TY - GEN
T1 - BDK Bileşenlerini Siniflandirmak için Verimli Bir Görü Dönüştürücü Modeli
AU - Surmeli, Cagri Can
AU - Ekenel, Hazim Kemal
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - Printed circuit board (PCB) assemblies in everyday electronic devices are mass-produced. As a result of this production volume, a fast way of visual inspection is necessary. An integral part of visual inspection systems is PCB component classification. In this paper, we have explored use of the Vision Transformer (ViT), which is a recent state-of-the-art image classification approach, for PCB component classification. We have employed several ViT models that are available in the literature and also proposed a new compact, efficient, and high performing ViT model, named as ViT-Mini. We have conducted extensive experiments on the FICS-PCB dataset in order to comparatively evaluate the ViT models' performance. The highest achieved accuracy is 99.46% for capacitor and resistor classification and 96.52% for classification of capacitor, resistor, inductor, transistor, diode, and IC. The proposed compact model's performance is comparable with the ones obtained with larger models, which indicates its suitability for real-time applications.
AB - Printed circuit board (PCB) assemblies in everyday electronic devices are mass-produced. As a result of this production volume, a fast way of visual inspection is necessary. An integral part of visual inspection systems is PCB component classification. In this paper, we have explored use of the Vision Transformer (ViT), which is a recent state-of-the-art image classification approach, for PCB component classification. We have employed several ViT models that are available in the literature and also proposed a new compact, efficient, and high performing ViT model, named as ViT-Mini. We have conducted extensive experiments on the FICS-PCB dataset in order to comparatively evaluate the ViT models' performance. The highest achieved accuracy is 99.46% for capacitor and resistor classification and 96.52% for classification of capacitor, resistor, inductor, transistor, diode, and IC. The proposed compact model's performance is comparable with the ones obtained with larger models, which indicates its suitability for real-time applications.
KW - Image Classification
KW - PCB Components
KW - Vision Transformers
UR - http://www.scopus.com/inward/record.url?scp=85173438273&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85173438273&partnerID=8YFLogxK
U2 - 10.1109/SIU59756.2023.10224039
DO - 10.1109/SIU59756.2023.10224039
M3 - Conference contribution
AN - SCOPUS:85173438273
T3 - 31st IEEE Conference on Signal Processing and Communications Applications, SIU 2023
BT - 31st IEEE Conference on Signal Processing and Communications Applications, SIU 2023
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 31st IEEE Conference on Signal Processing and Communications Applications, SIU 2023
Y2 - 5 July 2023 through 8 July 2023
ER -