Effect of color contrast to the accuracy of SSD-MobileNetV2
Machine vision with deep learning neural network is currently on the rise, specifically with the emergence of Industrial Revolution 4.0. It is further elevated with the advancement in the computational capabilities of modern edge computing to reduce the computational cost. Thus, making such technolo...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Ismail Saritas
2022
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/100579/7/100579_Effect%20of%20color%20contrast%20to%20the%20accuracy%20of%20SSD-MobileNetV2.pdf http://irep.iium.edu.my/100579/ https://ijisae.org/index.php/IJISAE/article/view/2133/716 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Universiti Islam Antarabangsa Malaysia |
| Language: | English |
| Summary: | Machine vision with deep learning neural network is currently on the rise, specifically with the emergence of Industrial Revolution 4.0. It is further elevated with the advancement in the computational capabilities of modern edge computing to reduce the computational cost. Thus, making such technology economically viable to the general manufacturing industries for industrial application. Visual quality inspection would be among the most relevant process to have such architecture implemented. This paper explores the feasibility of deploying deep learning model, SSD-MobileNetV2 to replace manual visual inspection for holes counting process after drilling on a carbon-reinforced fiber composite component. The drilled holes were set into three (3) different conditions; bare-holes and holes equipped with semi-transparent or red locating pins. We conclude that the contrasting color of the holes with respect to its surrounding plays a pivotal role in their detections. Holes with semi-transparent or red locating pins are with accuracy of 77.14% and 73.33% respectively; while bare-blackened holes are with accuracy of only 45.95%. |
|---|