The Transfer Learning Influence on YOLO v8 Performance For Motorbike Detection

• Main Authors: Pangestu, Gusti, Aziz, Baskoro, Pratama, Febby Candra, Pumomo, Agung, Puspitasari, Chasandra
• Format: Conference Proceeding
• Language: English
• Subjects: Computer architecture; Deep Learning; Motorcycles; Security; Testing; Theft Detection; Training; Transfer learning; YOLO
• ISSN: 2767-7826
• DOI: 10.1109/ICCED60214.2023.10425731

The number of motorbike theft case in Indonesia has increasing rapidly. This phenomenon led to the increasing of the criminal number especially motorbike theft case. Nowadays, the use of CCTV (Closed Circuit Television) camera has been widely applied. Nevertheless, CCTV does not has the ability to prevent criminal acts such motorbike theft case. Therefore, there is a need of a technology based to prevent motorbike theft acts. This research proposed an initial research of motorbike theft detection system that focused on the process of motorbike detection using Deep Learning approach. In this research, the YOLO architecture was utilized to detect the motorbike object inside an image frame. There are two testing scenarios for motorbike detection process, using trained model, and using pre-trained model. The trained model obtain better result rather than using pre-trained model by achieving 0.93 of its accuracy that proven the YOLO v8 model are working well especially when it is applied with transfer learning process for its based model.