Arief Setiawan
Program Studi Informatika, Jurusan Informatika, Fakultas Teknik, Universitas Bengkulu

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

Perbandingan YOLOv11m dan YOLOv12m pada Deteksi Plat Nomor Indonesia dan Pengaruhnya terhadap EasyOCR Arief Setiawan; Ernawati; Funny Farady Coastera
Jurnal Terapan Teknologi Informasi Vol 10 No 2 (2026): Jurnal Terapan Teknologi Informasi
Publisher : Fakultas Teknologi Informasi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.21460/jutei.2026.102.530

Abstract

Automatic License Plate Recognition (ALPR) is widely used in Intelligent Transportation Systems (ITS). This study compares the You Only Look Once (YOLO) models YOLOv11m and YOLOv12m for Indonesian vehicle license plate detection and examines how the detection results affect character recognition with EasyOCR as the Optical Character Recognition (OCR) engine. The dataset contains 2,819 vehicle images, consisting of 2,243 training images, 353 validation images, and 223 test images, obtained from Roboflow. Both models were trained for 50 epochs on an NVIDIA Tesla T4 graphics processing unit (GPU) using the same configuration. On the validation set, both models produced nearly identical mean Average Precision (mAP@0.5)values, namely 0.9782 for YOLOv11m and 0.9783 for YOLOv12m. YOLOv11m achieved higher recall (0.9554), a slightly higher test F1-Score (0.9540), and lower validation inference time (24.58 ms). YOLOv12m achieved higher precision (0.9483) and mAP@0.5:0.95 (0.6748). In the OCR evaluation, YOLOv11m detected all 31 test plates and obtained a fuzzy matching score of 0.6632 with a Character Error Rate (CER) of 0.3368. YOLOv12m detected 29 of 31 plates, with a fuzzy score of 0.6139 and CER of 0.3861. The preprocessing sequence used in this study reduced OCR performance for both models compared with the original red, green, and blue (RGB) crops. Based on these results, YOLOv11m provided the better overall balance between detection recall, OCR performance, and inference time in the tested configuration.