to YOLOv11 effectively improves the quality of medicine name region detection on packaging images and has the potential to support the development of more accurate and reliable image-based medicine identification systems.