IMPLEMENTASI TRANSFER LEARNING DENGAN FINE-TUNING PADA DETEKSI OBJEK MULTI-KELAS MENGGUNAKAN YOLO (STUDI KASUS CAR FREE DAY JALAN EL TARI KOTA KUPANG)

Abstract

Object detection is a computer vision technology used to recognize and determine the location of objects in images or videos. This study aims to implement the transfer learning method with fine-tuning on the YOLOv8m model to detect multi-class objects consisting of persons, vehicles, and umbrellas in the Car Free Day environment on El Tari Street, Kupang City, as well as to develop a web-based object detection system capable of automatically detecting objects in images and videos. The research dataset consisted of 315 images obtained through field documentation and annotated using Roboflow, which was then increased to 757 images through the augmentation process before being divided into training, validation, and testing datasets. The YOLOv8m pretrained model based on the COCO dataset was trained using Google Colab for 100 epochs with the transfer learning and fine-tuning approach. The results showed that the model achieved a precision of 0.903, a recall of 0.801, an mAP50 of 0.868, and an mAP50-95 of 0.625. In addition, the developed web-based system was able to automatically detect objects in images and videos and display bounding boxes, confidence scores, object counts, detection result graphs, and model evaluation metrics. The results of the study indicate that the application of fine-tuning to YOLOv8m is capable of improving the model's adaptability to the characteristics of the local Car Free Day environment, thereby potentially supporting more effective public activity monitoring.