Small but Diverse SEM Image Dataset: Impact of Image Augmentation on the Performance of AlexNet

• Published in: TEM Journal
• Main Author: Shariff K.K.M.; Abdullah N.E.; Al-Misreb A.A.; Jahidin A.H.; Ali M.S.A.M.; Yassin A.I.M.
• Format: Article
• Language: English
• Published: UIKTEN - Association for Information Communication Technology Education and Science 2023
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85161562212&doi=10.18421%2fTEM122-34&partnerID=40&md5=d739177f4a4b240a103a3df5061f6118
• ISSN: 22178309
• DOI: 10.18421/TEM122-34

To this date, scanning electron microscope has produced among the most complex and diverse images at nanoscale resolution. The highly magnified images of backscattered electrons reflected from the surface of samples are non-uniformed, even for the same class of images. The study investigates the impact of having a small but diverse dataset on the performance of AlexNet. A total of 160 samples from EUDAT Collaborative Database Infrastructure is used for the study. Compared to the use of new nonaugmented samples to increase the size of dataset, image augmentation has been significantly improved classification performance and generalization ability of the AlexNet. © 2023 Khairul Khaizi Mohd Shariff et al; published by UIKTEN. This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 License.