UPM Institutional Repository

Analysis detection of real-time metallic surface defect using MobileNetV2 and YOLOv3 on Raspberry Pi


Citation

Muhammad, Muhamad Sufri and Zainudin, Muhammad Noorazlan Shah and Idris, Muhammad Idzdihar and Kang, Soh Jun and Chee, Tan An and Xian, Teoh Yu and Alsayaydeh, Jamil Abedalrahim Jamil and Razali, Md Saifullah (2024) Analysis detection of real-time metallic surface defect using MobileNetV2 and YOLOv3 on Raspberry Pi. Journal of Advanced Research in Applied Sciences and Engineering Technology, 57 (2). pp. 105-117. ISSN 2462-1943

Abstract

This work presents an innovative solution utilizing a Raspberry Pi detection system to identify any defects on metallic surfaces in real-time. Manual inspection has several limitations, including time-consuming, subjective assessments, and a higher probability of human error could compromise product quality, lead to potential failures, and result in substantial costs for manufacturers. The primary focus of this endeavour is to enhance manufacturing efficiency and reduce labour expenses by automating the defect identification process. This objective is realized by employing the YOLOv3-tiny and MobileNetv2 algorithms which are subsequently deployed on a Raspberry Pi to enable precise and swift defect detection on metallic surfaces. The implementation process involves training and testing the models on a computer, followed by their deployment onto the Raspberry Pi. Upon proper setup, the trained models are employed for real-time inferences, effectively identifying defects. Notably, while the MobileNetv2 exhibits impressive accuracy in classifying defect types above 0.9, it is found to be less efficient for real-time detection on the Raspberry Pi. In contrast, the YOLO model proves to be well-suited for real-time detection on this platform with above probability of 0.8 for selected types of defects. The successful integration of this model significantly transforms quality control and inspection procedures across various industries.


Download File

[img] Text
119095.pdf - Published Version

Download (6MB)

Additional Metadata

Item Type: Article
Divisions: Faculty of Computer Science and Information Technology
DOI Number: https://doi.org/10.37934/araset.57.2.105117
Publisher: Semarak Ilmu Publishing
Keywords: Yolov3; Mobilenetv2; Raspberry pi; Defect detection
Depositing User: Ms. Zaimah Saiful Yazan
Date Deposited: 06 Aug 2025 04:08
Last Modified: 06 Aug 2025 04:08
Altmetrics: http://www.altmetric.com/details.php?domain=psasir.upm.edu.my&doi=10.37934/araset.57.2.105117
URI: http://psasir.upm.edu.my/id/eprint/119095
Statistic Details: View Download Statistic

Actions (login required)

View Item View Item