AI-Powered Vision Inspection System for Object Classification Application

Authors

  • Jing Song Ong Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, 75450 Melaka, Malaysia
  • J. Hossen Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, 75450 Melaka, Malaysia
  • Poh Ping Em Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, 75450 Melaka, Malaysia
  • Thangavel Bhuvaneswari Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, 75450 Melaka, Malaysia
  • J. Emerson Raja Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, 75450 Melaka, Malaysia
  • Min Thu Soe Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, 75450 Melaka, Malaysia
  • Nor Hidayati Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, 75450 Melaka, Malaysia
  • Gin Chong Lee Department of Engineering, School of Engineering, Computing and Built Environment, UOW Malaysia Penang University College, 32, Jalan Anson, George Town, 10400 George Town, Pulau Pinang
  • Sajid Abdullah Alam Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, 75450 Melaka, Malaysia

DOI:

https://doi.org/10.15379/ijmst.v10i1.1817

Keywords:

Transfer Learning, You Only Look Once, YOLOv3, Darknet53, Object Detection, Palm Oil Kernel Classification.

Abstract

Human operators are often susceptible to eye fatigue due to sleep deprivation and excessive workload, which may negatively impact their consistency and efficiency in performing repetitive and challenging inspection tasks. This paper presents the development of an AI-powered vision inspection system for object sorting applications, utilizing a You-Only-Look-Once (YOLO) version 3 pre-trained model based on Deep Convolutional Neural Network's (DCNN) transfer learning technique. Feature extraction for each data point is performed using Darknet53, which subsequently trains the YOLO v3 model. The dataset is partitioned into a training set and test set at a 90:10 ratio. The trained model achieves a mean average precision (mAP) of 99.146%. Enhancing the precision and recall values of the model can be accomplished by increasing the number of dataset instances used for training.

Downloads

Download data is not yet available.

Downloads

Published

2023-09-05

How to Cite

[1]
J. S. . Ong, “AI-Powered Vision Inspection System for Object Classification Application ”, ijmst, vol. 10, no. 1, pp. 363-376, Sep. 2023.