Image Detection Analysis for Javanese Character Using YOLOv9 Models

Abstract

The Javanese script needs to be digitized to improve access and usage, especially among younger generations. Digitizing Javanese characters is crucial for preserving Javanese culture and traditions in the long term. This study aims to detect and recognize Javanese characters using the YOLOv9 algorithm, known for its ability to detect various object types, including Latin and non-Latin scripts. The dataset used consists of 85 images of complete Javanese script arranged in a 4x5 grid of different characters. The dataset was divided into a training dataset (75 images) and a validation dataset (10 images). All data pre-processing was done using Roboflow tools. Two experiments were conducted, varying the weights of the YOLOv9 algorithm model: YOLOv9-c and YOLOv9-c-converted. The research results showed that the YOLOv9-c model outperformed YOLOv9-c-converted, achieving a confidence level of over 80% and an mAP value of 0.95 in recognizing Javanese script images. In other words, the YOLOv9 model succeeded in detecting and recognizing Javanese scripts