Analisis Perbandingan OCR dan Fuzzy Matching untuk Pengenalan Nama Obat Berbasis YOLO bagi Tunanetra

Individuals with visual impairments often face difficulties in independently identifying medications, increasing the risk of medication errors. This study aims to develop a medication name recognition system based on computer vision by integrating object detection using YOLOv11n, text extraction using Optical Character Recognition (OCR), and text correction using fuzzy string matching methods. The dataset consists of 350 drug packaging images representing 70 medication types collected from UMM Medical Center and expanded to 1,050 images through data augmentation. The YOLOv11n model detects the region containing the medication name and produces a Region of Interest (ROI), which is then processed using OCR methods, namely EasyOCR and TrOCR, to extract textual information. The extracted text is matched with a medication name database using fuzzy matching algorithms including Levenshtein, Jaro, and Jaro–Winkler implemented through the RapidFuzz library. Experimental results show that YOLOv11n achieves 0.997 precision, 0.996 recall, 0.995 mAP@50, and 0.801 mAP@50–95. In the text recognition stage, the TrOCR and Levenshtein combination provides the best performance with 98.94% accuracy, 96.62% F1-score, and 0.55% Character Error Rate (CER). These results indicate that integrating object detection, OCR, and fuzzy matching improves the accuracy of medication name recognition and has potential to assist visually impaired individuals in identifying medications more independently.