Radiator count monitoring system for logistic management system using YOLOv8

Ahmad Adnan, Muhammad Syukri (2023) Radiator count monitoring system for logistic management system using YOLOv8. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Radiator count monitoring system for logistic management system using YOLOv8.pdf Download (3MB)

Abstract

This project focuses on the development and implementation of an object detection system utilizing the YOLOv8 (You Only Look Once version 8) deep learning architecture for accurate identification of car radiators in the context of assembly line and logistics. The automotive assembly industry relies heavily on manual labor for part supply monitoring. Traditional methods for radiator detection often fall short in terms of speed and accuracy. Using the capabilities of YOLOv8, this approach aims to enhance real-time detection of car radiators, enabling swift and precise identification for logistic management system for parts replenish applications to assembly lines. By training the model on a comprehensive dataset of annotated car radiator images, the aim is to fine-tune the YOLOv8 architecture, specifically to recognize the complex features and variations associated with different radiator designs. The project's key objectives include optimizing the detection accuracy, minimizing false positives, and ensuring real-time processing speed to meet the rigorous requirements of automotive applications. This includes providing information on the quantity of logistic parts supply, specifically car radiators, within the field of view. The study demonstrates the progressive improvement of YOLOv8's performance metrics with increasing epochs. At 10 epochs, YOLOv8 achieves moderate precision (0.59862), recall (0.59321), and an F1 score of 0.59, which improves significantly at 25 epochs with precision (0.92441), recall (0.8916), and an F1 score of 0.91. By 50 epochs, YOLOv8 further enhances its performance, achieving a precision of 0.95921, recall of 0.95593, and an F1 score of 0.96. At 75 epochs, the model maintains high precision (0.96098), recall (0.96816), and an F1 score of 0.96, ultimately reaching outstanding precision (0.97493), recall (0.97277), and an F1 score of 0.97 at 100 epochs. Comparatively, YOLOv9 shows potential with higher initial precision (0.67825) and recall (0.6316) at 10 epochs but requires longer training times (6.651 hours) and further optimization, with an F1 score of 0.67. The shorter training time of YOLOv8, requiring only 1.664 hours for 10 epochs, makes it advantageous for rapid prototyping and iterative model refinement. YOLOv8's successful recognition and localization of objects signify its potential contribution to automating logistic part supply chains, offering real-time insights for streamlined operations.

Actions (login required)

View Item